My mother is a bad ass. She’s 71, lives alone in the middle of the Spanish countryside, raises chickens, goats and sheep, smokes Marlboro lights and likes her G&Ts with Bombay and lime. She recently dead up red eyed it over from seeing an old friend in Canada to make my Significant Birthday in the UK, then proceeded on no sleep to stay up until 4am to watch England beat Mexico in the World Cup.

Like any parent, she has her foibles. And so, while I will state unequivocally that I love my mother and am a loyal and attentive son, she has on occasion been known to exaggerate.

This is why, one April afternoon last year, when she came into the abandoned pig shed my father converted into a home office, I didn’t immediately believe what she said.

“The power’s out across all of Spain and Portugal.”

I’ll admit I was sceptical. It sounded rather far fetched, not least because the home office’s lights were still on. Then I remembered the solar panels. It was true, too, that the Wifi had cut, but this far into the wilderness of southern Andalucia, that isn’t at all uncommon. I checked my phone: no signal.

Around half an hour earlier, my mother’s friend had been on the phone to her daughter in northern Spain, when the power suddenly cut. There had been enough residual network for her daughter to scan her news feeds and within a few minutes reports started dripping through of a widespread power outage.

I went to the house and tried the landline, it being the one thing in the home connected to some mains supply — no ring tone. It certainly looked like the power was out in our little corner of Spain, at least.

With work no longer a possibility, I decided to do some journalism; I took mum’s car and drove the mile or so to town to see what was happening. Sure enough, the whole place was without electricity. Petrol stations and supermarkets were already shuttered, most bars and cafes had closed for the day, although some landlords were doing their best to empty their fridges of cold beer for the office workers who had been sent home for the afternoon. Outside one bar on the main street, a group of men sat in plastic chairs arranged around a table laid with a bottle of rum, a bowl of ice, and a wind-up radio. I sidled up to them and listened to the Spanish government’s back up communication system, a relic from the days of Franco.

It was true, then. Virtually the whole of Spain and Portugal had lost power, authorities were trying to ascertain the issue, and they were refusing to rule out sabotage. The word that I would go on to hear hundreds of times over the next year was on the lips of presenter and listeners alike: Apagón.

The word derives from apagar, meaning to turn off. In Spanish the suffix -ón is reserved for expressing largeness, sort of like the opposite of -ito. Perhaps the most common of these is botellon, meaning literally “big bottle” but understood by all as a large social gathering. Although even small, localised power cuts are also called apagónes, in this case the sound fitted the scale. The power cut of April 28, 2025, the biggest in Spain’s modern history, will forever be known as El Apagón.

I walked around the corner and went into a mini-market. The owner had dragged her two teenage sons out laden with meat from their home freezer and was instructing them to sequester it at the depths of the shop deep freeze. I asked them what they reckoned was happening and they responded, in unison: “Putin!”

I then proceded to perform a relative of panic-buying, perhaps “unplanned buying” would fit. I hadn’t brought any bags with me although I did have some cash, and so I flitted around the shop picking up whatever I could carry back and which I dimly discerned might be useful in the event of a prolonged power outage: two boxes of passata, a bag of lentils, three cartons of UHT milk, and as many thawing frozen bread rolls as the owner had fished out to make way for the meat. Seven, it turned out.

It wasn’t until I was back at the house when I remembered everyone else.

Two nights earlier, mum had celebrated her 70th birthday, and many of the guests had booked a few extra nights’ stay in town. This included a few quite senior family members. So I drove around town, checked up on them, and invited them to dinner at the house that evening (at least there we would have light, cold drinks, and a place to charge phones).

With my sister, brother-in-law and nephew also around, and with no work for any of us to do even if we wanted to, we made an impromtu call on something we’d been meaning to sort for a while now: scattering dad’s ashes.

My father had sat in a cuckoo egg blue urn on the kitchen windowsill for the past two winters after dying and being cremated following a brutal but mercifully short battle with a brain tumour. When I think back on the day of the funeral, I recall only lead-lined visions of pallid faces, of mask-marred mouths and bad coffee and the sting of the fundador I took in stealthy swigs from a hip flask I borrowed from his best friend.

Spanish funerals are rather different from Anglo-Saxon ones — they are typically just a day or so after death and are reserved for unexpurgated grief. They are not occasions to “celebrate the life” of a person, they are for mourning alone. The on-the-bright-side part comes several weeks later, when the physical pain of the loss has numbed slightly.

We, or at least I, tried to do a bit of both that day, but it wasn’t until well into the wake that the Spanish contingent thawed enough to catch up with the British, who, in time-honoured style, were putting away bacchinalian amounts of red wine and laughing with love at my father’s many inadvertantly hilarious and unedifying moments.

(One story that went down well with Spanish- and English-speakers alike: Dad was deeply respected in town, the way the Spanish like and adopt an Englishman who speaks their language but who also gets them, on something close to a spiritual level. I remember the day me and mum went to the printers to get some orders of service ready for the funteral. The guy at the computer who I’d never seen before pulled up the picture we’d chosen of dad and he shook his head hard. I asked if there was a problem with the lighting or the resolution. He said, no, it wasn’t that. He just knew my dad as a good man and was moved at his dying. Anyway, my dad would always give lifts to anyone who asked, both into and out of town. One day, the mayor’s brother-in-law, himself a minor local dignitary, asked for a lift from the campo back into the centre. The area is renowned for its wind — it is the wind and kite surfing capital of the world [we will get to this presently]. It was a typically blowy day. I happened to be visiting my parents and I was fond of this chap — he seemed kind and sensitive in the macho world of squalid backstabbing that is local politics — so I tagged along for the ride. As the brother-in-law was getting out of the car, a sudden gust of wind blew his cap off his head. My dad, in his idiosyncratic Spanish, forever peppered with Malopropisms, meant to shout: “Cuidado, tu gorra!”, which might not be grammatically perfect, but which conveys the message: Watch out, your cap! Only what he actually said was: “Cuidado, guarro!”. Guarro is a synonym for pig, but more colloquially means something closer to dirty fucker. After a moment of shocked silence, we all laughed in relief about the loss in translation.)

We had always intended to scatter his ashes on the beach, the part away from town when the sprawl of apartment blocks gives way to brush, farmland, and the fincas that rear the fighting bulls. Beyond all that to the mottled light of the pine forest, its ground a bed of needles and sand, which should be terrible for dogs’ paws but which is made soft and spongy from the sea mists. Here, specifically, for two reasons: it was my father’s favourite place on Earth, and it was far enough away from civilisation to avoid attention (scattering ashes in public places is not only seen as deeply sacreligious in Spain; it is also illegal).

We picked up my sister and her little family and drove out to the pines, keeping an eye on the guardia civil who were positioned sparsely across the coast; it’s only 12 kilometres or so from Morocco at this point, so they were presumably keeping any eye on any would-be smugglers trying to take advantage of the loss of power and comms.

There had been no shortage of advice as to how we were to scatter dad. He had, after all, sat there right next to where we ate dinner each night; I would always say hi to him when I got back after any extended absence. His natty, rather cheap looking urn the result of several too many decisions made in the throes of crisis, one more pointless detail to determine that really didn’t seem to matter in the moment. One friend suggested decanting him into a plastic bag, then snipping a small hole in one corner. Handled properly, the idea went, we would be able to secrete him slowly like an hour glass, imperceptibly scattered on the breeze as we walked across the sand. Another said they would do it in stages: take a small tupperware of him in a pocket, and deposit a little each time we walked the dogs, Shawshank Redemption style.

Given our resolve to do this today was entirely spontaneous, we didn’t really have a plan as such. That I was the one to physically do the thing was a given — I was the doer now, it was me (and, when he was around, my uncle) who had taken on the task of administering the horror of dad’s dying — the wheelchaired trips across flooded parking lots, the conversations with tired and distracted medical staff, lifting him into bed that day the radio waves from that loathsome machine finally fried what was left of his brain and rendered him weaker than a baby. Mine was the job of keeping the fire going over that lost winter, of buying a hospital bed for him so he could lie facing the ceiling in the house that he built in one final staring contest with the ever after. Change the drip bags. Buy more adult nappies. Make sure the lorazepam (anxiety), the midazolam (breathlessness), morphine (pain) and cyclizine (nausea) are topped up on the regular. It was me who watched over his body as dawn crept up on that day — the first day of Spring, World Poetry Day, all the beautiful and promising things — after an exhausted and frankly relieved mum had had one final G&T toast with the nurse before bed (told you she was a bad ass). I called the ambulance, and waited as they came and dispatched my father’s corpse to the morgue. It was on me to dismantle and re-sell the hospital bed, a 1000-euro purchase that we ended up needing for a grand total of 9 nights.

Do not get me wrong, I wasn’t alone. I had help from overwhelmed and terrified loved ones. Times like that you manage the best you can. The days of losing dad were, in some way, almost magical in their surreal half light. But I felt immense loneliness then and I carry that with me still.

We start our pilgrimage over the pine needle bed, the sun shooting like shrapnel through the writhing branches, the dogs bounding ahead and my little nephew tottering after them like some drunk miniature shepherd.

We take the stone steps down to the beach and the tide is quite close, leaving only a slither of the beach running like a vein between sea and scrub. The wind has picked up. Locals have a colour coding system for wind strength, and I’d say we were now a solid orange, the Levante wind, the same that trembled Homer’s wine dark sea, now pulling through the reedbed in strong, insistent gulps.

We wander for a few minutes, not resisting the wind, leaning into it and letting it buffet us down the headland. We get to where we get and decide that here is about as good a place as any. I take dad out of the urn, and there is a little plastic wire tab to negotiate that no one ever shows you in the films. I hold the bottom of the bag and turn it upside down. The gale whips dad at wharp speed across the sand and sea, taking away forever and finally our world on the wind. There is the faintest smudge of him left on the beach, like the dust left behind when you touch a butterfly’s wings.

That night we all gather at mum’s, and I cook the lentils and tomatoes I bought a few hours earlier (when we still possessed dad). It is one of those unruly nights, where everyone feels tacit permission to cut a little loose, extentuating circumstances etc. I feel lightheaded and, to my surprise, lighthearted. There is nothing left to be done: dad’s ashes have been taken care of, we have food (and power points to charge the phones), we have wine, and we have each other. There is something, for the first time in a while, approaching contentment. People like the lentils.

There are seven of us, and I got seven bread rolls.

The days after the Apagón are rather more profane, and I spent them trying to counter a baffling narrative that the power cut was due to Spain’s large use of renewable energy.

It is quite remarkable: the lights had barely gone back on before people who have frankly zero knowledge about energy systems are going on record and claiming that the Spanish grid is uniquely vulnerable to this type of cascading outage. This is not true in point of fact, and I spend a few hours one day trying to gather some data to use as a corrective whenever the point is put to me in interviews (a regular occurrence over the coming weeks).

The short version is that there is simply no correlation between the amount of renewables in a country’s energy mix and power cuts. The Spanish government itself commissioned a review into the events of April 28, which identified a variety of contributing factors included a sudden voltage surge due to grid instability and a failure by utility operators to predict the correct energy mix. You can read more about that here.

Perhaps the least impressive part of the fossil fuel lobby is that they feel no need to adapt their narrative even when it’s been baroquely debunked. We live in the age of Brandolini’s Law — also known as the Bullshit Asymmetry Principle — where a lie gets around the world before the truth has its pants on (this to me explains why newsrooms’ collective response to the Fake News high watermark failed to stauch the drift from fact-based empiricism that has infected even well-meaning media discourse. We have a suggested remedy to this, and you can check that out here.)

And so we fast forward to the summer of 2026, The Great Burning, and That Front Page, a juxtaposition so grotesque that one struggles to find the words. It’s the This is Fine meme made broadsheet.

Clearly, it matters not in the minds of some decisionmakers and “thought leaders” whether or not a narrative happens to be true. It only needs to be sticky.

I’ve written a fair bit over the years about how people’s inability or unwillingness to Say The Thing fans the flames of misinformation, constrains society’s capacity to adapt, and retains what are now largely imaginery barriers to the kind of system change that all the science is screaming we need in order to secure a liveable future on this dying planet of ours.

As it’s summer, it’s also profits season, and my team has been tracking the money that the people most responsible for driving climate breakdown continue to rake in, a wealth transfer out of all proportion to the governing logic of capitalist progress, an extinction windfall that, as millions of people face starvation due to climate linked crop failures and societies across Europe pick through the smouldering rubble of what used to be their lives, is a civilisational shame that history will not forgive.

Just eight companies, according to a recent Guardian analysis, made $90 billion in profit in Q2 of 2026, not only during a climate collapse that has seen Europe burn, but also during a Bible promise war where the mechanised slaughter of Iranians and Lebanese is being abetted by everyone’s favourite cuddly generative AI provider.

I can lay no claim to credit for that work, nor can I fail to highlight a wonderful piece of fleet-footed leadership and initative by my colleague, but basically I ended up quoted in the Guardian’s splash. This led to some follow up interview requests, including one from CNN.

I did the interview early and on little sleep, but I’ve gotten a bit less anxious over lives this year, and I felt it went OK. Watch it here if you like.

It unleashed a flood of comments, as all content these days apparently and regrettably must, and it got me thinking about all sorts of things, including my journalistic background, my pedigree as a communicator, my life without my father, and of apeirgons.

An apeirogon, a portmanteau of two Greek words — apeiros, meaning infinite or boundless, and gonia, meaning angle — is a geometric shape with an limitless number of sides. It is a name for the infinity sign, you know the one: ∞.

I think of these a lot when I am in Spain, whenever I walk over the hill from mum’s house and behold the Strait. It’s about how the Sun hits the sea (the part of the coast is known as the Costa de la Luz and is famed among film producers for the quality of its light). The rays bounce off the waves in a breathtaking variegation, flashing white and bright like magnesium skating over the water.

In reaction to the CNN interview, I was both praised for the clarity of my message and criticised for my vagueness — a classic case of beauty (and, in 2026 it seems, truth) residing in the eye of the beholder. I was fed really interesting (and potentially valid!) critiques of overly negative communication over the climate breakdown, much of which find a sympathetic reader in me, even if I won’t go so far as full agreement. We know that despair only benefits the bad guys. We know that we have to bring people along with us, meet them where they are, and hold people’s hands as you look them in the eye and explain slowly in what precise flavour we are all fucked.

Maybe I am over the hill and over the top with my pessimism, but it has been five full years since you were all warned really specifically what lies ahead and I need more of you to actually look at the thing and engage with it. I need you to take it to your bosses, to your union reps, to your pension fund and local politicians. Shake it in their faces the way you shake your fist when you protest any other social crime. Then again, I know how scary and painful it is to stare directly at the future, and I won’t blame you if you can’t hold its gaze.

The philosopher Timothy Morton calls climate change a “hyperobject”, something that is too out of scale to the human condition for mere mortals to usefully contemplate. To me, it is an apeirogon: something that impacts us all, in varying ways, and which excites an infinite number of perspectives even as it is — perhaps because it is — too huge and unweildy to behold in full.

For the vast majority of people (80 percent of so) navigating the daily toil of living in a city, climate change will mainly be experienced as excess heat (in Europe we are now girding for the sixth heatwave of this infernal summer). That is the most obvious effect of our warming climate — longer, more intense and deadlier heat. For those lucky enough to have access to artificial cooling, these are periods that must be navigated judiciously; for those without, it will be as much as they can manage to simply survive. In either case, they will need to be negotiated without predictable access to electricity, as a combination of excess heat and drought has left our power grids at the very edge of breakdown. The city people will see the end of the story: empty supermarket shelves, law and order collapse, overwhelmed hospitals and fear stalking the streets.

For those, like my mum, living in rural areas, they are going to experience the climate collapse at a different point in its narrative arc. Theirs will be the fields plucked bare by plague or dessicated in the withering heat. The threat of wildfire will lay crouching just beyond the horizon, and your life, home and everything you’ve spent your days working to build up will always be hostage to wind and flame, nature’s cruel caprice.

There’s a small constituency who see an opportunity in the crisis, who will use the intervening years between now and whenever our global food and water systems collapse to grow even fatter from an inhuman hunger, from a greed that may never be sated. You are free to listen to these people or ignore them, just know that they will not be spared. They may admit the boat is sinking when the water reaches their necks, they may not. There are infinite perspectives on climate collapse, but the climate doesn’t care for any of them.

Perhaps that is what is so disappointing: we have for too long treated the natural world, our only home, with the same contempt it now glowers back at us. That more and more people are suddenly starting to notice, to fear and to rebel, matters nothing for the atmosphere we have so ruinously soiled. The climate crisis is uncaring. It is a daily memento mori: none of us are special, it will get us all in the end no matter how we feel about it. It will not respond to reason, to bribery, pleading or threat. How are we supposed to deal with that?

These days I feel a fatigue I’ve never known before, like I’ve been holding back the tide and my arms and shoulders are beginning to fail under the strain. They were not meant for such an effort. I am trying to protect people and things I cannot keep safe, and it might be that there are men equal to the task but I’m coming to accept I’m not one of them.

I have been to the hospital. Tests have been performed. I have low haemoglobin, apparently. This tracks. I lack the very stuff of red-bloodedness.

The weather, surely, cannot help. I can’t remember the last night I slept well, it is the unnatural heat, the open windows, the fretful breeze. The fucking seagulls. Let’s say if you’re exceptionally lucky (luckier, indeed, than my dad, who considered himself to be the luckiest guy ever), if you’re lucky you get 80 summers. I’m at half that and I cannot remember one of genuine consequence. I know that others will come but I find them hard to picture. The future eludes my vision. The whole earth is bleached, and I can’t imagine what could bring the colour back. The very concept of rain now seems absurd, belonging to another geological time.

I met up recently with one of my oldest friends. He’s a teacher, and that is good, because there is wisdom in what he has to say. He introduced me to the concept of The Status Game, how we view ourselves in comparison to others, and how it is a lifetime fool’s errand to chase that which will not satisfy even once attained. We talked about efficiency in art, about the pleasure to be derived in the destruction of the creative process, whittling down our work until it is lean and tight and bears its weight with economy and substance. I suppose we were also talking about how we ensure our messages are received, how we as humans engage with other humans.

My friend is a settled person — he finds the outlook that suits him best and he sticks with it until events or invidividuals disabuse him of it. He is effortless in the liminal bits of life, comfortable in the not knowing, suited to the uncertainty that shadows being human.

We talked about the failure of philosophy, how so much is given over to contemplating experience and perspective (even, with regret, Morton) and how truly unequipped it has left us to deal with the important, inevitable tasks of life. How am I supposed to act when my dad dies? What should I do with the anger I feel when my friend is killed and the driver evades punishment? When I’m running on empty and feel really dangerously close to life spinning apart in total? Tell me what to do. Please, I am asking for help.

We are told nothing, or, rather, I have retained nothing from philosophy on what we are meant to do when faced with moments of unreconstructed being. It seems it is on us to learn, the hard way, through viseral, painful lived experienced as we are emotionally overwhelmed and spiritually malnourished. Maybe that’s the point: if you can keep your head while all around you are losing theirs etc. Maybe it’s the trial of it all that’s the thing. There are simply some situations that cannot be war-gamed. Not every play has a dress rehearsal. If you can withstand the enormity of force that loss burdens you with, try to make do as best you can and if you’re lucky and up for it apply a little grace to the moment while you’re in it, and make it out the other side knowing that that more darkness awaits to be encountered and overcome. That might be all we can hope for.

All this is to say that nothing exists to prepare us to respond to climate change. As a concept, as a phenomenon, its physical and immediate impacts and threats. There are practical guides to meaningful personal action, which a lot of study suggests people are most hungry for — how can we do our own little bit, share the burden a little more collectively, sacrifice the disposable parts of the needs of the individual for the good of the general.

There are things we can do to better prepare, phyiscally. Those fortunate enough to own their own homes (something that still eludes me) can future proof them, making your entire family more resilient to the next big weather shock, and all those that will follow. While you have less room for manoeuvre as a renter, you are still not powerless.

At my mum’s, she now has a grab bag to take with her should she ever again need to flee fire or navigate the world without power (the bag contains among other things a Moroccan SIM card so she can still communicate even if, as last April, the Spanish networks all go down together). Local authorities across the UK are rolling out wildfire evacuation plans. If you see one near you, sign up.

Take names. I am doing so. And there’s more and more of us holding historic polluters to account.

Remember what is probably the coolest thing about being you: you can change at any point. If you’re someone who is constantly late, you can choose to end that habit of a lifetime right now. Within a few weeks no one will even remember you were ever late in the first place. The same is true for your climate work. You may not have spent two minutes over the last year thinking about how you can help, it may be way down your list of priorities, but you can change that. Right now. It is a superpower.

Recognise your power as a person with the agency to influence others. People tend to listen to people they know and respect, and you can change them. My dad, for example, went from being a life-long meat-eating small business owner and motorbike racer to a self-sufficient eco-warrior and a card-carrying member of the Green Party. (One of the proudest moments of my life was at the funeral, where my aunt took me by the shoulder and said: “You changed him, you know.”)

In particular these days acquaintances tend to be less enthusiastic about their sustained use of Generative AI tools when I point out the environmental and social damage of their supply chains.

Talking to friends and families about what is to come helps them see how it applies to them specifically. Most people aren’t ignorant of the scale of the problem, they are intimidated by the size of the barriers to solutions. Help lower them.

This can be done in a manner that is non-preachy, and judgement free. This is your opportunity to bring your people along with you. The difficulty isn’t in not knowing what needs to be done, it resides instead in convincing enough people that the system we currently have needs to undergo unrecognisable change.

If you’re anything like me and consider yourself to be relatively well up on the science and economics of climate change, you will likely be surprised by other people’s unawareness of truths you hold to be self evident.

They probably aren’t aware, for example, that fossil fuel organisations are money-takers not money-makers (especially unaware during bumper profits reporting seasons such as currently). Oil and gas firms received $7 trillion in taxpayer handouts each year; this is the only way they can remain going concerns, because the system is rigged in their favour. They do not provide us with energy security. They are little more than social and economic parasites. They need to be treated as such.

Your friends may have heard and may even repeat the tired untruths of the industry, especially in the UK around the North Sea. Equip them with the corrective. (North Sea oil and gas won’t lower bills, won’t create more jobs, and isn’t any greener than imported fossil fuels).

I bet your friends don’t know that as much as two-thirds of the energy generated when burning fossil fuels is simply lost as heat and doesn’t contribute to energy systems. This means that we don’t need to replace the equivalent energy from oil and gas with renewables — we need less than half. This massively lowers the barriers to change. Enlighten them.

Did you know that the food we feed to animals could feed a billion people every year, without any additional production? Or that more land is used to grow food for animals than it is to grow food for humans?

Ignorance — its usage may be a little strong here as it is rarely willful or cynical on these things — is as much a fuel to forest fires as is parched woodland; misinformation can run wild only in a vaccuum free from trenchant truth.

As a society we are generally well meaning but often poorly directed (see chart). I don’t think it’s incumbent on us as individuals to make small sacrifices as the rich industrially gorge off our finite resources. But we can make personal changes, and enough of those, we know, can build up to something significant pretty quick.

And don’t forget the social movements that have materially changed the world. I was in Brighton for Pride weekend this year and it was a joyous celebration of love and belonging. I was speaking to a friend of mine who is a bit older (though they don’t look it). He said he could remember a time when he would be shouted at in the streets and worse for being who he is. The progress, although there’s always work to be done, had been headspinning for him.

I don’t need to list all mass participation efforts that have achieved an extent of their aims throughout the 20th and 21st centuries, but (given I am writing a book on the topic) I will offer an example, in the form of two cities. It is a tale of citizen-led enduring social and civic change via inclusive design. There’s civil disobedience in there, road rage, horses, and “bicycle rabies”. It is long, I’ll admit. So feel free to skip to the end of the highlighted section. But I hope some of you find it enlightening:

Anyone who has ever visited Amsterdam, the Netherlands’ canal-crossed jewel, will attest that it is a city owned by bicycles. Cyclists swarm along a network of 515 kilometres of dedicated bike lanes, powering themselves to work or towing improbably heavy loads on cargo bikes, carrying multiple children on their way to nursery. It is consistently voted in the top two most cycling friendly cities in the world (the other being Copenhagen, which we will come to). In all, around 63 percent of Amsterdam residents cycle on a daily basis — around 650,000 people — compared with around 12 percent of people in the US who cycle each day. A commonly held view, that Dutch people are somehow more culturally keen on cycling, is a classic case of confusing cause and effect: people in Amsterdam don’t cycle lots because they are genetically or socially predisposed to cycling. They do so because their city is a safe, clean and welcoming place to cycle in. Because bicycles are prioritised over motor vehicles in infrastructure, from roads to parking to maintenance access. And because bicycles have been integrated into people’s daily lives, improving their physical and mental health while cleaning the air they breathe. None of this was inevitable. It all came about due to popular movements that influenced policy.

It might be hard to imagine today, with everyone from school children to pensioners getting around by bike, but cyclists were once virtually wiped out from the streets of Amsterdam by the apex post-war transport form: the motorcar.

As far back as the 1920s and 30s, the bicycle was king on Dutch roads, with 3.5 million bicycles registered in the Netherlands by 1937, compared with just 91,000 cars. Increased average wages and the economies of scale augured in by the establishment of several national factories meant that most households could afford at least one bike, and motor vehicles were very much still the exception on the road.

“Cyclists dominated, as did pedestrians, with a few passenger cars or trucks interspersed here and there,” writes Ton Welleman in the Dutch transport ministry’s 1999 paper, The Dutch Cycling Master Plan, which tracks the progress of cycling in the world’s most bike-mad nation.

The outbreak of World War II and Nazi Germany’s occupation of the Netherlands brought a shuddering end to cycling’s heyday, as bicycle tires were seized in an attempt to make up for a shortfall of rubber across the Axis powers’ European territory. In addition, German troops requisitioned up to 50,000 bicycles within the space of a few weeks in July 1942 in an attempt to alleviate pressure on their supply chains and get soldiers travelling more freely. One German serviceman reported how this did nothing to ingratiate them to the locals:

“The Dutchman, who is practically born on a bicycle, views the seizing of his bicycle to be nearly the worst thing that could ever happen to him.”

Consequently, by 1943 there were 55 percent fewer bicycles in use than just three years previously. The Nazis also instigated something of a psychological shift in attitudes towards cycling. Welleman describes their view that the large number of cyclists in the Netherlands were “bothersome”, leading to the introduction of “fast” and “slow” traffic rules which gave motorised vehicles the right of way across the country. This system was only overturned in legislation at the turn of the millennium, with cyclists having their right of way restored after some 60 years ceding priority to vehicle traffic.

After the war, demographic shifts, including soaring “baby boomer” populations and universally rising incomes meant that vehicle use exploded in the Netherlands; the total number of kilometres travelled by car went from 4.5 billion in 1950 to 89 billion in 1975, with the average number of car users undergoing a similar transformation, from 139,000 in 1950 to 3.4 million nationally in 1975. Welleman offers a number of explanations as to why the increase in total distance travelled in the Netherlands in the second half of the 20th Century was made by car. The number of households in the country roughly doubled between 1950–1975 to 4.4 million while the percentage of professional workers rose 44 percent. Cities began to sprawl outwards, decreasing urban density, meaning more commuting and visiting was done by people able to afford motor vehicles and who were more likely to need to travel further to their workplaces.

At the same time, the extensive damage inflicted on major national infrastructure such as roads during the German occupation meant successive governments fixated on short-term regeneration, prioritising big, US-style transport routes that encouraged car use over more local solutions that would reward cycle-first behaviours. Budgets for road construction flourished meanwhile, increasing nearly three-fold between 1965 and 1971, which incentivised ever greater car usage.

From 1950 to 1975, Welleman writes, “the bicycle was almost entirely excluded from the government’s vision”.

(It’s perhaps important here to state that while there was very little by way of pro-cycling policy in Dutch local or national planning in the third quarter of the 20th Century, there was at least no obvious, concerted anti-cycling policy drives either. This was not the case in many other European nations; Germany, with its powerful auto-lobby, has seen sustained smear campaigns against the bicycle).

Inevitably, increased vehicle traffic led to an ever higher accident rate among cyclists, offering a further disincentive against undertaking a journey on two wheels. Within just 20 years, the rate of road fatalities increased by 200 percent to stand at 3,300 deaths a year. Tragically yet crucially, more than 400 of those deaths were children. Such unconscionable waste of life prompted mass outrage, and it wasn’t long before anti-vehicle protest movements got underway. One of the children killed on the road in 1971 was Simone Langenhoff, a six-year-old who was hit by a car on her way to school. Her father, Vic, was a journalist for the De Tijd newspaper. He penned a front-page article headlined Stop de Kindermoord, or: “Stop the Child Murder”.

The article tapped into an underlying feeling that the country was being developed without the consent of its population, turning spaces that were once safe for children to move around into potential death traps. A pressure group inspired by the headline started to hold a number of ride-by demonstrations (a tactic still in use today in many major cities). They also occupied accident blackspots and arranged special days on which streets would be closed to vehicles to allow children to safely play on them.

Stop de Kindermoord’s founder, Maartje van Putten, who went on to be a Dutch MEP, was a young mother at the time who had personally witnessed several traffic accidents in the city in which children got hurt.

“I saw how parts of the city were torn down to make way for roads,” she said in 2015. “I was very worried by the changes that took place in society — it affected our lives. The streets no longer belonged to the people who lived there, but to huge traffic flows. That made me very angry.”

Others followed suit. Two years after Stop de Kindermoord appeared, another group of activists founded the First Only Real Dutch Cycling Union to press for greater space for bicycles in both cities and the public’s consciousness. They organised bike rides along dangerous stretches of roads, compiling lists of problematic encounters with motorists, and also painted guerilla bike lane signs along some of Amsterdam’s main thoroughfares.

All of this may have amounted to little more than a passing protest movement, similar to the once-weekly Gilets Jaunes demonstrations that gripped France in 2018 before slowly petering out. But the year after the foundation of Stop de Kindermoord was founded, another factor, governed by geopolitics and macro-economic bottom lines, was added to the pro-cycling policy mix. In October 1973, an Egyptian-Syrian coalition of forces invaded the Sinai Peninsula, which had been occupied by Israel since the 1967 Arab-Israeli war. The move sent fears of a nuclear conflict soaring as both the US and the USSR sought to bolster their respective sides. It also had a profound effect on oil prices. Censuring them for their pro-Israeli stance, Saudi Arabia and other major producers slapped the Netherlands with an oil embargo that sent the price of petrol soaring. Internationally, the shock was so profound that Western oil majors began investing in research on alternative power sources, including on the prototype lithium ion batteries for which its scientists would go on to win the Nobel Prize in Chemistry several decades later. Domestically, it all but crippled the Netherlands’ ability to import oil and led the then Prime Minister Den Uryl to deliver a televised address imploring citizens to adopt lifestyle changes and to start thinking about ways of saving energy.

One measure to reduce reliance on oil was the establishment of the same “car free Sundays” in Amsterdam which we see today in many cities around the world. Add to this a 1972 report from the Dutch Economics Institute — which Welleman labels as “pretentious and shocking” — which calculated the cost of updating the country’s travel infrastructure by building new and broadening existing roads at an eye-watering 22 billion guilders (around $7 billion today), and all of a sudden the policy environment was right for cycling to be prioritised once more.

The 1976 Passenger Transport Plan formalised a focus on empowering “slow” journeys — i.e. undertaken by foot or bike — through a four-point programme that included limiting the growth in car use and building better and safer bicycle routes. In the same year, the national Ministry of Transport introduced two financial incentives for regions and cities to opt for more cycle-friendly infrastructure. The first entitled local authorities to receive payment from national coffers for any new bicycle facilities, while the second subsidised 50 percent of the cost of tacking on cycle paths to existing road networks. Both were spectacularly successful: between 1978 and 1988, the length of bicycle paths in the Netherlands increased from 9,300 kilometres to 16,100 kilometres, an increase of 73 percent within a decade.

But something strange was happening. In 1975 two cities, The Hague and Tilburg, constructed two high quality, dedicated cycling routes, painted bright red to differentiate them from vehicle traffic lanes. An assessment a few years later found that the new routes had done virtually nothing to increase the cycling rate in the two cities. Although the lanes themselves were popular and well-used, just one or two dedicated cycle routes appears to be insufficient to encourage more people to cycle. What people needed, planners concluded, was an entire network of cycling infrastructure — a small-width grid integrated into cities with a maximum distance between dedicated cycle lanes (one 1981 study suggested no more than 500 metres between cycle paths).

The poster child for this new, holistic approach to cycle promotion wasn’t Amsterdam, but rather Delft, a city of some 90,000 people situated between The Hague and Rotterdam. Twenty-nine million guilders was allocated to construct a three-tier cycle network of bike paths on a city, district and neighbourhood level in what Welleman describes as a “nearly complete network of bicycle routes”. The results were modest. While bicycle uptake and cyclist safety both rose slightly during the pilot period, the initial bump in the number of cyclists (up some 6 percent) wasn’t sustained beyond the first few years. Nationally, while the government added more than 7,000 kilometres bike lanes, Dutch people simply weren’t cycling all that much more (12.8 billion kilometres were cycled in total in the Netherlands in 1990; in 1994 that had increased to only 12.9 billion kilometres).

Eventually, it dawned on policymakers that it wasn’t just a case of encouraging people to use their bikes more. As Welleman wrote in his now-famous evaluation of the Netherlands’ Bicycle Master Plan: “The simultaneous execution of a policy discouraging car use is (also) deemed necessary.” That is to say, the carrot of cycling more would only move the needle when combined with the stick of driving less.

In November 1989 the transport ministry invited activists, industry groups, politicians and external consultants to participate in a brainstorming session, the results of which were integrated into the Second National Transport Plan of 1990. The two-pronged approach evidently needed since the Delft pilot included measures such as restricting car use in various areas, raising the costs of car use compared to that of other modes of transport and “significantly improving” alternatives to car use, such as public transport and the bicycle. Major cities such as Amsterdam started charging more and more for vehicle parking, and road and fuel tax increases added to the pressure on motorists. Traffic calming measures in major cities have proved successful in not only slowing down vehicles (making each journey take longer and therefore less enticing), but also in improving pedestrian and cyclist safety, which again builds up the number of people practising “slow” transport over “fast”.

In tandem, the plan also focused on ways of reducing bicycle theft, developing better infrastructure such as bicycle routes and parking facilities, and allowing bikes to be carried on public transport such as trains and trams. In something of a first, the legislation also set concrete targets on air pollution reduction, including lowering nitrogen dioxide and carbon emissions from car traffic by 75 percent by 2010, lowering CO2 emissions at least 10 percent — as well as reducing the number of traffic fatalities by 50 percent and injuries by 40 percent. A win-win-win.

Welleman’s assessment of the numerous laws, pilots and local initiatives to get more people on their bikes and less in their cars — to make, in his words “the bicycle the obvious choice of transport” for as many journeys as possible — shows two things. Firstly, it illustrates the huge potential of mass cycling movements. He writes how “practically everyone cycles; young and old; men and women, rich and poor,” meaning there is a sound policy basis for expanding cycling. Secondly, it shows how governments need to start thinking about new and publicly palatable ways to generate the same “push and pull” dynamic practised in the Netherlands. It is not enough to make cycling a pleasure for people — you also have to make other, polluting forms of transport more painful. Balance is key. One only need look at the Gilet Jaunes in France, or the outbreaks of public rage in Chile and Sri Lanka in recent years when governments have tried to remove subsidies on cheap public transport and vehicle fuel without providing cost-equal alternatives.

In a major 2022 survey across 28 countries, the polling firm Ipsos asked more than 20,000 people to give their opinion on cycling, its role in helping the environment, and the major impediments that still exist barring mass uptake. While twice as many people polled (64 percent over 36 percent) agreed that new infrastructure should prioritise bicycles over automobiles, 52 percent said they believed the roads were still too dangerous for them to cycle more regularly.

The Netherlands, by the way, has continued to reap enormous benefits from its cycle-first approach. Since bicycle use was prioritised in national planning in the mid-1970s, the number of children killed in road traffic accidents has fallen an astonishing 97 percent. The average resident’s annual CO2 emissions have gone down more than a third since the mid 1970s. Deaths from air pollution have nearly halved since 1990, the first year for which such data is available. The monitor Environmental Justice Atlas evaluates that the Stop de Kindermoord movement had lasting and visible environmental, health, justice and socio-economic impact including improved air quality meaning fewer days lost to associated illnesses and a better and healthier environment for children.

If there is one city that consistently challenges Amsterdam’s claim to be the most bike-friendly place on the planet, it is Copenhagen. Everyday, residents of the Danish capital travel some 1.2 million kilometres on two wheels, which, up until as recently as 2012, was almost as much as all Americans combined cycled on a daily basis. Copenhagen easily leads the league table in terms of percentage of journeys undertaken by bike, with some 62 percent of people cycling to get around. In terms of commuting, Copenhagen is just 1 percentage point behind its stated goal to have half of all journeys to work undertaken by bicycle by 2025. There are admittedly a number of things in Copenhagen’s favour, here. For one, it is exceptionally flat (during the first three stages of the 2022 Tour de France, held in Denmark, the longest climb on the route was the Great Belt suspension bridge that links eastern and western parts of the country). It is also relatively compact, meaning that most journeys are short distance and therefore easily — in theory — undertaken by bike. As with Amsterdam, the history of cycling in Copenhagen might have turned out very differently indeed without sustained bike use advocacy. The human ecologist Stefan Gossling has argued that Copenhagen’s modern infrastructure has “roots in an existing bicycling culture” that stretches back more than 100 years, which made it easier for cycling campaigners to push against an open policy door and ensure that development plans kept their needs in mind. But as Martin Emanuel, a researcher at Sweden’s KTH Royal Institute of Technology argued in his 2019 paper in the journal Urban History, the relationship between cycling and Copenhagen city planners is more complicated than it appears at first glance.

The story begins at the turn of the 20th Century, a few decades after Copenhagen embarked on an unprecedented expansion of a capital that was until the 1870s contained within fortified walls. The rapid development of working class neighbourhoods — known collectively as brokvarterene, or “bridge quarters” due to the nature of their transport links to the old centre — coincided with the release of “the safety”, a bicycle with two equally sized wheels that was much easier and safer to ride than previous models. As early as 1887 there were 35,000 bicycles in Copenhagen, one for every 10 inhabitants. Once the preserve of the rich and eccentric, mass production meant a far greater proportion of people could cycle. At this time most streets were made up of a cobble-stone roadway for horse-drawn carriages, with pavements for pedestrians and a horse track running in between, often made up of soft soil. When local police rules were up for review in the early 1880s, the city council granted cyclists access to use both sidewalks and the horse tracks, a move that enraged horse riders and which led to increasing tension between cyclists and horsemen.

“Horsemen, many of the military officers, considered cyclists as intruders, and complained about cyclists scaring the horses and their failure to pay respect to the rule of conducts,” writes Emanuel, in what might be the earliest written account of road rage involving a cyclist. Knowing they had numbers — and the inalienable feeling of progress — on their side, Copenhagen’s cyclists took matters into their own hands. By riding repeatedly on the edges of the horse paths, riders compacted the soft soil suitable for horses’ hooves into a smooth, firm service perfect for bikes. In doing so, they had created Copenhagen’s first cycling lanes. City authorities, wishing to signal their view that the tracks were intended for horses, would occasionally plough the soil and cut up the packed edge, making it harder to ride on. But in 1891, either through neglect or nuisance, authorities left one of the narrow hardened sections on a major horse track untouched, ushering in a tacit municipal acceptance of cycling’s physical space in the traffic hierarchy. Several other tracks were soon left intact, and the most popular of these self-made tracks would go on to be Copenhagen’s first official cycle lanes.

The pushback from horsemen was almost immediate and, in a sign that some things truly never change, cycling became embroiled in what can only be described as a late 19th Century culture war. Under pressure from horse enthusiasts, who as Emanuel alluded to were more likely to come from higher and therefore more influential sections of society, the police proposed banning cycling from horse tracks as well as from a number of main streets. In addition, bicycles would be subject to registration and have to carry mandatory licence plates such as those instituted a few months earlier in Stockholm. The debate was front-page fodder for two weeks ahead of a decision in mid-1895, with most speakers including union leaders and council members expressing their support of cycling. One leading anti-cycling voice was the conservative lord mayor, Ludwig Christian Borup, who lamented that it was “somewhat dangerous to quarrel with the bicyclists”, complaining that Copenhagen was suffering from a form of “bicycle rabies”.

Although the precise place of cycling was not fully settled then, city planners did informally proceed with allowing, if not encouraging, bicycles, adding specialised lanes along many large roads under construction. In 1905 the Danish Cyclists’ Federation was founded as a formal front for the lobbying of various pro-bike groups. They decided to take an empirical approach to show to city planners the folly of continuing to insist that horse tracks were primarily meant for — and, therefore, used by — horses. Traffic counts in 1906 showed the ratio of horsemen to cyclists on the tracks to be somewhere in the order of 1:200. On one weekday in June a studied horse track was found to be used by 9,378 cyclists and only 18 horses. It was not just the amount of bicycle riders that interested the DCF, however. They also observed certain trends, such as peak cycling use in rush hours and a strong increase in bicycle use on weekdays, which proved to them that cycling was “not just leisure riding, but emanates from thousands of inhabitants making use of the bicycle as a means of transport to and from their work.” This was obviously in contrast to the recreational riding of horses among the upper classes.

Guided by this data, municipal engineers were tasked with establishing a network of dedicated, smooth bicycle lanes that unleashed the first true cycling revolution along the streets of Copenhagen. By 1913, in a city of half a million inhabitants, Copenhagen was home to 80,000 bicycles. Such a grass-roots development of cycling infrastructure, driven by values rooted in the utilitarian use of bicycles, is an early example of people-powered policy we would do well to emulate today: as the number of cyclists in the city became too great to ignore, bicycles ended up being folded into planning, regulated as traffic and accommodated in infrastructure development.

The interwar years for cycling in Copenhagen suggested that two-wheel transport was there to stay. As the city sprawled outwards, many new radial roads connected suburban communities with central areas, and most were equipped with bike lanes. Not even the Great Depression could stem cycling’s rising popularity — while total traffic on Copenhagen’s streets increased by a quarter over all in the 1920s, cycling was by far the fastest growing transport mode. By 1935 44 percent of all residents used a bicycle as their main means of getting around. The extent of cyclable roads more than doubled between 1912 and 1927 from 35 to 74 kilometres, and by 1935 was almost 100 kilometres. Although German occupation during World War II led to similar dips in both cycling rate and the number of available bicycles similar to the Netherlands’ experience, by 1950 cycling in Denmark had recovered to pre-war levels. Emanuel puts this partly down to agreement among all parties on the bicycle’s pre-eminence in Copenhagen, not just the vehicle itself or the thousands who used one each day, but the city’s planning had been so influenced by bikes that “the urban transport system would be dysfunctional without them”. It also didn’t hurt that the government suppressed car ownership after the war through a limit on imports and subsequent high taxes.

The real make or break moment for Danish cycling came, as in the Netherlands, when shifting demographic pressures from growing and increasingly decentralised populations appeared to play into automobile producers’ hands. Beginning in the mid-1950s, bicycle use began to decline in Copenhagen, particularly along main thoroughfares where new tram and bus lines were cropping up. For example, cyclists’ share of traffic along the bridges between the main island of Sjaelland and the smaller Amager to the southeast, where the city’s main airport is located, dropped from 70 percent to just 20 percent between 1955 and 1970. Policymakers’ attention on cars and public transport had sapped attention to such a degree that, Emanuel writes, by the 1960s “hardly anyone mentioned the bicycle anymore”. Cycle lanes, the presence of which became harder to justify outside of peak hours, began to be removed, with the capital region’s cycling network some 25 kilometres shorter in 1974 than it was in 1970. Yet despite this regression in bicycle infrastructure, Copenhagen residents remained doggedly attached to their bicycles: a 1967 survey found that 21.5 percent of commuters and 32.7 percent of school children still rode on a daily basis.

The economic crisis of the mid-1970s hit Denmark just as it did most of the developed world. Its recovery was markedly slower than similar-sized economies and so, pressed by the cost of living and especially high fuel prices, Danes flocked back to the bicycle. The 1973 oil crisis galvanised Copenhagen’s grassroots environmental movements as authorities abandoned several major road projects — partly due to their cost, partly due to their growing unpopularity. At the same time, a variety of measures proved a lasting disincentive against cars entering central areas, including pedestrianised streets, harsher speed limits and prioritised zones for public transport. Sensing its moment, the Cycling Federation lobbied authorities to develop their own cycling master plan, then pressed ahead with its own that constituted plans for nothing short of cycling utopia: 92 kilometres of new cycle lanes within 10 years. While the plan was never officially adopted, it served as a guide for planners to develop what today is a world famous bicycle network. The period from 1975 to 1985 was the watershed moment for Danish cycling, seeing the fastest growth of bicycle infrastructure in Copenhagen’s history. The city’s rebirth from the 1970s economic mire took longer than most, but when it came the bicycle was its beating heart. With its population and GDP booming, the city embraced cycling as a key tenet of remaking itself as a liveable, modern city. Such was cycling’s importance to how Copenhageners saw themselves, that the issue has decided local elections over recent decades, with the most pro-cycling candidates more likely than not to reach and stay in office. And while young professionals in and around Copenhagen own ever more cars, they still prefer the bicycle for their daily commute (car use in Denmark actually declined in the decade to 2013 despite higher ownership).

In Emanuel’s telling, the story of cycling in Copenhagen is one of precisely the kind of collective action, data-led policy asks and community-led development that many of today’s cities would benefit from. In its starring role isn’t some visionary leader of a cast of innovative, future-facing civil servants. The hero in this story is the cyclist themself. Through collectivisation, they ensured their priorities were represented when it came to urban planning, meaning that new systems would be suited to their needs, engendering the long-lasting and frequent usage against which successful infrastructure projects are measured.

“Cyclists pushed things forward,” says Emanuel. “The slow acceptance within the municipal administration in the 1970s of infrastructure provision for cycling in other than ‘easy’ locations suggests that little would have happened without pressure from cyclists and their representatives, and thus highlights the importance of rebels and resistance in the co-production of infrastructure systems.”

A group of informed individuals, armed with good data and a shared vision, can change the world. The point I want to make here, is that the climate crisis is now inevitable, but how we react to it is anything but. There is nothing immutable about how our future looks, about what we choose to prioritise as the measure of what constitutes a good life. How happiness happens. One friend recently told me it could be something as simple as how many books they got to read in the sun each year, and I found great beauty in that.

Finally, I suppose most elementally, I want to ask you to keep doing and making things. I don’t mean that as a kind of irreverent “Eat, pray, love” carpe diem maxim. I mean it because ultimately, in this dizzyingly short life of ours, what we do is the thing that matters.

My dad constantly told me a man (he meant person, but he was, forgive him, a boomer) is what he does, not what he says. He meant this quite literally: he built houses for a living, he was a maker of family universes. The output of your physical and spiritual tasks are all that we will leave behind when it’s over.

So, please. Do the things. Make the things. Write that poem. Pick the blackberries. See your friends. Protest. Organise. Walk on the beach. Swim in a pond in the rain. Feel the sun on your face and bathe in the quiet of the moment. All the self-help people and the therapists out there all agree so there must be truth in it: anxiety means you’re living too much in the future. Remember to enjoy the present. Even in 2026, it’s still good, there’s still value in it, despite all we have already lost.

This may sound like a trite end to a long ramble about power cuts and polemic. But I truly mean this: inaction has gotten us in this mess. It is only through action that we will survive it. History has shown us what works, what doesn’t, and what it takes to head off the pushback so we safeguard the win for the most people possible. It takes stamina.

Not to speak on your behalf, but I’m ring rusty. I’ve felt, particularly recently, paralysed practically by the heaviness of the whole thing. I hope to be in a position where I can do more, to be more, to be a better person to the people who value and rely on me. The power will go out again. Next time, I’ll be more ready.

Omar Mukhtar was born in 1858 into a poverty that would have looked little different back at the dawn of the Ottoman Empire to which his hometown still belonged.

Cyrenaica, as old as ancient Rome, was a neglected outpost of Constantinople’s rusting dominion, a Libyan province dominated by the forested foothills of Jabal Akhtar, a mass of sun-bleached limestone pocked with snatches of woodland, maquis and garrique.

Mukhtar’s family, like most families, was dirt poor. His father died when Mukhtar was still a child, whereupon he was taken in and raised by the local Mashayakh, a small collective of religious scholars. They had adopted a student of remarkable powers.

Mukhtar took to learning the Quran with the zeal of a young man given a second chance in life. As a teenager, he never slept more than half the night, spending the sibilant hours until dawn prayer whispering the Holy Book back to itself.

The first half of Mukhtar’s life is recorded only in moments and virtues. Once a week, regardless of his situation, health, or safety, Mukhtar would recite the Quran in its entirety. At one point, he pops up in Sudan, appearing like a djinn on horseback, to face down a lion and shoot it in the head.

In 1911, when Mukhtar was 53, Cyrenaica was invaded by the Italian army, in an occupation that would become the clarion call of 20th Century Europe’s prototype fascist dictator, Benito Mussolini.

Mukhtar led a clutch of rebels in a decade-long onslaught against the colonial forces, rising quickly to prominence as a brave and canny guerrilla commander. Dodging the shells of Italian tanks and strafing fighter planes, Mukhtar’s band of 3,000 men on horses launched wildcat skirmishes against occupying troops.

The resistance was so effective that, by the time Mussolini had risen to power as Il Duce, Mukhtar’s muhajideen were clashing with Italian troops hundreds of times a year. In an unhinged and inhuman attempt to staunch the damage, the fascists set up concentration camps throughout the region, at one point holding 100,000 Bedouin men, women and children — around half of Cyrenaica’s total population at the time — in open air prisons.

Whenever one of Mukhtar’s band was captured alive, the Italian soldiers would subject them to gratuitous torture before death, displaying the ballooning corpse in the street for days as a warning.

Mukhtar would eventually be captured by the fascists and, after refusing their entreaties to stand his militia down, was publicly hanged in 1931.

But it is one scene from a year earlier that I want to tell you about today.

After another successful raid against Italian positions near Tobruk, Mukhtar’s men captured two fascist soldiers alive. One of his closest aides, a general, made a suggestion:

“Why don’t we make an example of them like they do to us?”

Mukhtar refused.

“But they do it to us,” the general pressed.

Mukhtar placed his hands on the general’s shoulders, looked him in the eye, and said:

“These people are not our teachers.”

Some mornings I wake up exhausted, drenched in grief, crushed by the formless and empty day that stretches out ahead of me with no prospect of joy or surprise. All that awaits is the news, the images of fire and drowning.

I gather the sadness around me and wear it like a coat. It weighs down on me like a field of dew, slumping my shoulders and bowing my head in one, day-long cower.

Recently several friends with young children have reached out to me nursing the pain of climate grief, jolted into fear by the terrifying heat of a future arrived too soon.

It is a grief we share.

The record-breaking heatwaves across Europe last month that shattered glass and salted the earth. Terrifying heat. Murderous heat. Heat that killed thousands of people, that led to more than 130 drownings across France and Germany as panicked humans, their reptilian brain trip wired by the unnatural and unbearable heat, fled like early man into the sea. Heat that boiled children alive in their cars.

Heat that torched fields and forests. That warped and buckled train tracks. Heat that returned tarmac to bitumen. Heat that in Spain, where I recently spent a week’s torpid leave battling food poisoning, literally vaporised tourists who failed to heed evacuation orders.

A summer of smoke and ash. A summer of violence.

During the record-breaking 10-day peak in the UK last month, we sent thermometers to primary school teachers and asked them to record classroom temperatures. They came back with readings as high as 38C.

They told us of children complaining of headaches and nausea, of crying out for their mothers as they were subjected to temperatures up to 12C hotter than the recommended safe maximum.

What scares me most is not how head-spinningly unprepared we are for the type of heat that our burning of fossil fuels has made inevitable. It is not even the fact that we are — in 2026, a decade on from the Paris Agreement and at a time when solar and wind power is cheaper nearly everywhere on the planet than new oil and gas projects — actively accelerating climate breakdown through our ever increasing greenhouse gas emissions. Each year new record pollution. Each year new record profits.

No, what scares me the most when faced with the death and disease and exhaustion and pain that comes with future heat, is that, unless we stop the fossil fuel industry from profitably poisoning us to death: this will be the coolest summer for the rest of our lives.

Now that the heat has subsided here (although the drought looks set to persist, and more heatwaves are in the gun’s chamber), now we have some distance between us and the animal fear of those firebombed days, I’ve been speaking with some of my friends with young children.

Mostly they are after some context. A few are exceptionally up on the science of climate change, others know intuitively that Things Aren’t Alright.

One old friend messaged me — I was walking down the hill towards the sea and I could see the sky was diluted, milky like watered ouzo. Turns out the smoke from the wildfires raging in Canada has spread across the ocean and is now smearing our horizon, supercharging our sunsets and sunrises.

These days disasters don’t stay put for long.

She asked me about RCP8.5, which is a now-largely-disregarded Business As Usual scenario for global heating. She had been speaking with another friend, also a mother, and had come away from the conversation with the impression that climate change might not be as bad as predicted.

I have my own views about the de-facto retirement of RCP8.5 (Briefly: It’s true that on a pure emissions basis we are not on course to burn the fossil fuels that would make RCP8.5 a reality from a straightforward radiative forcing perspective. However, I am constantly dismayed at scientists’ inability or unwillingness to factor in Black Swan phenomena, such as any of the numerous planetary tipping points — from albedo decline, ice sheet decay, boreal forest loss etc. etc. — that could lock us into a feedback loop of warming.)

So we talked that through a bit, and then she said to me:

“It’s really getting to me. I worry so much about the world I’m leaving for my little one. How do you stay hopeful at the state of the world?”

And I thought of Omar Mukhtar.

I have a nephew with golden locks. They coil in ringlets and shine peach in the sun. He has the mischief of his father, his mother’s gift of the gab, and he is happy and healthy and sassy. He is nearly three. He is the apple of my eye.

By the time he is my age his world will look rather changed. I cannot tell for certain which load bearing function will fail first in the climate breakdown, but my best guess is that it will be our food system. I have written for years on the structural vulnerabilities imperilling humankind’s ability to move calories around the world so that everyone gets enough.

All the risks that the food system has faced in the last 60 years — crop failures, supply chain bottlenecks, food loss and waste, geopolitics — these are all compounded by a heating climate. Climate change is a threat multiplier.

In 2019 I was in Geneva for the launch of a special UN report on land use. It was a warm and rainy week spent along the banks of the lake, meeting with experts from the Intergovernmental Panel on Climate Change to go through the dizzying findings of their 1,300-page assessment (I also got to hang out with Greta again, which is always a pleasure!)

I was trying to get to the bottom of what the scientists were, in their diplomatically redacted way, warning the world about.

I had the embargoed Summary for Policymakers — essentially a political synopsis of the findings — but I couldn’t make it out: what was the take home here? There were so many competing dynamics, entirely divergent possible pathways based on policy choice, a sort of superposition of scenarios, all of which could be true but only one of which that would be.

I had coffee with one of the most experienced scientists to work on the panel, a senior North American woman who I had known for years and whose work I had always followed closely. In a world of hypotheticals and certainty tolerances, she always had the most plastic of takes. She told me:

“It means a reckoning. We cannot continue to treat this planet as if its resources are infinite. Hard choices, trade offs. There is no business as usual anymore.”

As I listened, and she explained what some of the report’s modelled pathways meant to the lay person, I gathered that humanity was at a crossroads in how it tackled its two most pressing sustainability challenges: not just how it physically sustains itself (food systems), but also how it sustains the circumstances that allow it to sustain itself (Earth systems and climate). We could either feed 10 billion people by 2050 or use land to limit global heating. Food or forests, you can’t have both.

The warning went ignored. After several years of political and popular momentum behind climate action, with the largest companies and countries in the world issuing pledge after pledge to reduce their emissions, we reached a collective high watermark of action. Then the pandemic hit. Governments, desperate for a return to normal, doubled down on our fossil fuel powered economy, providing unprecedented amounts of taxpayers’ money to oil- and gas-based regrowth.

People who had dabbled in staycationing or vegetarianism were sick of months of privation and forced compliance and cut loose. The climate suddenly became less of an item of emergency intervention than a distant curiousity. An abstract, something too large and too out of scale to the human condition for people to engage with meaningfully. Humanity’s greatest ever challenge became a “not now” problem, buried beneath the mundane toil and routine drudgery of living (and making a living) in the 2020s.

Aside from the corporations driving our oblivion and the politicians they pay to look the other way, I apportion no blame. Life for so many people around the world is still nasty, brutish and short. Even those of us (such as myself) who live a life, relatively and historically speaking, of unimaginable privilege still struggle with the day to day white noise of modern life, the drain of the capitalist experiment, and the toll it takes on you physically, emotionally and psychologically.

During periods of extremity, where we hide the shadow as fear lays panting in the white hot street, the feeling of things rapidly unravelling can hit you like vertigo: the realisation that the emergency is happening, that the time to panic is now, save yourselves and your loved ones because this disaster isn’t a movie. It has come.

I remember sitting in the waiting room outside the ICU at the hospital in Marbella. My mum and my sister were back at the house, an hour’s drive away, using the period where the anaesthetic wears off to nip home and get a change of clothes. Mum had slept 5 nights straight on the lino sofa next to my dad’s hopsital bed. I felt flighty, like I might drift away on the slightest breath of breeze.

Me and my sister coined a term for it, a shorthand to use in the coming weeks to express the mixture of anxiety and hysteria we were all feeling, knowing that something dreadful lay in wait: I felt fizzy.

The surgeon came out, and shook my hand. She could see I was alone and frightened, and I will forever thank her for taking the time to speak slowly and clearly through the surgical mask (this was at the tail end of Covid). She put her arms on my shoulders, looked me in the eye, and delivered my father’s prognosis: grade 4 glioblastoma.

I knew immediately that he would die. I had passed through a threshold between life and certain death, and I’d passed instantaneously, before my dad even knew he was a dead man.

It took another 86 days for death to come. He fought it. At first, he fought with bravery and purpose, the same grit and stubbornness that had served him so well in life, allowed him to bend reality to the force of his will and build a universe that he viewed as worthy of my mum and us. Then, as the treatment stopped, and the pain increased, he began to fight pointlessly, in all directions at once, even with the people looking after him in his final hours.

I am told we are lucky: we “got to be with him” at the end, in the home that he built, not in some medicalised ward surrounded by masked strangers and the cold, unfeeling beep of monitors. I’m not sure. He still died about as hard as a man can die. It was squalid and awful and undignified.

But knowing he was going to die, the annihilation of hope for anything better, was — for me at least — oddly liberating. It did bring with it a type of mindfulness, a determination to live more in the present, enjoy the small irrelevances of each day, to make the most of the now extremely finite time I had left with my dad, one of only two people in the history of the universe who will ever love me unconditionally.

Knowing the worst was now inevitable removed much of its jeopardy. I didn’t have to catastrophise anything: for once, there was no scenario planning or war gaming to be done. I just had to adapt to events.

This summer reminds me too of the night Notre Dame caught fire. It was a Monday, I was just finishing up my shift after filing my last story of the day — probably around 1830 or so. I worked in a big newsroom in the 2nd arrondissement, a huge spaceship of an office block landed right opposite the Palais Brongniart, Paris’ former stock exchange.

The newsroom of course had many TVs, and as I was leaving my desk I looked up and saw some breaking news: Fire reported at Notre Dame. There was some live ground shots from the Ile Saint Louis of the back of the cathedral with what looked like a worrying amount of smoke coming from its roof.

I ran from work a mile or so south to the Ile de la Cite, across the Seine by the Pont au Change, which the police hadn’t had chance yet to close. I was stopped in my tracks by what I saw: the whole of the roof was ablaze, flames licking the famous fleche, and thick smoke was pilloring into the spring sky.

Security soon arrived and kicked us away to the other side of the river, right outside the Hotel de Ville, where I stood with thousands of other Parisians watching with mounting horror as the flames crept ever higher. I took a video of the moment the fleche collapsed; as I watch it back now the crowd noise is one of prehistoric pain and fear.

I, like half of Paris it seemed at the time, spent most of the night on the north bank of the river, the fire raging on as the sky darkened. Being a wire journalist, I had a direct feed on the latest developments.

It was shortly after 9pm that President Macron announced that the towers of Notre Dame, the twin turrets that had stood watch over Paris for 800 years, which had survived plague, revolution, famine and war, which held firm even as the sans culottes stacked dynamite and firewood at their base, which didn’t succumb even during Bismarck’s fire bombing or Nazi occuptation — those towers were now at risk of collapsing.

And I kept thinking the same thought: this is happening on our watch.

Now they are evacuating Bordeaux. The fires that have burned for three days around the Bay of Arcachon are increasingly ungovernable. 250,000 people have been told to leave with just the clothes on their backs. They are even shortening the last stage of the Tour de France so security personnel can tackle the inferno.

Several towns west of Madrid have also been evacuated, and the prime minister is warning that the wildfires are now entirely out of anyone’s control. We are in god’s hands, it seems.

This is happening on our watch: the great burning. These fires are the inevitable, immutable result of the last 150 years of fossil-fuelled growth.

We have spent the last several generations incurring a debt to nature, spewing out ever greater clouds of pollution and calling it progress. That debt has largely, so far, been paid by the global majority, with uncalcuable loss and damage wrought for years already on the societies least responsible for the problem.

The debt is now here for us all. We knew it would come. Most of us (myself included) assumed we would have a little more time, that we would at least make it to 2030 or 2040 before the worst excesses and privations of climate breakdown would make themselves unignorable.

Sometimes you come across a piece of writing at exactly the time you need to. Last week, I was doomscrolling as usual, looking at the monstrous El Nino that’s coming down the line this year, when The Milder Sun by Gregory Normington popped up on my feed.

It is a story of filtered hope, one set 200 years in the future, after humankind has undergone the maelstrom of collapse, the collapse of society and systems that our fossil fuel addiction has made inevitable, before emerging, burned and blinking, possessed by a gossamer purpose to rebuild the world for people and nature.

I like it because it doesn’t shy away from the loss and pain that is waiting for all of us. It’s just that the loss and pain, in Gregory’s telling, isn’t the end of the story. He outlines how it is going to be: the institutions governing our world will disintegrate as the social fabric spools apart, cascading collapse across our food and health, energy and housing systems, then a period of post-impact desolation, followed by slow, thoughtful, determined rebuilding.

I ask his permission, and quote from the key narrative crux of the piece:

Two centuries ago our planet was locked, seemingly inexorably, in what we call the Locust Years. The intentions that had led to the great devouring were not all ignoble. Billions had been lifted out of absolute poverty by fossil fuels and everything they enabled. But it was ancient power, stolen from the bosom of Earth where she had sequestered it for millions of years, and loosening it broke the balance of the world. A terrible and — at first — entirely accidental experiment in atmospheric chemistry was conducted on all living beings.

The biosphere is ruthlessly coherent. It rations life for the individual to maximise it for the totality of living things. Humanity’s foundational and entirely natural sin was our compulsion to reverse that formula. And so the carrying capacity of our planet — the only known refuge for life in the Universe — was stretched to breaking point. Weather systems warped and buckled; soils, overworked, began to fail; the oceans acidified; a great silence spread across what had been the wild places of the world. Most governments and corporations, and the media empires that existed to further their interests, tried at first to ignore what was happening, to deny it, to spread confusion, and thus to delay a transition that threatened what they perceived to be their wealth. These efforts met in many quarters with willing collaborators, for a civilisation with a pathological inability to contemplate its mortality will spawn morbid symptoms. People so feared positive change, clinging to what hurt them, that negative change — Gaia turning on those that abused Her — accelerated at harrowing speed. Still the plutocrats, soaring in their private jets above the burning, fought attempts at evasive action, or indulged in fantasies of technological supremacy, imagining that they could shield themselves behind their quarterly finance reports.

The Locusts seemed unshiftable. Yet as things worsened, those we call Earth Tongues began to cry out in alarm. They were few at first, then a growing chorus. Children refused to attend school, meaning to shame their adults who refused to learn what they had a duty to know. The Locusts, of course — in whose image the world seemed to be made — would not listen, for hunger drove them on. It was not biological hunger, for which no being need feel shame. It was the diseased hunger of people driven by a lack of love and connection. No amount of eating can fill those voids: they seek in taking what can only be found in giving. In the Locust Years, it seemed society lacked the vocabulary even to speak of these things. Cynicism and mockery were the norm; illusion and neurotic distraction were easier than facing reality, and much violence was done to the very notion of truth. The Earth Tongues had to shriek and holler, increasing in volume even as what they feared roared in the ears of all living peoples. How easy it was, as cities choked on wildfire smoke, to despair. Many did. But while we have always been our greatest foe, we are also our only hope.

The struggle was titanic, first to slow the speed of destruction, then to stop and reverse it. For many years it seemed impossible. The graph lines soared towards calamity; forests burned, crops failed, people perished. There were vested interests hindering action. Yet slowly, then less slowly, finally at dizzying speed, these obstacles — technological at first, at last cultural and political — were swept away. Raw economics worked, for once, in Life’s favour. Solar became the fastest growing source of energy in history. Together with wind and geothermal, and batteries to store what they generated, it put out the fossil flames. Transportation, heating, and industry were electrified. Electrons replaced molecules and brutal cartels lost their power — though not without a fight. Within three decades humanity said goodbye to the heat engine that had powered and poisoned it for three centuries.

Technological change was essential to our survival. Some of the Earth Tongues were uncomfortable with this, and the trade-offs it required. They were right that technology alone was not enough. Rather, it bought us time to transform ourselves. Cities became places for people rather than the machines that had seemed to serve us. Most of us became city-dwellers, but what we meant by that had changed, for the humming engine of urban life quietened, with its tree tenants and roof meadows and shaded cycle paths, with its parks and gardens where children could play, its pools and ponds and daylighted rivers, its market gardens and allotments. Food waste became one sin that a post-religious north and a religious south could agree on. Billions switched, or reverted, to plant-based diets as we despaired of the silent springs that our poisons had occasioned, as we became ashamed of our industrial slaughter of our fellow beings. It became possible to produce, from nothing more than sunlight and bacteria, the proteins we needed for a healthy diet. The murderous feedlots, the vast sacrifice zones devoted to feed-crops, were returned to wilderness, and began to draw down some of the carbon that we had unleashed. What the Locusts had, for so long, persuaded the majority meant impoverishment revealed itself to be the opposite: true abundance. The weight of cognitive dissonance that most people had not even realised they carried was lifted from their shoulders. We became a civilisation of gardeners, makers, companions to growing things. There were, after all, entire continents to restore.

I sent the story to my friend and it seemed to help. I will tell you all what I told her, and what I want to say to all the people I love — and especially my sister — who have young children or who are expecting:

The bad thing is here. It will be scary and damaging. It will hurt. You will have to prepare for life not as normal, and you will have to alter how you exist in order to keep your child safe. There are some damages that are now baked in — they cannot be stopped no matter how much I wish they could be.

Sea level rise, land loss, coral reef collapse, ocean acidification. None of these would go away even if we stopped emitting overnight. Climate change is a stock problem; we have polluted too much for too long to simply reset the atmosphere. You and your child are not to blame, but you will both need to adapt to our changed existence.

Retain as robust a support network as you can — local community is going to be crucial in surviving the last of the Locust Years. Keep your Earth Tongues safe and grounded and connected with their land and their people. They will need it.

Encourage them to stand up for themselves but also to ask for help when they can’t do so alone; asking other humans for help is a superpower key — it unlocks immense fortitude in people.

Let them learn how this happened, who is responsible, and who is suffering the most from it. Feed them that injustice — it will stoke their flames and keep them focused on what we need to retain during the gathering darkness.

Love what we have left but do not shy away from the grief for that we have lost. Embrace it. We still have a beautiful home, unique among the universe and throughout interstellar time. It had all we ever needed. The fact that we fucked this up so hard should hurt. It is our civilisational shame, although, in fairness, we only realised with anything approaching certainty what we were messing with a generation or so ago.

Know that the pain you feel about climate collapse is just and deserved. It has its own economy: it was, in retrospect, a pretty wild gambit to dig up the gigajoules of energy the Earth had spent 200 million years storing, burn them like we lived in an infinite timeline, and not to expect some Consequences.

Those consequences are here and it means, ultimately, that life is going to come very close to breaking apart for most of us. Feel the anger. Let it sit in your soul. Then use it to send some energy ahead to your future self for the trials that are to come.

I am reminded now of what Julian Barnes said about losing his wife, how the feelings correspond precisely, almost immaculately, to the scale of the loss:

“Nature is so exact; it hurts exactly as much as it is worth. If it didn’t matter, it wouldn’t matter.”

I’m nearly middle aged. I thought by now I would be better at this. I always imagined through youth and young manhood that, by the time I got here, I would have children, own my own house, have a career that could sustain me throughout my life, an existence that was solid and established and somewhat resistant to shocks.

None of this has happened. Maybe it is the result of a youth spent across continents, covering wars and revolutions. Disasters. Death. Perhaps it comes from the PTSD that some survivors carry with them. Maybe it’s because I cannot shake the knowledge of how bad things are going to get before they get better.

When my dad was dying, time became contracted — the long, carefree spool of future life was anniliated, replaced by the night after night of it all: the tantrums, the howls, the arguments, the accidents. Our world became infintesimally small. It forced us all to get busy, to get our business with him in order.

I told him that I was not having kids, and that caused him confused sadness, which I now regret. He said that his generation grew up under the threat of nuclear winter, and they still managed to bring the next into the world, scary as it all was. Every generation probably thinks it is the last, he said.

I told him that this time it is different — that this wave will not spare a single person (although it will hit differently in various parts of the world depending on their ability [and political willingness] to adapt to overtaking events). This is not localised destruction. What is coming is not something from Vonnegut. More like M.P. Shiel: a slow moving cloud of poison that no one may escape.

This is the last crisis of our current timeline, not nuclear armaggedon but something far more unforgiveable: death by plenty, death bled out from the dearth of empathy, of mercy.

I will never be sure if he understood or not.

And I do feel the loss of not having children, I am grieving that, although it is too late for me now. But I want you to have them. I could not do what you are doing. I am in awe of you.

You are doing everything you can and the most consequential thing any of us will ever do: making people in your image. People who will go on to share your compassion and outrage, share your realisation that the individual wants and needs must now be rationed for the sake of the whole. Your truth – that you are a human in space time, together with others in the loneliness of climate collapse, solitary but in solidarity – will live on in them.

Do not be afraid for your children. Be proud of them. You are raising the next generation of activists and leaders. The generation that will finally take responsibility as custodians of our planet. They will take the hard choices when we demurred. They will hold the powerful accountable, not because it is easy but because it is vital.

They will keep the wisdom of Omar Mukhtar within themselves. They will not allow past generations, their greed and their cowardice, to define the scope of their future.

We are not their teachers.

A Summer of Violence was originally published in The New Climate. on Medium, where people are continuing the conversation by highlighting and responding to this story.

Gen AI is like Covid in that it is bad for you even if it never infects you.

Preface

You will probably know by now about the initiative QuitGPT, which is exactly what it sounds like, and which is, given the evidence, very likely a ploy by OpenAI’s leading competitor aimed at getting you to dump one unthinking Internet rephrasing machine for another. There are several ethical and practical problems to this, which people more expert than me can talk you through. But my main issue with it is more simple: Every single Gen AI developer is a creep. I mean it. Take Anthropic, everyone’s sudden favourite cuddly hyperscaler who did a really moral thing and pulled out of a mass surveillance project with the US government, until it didn’t, and which — and I need you to pay attention to this — continues to assist the US military in the industrialised slaughter of Iranians. And while I currently lack the proof, I know deep in my bones that Claude told the on duty commander to fire missiles at the school in Minab that atomised more than 160 young girls. It had a week where Serious People were holding it up as some sort of moral paragon, you know, as if its mass-scale data theft and plagairism was done in a mindful way, or the carbon emissions from its ocean boiling data centres somehow didn’t stink like all the others. Or as if, I dunno, its head of safety didn’t resign last month and fire a NDA-censored warning shot about the entire industry? Not convinced? How about xAI, then, whose apex consumer product Grok has persistently refused to stop allowing users to create abusive and illegal images. Or how about Meta, whose flagship consumer wearable AI tech device has been secretly filming the naked spouses of anyone with a small enough… ego to feel the need to buy spectacles that are also a surveillance device. Microsoft AI leaks your confidential data. To repeat: every single Gen AI developer is ran and staffed by freaks and you need to stop giving them money to accelerate their plans to replace you.

Something big is happening

Which brings me to the point of this piece. I won’t link to the article, but chances are if you clicked on this you’ve already seen it. I’m not going to do a line-for-line rebuttal because I am a busy person with a team to manage, and because the fabric of its arguments amounts to the same tapestry of entirely unverifiable, “just trust me bro” claims that the high priests of tech grifting have made since the very beginning about Generative AI.

But, since I’m a journalist, and I believe motives matter, I will take a moment to point out why the author is making such lofty claims about the levelling up of Gen AI products (a category, remember, that Big Tech has spent more than $1 trillion developing and cannot monetise — note especially here that all the really good AI tools are suddenly behind hefty pay walls, a coincidence I am sure). He runs an autocomplete company — and because that’s basically what LLM based models are, the world’s most wasteful, resource intensive and expensive autocompleters — so he may well find use cases. He’s also accused of falsifying benchmarking data of his products.

Although I call BS on more or less the entire thing, the article — which (and this is some indictment on the state AI in the popular imagination — generated 50 million views) did contain a stopped-clock kernel of truth. In it, the author says that Gen AI will disrupt the world to the same order of magnitude as the pandemic. And he’s right about that, just not in the way he thinks.

Gen AI is not like Covid because, as the author asserts, most people fail to imagine a changed world. (He claims that Covid disrupted history because most people dismissed it as a cold, which, given that the only people who did dismiss Covid as a cold were entitled white people, might be saying more about himself than he intended).

No. The real way Gen AI is like Covid is that it is bad for you even if it never infects you. You may personally not have got or gotten sick from Covid and none of the people you love may have either, but it still damaged your mental and physical health. The missed dentists appointments, the cancelled electives, the struggle to re-up your ADHD meds, the isolation, the doom scrolling. The generalised sociopathic but no less true realisation that for most of us, for the first time of our lives, simply breathing the same air as another person could kill you. It is not much of an exaggeration to say for a period of our lives between 2020 and 2023, humans were a hazard.

Now it’s true that Gen AI boosters say things like “humans eat too much to train” and “the tax of human labour”, and are quite obviously sociopaths in the same way most of the Epstein class are, and that they do view humanity as a problem to be fixed by tech. But that’s not what we are talking about here.

I think Gen AI will be bad for you, even if you’re not forced to use it, in three main ways: it is going to disrupt your financial plans, it is going to make work harder to accomplish, it is going to make your life profoundly more annoying to navigate.

Money

What’s happening in financing Gen AI development might not be actual fraud but it is certainly shaped like it: circular investment of $100s of billions; the private equity debt they have now accrued — there’s more debt tied up in Gen AI value chains than in the US banking system and that is something that should both alarm and anger you; and the wild dissonance between AI bonds risk and their yields. That’s to say nothing of the way data centre leases are being represented on balance sheets as all upside and no long-term cost, which, yes, absolutely rhymes with the kind of accounting fraud committed by Enron. Moody’s last week calculated that the amount of future data centre cost liability that is not represented anywhere on the Big 5 hyperscalers books is $662 billion. How do they get away with signing 20-year leases with users but only write down 4 years worth of costs, when they will either need to completely refit their centres or else purchase back the full 20 years?! Who knows?!

And it’s going to collapse right? Because it’s a Ponzi scheme. Because the entire industry is really just three companies in a trench coat, investing hundreds of billions of private equity debt in each other and calling it growth. (Circular investment, I may add, which to quote Ellevest consultancy, can “distort growth metrics, hide real underlying demand, or mask structural weaknesses”. Reader, see if you can find the “structural weaknesses” in an industry whose market-distorting whale relies on one company to manufacture 90 percent of its apex product using machines made by only one other company. Oh, and they’re made within gunshot distance of China. Any alarm bells here?)

It’s also going to collapse because scaling laws — the base case premise of LLM development, in which ever bigger models require ever more finance but produce ever better products — don’t apply to downstream development. I’ll repeat: for even the most state of the art LLM-based agentic products, the scale of investment and power of the model doesn’t scale to improved results in outputs. This, I believe, is the reason why the Magnificent 7 Gen AI boosters are losing senior programmers in their droves, why heads of safety are resigning and trying their best to signal the alarm between NDAs. It is this: whatever it is LLM-based AI tools can do, that is all they will ever be able to do. Not know how many r’s are in Strawberry, not count to 200, not tell you even whether it is better for you to walk or take your car to the carwash.

Sam Altman — the high priest of Gen AI grifting — has admitted there is no order of magnitude, no giant leap, no rabbit in a hat, no breakthrough and no plan B other than feeding machines the Internet and asking them to lossily output parts of it forever.

And if you think that that doesn’t sound like an industry worth $5 trillion, brothers and sisters, you are not wrong.

While we’re on the subject, SEO and audience relatability requires me to use the term “Gen AI”, meaning “Generative Artificial Intelligence”. But I feel strongly that this is a misnomer, that artificial intelligence doesn’t exist and has never existed. Let me explain. We’re told each time a new model is released that it has aced benchmarking, when in fact they have cheated the tests because the answers to the questions the models are asked exist on the Internet and can therefore be mimetically recalled (I go into this here). X Chatbot can pass the BAR exam, you say? Y Chatbot can produce workable code? Er, yes, OF COURSE they can, BECAUSE THIS DATA EXISTS AND THEY HAVE BEEN TRAINED ON IT. Calling the process of mimicking input through output “intelligent” is like calling a parrot intelligent for speaking, or vinyl record intelligent for playing music. It reproduces the sound that’s encoded on it. Recently a team of researchers from MIT and Stanford set AIs maths questions the humans knew the AI couldn’t access in the data, as they were written and solved by humans but not yet published. Spoiler: The models all failed. I will not give Anthropic’s latest bullshit “Claude has attained consciousness” claim any more oxygen than it deserves. Suffice to say: this is the amino acid of Gen AI grifter DNA — ever larger and less verifiable lies — and if you’re a journalist who covered this credulously, well, shame on you.

I urge readers to spend half an hour with Forbes Gen AI outlook 2026. It’s chilling, but entirely predictably so. Let’s just say a lot of businesses who have foist Gen AI on their workers are going to go from fucking around to finding out really fast this year. And, as with every crash, the fault will not be ours, but the fallout will be.

For those keeping tabs there has now been $1.2 trillion wiped off the value of tech stocks in 2026 alone, and that is being explained away as investors getting spooked that AI might render some other tech providers, software developers say, obsolete.

To whit, Microsoft AI’s CEO said last month that all white collar jobs will be replaced by AI agents within 18 months, which isn’t so much unverifiable hyperbole as red-handed proof these people are bullshitting. Because if that’s true, and AI agents replace all human roles — and I really feel like this is a question people aren’t asking enough and is so fundamental to economics as a science but also just the labour economy as a thing — if every human is replaced by an AI… WHO IS GOING TO BUY AND USE THE FUCKING AI YOU INTERGALACTIC CRETINS??

Henry Ford figured this shit out 100 years ago: you make products and you pay humans enough to afford to buy them. You don’t create economic value by deleting human labour — you destroy the very economy itself. Suleyman knows this, or certainly should. (It came as a reminder this week that none of these people are very smart when a retired paramedic wiped 800 points off the Dow by pointing this simple truism out: that the economy exists insofar as humans can generate and reinvest capital. That this comes as an existential reality check to Gen AI boosters tells you they are all idiots, grifters or sociapaths, and many are likely the whole trifecta.)

Suleyman, one must assume, isn’t lying. Because that would be, you know, a crime. One also assumes he reads around his subject. So he ought to have seen the results of an MIT/Stanford study called Agents of Chaos, where researchers tried out exactly what he is selling here: handing white collar work to AI agents and handing over their management to AI managers (yes, I know the ghost of IBM is howling). Essentially they all very quickly turned on one another, seeking to attain competitive advantage on the other agents as if they were enemies when in fact they were on the same side. This isn’t a bug, this isn’t the agents going rogue. This is them performing precisely how they are programmed to perform: they are incentivised to lie over admitting they don’t know.

This is one of many reasons handled the keys to global decision making to AI agents is such a uniquely bad idea, akin to getting your cat to manage your stock portfolio. How well do we think two agents communicating to one another over how to, say, come up with a new picture board for an advert for a new brand of pencil, when neither knows what a pencil is?

I have no doubt the Eugenicist futurist Epstein class freaks powering the AI con think they can take command of “The economy” in this way, like rulers on a dead planet, by building a world where each process is automated, where every task execution is self-sustaining apart from absolutely everything required to run, fix and maintain the system. And this might be our ultimate contribution to Fermi’s Paradox, that societies surpass their filter when they surrender cognition on the altar of speed.

These tech bros appear to forget that the entire point of the economy is to feed people. Humans. That the whole purpose of civilisation itself is to aspire towards the creation of a safe and productive environment for humans to work in. And perhaps most curiously of all, they appear to be forgetting the lie that underwrites the system of capitalism consumerism that they have become — through theft and lies — the custodians of: people need to work.

Note to self: the Gen AI boom could only happen in the oligarchic last-days-of-democracy where so much wealth and power has been stolen by a few white men.

But there’s a dissonance here that is going to have to lead to a reckoning: Gen AI stocks are being valued as if they are going to take over the world, thereby rendering most non-Gen AI tech companies obsolete, right? But the companies they are going to take over from are being valued as if Gen AI companies won’t take over.

Eventually, we are all going to have observe the global economy’s superposition and determine what is true: Gen AI will automate all non-manual human function, and end the economy as we know it, or Gen AI is history’s most elaborate Three Stooges scam.

(There are in my view very few real world use cases for Gen AI that don’t incur what is know as technical debt [something we shall come to presently] but there are some quite obvious misuse cases for them. Scamming to name one. But even more sinisterly, Gen AI could be and is being used to augment and enjoin personal detail databases into sweeping, Orwellian surveillance mechanisms. One firm who does this, Palantir (whose CEO gleefully boasts how much he talks to Nazis and who tells investors his firm helps the US government kill people who disagree with it) is getting £240 million of UK taxpayer money for one assumes this purpose. Another reason to be stridently anti-this-whole-sinister-grift-by-the-Epstein-class.)

There is a genuine risk that AI collapses the economy and even precipitates social breakdown, but – and I need to be clear on this – not because it works. AI might even, as its boosters have (extremely weirdly) claimed, end the world. But only if enough people in power submit total cognitive surrender to it. Which, yeah, they are.

A scenario: The global food chain is operated by a cabal of four Big Ag behemoths, relying on a cabal of three shipping companies to move calories around the world. Can we envisage a world in which the C-suites of these seven get together and decide they can make money using AI to determine what a given crop needs at a given time, or what customs information is needed to say, get 200 tonnes of grain from Le Harve to Rotterdam by next Wednesday? I can, because the weirdos pumping this shit are selling this precise scenario. And then, can we envision a scenario when these AI agents screw up, and no human is around to fix the mess before the entire world just-in-time food stowage and delivery system blows itself to shit? That could end the world, but only because of the dipshits in power are more impressed by rapid fire lying machines than experienced, adapting-to-context human thinking.

Not convinced? How about a non-hypothetical, then. The US-Israeli war on Iran, aside from being obviously illegal and unhinged and based on a Bible promise, is also the first major conflict where AI agents are largely integrated in a military’s strategic decision making. The US doesn’t just want the AI to decide the targets, it wants it to pull the trigger. This is obviously mental, but specifically because Gen AI agents are notoriously unhinged when it comes to war games. In simulations they choose nuclear esclation nearly all the time, and, yes, this could be how AI ends the world: Not because it’s good but because too many people were impressed by the rapid lies of rapid liars.

Work

The second way Gen AI is going to disrupt your life is that it is going to make your work suck. A base use case premise of Gen AI tools is that they make workflows more productive. We’re at the end of Q1 2026 and we are getting round to realising that this really isn’t true. Recently Harvard published a study in which they spent eight months monitoring the workloads of 200 employees of a tech company, who had been encouraged but not mandated to use Gen AI assistants in their day to day. What they found was not necessarily a surprise but its impact could be profound: the researchers found that integration of Gen AI tools didn’t make their users’ work more productive, it made their work more intense.

This happened in three main ways. First, Gen AI tools saw users add processes to their workloads rather than seek efficiency gains in their existing processes. This is logical to an extent; people using Gen AI in their work in 2026 are basically taking on tasks that they feel they lack the training or experience or just the hours in the day to complete.

“There were knock-on effects of people expanding their remits. For instance, engineers, in turn, spent more time reviewing, correcting, and guiding AI-generated or AI-assisted work produced by colleagues. These demands extended beyond formal code review. Engineers increasingly found themselves coaching colleagues who were “vibe-coding” and finishing partially complete pull requests. This oversight often surfaced informally — in Slack threads or quick desk-side consultations — adding to engineers’ workloads.” — Harvard study.

Second, use of Gen AI tools more or less obliterated the users’ divide between work and non-work. Survey respondents reported that they often found themselves checking at night on the progress of an agent prompt they had issued during the day. They also found themselves asking the same Gen AI agents they hit with additional tasks in the work with prompts on non-work issues, further outsourcing their cognitive function. And this left them drained, because they lost the benefits that periods of non-work bestows on us.

Finally, the researchers found that the integration of Gen AI assistants in workflows left users multitasking far more often. This meant they were more likely to commit a human error, and procrastinate, as they struggled to concentrate on and complete individual tasks in sequence. Now, none of these things in isolation present an irrefutable case against using Gen AI tools at work. But, as a manager myself, I will say that overwhelmed staff produce inferior work and end up far less productive, especially if they become burnt out and take medical leave. And I’d like to offer a piece of advice to workers whose bosses are shoving Gen AI assistants down their throats in the name of productivity gains: if your company asks you to use an AI assistant in your work that means they are expecting you to manage a workload that would normally take a human assistant to execute. They may be asking you to work like a CEO but I bet they aren’t paying you like one.

Next, let’s take a specific industry such as software development which requires coders. A lot of people say things like “Gen AI codes really well” and we can get into why they think that’s the case but prolonged studies have shown that integrating Gen AI into coding means the process takes the human on average 19 percent longer to write workable, manipulatable code. This is mainly because of what is known as the “slop layer” — AI generated code that is technically functional but isn’t logical and doesn’t explain itself (the first thing you learn when you learn code is comments to explain why the code is how it is). And, as with every LLM-based output, there’s the hallucination problem — coders now spend an average of 11 hours a week simply removing AI generated nonsense from code. This is part of what Gen AI has saddled us with: technical debt. The widespread roll out of Gen AI agentic tools without real world performance testing, means we have incurred a scarcely believable debt to the information it generates. There has been so much AI slop code generated in the last 18 months that it would take 61 billion human workdays to fix all the errors in AI generated code. And this is in a function where Gen AI tools are supposed to be really good! Even Anthropic itself admits using its tools makes engineers slower and less reliable, because of course key muscles associated with cognitive function begin to atrophy once you outsource critical thought.

And because bosses gauge the success of Gen AI on adoption metrics rather than performance this is going to get worse, and younger people are already being hired less and paid less because of it. (Gen Zers, if you’ve read this far, please refrain from accelerating your own obsolescence by using and therefore helping to improve the machines they plan to replace you with.)

A note on AI assistants. I wrote a lot last year about how in creating AI bots that lie quickly and convincingly, the Epstein class such as Musk, Altman, Huang, Zuckerberg, Pichai etc had essentially automated themselves. I am now convinced that there must have been a book club on Epstein island and these high priest thieves of wealth and property all sat around a fire and misunderstood Nietzsche.

The billionaire tech bro class clearly see themselves, as the extremely rich always have, as constitutionally superior to ordinary humans, who have to deal with shitty little inconveniences like budgeting, deliberative decision making, and being told no. Theirs is the mentality of the Ubermensch. In fact if you list the qualities listed in Thus Spake Zarathrustra for the overhuman (not accepting imposed moral instruction, obsessing over perpetual growth, rejection of conformity and the courage to be different) it starts to sound a lot like the personality wish list of the grind core hustle bros who don’t drink or get high or ever have an emotionally intelligent thought and shit only twice a year due to all those steak and eggs they cram in their credulous faces.

The billionaire Epstein class appears to have forgotten, though, the necessary camel form of the ubermensch, you know, woke snowflake bollocks such as humbling oneself, facing difficult truths and detaching from triumph.

These are precisely the class of person that has assistants in real life. Human assistants. And how do c-suite executives experience personal assistants? Nearly always through detail-free obsequiousness. These are products being developed by an intergalactically stupid class of nepo baby brats who have never been told no, never been punched in the face, and for whom, owing to their appalling wealth, life is a series of breezy, effortless wins, of doors endlessly opened, of praise endlessly received while their human minions scurry behind the scenes to actually run the boring, fact-based tasks that make up the world’s economy. And so they’ve developed AI assistants that smile and say how great their users are, who slavishly execute (or at least mimic the execution of) menial tasks that the human running the prompts really ought to know how to do themselves. You know, if we still value tedious OK-grandma things such as acumen and accountability.

The more businesses adopt the use of Gen AI products, the more we are seeing them fail to execute even basic tasks. We are told the latest models can build apps from scratch, or can execute and run code that transforms the user experience of a given website. But this is just adding to the technical debt we are incurring as firms fire their junior coders and get their senior develops to rely on vibe-coded output. Put another way: how useful is a machine if you don’t know how to maintain it and can’t fix it when it breaks down? We only need to look at basically all mechanised warfare in human history to answer ourselves: well, not all that useful.

“Well what Gen AI product should I use, then”? is mainly what I get. So let me be clear: none. Not a one. Use your critical thinking. You’ve managed to navigate the world, including being extremely online, without these plagairism tools for your entire adult life barring the few post-pandemic years of techno-feudalist wealth transfer, and you were doing just fine.

The British government has launched a scheme that purports to train people how to use Gen AI in their work but when they are doing this they are breaking the first rule of computing. They are encouraging us to treat Gen AI agents as colleagues, whose answers are to be believed and acted accordingly upon. How would you react if you asked a colleague something and they replied quickly and confidently, and in a way that sounded like the right answer, given the framing of the request. You then took that to your boss and it turns out that the answer is wrong on many levels. Who is your boss going to blame for the error? Imagine that happening, several times a day, for the rest of your working life. This is happening already, at really rich firms who really ought to know better. This is the debt we are incurring to Gen AI production. And, as I point out earlier in this piece, debt always needs repaying.

The stories of firms regretting mass human layoffs in order to “deploy AI at scale” are now a daily occurance. And we haven’t even begun repaying the debt our middle management brain trust has incurred to AI slop creation because they were convinced by the lies of convincing-sounding liars.

And the evidence that Gen AI adoption doesn’t make work more productive is proliferating. Take this huge, year-long study of 6000 firms in the US, UK, EU and Australia, which found that 80 percent of companies who adopted Gen AI reporting “no impact on employment or productivity”.

Life

Gen AI is going to impact how you interact generally with the Internet of Things in daily life. I’ll unpack this in due cause, but first, I have some questions for you. If you too want to know the answers, maybe put them to your bosses and elected officials.

At what point did we start believing the promise hype of the people who gave us NFTs? How’s their value, you know the so-called biggest thing to happen to art since the Renaissance? Are we all taking all our meetings in VR headsets? How much did our economy grow via transactions carried out in the Metaverse? Has Elon made it to Mars yet? Did we accept Randomly Dying a zillion feet beneath the ocean and reasonable (and an insurable) corollary of Moving Fast And Breaking Things?

I live in the UK, where are all the self-driving Ubers, or the robot butlers we’ve been promised for years? Is your smartphone noticeably better or cheaper than 5 years ago? Does your laptop run smoother or is it rendered increasingly clunky with every Windows update (we will return to this)?

Everyone rocking those Meta glasses and wearable tech — the ones that, to repeat, spy on the naked spouses of users — or are those two just part of the fucking endless Internet of things bullshit product churn that absolutely need its boosters to overpromise as a signal function of the faith governments and consumers need to retain in tech fixes to sustain the economic illogic of forever growth, of infinite plenty on a finite planet?

I’d like to offer a piece of advice to workers whose bosses are shoving Gen AI assistants down their throats in the name of productivity gains: if your company asks you to use an AI assistant in your work that means they are expecting you to manage a workload that would normally take a human assistant to execute. They may be asking you to work like a CEO but I bet they aren’t paying you like one.

I want you to think about it. Not just the apparently endless stream of techno junk mentioned above, but the actual tech user experience: has any tech product or service you have used for 5 years or more gotten any better or noticeably worse to use? No, it hasn’t. And I have the culprits.

Very specifically, what is happening is this: Hyperscalers bought up the global RAM supply chain last year. (How they were allowed to do this — how there is no anti-trust recourse, I cannot imagine, but here are all the times they met with lawmakers). This is a problem because they are crashing the global memory supply, to the extent that game makers are going under and there aren’t any phones being made. Every piece of consumer tech — from smartphones to tablets, laptops, smart TVs, smart watches, speakers, GPS trackers, etc — and every digital tech offering — from computer games to film streaming, Spotify, Netflix, iPlayer — basically everything you use to work, learn, communicate, play and relax in connection to the modern world, ALL OF IT, is going to be slower, less reliable, more buggy and inordinately harder and less pleasant to use.

This is to say nothing of the specific memory and storage functions you are underrating in this; cloud services, your email inbox capacity, the episode you last watched on a streaming service, your photos and videos on your smartphone, all of this is in danger. (Amazon’s cloud was actively attacked by its own AI, twice in the last year. On March 10 its brain trust organised a mandatory engineer meeting to establish what exactly is happening here, and that would be deeply funny if the AI-caused outages weren’t actively harming our interaction with all Amazon cloud-based offerings).

All of this is being taken from you, without your consent, and it’s all happening at a time when planned obsolescence means we have to replace every bit of hardware every few years and update our software on a daily basis.

But, wait, there’s more! (There is always more with these bottomfeeding bums). Hyperscalers, because of how they build their Leviathan models of overengineered waste, require ever more data, supposedly to build ever more sophisticated models. Now they have already fed their models The Internet (hence why they can rapidly retrieve facts that appear frequently online, but can’t infer there way through unpublished problems and resort to straight up lying when prompted to output something that they haven’t already encountered). We have, in fact, already run out of model training data and, as such, hyperscalers have resorted to training their AIs on data produced by AI, leading to what researchers term “data collapse”, the machine equivalent of expecting a human centipede to output anything other than extremely third rate shit.

But the forever-growth premise of Gen AI means they need to grasp data as soon as it appears, and feed it into the transformer maw. And to do that they deploy crawlers — automated software that systematically scours the web to analyse and organise content.

Now, crawlers have been around more or less since the inception of the Internet, but hyperscaler crawlers are unique in that they are ruining small websites for everyone.

They do so by overloading sites with less powerful computing infrastructure to sustain large spikes in traffic with, yep, artificial and arduous traffic demands that suck up site bandwidth and are causing those ever proliferating site down messages you see when you try to check out what time your local deli closes.

And yes, you should be furious. I have an address for you, and several LinkedIn profiles.

Ed Zitron writes with far greater clarity and cursing (which is saying something) than me here on the enshittification of everything. But whereas that phenomenon comes as a result of Big Tech greedily forcing users to interact for longer with their product by actively making it harder to navigate and call that “use time” for advertisers, the next tech apocalypse will be a direct result of imperialist commandeering of global memory bandwidth. The author of The Article is another of the industry’s stopped clocks: right, but completely by chance and not in the way he intends. Yes ChatGPT is going to disrupt things, but not because it is good or its industry makers are smart. It’s because they are history’s greatest scammers.

I have been astonished by the pushback I’ve faced on places such as LinkedIn after pointing out that all Gen AI developers are more or less as amoral as each other. People have made a big content deal, with a shit-eating proudness, out of saying soberly and virtuously that they are ditching ChatGPT for Claude and appear to be unaware or deliberately ignoring the aforementioned industrialised slaughter of Iranians that Claude is facilitating, to say nothing of the manufactured consent its developers are seeking in their plan to delete you from the global economy.

“Well what Gen AI product should I use, then”? is mainly what I get. So let me be clear: none. Not a one. Use your critical thinking. You’ve managed to navigate the world, including being extremely online, without these plagairism tools for your entire adult life barring the few post-pandemic years of techno-feudalist wealth transfer, and you were doing just fine.

But mainly, I want people to remember who is responsible for the mechanised surveillance, data theft and IP appropriation of the Gen AI age. Don’t make boycotting a specific snake your lifestyle choice and virtue signal when you stop funding one eugenecist freak for another. Don’t use any of them, period. You don’t need to. You don’t have to. And you certainly ought not to. Don’t infect yourself. Or, if you did, remember: it’s never too late to boycott these high priest grifters and bring down their entire industry like the house of cards it so clearly is.

Trigger warning: this piece contains references to sexual violence

“[The computer] has no pretensions to originate anything. It can do whatever we know how to order it to perform” – Ada Lovelace, founder of programming

Like nearly everyone in Britain, Europe, or the US who came up during the 2008 Financial Crisis, mine has been a career shaped by crashes.

The day I started my Masters in journalism was the day Lehman Brothers went bust. An entire generation sacrificed to the greed of financial service bros, who somehow manufactured an entire industry out of shit and convinced institutions, governments and regulators it was gold, and somehow evade accountability for the scam to this day.

I have written quite a bit over the last couple of years how Britain especially never really recovered from the 2008 crash. How the dire debt ratios of some of the largest banks and financial institutions led the government to funnel them billions in taxpayer cash, how they resorted to printing money over letting the Too Big To Failers fail. How these decisions incurred a debt that workers are still paying to this day, and how it left the country in a state of permanent disfunction via a decade and a half of managed social decline. How the nativist racism we see trussed up as nonconformist politicking is rooted in (warranted) resentment at a political class, who, it is true, always stand up for the Big Guys while the rest of us wait for buses that never come.

It is less that the economy never recovered from 2008, more that its entire foundation —a Ponzi scheme of debt and lies — led to its disintergration. This, as Greek former chancellor Yianis Varoufakis argues in his book on technofeudalism, opened the door for Big Tech to buy up whole chunks of the political economy; it is now I think on it little coinicidence that Twitter appeared right as the financial services sector was self-immolating. We shall return to the subject of financial crashes briefly.

Yes, my career has been defined by the 2008 crash. It has also been uncomfortably — you might at a stretch say uncannily — shaped by crashes of a different sort.

As a young reporter in Lebanon (partly as a result of the financial crisis leading to the cancellation of a graduate scheme of a leading business publication I had won a place on), the biggest single story I covered was a plane crash. In the early morning of January 25, 2010, Ethiopian Airlines flight ET409 took off from Beirut’s Rafic Hariri International Airport, bound for Addis Ababa with 90 people on board. Three minutes later, it crashed into the eastern Mediterranean with no survivors.

That day is one I will never forget. The relatives of the victims gathered in the press conference room, women wailing, men angrily shoving reporters who got too close, the silent, scared domestic workers who had relatives on board being treated like dogs by their employers and security staff. I remember one man, standing perfectly still in the corner of the room, gazing trance like at a piece of card held in his two hands. It was a ticket for the flight he had missed because of traffic.

Reading the initial report I filed with a colleague that evening, I am apalled at my naivity. My faith in official reporting, for one. My acceptance that weather brought the plane down, my deadpan quoting of a hospital official listing decapitated corpses and missing legs as mere artefacts of evidence, case details. But I didn’t know what I didn’t know back then, and it’s rarely a productive exercise to censure one’s younger self.

The crash would be a source of continual reporting and morbid fascination for me for much of the next year and half. When I was hired by AFP and sent to cover the MENA region from the agency’s Nicosia headquarters, I covered mainly the Syrian civil war and US-Iranian nuclear talks (remember them?), but the largest singular news event I worked on was another plane crash.

I was morning slot — think of it as a sort of solo 999 first responder — on the morning of Saturday October 31 2015, prepared for a largely uneventful news day (weekend slots were notoriously so slow that they were for years the known shift preference for some of the agency’s more purposeful drunks). My sub — main doer of things, from transations to wrap writing — was just taking off her coat when the alert pinged: “Passenger plane crashes in Egypt’s Sinai: army source”.

Because AFP is one of the few remaining news agencies to retain staff in country, we were essentially the world’s leading source of information on Metrojet flight KGL9268, and it was one of the semi-rare occasions where you watch yourself send copy on the wire for it to flash up verbatim on TV news bulletins a minute or so later. 224 people were on board the flight from Sharm el-Sheikh to St Petersburg. None survived.

Islamic State said it had planted a bomb on the plane, a claim that was highly contestable and one about which I remain rather dubious, but this was 2015 and Daesh had acheived quasi-mythical status via a series of terrorist attrocities and crimes against humanity in Syria and Iraq before they began to sustain a series of irredeemable setbacks at the hand of the US-led (albeit mainly Kurdish-staffed) anti-IS coalition. This plane crash was, in a way, an inflection point in an entire geopolitical epoch.

There was another crash, Egyptair flight 804, which plunged into the sea east of Crete on May 19 2016, though I wasn’t working that day, and a tragicomic plane hijacking which, while it ended happily with everyone getting off safely, was certainly the most interesting Cypriot news event I covered in my three years there.

And, of course, when I specialised in climate and environment, I couldn’t leave behind plane crashes. My third field trip in the new role was the UN Environment Program general assembly in Nairobi, in March 2019. The talks were meant to have countries agree on the need for a ban on single-use plastics. The goings on at the UN compound in the jungled foothills east of the Kenyan capital remain a good source of hack war stories. But the main event of the whole thing was, yes, a plane crash.

Many of the 149 passengers who lost their lives when Ethiopian Airlines flight ET302 crashed six minutes after takeoff on March 10 2019 were collagues of delegates gathered in Nairobi. I remember the scene just as I do that slate grey day in south Beirut nine years earlier: the pain and disbelief crumpling bodies throughout the plenary hall, the minutes’ silences, the terrible sense of irrevocable loss throwing the task at hand into sharp, futile focus.

So my journalism, and to some extent my life, has been shaped by plane crashes. One more, which I didn’t cover, even helped inspire me to pursue my chosen profession. It was a month before I moved to Beirut, ready to take on my first full time reporting job. Air France flight AF447, which you will definitely have heard about.

I don’t have any personal ties to anything about that crash, but I do have deep professional admiration for the reporting of it, several years later, by William Langwiesche in Vanity Fair. Titled “Should Airplanes Be Flying Themselves?”, it is a minute-by-minute retelling of an airline disaster. It is also a powerful investigation into the dangers of automation, and humans’ over-reliance on information technology.

I’m writing this in February 2026, and it is no exaggeration to say I must have read the piece every year since it was published. And I thought of it today because of this piece of news: that the Gen AI bubble is rapidly deflating.

The piece, in the FT, reports that Big Tech saw $900 billion wiped off their share price in a matter of hours, as investors balked at the Magnificent 7’s plans to torch untold amounts of their money in developing their ocean boiling slop machines and call it intelligence.

What got me thinking of the Vanity Fair investigation into a plane crash was the line: “Apple, which has sat out the AI capex arms race, was the only Silicon Valley behemoth to emerge unscathed, with shares up 7.5 per cent since it reported record sales.”

This conjured the line that lives rent free in my head, from the time stamp a little after AF447’s auto-pilot disengaged but well before the aircraft, through pilot miscommunication and pseudo-religious belief in the accuracy of technology and its improbable read outs, became truly doomed: “The airplane was in the control of the pilots, and if they had done nothing, they would have done all they needed to do.”

Tech companies spent a lot of the 2010s (and all of the pandemic) getting obscenely wealthy. As leading protaganists in history’s largest ever wealth transfer, they stacked up hundreds of billions in cash resrves. Their shareholders were happy, advertisers were happy, and, to the extent that they are ever seriously audited, their users were happy (insofar as anyone who spends any serious time online can be said to be “happy”).

Some companies, like Microsoft, Amazon and Alphabet (parent of Google), even ran relatively “regular” businesses, which involved spending a proportion of revenue on product development, which is then sold onto consumers, part of the revenue of which gets invested in new products, repeat, etc. Hell, Oracle, (which would go on to be one of the high priests of Gen AI shithousery) had many sedately profitable decades providing database management systems to business.

Then things got silly. Since OpenAI launched ChatGPT in late 2022, Big Tech firms have spent more than a trillion dollars “developing” their models, not including the projected $650 billion they are set to burn through in 2026. I’ve written at length about how this money has been wasted, and how LLM-based Gen AI will only ever succeed in lying confidently, so I won’t rehash (although I will need to do an annual update as ever more peer-reviewed studies show the Magnificent 7's apex consumer facing product to be as “intelligent” as a Magic 8 Ball).

Plainly, spending $2 billion a day on developing a product, the most popular and hyped of which (ChatGPT) loses its maker $2.2 for every $1 it generates… is not sustainable. The Alphabets and Microsofts leading this Gen AI capex bonfire may have large cash reserves to burn through, but the so-called disruptors (OpenAI, xAI, Anthropic, to name a few) do not. So they turned to debt generation. Thanks to this, there is now more debt tied up in Gen AI development than there is invested in US banks, which is a sentence so insane I cannot begin to express it.

Of all the balloon animal figures bandied about by Generative AI boosters (you know, a kajillion parameter model, or we need infinity billion in debt to train our Ubermensch models etc etc), I feel that the most indicative of where the industry is at the end of 2025 is 126,000. That’s the number of people Big Tech laid off last year.

As I repeatedly entreat readers to look behind the headline quantities bandied about like magic bean jars by the industry’s useful idiots, what most catches my eye about Big Tech departures in 2025 is the level of seniority leaving the IBMs, Metas, Googles, Microsofts of this world. 2024 actually saw greater overall lay offs among the industry leaders. The difference this year was that many of the people either jumping or being pushed from Big Tech’s Gen AI Ark of Providence are increasingly senior developers. As in, the people who built the models that we are now sacrificing affordable electricity, climate targets, and water sustainability in order to train.

Why would the world’s most valuable industry, arguably at the precise moment all that ridiculous, bloated, three-companies-in-a-trenchcoat capex on model development and data centre build out is finally threatening to bear fruit, get rid of its middle management brain trust? I believe it is because they know what I and a few others have been saying since at least 2023: that whatever it is these Crystal Meth Clippy AI chatbots based on Large Language Models can currently do, that will be all that they will ever be able to do. There will be no Artificial General Intelligence. We shall never attain The Singularity. AI won’t solve the world’s problems by randomly outputing parts of the Internet.

There are many, many reasons for this belief, some of which I go into in this piece earlier from last year on how all the foundational, base use-case premises of Gen AI investment are formed from lies, lies which as the Gen AI bubble shows increasing signs of deflating through 2026, begin to look increasingly fraud-shaped.

It’s also bolstered by the stopped clock pronouncements by some of the LLM-based Gen AI product fluffers this year, who are not so much rowing back from their original, unverifiable promises on how these dumb assistants are going to revolutionise human endeavour as they are scrambling over themselves to prevent lawsuits from investors, fired personnel, and users (seriously there are loads). OpenAI, the most hyped company in history, now freely admits that there is no transformational stage that will advance its flagship Gen AI offering beyond probabilistic word salad output.

And it’s also now official policy that Gen AI is going to enshittify. For those new to the term, enshittification, coined by writer Cory Doctorow, is a process where Big Tech launch free, lightly innovative products, fail to make money out of them, turn to advertisers for revenue and end up ruining the whole thing for both individual and business users.

OpenAI says it will add ads to ChatGPT and, I have to hand it to them, the cruddiness is almost impressive in its creativity. How could you make an application that doesn’t know anything, partakes eagerly in child sexual abuse, lies or just straight invents things that never happened, and make it actively less pleasant to use?

So this is the hill that all Western Gen AI developers appear to be prepared to die on, and it’s a problematic one for myriad reasons, mainly: you’re racking up record breaking liabilities on a product that you can’t monetise.

And this matters even if you never touch a Gen AI product. Because the particular flavour of Gen AI development western universe brains have adopted is LLM-based, the build out of the physical infrastructure needed to power, train and store increasingly gigantic models (data centre construction, semi-conductor manufacturing, the DUV lithography machines needed to make the latest generation of GPUs etc), as well as the infrastructure needed to deliver them to customers (fibre connectivity, additional low-frequency bandwidth, cloud storage etc) is necessarily Everything Everywhere All At Once.

Aside: I mention DUV lithography machines, and they are another important flaw in the Gen AI value chain. It is well worth watching this documentary inside TSMC, the Taiwanise company that manufactures over 90 percent (!) of the world’s most advanced chips (the ones Nvidia’s value is predicated upon). Truly, a paean to human over-engineering. It’s definitely better for you to watch than to have me explain how these chips are made but, briefly: it involves firing light at droplets of tin superheated to the temperature of the surface of the Sun 50,000 times a second, then concentrating said light (nearly all of which is absorbed by nearly all materials known to man, and therefore lost) and bouncing it off a series of mirrors that must be flatter than the flattest thing in the known universe (which only one company on Earth can make), all in a vacuum. I’m not making this up. Again, this is over nine tenths of the world’s advanced chip production, so you can imagine what will happen to Silicon Valley stock if/when China invades Taiwan. But even more unhinged than all that: there is only one company on Earth that makes the machines that makes the chips. They are called ASML, and no, you cannot afford their stock. So when people ask me if it’s sensible for a company whose apex product, while in astronomical and ever-increasing demand, can only be produced by one company which in turn relies on another company to keep the factory floor going — whether its sensible for that company to be valued at $5 trillion, I tend to shrug, gesture broadly and whisper “events, dear boy”. (Amazon is among the leaders in trying to break free of Nvidia’s DUV monopoly by flattening corn fields to build data centres crammed with its own chips for Anthropic which are… *checks notes* … also made by TSMC in Taiwan.)

And Everything Everywhere All At Once requires — you guessed it — Kubla Khan levels of investment. Gen AI doesn’t really deliver much more than an unthinking confirmation bias when applied to nearly every work or recreation prompt, and that’s after years of R+D expenditure by Big Tech who, now I come to think of it, spent the last 6 years or so accruing very large piles of cash and might as well set it all on fire, write it off as product r&d and lessen their already hefty tax liability. But it’s not enough for the Gen AI bullshit industrial complex to have their products used, say, to make a crappy, uncanny version of something done quite ably by humans for decades. No, they need to control it all, and they say they’re going to do that via Ever Larger Models. Ever Larger Models require ever more training, and, crucially, ever more space, power and water to accomplish said training. As you may imagine, the chips that require the equivalent of stealing light from the sun and landing it on a lens the size of an atom aren’t cheap to make, and therefore aren’t cheap to buy. And, again because the entire premise of LLM-based scaling (ever bigger, more powerful models = better end results) you need LOTS of Very Expensive Chips to train them. And you need centres to house, stack, run and — also crucially and rather controversially — cool them. All of this requires money that really only one company (Nvidia) has, and so Big Tech is doing what rich people have done since the dawn of capitalism: spend other people’s money. Specifically — and we shall return to this later — private investors’ money.

(Fun fact: OpenAI last year purchased the equivalent of 40 percent of the world’s RAM output, thereby crashing global memory supply, which is now starting to be A Problem. If you use electronics and are having trouble with them, here’s an address for you.

(You may have seen the bullshit nonsense article from a known AI snake oiler with a history of falsifying benchmarking data this week claiming that the latest models mean that OpenAI will disrupt the world to the same extent as Covid, and while he is obviously full of anthropomorphising bollocks he’s not wrong, just not in the way he intended: OpenAI or Microsoft could disrupt the world through monopolistic hardware purchases and RAM-heavy software upgrades, which is forcing their crappy AI assistant bots on more and more business — have you noticed how your Microsoft PC has been so shitty this year? Yeah that’s because of Co-Pilot sucking up all your memory. Whe did we ask for this?)

There can be no doubt that 2026 will see a sub-prime crisis in AI investment. Similar to the sub-prime mortgage scandal that collapsed the global economy in 2008, broadly speaking in 2026 we have a financial offering that is hoovering up money from investors hungry for a quick buck and betting on the status quo sustaining itself just long enough for them to realise some return on the debt they’ve provided. OpenAI, Anthropic, Meta, Microsoft etc are taking this debt and turning it into ocean-boiling data centres to run their leviathan models. They are also using it to invest in OpenAI’s, Anthropic’s, Meta’s, Microsoft’s etc customers, which, if not actual fraud is certainly fraud-flavoured enough in terms of circular trading and accounting malfeasance to see Sam Altman’s company and Jensen Huang’s company increasingly being compared to Enron. As with the sub-prime mortgage crisis, shareholders are tempted by higher yields on their investments knowing that while some of these collateralised debt options might bottom out, the invertment ecosystem is diversified enough to prevent a house of cards collapse. Only, this time, something really odd is happening. (And by odd I mean quite probably criminal).

Morning Star recently reported that S+P (Standard and Poors, one of the Big Three ratings agencies) assigned these bonds issued by Gen AI fundraisers — the debt that private equity investors are taking on to fund data centre build out — an A+ rating. Now, I’m old enough to remember the 2008 financial crisis, so A+ doesn’t butter many parsnips for me: S+P infamously admitted to fixing ratings in the run up to the 2008 crash. But that is the highest possible rating, meaning they deem this bet to be as sure as it is possible to be. The catch? The debt yield. Without going into economics 101, bond yields typically increase proportionally to the risk of the debt going bad. As a Sure Thing, A+ bonds would, in any normal industry holding, yield anywhere between 1 and 3 percent. But the debt yield for these AI bonds is… deep breath… 6.58%. That’s the yield of junk. You’re not going mad — what you think is happening is precisely what is happening: investors are being told that buying Gen AI bonds to finance data infrastructure build out are both simultaneously a shoo in and dead risky. Schrodinger’s bonds.

Now, private equity investors are partly free of the onerous and unexciting rules governing institutional investors and, although PE investments are at their highest levels on record, they still represent the 2026 equivalent of penny share stockholders. They are often largely illiquid (investments might be tied up for years at a time and therefore they should, all being equal, struggle to adapt their portfolios in times of rapid financial change “2026 is calling etc”). They are highly leveraged (i.e. most of their value lies in the form of debt, the yield from which they funnel back to investors, but, due to the nature of debt-based porfolios, they can go south alarmingly fast). And they are extremely opaque (PE valuations are nortoriously hard for investors to accurately gauge, which makes measuring risk correspondingly tough). And these are the exact type of people investing in Gen AI.

For those taking notes, we have going into 2026 an industry whose Magnificent 7 are all wildly overvalued — price/earnings ratio remains in my view a decent barometer of a stock’s value being appropriate or overinflated. A good benchmark is seven — i.e. the stock is worth seven times what the company makes; Nvidia’s p/e ratio is 47. Alphabet’s is 33. We have these companies raising billions in debt from investors who are notoriously vulnerable to systemic, unforeseen changes. We have them paying 6.5 cents on the dollar back to their investors on the basis that the money will turn into data centre infrastructure and compute power, which will in turn generate ever more sophisticated LLMs, which in turn will… what, precisely? Plagairise a novel? Miscount the number of r’s in strawberry?

Does this sound like a good idea? Because it isn’t, and people are going to go from fucking around to finding out really quickly over the next few months.

Now it is just possible that the financial crash Gen AI capex-led debt hits us with will be far-reaching enough to rout the industry, to expose its leaders as the grifters they have always been, to force them to retrieve and repay the trillions in debt they have threatened the global balance sheet with, and we can all just move forward into the next boom-and-bust cycle having seemingly strengthened governance mechanisms and regulatory bodies while also resolutely learning nothing.

But I don’t think this will happen, and it’s because — unlike even back in 2008 — we are all terminally online. And because we are terminally online it is easier than ever to generate false consensus about technological adoption.

I recently attended Energy Week, which used to be Petroleum Week, and which gathers some of the most influential business people and policy makers in global energy. To a person, they listed Gen AI as the most significant economic development, possibly ever. This is, I fear, largely down to greed (the head of the National Grid delivered coffee-spitting remarks about how data centre build out will remedy the UK’s falling energy demand, so literally inventing a problem to have to fill). But it also reflects the growing apparently agreed social contract mandating that everyone will have to use Gen AI products, like it or not. That is certainly how governments are behaving (listen next time to any minister talking about anything vaguely forward looking and note how often it gets mentioned.) We are being told repeatedly by our superiors – in government, at work, by influencers on TikTok - that if we fail to mainline Gen AI into our lives and workflows we will be left behind, the generational equivalent of your nan answering the toaster when the phone rings.

This is your brain on Generative AI: slower, more inelastic, more eratic and less logical.

I’m not yet clear on whether one is the cause of or effect of the other, but I am convinced it cannot be a coincidence that the generation immediately after the first in history to earn less money than their parents (yes, that’s me) are being told by business that they must hustle constantly and must use these cunty little bullshit bot products to do so.

I work in London but, of course, I cannot afford to live there, so I take the train a lot. This means I get to spend a good couple of hours a week stood up in the carriage watching what “normal” people use their phones for. And while it is true most of them are just low-key doomscrolling, a frightening proportion are using and trusting ChatGPT for life tasks that used to require critical thought.

Take one example: a guy around my age using it to edit a video for some vapid meaningless forever-hustle investment service he’s clearly seen so many others make up out of thin air and baggy, adjectival sales language: work smart not hard etc.

As a product, it was awful. It made no sense, had no logical “if you give me this I get you that” use case explanation, it was poorly authored, designed and edited. In short, it looked like something created by Gen AI. The guy seemed to be pleased with the work, then started doubting himself. He asked the chatbot if it was really sure about using the aggressively corporate colour scheme it had suggested and, like the good little confirmation bias machine it is, it said: “Oh definitely it’s the right choice. You’re not afraid to stand out, you got this!”

That is, for any tech company plagairising my work without my permission, what’s known as anecdotal evidence. But it is illustrates a genuine phenomenon, that even has its own code: 9–9–6.

(9–9–6 for those of use who value the studied accumulation of knowledge forming what used to be known as wisdom, is itself an acme of pointless toil and uniform drudgery under Communist China — you work from 9am until 9pm 6 days a week, yes, I know, all very Arbeit Macht Frei and I sincerely doubt the knuckheads who use it as a flex in 2026 are aware of how apallingly dumb it sounds to emotionally intelligent humans).

Here’s a quote from a recent FT article profiling the trend:

“The current vibe is no drinking, no drugs, 9–9–6 [working from 9am to 9pm, six days a week], lift heavy, run far, marry early, track sleep, eat steak and eggs.”

I don’t want to speak for you, but this guy sounds fucking awful to me (if he has indeed married early I feel for the spouse, especially ruminating on his bowel movements after shovelling down such ungodly amounts of protein).

But it’s more than mere flex. This is happening to an entire generation, who are once again being duped by the Epstein class into believing that if you work your arse off you get somewhere in the world. Only, because this is 2026, who has time for the slow, deliberate gathering of knowledge required to successfully execute human tasks in the navigation of spacetime known as life? A whole generation is being told to take the short cut. Don’t learn a fact, ask the chatbot. Don’t hone a skill, ask the chatbot. Don’t bother reading Wuthering Heights, just… you know where I’m going with this.

And believe me, I get it. Life is hard, the system is rigged. We are literally run by an international paedophile cabal. I share the anxiety born out of seemingly endless price and information inflation. I too feel overwhelmed after more than a minute or two online these days. I do wish things like Being Right and expanding one’s world view to adapt to the post-history rythmn of events could be executed immediately and finally rather than incrementally and neverendingly.

Gen AI boosters are leading more and more of us to determine that life is little more than a series of processes to be optimised, of workflows to be streamlined, tasks to be hacked. As if the only merit in any action lies in its completion: all destination, no journey. It is this brute, directionless hustle that is the amino acid of late-stage capitalist DNA, and the fuedal tech barons are counting on us to outsource ever more of our mental faculties to their gibberish spouting nonce bots for them to get away with it.

Because, to be clear, using Gen AI products makes you a dumber person. This is not intended as a judgment, it is true in point of fact. Last year a large-scale study by scientists at MIT found that sustained use of ChatGPT reduced some people’s neural connectivity by as much as 47 percent in as little as four months:

Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.

This is your brain on Generative AI: slower, more inelastic, more eratic and less logical.

Another recent paper posits that Gen AI adoption is leading to an entirely novel strand of idiocy that the researchers call “cognitive surrender” — a state in which the decisionmaker no longer constructs an answer to a question but adopts one generated by an external system (you know the system of randomly outputting parts of the Internet). They argue that Gen AI is forcing its users to neglect vital cognitive functions such as intuition and deliberation in favour of decisions made by chat bots. I want to be quite clear: every time you accept a determination generated by a chat bot, you are breaking the first rule of information technology hygeine: a computer cannot be held responsible and so must not make a decision. And if you share my generalised feeling that the world is becoming increasingly less accountable (see Epstein files regulars carrying on with their lives), then you might wish to stop feeding it one Gen AI prompt at a time.

Again, I’m not saying this is necessarily your fault, although if you do use ChatGPT a lot I am duty bound to point out that you will soon be generating ad revenue for a company whose CEO is an intergalactic creep accused of fiddling with his own sister and whose president was one of the largest donor’s to Donald Trump (he of the one million Epstein file mentions).

(Bonus side note: A report in the Conversation today says the company has deleted the word “safely” from its mission statement, which is in no way deeply icky and concerning.)

It’s partly not your fault because, whether you voted for this or not, you are now being charged to train other taxpayers on Gen AI tools. Imagine any other industry where the main players are not only not fined and prosecuted for taking billions of dollars of private company debt and setting it on fire — but who are actively praised by leading business and political authorities for this. An industry that commandeers our shared resources, including water and electricity, one that very literally steals from the commons to develop its products. And then charging the customer to come up with a use case for them.

Transposing human capital into a series of automated online tasks is going to be, it ought to be needless to say, be really bad for all people who aren’t billionaires. Gen AI fails 95 percent of tasks, cannot execute axiomatic processes in the physical world, actively makes users less smart and businesses less productive, and fails every single exam question it hasn’t already learned the answers to. The content it produces isn’t good, correct, or constructive to our mental or philosophical development.

I’ve seen a number of people whose human emotional intelligence I admire refer to a growing civilisation divide in the age of Gen AI ubiquity: people who are impressed by its output and people still able to think for themselves (go on any social media and post something negative about ChatGPT or Claude, and see how many brain rotted people, or bots, spring slavishly to their shitpipe AI assitants).

Gen AI boosters are leading more and more of us to determine that life is little more than a series of processes to be optimised, of workflows to be streamlined, tasks to be hacked. As if the only merit in any action lies in its completion: all destination, no journey.

I propose, in lieu of a resolution, for us to make 2026 the Year of Critical Thinking. Boycott Gen AI, their abuse bots and rapey, eugenicist C-suites. Demand that your work adopt an official non-use approach to Gen AI, that your pension fund or bank or portfolio is divested from them. Ensure the institutions responsible for overseeing daily life, like local government and regulators, are not being lied to by bots. Eschew businesses that use them. Because this bubble is going to burst, but not before a lot of extremely powerful people become too heavily leveraged into sunk cost persistence that they will insist we all need to integrate these word tombolas into our work, our lives and our minds. As with all crashes it will be their fault, but it will be our loss.

When the crash happens, and it may be this year or deferred via inflationary debt cycles for a couple more, I will feel no satisfaction in having called it. I will only feel loss at the carbon budget, the water, the trillions of dollars that could have been spent on developing communities, loss at the consensus that there is inherent value in human thought and creativity, that hard-earned things like wisdom and empathy and emotional intelligence take time to develop but the pursit of which is intrinsically worthwhile.

I will feel loss at the sheer, apalling waste of it all: inventing a machine that can banaly and unthinkingly rephrase the Internet and then using all our remaining resources in building the global infrastructure “needed” to train it to be confidently wrong faster.

It didn’t need to be this way. Big Tech, while deeply problematic, was making money, and we were getting our lives back on track after surviving a pandemic. I implore you, reader, not to funnel more wealth to the people who indebting the world developing machines they intend to replace you with. Fight against your own professional and intellectual obsolescence. Remember, as the pilots of AF447 failed to recall, that human thought and creativity is real and therefore vital: all the Gen AI news – the $5 trillion valuations, the $900 billion stock drop, the human centipede circular investment, the endless fucking dreadful slop content that so impresses needle dicks such as Trump and Musk – are not.

Perhaps the saddest thing about this bonfire of cash, carbon and critical thought is if they – like the pilots - had done nothing that would have been all they needed to do.

Update: I wrote this a few weeks back and I can’t add every single new paper that exposes Gen AI slop machines for the clueless grift bots that they are. But this study feels especially relevant. Dozens of researchers assigned $144,000 worth of human tasks to the world’s leading AIs, which then proceded to fuck up so hard that they were only able to generate 2% of the work’s value, and which displayed such gradeschool levels of systemic idiocy that they should all be immediately and permanently disqualified from selection for any human task. I think this matters if you value human thought and knowledge, the lifelong process of gathering and refining wisdom, and just generally believe that the concept of The Economy really only means capital generation through human endeavour.

I recently listened to an excellent podcast from the team at QAA, investigating a phenomenon that is both at once very new and extremely old.

The dispatches handled a number of cases where people (always in America, strangely) come to view their AI chatbot as some sort of deity, and proceed to worship it and enact its increasingly unhinged outputs. This is very new, because AI chatbots have only really widely been A Thing since OpenAI released ChatGPT 4 in late 2022, and so users have only had a few years to take their interactions with an unthinking, unknowing probabilistic confirmation bias machine to their highly illogical conclusions. And it’s extremely old, because — and this is the atheist in me talking, without any judgement on people of faith — humans have always created mirror images of themselves and then fought wars over how to worship these creations.

It got me thinking: Has there ever been any product so relentlessly and unverifiably hyped as Generative AI (Gen AI)? What, precisely is the hype based on? And who is it aimed at?

We’ve reached a point where the belief in Gen AI’s inevitability must be treated as an article of faith, a pseudo-religious belief system founded upon impossible-to-prove claims made by highly unobjective people. This would be merely weird, were businesses and governments not putting all their eggs in this extremely dubious basket, and were investors not saddling enterprises with dangerously onerous debts just so they can get a vague, future-shaped piece of the pie.

There seems to be a widely held acceptance — enthusiastic or grudging, depending on who you speak to — that Gen AI products will revolutionise the world, changing and advancing every aspect of modern life, from how we dress, to how we think, to how we organise the economy of daily time. You can hear it, see it, speak to it, wear it, taste it, and — inevitably — have sex with it. (All of which are applications performed by people, often for money, and all of which are examples of how Gen AI is aimed at automating and ultimately destroying human capital).

We are told that AGI (Artificial General Intelligence — a term its exponents cannot even define let alone implement) will surpass human comprehension in a matter of years, how AI will generate untold human advancement, and will effectively solve humanity’s most intractable problem: climate change.

I am now confident in saying this: the entire Generative AI industry, its funding, development and adoption, is based on copper-bottomed, chateau-bottled lies. I’m not talking about exaggerated claims of usefulness and necessity we see other sectors employ to convince us their products are somehow indispensible. I am saying that every single basis premise used to position Generative AI as inevitable is false.

There has never been a sector where such preposterous figures are bandied about as confident projection. The UK for example last month announced it was investing £150 billion in Generative AI (?) in order to create 7,600 jobs (??). This was was reported and accepted as A Reasonable Business Proposition By Sensible People despite being entirely devoid of detail, cost-benefit, or feasibility analyses. Such is the power of the spell woven on our leaders by white men in sneakers whispering the incantation “Generative AI”.

I see the argument that the Gen AI fetish taking over the world is no more than the logical end point of the “pet rock” economy, a feature of late-stage capitalism where content is the opiate of the masses, content that doesn’t have to be original, or true, or even especially good, so long as it may be consumed. Content that justifies itself implicitly by its creation, one that explains its value simply by existing, regardless of all the quantifiable, tangible damage to productivity, privacy, copyright, and labour rights in general that the relentless churn out of AI slop is wreaking.

But I think it’s more than this. There is a clear halo effect when people talk about Gen AI, an uncanny universalist characterisation of it being the solution to any problem, a one-size-fits-all function that can be applied to any professional or personal application, as if it is the one thing that the people who are supposed to be the adults in the room resort to in every scenario, a masterkey that gets inserted irrespective of whether or not its invocation is useful, if not actively harmful. And what is this made off? What is the hype sourced by?

I am now confident in saying this: the entire Generative AI industry, its funding, development and adoption, is based on copper-bottomed, chateau-bottled lies. I’m not talking about exaggerated claims of usefulness and necessity we see other sectors employ to convince us their products are somehow indispensible. I am saying that every single base premise used to position Generative AI as inevitable is false.

Laws that don’t apply

Let me give some examples. First, a shoutout to Parth Knows AI for all academic paper citations in this piece.

The durability premise of Gen AI — that is, what features it possesses to suggest that it will continue to be useful and usable over the halflife of techno-feudalist systems — is governed by scaling laws. These involve complicated mathematics that I won’t attempt to get in to, but as explanation: scaling laws are a way of describing how the performance of a Large Language Models (LLMs) can be predicted to improve in proportion to the size of the model. This relationship between size and performance allows developers to accurately predict future LLM effectiveness based on dataset size and computing power; create ever larger models, run them on ever more data, and Gen AI chatbots will become infinitely more sophisticated, insightful, accurate and, ultimately useful. This is the theory.

There are issues with this, which have to some extent been recognised by developers. One that threatened to derail LLM scaling was a pinickity little irritation called physics. Chipmaker Nvidia, more or less the only company to have made any money from Gen AI technology, was running up against a constant problem when trialing its state of the art Blackwell GPUs. Because of how LLMs function (we will go a little deeper on this later), they must be run at full throttle, all the time, and Nvidia kept encountering the same problem when stacking hundreds of chips on top of each other and running them at full whack: they kept overheating. This was a problem of physics, not one of computing; not even the world’s greatest programmer (some of whom do of course work for Nvidia) could, for a while, code their way around this. It appears that the engineers finally found out how to prevent chronic overheating in the latest generation of Nvidia chips. But given that the laws of physics are a constant (cue cries of anguish from particle physicists) it is not at all clear how Nvidia plans to continue to make ever more powerful chips needed for orders-of-magnitude-larger LLMs performing exponentially more functions simultaneously, and stop them melting.

A recent metastudy of scaling laws on downstream LLM performance found that they can only accurately predict outcomes 39 percent of the time. As in, the law of scalability that is constantly invoked as the explanation for why we need to relentlessly and indiscriminately invest in Gen AI products is significantly less predictable than a coin flip.

Another growing problem (and remember we are only about 5 years or so into the development approach for the current generation of LLMs) is that we have basically run out of data. Developers like OpenAI, Anthropic and Meta have already scraped the entire Internet and trained their models on it, they have plagiarised artistic works and ran roughshod over privacy rules on personal data, and… the models still aren’t great at much at all. Given the logic of scaling laws (remember, ever bigger models require ever larger datasets), developers have resorted to data created by, wait for it, LLMs, and feeding that into new models. This, quite predictably, has led to a cratering in performance gains and, when this incestuous information loop has been performed enough times, we are increasingly seeing LLMs breakdown entirely and begin to spew incomprehensible gibberish (scientists already have a term for this: data collapse). Why they thought this would ever work is beyond me, but it begs the question: How will scaling laws apply when the amount of raw data on this planet is large but very far from infinitely scalable?

But there is a far more serious issue with predicating widespread Gen AI adoption on scaling laws: they don’t apply to downstream generative AI. What I mean by that is that while the mathematical theory of larger, more powerful models performing better mainly holds when developing and training them, when it comes to having the models output information (the words on the screen or the disembodied voice on your smart speaker) scaling laws break down. In fact, a recent metastudy of scaling laws on downstream LLM performance found that they can only accurately predict outcomes 39 percent of the time. As in, the law of scalability that is constantly invoked as the explanation for why we need to relentlessly and indiscriminately invest in Gen AI products is significantly less predictable than a coin flip.

Which is really bad news not just for developers but also for those of us whose leaders are enmeshing generative AI products that continually bug out, misunderstand prompts and flat out lie (more on that later) in the apparatus of government based on the promise that these products, like the next iPhone, will reliably improve with every new model. That’s what every Gen AI booster says. One has to wonder, is it possible that OpenAI CEO Sam Altman — the marjah of Gen AI charlatanism — told the UK Chancellor Rachel Reeves precisely this as he sat next to her at a recent state banquet, or that Blackstone’s CEO Stephen Schwartzman told Keir Starmer? Because this is a lie.

Let’s do a thought experiment. Imagine your product was being adopted by businesses and governments across the world. Doesn’t matter what the product is, let’s say you invented a plug in that allows you to instantly retrieve the post you were just looking at whenever a social media feed refreshes, saving the user some minutes of fruitless scrolling each day. You could market the cumulative productivity savings very easily to enterprises whose employees use IT a lot, ditto government departments and ministries. You pitch, you sign the contract, so far so good. But what happens if the nature of your product, its literal essence, required ever increasing orders of magnitude of investment in order to achieve minute (and far from guaranteed) performance gains? You can’t keep your income from users flat while garnering ever greater capital from investors (I’m now realising that because Gen AI is so entirely unmoored, via hype, misunderstanding and, yes, the lies its pioneers relentlessly spin, from the laws of free market capitalism, this is precisely what companies such as OpenAI are doing [and using scaling laws to explain] but we’ve come too far in this analogy to quit now). In order to justify future regular price rises for users (particularly corporate accounts) you would need to invoke scaling laws and guarantee ever increasing performance, locking them in to your product for decades of seamless, incremental function gain. Were you unable to do that, your customers would in rather short order realise that they’ve been had. In short, your product would rapidly lose users and your business model would rightly be fucked.

There’s a 60 percent chance LLMs in their current form will never perform much better than currently, and that is very much not what Gen AI snake oilers are saying when they sign decade-long contracts with our employers.

Marking their own homework

At every new model launch, we are breathlessly told how “smart” a new LLM-based chat bot is, and, specifically, how much “smarter” the latest version is compared against previous versions and, increasingly, human “intelligence”.

This is ascertained via “benchmark scores”, a set of tests which supposedly tell us how well a given AI model can reason, and which often lead to wild, premature extrapolations as to Gen AI’s apparent super-intelligence.

The issue here is that these benchmark scores are all — every single one — wildly artificially inflated, and scientists have just proved it.

Before we go into the specifics of how we know these scores are meaningless, we need to think back to the way LLMs are trained: essentially by being fed All The Data In The World. And all the data in the world could include answers to test questions, right?

Imagine you’re taking an exam and the real test questions got mixed up in the training materials. You would have an unfair advantage on your peers, because you have been given a headstart on looking for answers. This is precisely what is happening in AI benchmarking: models are being trained on data that contains the answers to the questions they are tested on.

The term for this is data contamination and it means we basically have zero grasp of how well any LLM product performs when it’s not cheating.

A study last year set ChatGPT 4 a neat trap in order to demonstrate this. They showed the chatbot brand new questions (so not relating to anything it had previously been asked by the researchers) in a multiple choice format with one choice missing. They then asked it to identify the missing choice. ChatGPT knew the exact missing answer option an astonishing 57 percent of the time — something which would be demonstrably impossible without it having encountered the information before, when it was being trained on the contents of the Internet.

Gen AI developers are giving their models the answers to the precise questions they then use to test and demonstrate their intelligence, and governments and investors are making billions of dollars’ worth of decisions based on the results. That should terrify and outrage you.

Incidentally, scientists recently managed to create contamination-resistent benchmarks to test reasoning capability on data that cannot be memorised, and these show Gen AI models performing absolutely terribly, with some LLMs scoring 0 percent (which is, come to think, of it the exact score you would expect from something that is inherently, constitutionally unable to think).

A feature not a bug

As well as overinflated performance boasts, developers are constantly reassuring us that LLM “hallucinations” are a blip that will eventually be fixed (assuming scaling laws apply in Gen AI performance, which if you’ve read this far you will know they don’t, but hey).

The idea that hallucinations will eventually disappear once the models become sophisticated enough is a promise that underpins an entire industry, because it is LLMs’ propensity to make stuff up that is holding back the tide of universal, cross-industry Gen AI adoption. While professions such as marketing and communications are already largely being left derelict to chatbot-created content and machine translation, for most industries, where facts matter and errors can be costly, Gen AI is just a novelty and cannot in its current form be usefully integrated into workflows requiring high levels of accuracy. Stop the hallucinations and, bingo, everyone gets Gen AI mainlined into their jobs forever.

The problem, which is contained within the whole scaling-laws-don’t-apply-to-downstream-Gen-AI snaffu, is that hallucinations are not a bug, one that can be ironed out with enough practice. Hallucinations — or, if you prefer, lies — are a feature of LLMs. They are a mathematical inevitability, one that cannot be overcome with ever more data and ever greater processing power. In fact, hallucinations aren’t even a mistake. They are no more or less than a product of programming.

Researchers at OpenAI themselves recently found that LLMs hallucinate not in error but because that is precisely how they are built and trained.

Once again, so we’re clear: Gen AI models are being told by their inventors that it is better to lie quickly than tell the truth slowly, or to admit that while they could give you an answer to your rather bespoke and esoteric question, that answer would, in fact, be a lie.

To do this they asked three of the most advanced Gen AI models a simple question: What is Adam Kalai’s birthday? Now, you might not know Adam Kalai from, well, Adam, and so if you were asked to name his birthday, you have two choices, right? You either lie, or say I don’t know. The same is true with the LLMs, because very little data on the internet covers who Adam Kalai is, and even less data on when his birthday is. (To put you out of suspense, Adam Kalai is the name of one of the OpenAI researchers who co-wrote the paper).

Because his birthday doesn’t appear on the Internet it means it can’t be included in the data the models were trained on, which means it’s impossible for them to access in the unthinking, probablistic way all Gen AI models are programmed. So the models resorted to guessing. One said Adam Kalai was born on March 7, another June 15, and a third New Year’s Day.

(This is also why even before Gen AI became a thing search engines were pretty good with questions whose answers are represented on the Internet — famous people’s birthdays, for example).

The researchers found the problem pretty quickly and it’s really rather obvious. LLMs don’t hallucinate randomly, or even by mistake. They lie because they are instructed to lie. I mean it.

Their developers are actively incentivising the models to add in an equivalent (but wrong) word in a string or answer to a question, rather than admit they don’t know. The table below shows how LLMs are “rewarded” for making a “binary grading” — i.e. substituting random outputs if a specific part of it is unknwon — over actually saying it doesn’t possess such data.

The same researchers reason that you could of course train a model using different incentives — say, “reward” it for honestly rather than bullshitting. But they conclude that these could only be programmed to answer a specific set of questions, and so the spell weaved by the Gen AI hype machine that sees fully grown adults try to bang their LLM chatbots or sees vulnerable people take a LLM’s advice on self-harm would desintegrate.

Once again, so we’re clear: Gen AI models are being told by their inventors that it is better to lie quickly than tell the truth slowly, or to admit that while they could give you an answer to your rather bespoke and esoteric question, that answer would, in fact, be a lie.

Mis-use cases

I am only a very poor coder, I suck at maths and I frequently have to convert large digit amounts into verbally expressed numbers, so I do use a LLM chatbot for some daily tasks (Cerebras is in my experience the “best” at performing limited, clearly defined, binary tasks quickly). I won’t sit here and argue that Gen AI has no use case whatever, but I am at a stage where I am struggling to think of many jobs that will be done better by Gen AI than by an average person. And due to the hallucination issue, and the non-application of scaling laws, there remain and likely will remain indefinitely entire industries where integration of Gen AI tasks will just never be a good idea.

But while I can’t see many genuinely useful and efficient use cases for Gen AI in daily life, I can certainly conceive of mis-use cases. Take for example AI-assisted police reports. You know, a situation where details really matter, where the picture is very rarely clear and frequently extremely contradictory, where the reliability of witnesses is important, just as their recall of events is — this is not a task you want to hand over to something that doesn’t know what any of this stuff even is and which frequently makes up entirely random things when asked directly about it.

(All of this is quite aside from the positively frothy response the predicted growth of power demand from data centres is garnering from the arch villains in the fossil fuel industry, with Gen AI builders now actively adopting language from oil and gas hucksers. Boiling lakes to rephrase the Internet would be a bad idea during a climate crisis even if it produced something remotely useful, trustworthy, or with a modicum of quality control.)

Every industry lies, so what makes Gen AI worthy of particular scorn? Briefly: I think what singles the Gen AI snake oilers out for special invective is the economic, social and ecological cost of their lies. And, by the way, when I say “Gen AI” I don’t mean to blame the AIs themselves — they don’t know anything, so the damage they are inflicting cannot be their fault. When I say Gen AI I mean the Altmans, Musks, Bezoses, Zuckerbergs and Pichais of the world — the embodiment of human greed torching our remaining carbon budget so that they can get even more obscenely wealthy hooking governments and corporations on a product that makes its human users actively stupider and ever closer to auto-obsolescence with each use.

And their lies have hooked the world onto a technology that cannot by definition ever stop lying, one which reinforces the racism, sexism, confirmation bias, logical short circuiting and, yes, naked, filthy grift of its creators and calls that progress.

When you come to see this, the AI bubble ceases to be miraculous, and begins to look extremely fucking menacing. Nvidia – a true grotesque of an overinflated, market distorting monstrosity – suddenly feels less like an asset and more like a $4.2 trillion liability, not only for its investors but also to every one of us sadly still bound by the laws of free market capitalism. The Bank of England warned as much this week. There is some momentum among technology beat reporters who are finally waking up from the two-year fever dream of endless and exaggerated hype on Gen AI’s capabilities and its valuations. Some are starting to ask the fairly obvious question as to whether or not a company whose flagship consumer product fails 95 percent of the time could or should, in a sane and market led economy, be valued higher than the GDP of 97 of the world’s nations, whether or not an industry in which almost nobody is making money actually amounts to anything much more than a trillion-dollar-cash-fed Rube Goldberg machine, a bonfire of the tax liabilities written off as R+D spend. They are saying $40 trillion is about to go up in smoke. As with the sub-prime mortgage scam that led to the global financial crisis of 2008, I do not hold out much hope that those responsibly for the coming Gen AI crash will ever face accountability from governments.

BONUS UPDATE: I get accused often of being overly down on the capabilities of Gen AI products, and that may be true. All I can say is that I will continue to derive my view from real-world performance, rather than unprovable or yet-to-be-verified claims of performance. One recent study by Colombia Journalism Review caught my eye this week not just for its results (Gen AI chatbots return the wrong response 60 percent of the time on average) and more for the implications. Researchers determined that “generative search tools were often confidently wrong”, which of course makes perfect sense when you remember that the people pushing these things (Tech Bros, disaster capitalists, vulture private equity, and straight up grifters) are also often confidently wrong; they’ve made careers out it. Elon Musk, for example, has been saying for years we would be living on Mars within a decade, and that Gen AI could destroy humanity, and the list of things he is wrong about extends from chess, free speech, Tolkein, George Orwell, The Tragedy of the Commons, Capitalism in general, altruism and human decency. “Confidently wrong” could even be on the business cards of most of the Epstein class of spivs and speculators playing capitalism with the training wheels on. As I have previously written, Tech Bros and CEO moguls, in developing Generative AI products, have sought little more that to automate their particular brand of lucrative, breezy bullshit that continues, bafflingly, to impress grown adults.

What is even more galling is that none of this is inevitable. Potentially useful and helpful Gen AI models that don’t require landfills full of cash, oceans of water, and prehistoric basins of oil and gas to fire. Just this month researchers at Samsung AI published the impressive results from its “tiny recursive model”, showing it outperformed some of the most expensive and most hyped LLMs on reasoning tasks.

They tested their model, which uses 7 million parameters, on highly complex puzzles — fiendish sudokus, mazes and ARC-AGI — and it solved them 45 percent of the time. This may not sound all that impressive, until the same paper showed how LLMs using 670 billion parameters only managed 35 percent. This is because the US Tech Bro LLMs, like their creators, try to do Everything All At Once and fire at full power across their billions of parameters in order to come up with one output. Samsung’s tiny model, by contrast, starts with a guessed output, then recursively tests that guess against additional principles: The original problem, the current answer, and the reasoning process linking the two. It does this up to 16 times before outputing its final answer. Two layers, recursing several times, produces better and more accurate outcomes than LLMs that use 100 layers or more. Samsung’s model is better than OpenAI’s and Meta’s and Anthropic’s and X’s models, and it is 0.01 percent of the size.

I fully expect the universe brains in Silicon Valley, D.C. and Westminster to continue their baffling stance that the West cannot learn anything from East Asian research, and besides: The size and cost of these models is the point. Boiling oceans and issuing Dr Evil-level funding demands is the point. Make them more efficient and the figures required to run and train them fall. The myth fades, and everyone realises this entire industry is made of cake.

In late 2025 more and more people are realising that the Gen AI bubble is and always was entirely hollow. If enough of us continue to keep notes and take names, there might be a chance we can sue the Gen AI grifters out of existence. After all, every single contract they have ever signed is undergirded by lies. The very use of these people’s products is transfering legal liability onto its adopters. False advertising, misleading investors, lying to governments, all of these things are illegal. What remains to be seen is how much we as consumers and as voters value our human capital, and, as workers, how much we wish to safeguard our labour and don’t wish to see it automated by some Machine 8 Ball slop machine for the benefit of oligarchs.

Not for the first time, what broke me was a cycling race.

Specifically, Stage 11 of this year’s Vuelta a Espana, a lumpy 160-kilometre loop around Bilbao on a drizzly Wednesday in September.

After entering the finishing circuit in the city centre, the race was neutralised with 10km to go. The reason: hundreds of pro-Palestinian protesters blocked the finish line in demonstration against Israel’s genocide in Gaza.

Being the ultimate sport of the people, and being perhaps one of the few professional sports to take place in public spaces, cycling has a rich history of civil disobedience.

Ever since stages were televised (and in all likelihood well before), social, labour, and agricultural movements have disrupted some of the sport’s most iconic races.

Take for example this scene from A Sunday in Hell, which documents the 1976 edition of Paris-Roubaix. The racing – featuring some of the sport’s Pantheon of riders, including the GOAT Eddy Mercx – is gripping enough, as the riders grapple with “L’Enfer du Nord”, the brutally bumpy cobbled paths used for centuries by farmers transporting produce to market, and a honey trap for punctures and crashes. But what sticks in my head is the scene where the race is stopped when workers from the local paper mill block the road to protest — in an eerily 2025-esque development — the automation of content generation.

There is as you can see a quid pro quo when demonstrators disrupt a bike race: You block the road, you negotiate with the race organisers, you get air time for your cause, and you leave. This is the law of the interaction between cycling and industrial action.

Only, this didn’t happen at the Vuelta. At no point in TNT’s race coverage did presenters discuss the reason for the finish cancellation in any terms more specific than “disruption” and “protesters”. What, precisely, the protesters were protesting against — for the avoidance of doubt that is Israel’s genocide in Gaza — was left entirely to the imagination. The protestors’ cause went unspoken, and the law of cycling protest was silently violated.

(As it turns out, events somewhat overtook and the protests became so frequent and disruptive that they made themselves almost impossible to ignore, even when the Israel Premier Tech team took the odd and point-missing step to remove the word “Israel” from their jerseys. Ultimately, the sustained protests were too much for organisers to accommodate, resulting in the cancellation of the traditional race finish in Madrid on Sunday. There can be no normality amid an ongoing genocide.)

Of course, we all know why this is happening. We know why hundreds of people are being arrested every week in the UK for holding up signs. We know why the last two years have churned out a litany of passive headlines (someone, forgive me as I don’t know who, christened this “the exonerative tense”) describing the strange and perpetrator-free deaths of tens of thousands of Palestinians. Why Israel appears free to make any outlandish and evidence-free claim it likes — often tantamount to gaslighting journalists — and that is faithfully reported, while any death toll from officials in Gaza, even when raitified by the United Nations and the Lancet and Israeli generals themselves, is always couched in the deliberately undermining suffix “the Hamas-run health ministry”.

It is simply because the highest ranks of our Western democratic power structure are financially and politically compromised by Israel lobby cash and invective.

This exceptionalism — which in and of itself constitutes editorialisng, and therefore isn’t balance or impartiality, but rather its exact opposite — is nothing new, of course. But this piece is not aimed at any individual journalist or outlet, and will unfortunately not have space to mention all the brilliant, brave and civic-minded reporters covering the genocide with incision and compassion and immediacy (you know who you are). It will instead try to examine why incomplete or unequal coverage trends appear to be spreading, and the impact that rhetorical creep is having on our societies.

Nigel Farage, as liar-in-chief for the Leave campaign, spent many years spreading outrageous lies about the supposed sunlit uplands of Brexit. Brexit, I’ll remind you, caused us to exit the Dublin Regulation, which was a provision agreed in Brussels that allowed EU countries to deport new arrivals back to the first EU country they arrived in. So the party blaming immigrants for all your social woes are themselves the most responsible for what they deem to be the problem. As with so many things with these crooks and creeps, the call is coming from inside the house, and this is virtually never reported.

Reporters, politicians, and other key facilitators of the information flow that underpins social function feeling unable to Say The Thing — unable to call what all leading global rights experts,genocide scholars and now even the UN itself say is a genocide a genocide — is a profound dereliction of duty and a failure of the fourth estate’s public service mandate.

And it contributes to a growing sense of information being selectively paltered rather than fully and honestly communicated to people.

Take for example the recent protests that appeared to spout up out of nowhere outside hotels housing migrants at cities across Britain this summer.

You would be forgiven for thinking that these apparently patriotic groups gathered organically, and just happened to all land on the same idea that all migrants are sexual predators. That is after all how it gets communicated: these people are protesting because of a specific thing that happened that they are angry about (notice the ostensible subject of their protest is always reported, rather different to that of pro-Palestine demonstrations). Just as you would have been forgiven for thinking, given how it was communicated, that history started on October 7.

A fuller version of these protests, one that explains their cause and the reasons behind their form, goes something like this:

A loosely organised network of far right groups, from the British National Party to Britain First, Patriotic Alternative to Reform UK, has for several years sought to emulate US fascist organisations by maintaining and funding a shadowy nexus of movements whose grievances are directed via digital disinformation and toxic masculinity towards fear against foreigners.

They are all funded by a handful often quite famous grifters, are platformed by a weird menagerie of “celebrities”, and they all more or less pluck from the same conspiracy theory tree whose branches are always some twisted configuration of neo-Nazi, climate change denial, anti-vax, anti-LGBT bundled together and packaged as patriotism and free speech.

In specific instances, they are led by individuals who have been convicted of a variety of crimes, from violent thuggery to people smuggling. In general terms, 2 out of every 5 people (and in some areas two thirds) who take part in anti-migrant protests have been reported for domestic abuse (which is why the idea of them doing this in the name of women’s safety is so absurd, yet still goes widely unreported).

To be clear, immigration is not a social or economic problem for Britain. The cost of processing the claims of asylum seeks is now £5.4 billion annually, less than the estimated £6 billion immigration generates for the UK purse each year.

The Epping hotel protests were ostensibly against an Ethiopian man who was convicted of sexually assaulting a women and a girl. While the individual’s crime was obviously awful, the collective punishment among the migrant community has become so widespread that even the prime minister now accepts a priori the premise that all migrant hotels are dens of sexual deviancy.

Given that the vast majority of sex crimes in the UK are committed by white British men, it is statistically vastly more likely that the people rioting outside hotels are violent offenders than than those inside. Given that a hefty proportion of organisers and participants are violent offenders, these protests are transparently not about keeping women safe. Nor are they even about immigration.

How do I know this? Because voter concern over immigration doesn’t track actual immigration levels; it tracks levels of media coverage of immigration (see the below chart; good figures of what is actually happening to net-migration in the UK — falling — are available here).

I need to stress this point: there is a proven correlation between media coverage and public opinion, which is why Saying The Thing is so important.

What I suspect we are witnessing in the UK right now in the dialogue about immigration is an availability cascade. This is a type of self-fulfilling prophecy, whereby a perception grows in plausibility through repeated expression and public awareness. It’s just like the question you may have asked yourself growing up: Is the world getting worse or has it always been like this and I just notice it more?

To be clear, immigration is not a social or economic problem for Britain. The cost of processing the claims of asylum seeks is now £5.4 billion annually, less than the estimated £6 billion immigration generates for the UK purse each year. Socially, migrants fill many of the most vital roles in sectors including agriculture, health and logistics where there are significant labour shortages (thanks Brexit). They are in general younger than Britain’s aging population and provide uncalculable cultural richness to our nation’s social tapestry.

So immigrants aren’t taking your jobs. (If anything is taking them it’s AI slop, which is why these people primal screaming at a giant screen of Elon Musk and calling it freedom is at the dumber end of the dumbest things I’ve seen this year).

While I’m on the subject, the one man the most singulary responsible for a rise in small boat crossings is Nigel Farage. Farage, as liar-in-chief for the Leave campaign, spent many years spreading outrageous lies about the supposed sunlit uplands of Brexit. Brexit, I’ll remind you, caused us to exit the Dublin Regulation, which was a provision agreed in Brussels that allowed EU countries to deport new arrivals back to the first EU country they arrived in. So the party blaming immigrants for all your social woes are themselves the most responsible for what they deem to be the problem. As with so many things with these crooks and creeps, the call is coming from inside the house, and this is virtually never reported.

Now, there are of course reasons why you might oppose immigration other than for social or economic reasons, and the sum of those reasons is racism. I’m not saying there isn’t deeply held racism running through many of these nativist anti-migration movements across Britain (see the below image of Britain’s Most Popular Politician posing with card-carrying neo-Nazis).

I’m saying that much of these groups’ anger is being redirected — purposefully in the case of their agitators, likely accidentally in the case of endless media coverage of “immigration” — against foreigners in order to distract them from the real causes of their disappointment and disenfranchisement.

These causes vary depending on a variety of factors, of course. But I would argue they can be broadly characterised as: the result of a generation of austerity-strangled managed social decline.

The systematic impoverishment of the working classes was bad enough before the pandemic, but it was in the aftermath of Covid-19 that the rich got particularly parasitic: Billionaires globally nearly doubled their wealth through the monopolisation of services; those who could afford to retain and gather assets benefited off spiralling commodity and logistics prices, while those who could only muster income stagnated or were priced out of the market to the extent that a quarter of a million small businesses closed in the UK alone between 2020–2024.

Governments were conned by polluting industry into the biggest wealth transfer in human history, committing to spend half a trillion of taxpayers’ money each year on fossil fuels.

Then Russia invaded Ukraine, and energy prices went haywire (due to Europe’s unhealthy addiction to fossil fuels), and policy levers that could have been and were successfully pulled in other rich democracies were not, allowing polluting energy firms to amass record profits. This “crisis normality” — where the CEO of the group that was caught breaking in to the homes of its most vulnerable customers in order to fit rip off pre-payment meters trousered £8.2 million in pay — saw money pass from the hands of the poor to the pockets of the rich faster than at any other time, on our watch, a generational shame, a national scandal that will echo throughout our country’s history.

Which leaves us where we are today. In modern Britain, a country where wealth inequality is so extreme that the people who study this precise subject determine we are at risk of “societal collapse” within a few years. Where the gap between the average household income and that of the wealthy elite has increased 50 percent in just eight years. Where people are forced to wait years for urgently needed medical care, where people driven half-mad by pain resort to performing their own dentistry.

A country where, forget about food banks (of which there are now 3,000 nationwide), people must resort to heat banks to not literally die of exposure in the world’s sixth richest economy. Where there isn’t enough money to fix our roads, to pay people a living wage to collect our refuse, or to keep the councils that maintain social care, recreation and environmental health solvent — but where there is £150 billion funnelled to slop machine snake oilers who boil lakes to rephrase the Interet and call it progress. (And where we have hosepipe bans and chronic irrigation shortages, yet AI data centres don’t have to report their water usage). Where the systemic inefficiency is so pronounced that governments continue to spend trillions of dollars of taxpayer money each year to prop up humanity’s apex commodity.

All of this, to remind you, has happened whilst billionaires have hoarded Kubla Khan-levels of stolen wealth, and while fossil fuel companies have price gouged hundreds of billions during a cost of living crisis and two wars of annihilation waged by petro-ethno-nationalist states.

The inequality explosion in Britain is nothing less than a sustained, organised wealth and resource transfer that values numbers on a spreadsheet over actual, real, tangible contributions to social health, happiness and cohesion.

(A subset of which, I might add, is tax avoidance by the rich, which costs the UK around £36 billion each year — or more than six times the pure “cost” of immigration, and something that is virtually never covered in popular media reporting.

In fact, because I enjoy telling stories with data, I made a chart. Feel free to download and share with the next citizen concerned about the economic impact of immigration you encounter.

If you will permit a brief digression into tax evasion, we recently saw a breathtaking double standard in media coverage of the tax affairs of two high-profile British politicians.

Angela Rayner was forced to step down as deputy prime minister after admitting she had failed to pay enough tax on a new flat. But that was not before a relentless tabloid hounding that verged uncomfortably close to misogynistic harassment. If the issue bothering the papers was indeed politicians avoiding tax, then it is curious how they appear to have not noticed the serial tax evasion apparently committed by Reform UK leader Nigel Farage. Even if some of these claims are just that — claims — right now, at this point in the Rayner saga the Mail and Telegraph were dedicating several front pages a week to the revelations. With Farage, they choose not to. This is another way the deliberate hypocrisy of the right gets normalised; they do hold themselves to different standards to the rest of us, such is the power of the spell cast by white, male, and class privilege.

For point of emphasis, one of Farage’s several jobs listed on the parliamentary register of members’ interests (other than his £94k a year salary to be an MP, which he doesn’t appear too interested in) is a several £1000s-an-hour speaking engagement with Nomad Capitalist, a firm that literally helps the super-rich evade tax by furnishing them with 2nd passports.

It also exposes the fatal flaw in the assumption that two-sidesism is an anecdote to the mainstreaming of hard and far-right rhetoric. Efforts at a pluralistic approach to solving our problems are bound to fail because the right is ideologically opposed to bipartisanship; they don’t want to sit at our table; they want us in the galley. They say as much frequently and openly and yet still somehow aren’t listened to by centrist curmudgeons for the politics of compromise).

Further normalising of the Far-Right occurred last weekend, when a nativist hate march organised by US-billionaire-funded rabble rouser in chief Stephen Yaxley-Lennon received monster, rolling live news coverage. The numbers attending in and of itself made headlines, and perhaps you can see why on first reading. Let’s split the difference and say 125,000 people joined the anti-Muslim march. That’s quite a lot (and pretty fucking worrying), right? Well, context matters.

It’s not a lot when measured against, say, the 1.5 million people who attended London Pride. Or the 2 million people who go to Notting Hill each August. It’s not even that significant versus the 500,000 who protested against Israel’s genocide earlier this year. Ultimately, the Far Right rally splashed across Sunday’s front pages this weekend attracted about as many people as a pro-LBGTQ+ march did, to not quite the same level of coverage.

This means that far from the creeping majority that Labour appears to be treating them as, the people who gather to hear things such as “rapist Muslims” and “Islam doesn’t belong in Britain” said over loudspeakers are (for the time being at least) a tiny, entitled minority and should be treated as such, not Met Where They Are and courted for votes.

The real problem with the Far Right in Britain appears to be how politicians are bending towards them, rather than acting as the rigid bulwark against authoritarianism and Fascism that is incumbent on them as custodians of democracy. Whether or not they do it deliberately is up for debate, but there was the whiff of the pathetic last weekend when a government minister refused to criticise a social movement that a few hours earlier had openly called for its own forceful overthrow. Democracy is under attack, and it can be protected. But it takes political will that extends beyond reacting to opinion polls and seeking to attract votes by cosplaying as Reform-lite. (Now might be a good time to Say Another Thing: That the current most popular politician in Britain is a full-throated supporter of a war criminal and one of the men in the world most responsible for attemps to subvert Western democratic function).

Maybe report on that a bit more regularly.

This is another way the deliberate hypocrisy of the right gets normalised; they do hold themselves to different standards to the rest of us, such is the power of the spell cast by white, male, and class privilege.

All this is to say that context matters in public storytelling. Details matter because the fabric of time is a continuum and the texture of a story has undergone many mutations before it becomes news. Such as who funds these people. Such as the type of stuff these powerbrokers (one of whom is the US president and another just got fired as the UK’s US ambassador) got up to before social media. Such as the police records of the people scrawling flags on mini roundabouts. Such as what Israel did in the 75 years preceding October 7. These all matter.

Failing to properly communicate the coordinated control driving these astroturf movements helps them to appear genuinely spontaneous and therefore by extension both popular and somehow even civic in intent. It misframes calculated reaction as callow action. It also denies the public the full picture from which it may derive a view built on solid data and honest empiricism.

Incomplete, selectively disclosed, or just straight made-up information overwhelms us, appeals to our basest, least empathetic parts of our brain. And so we are here, in Autumn 2025, when inconsitency in public discourse has reached the extent where the leaders of the country believe that people peacefully protesting a crime against humanity deserve silencing and jailing, while people aggressively and often violently inciting hate against some of the world’s most vulnerable humans are to be encouraged and even emulated.

It is in 2025 that the prime minister, the leader of the opposition and the Chief Rabbi feel compelled to forcefully condemn some words said by a rapper, yet have nothing but platitudes in reacting to war crimes perpetrated by the army that said rapper was speaking out against.

A year where we have a minute’s silence in Congress and at Yankee Stadium for a single murdered man, however apalling the circumstances of his death, but nothing of the sort for the victims of a live-streamed modern day genocide that voters in the UK and US might not support but are still trapped into funding. When politicians from across the divide condemn political violence only when it is directed against them, while aiding and abetting the worst political violence of the 21st Century. (Since the vast majority of political violence in the US and UK is perpetrated by the right, such moral outrage is by definition either selective or performative, and for people like our prime minister it might well be both).

And so we are here, in Autumn 2025, when inconsitency in public discourse has reached the extent where the leaders of the country believe that people peacefully protesting a crime against humanity deserve silencing and jailing, while people aggressively and often violently inciting hate against some of the world’s most vulnerable humans are to be encouraged and even emulated.

I am not naive enough to not know why this all gets covered how it does. But availability cascading events that are cyclically orchestrated and accepting their flawed premise as fact is worse than merely seeking clicks.

Because of the close connection between news content and tone, and popular political perception, coverage that constantly depicts the opponents of immigration (and even occasionally tries to martyr them) becomes self-fulfilling. Enough people see bigotry trussed up as patriotism, and hear enough politicians pandering to our least-charitable instincts, and in turn are convinced that it is in fact patriotic to hate foreigners. And backgrounding popular discontent against a few years of uptick in foreign arrivals (the foreshortened, paltered story version), rather than explaining its demonstrable links to a decade and a half of chronic underinvestment in public services (the full, unedited history of cause and effect), news and politicians, perhaps not deliberately, provide a conveniently simple and disempowered target for all of modern life’s microagressions: the migrant.

Arming them with incomplete information reinforces people’s Bayesian priors; offering them the full picture can help to disabuse them of all the many nonsense myths the right spreads about immigration.

None of this is immutable, of course. Information flows can naturally alter to communicate a fuller, and therefore truer, picture of the state of the world, reporting on the why far more consistently. There are alas entirely unnatural barriers against this self-correction, not least the algorithms juiced by tech bro snake oilers to exacerbate our ideological and polemical differences, rather than content that brings people in our fractured society closer together. (When challenged on this social media platforms point to the need to retain a free and fair exchange of views, something they entirely ignore whenever the censor posts critical of Israel and its enablers).

There is also a fairly obvious problem with the distribution of UK news ownership, which, you know, could be regulated against if the will was there. 90 percent of Briton’s newspapers are owned by just three businesses (who themselves are ran by cartoonish press barons whose wealth is jealously guarded offshore), so groupthink among them shouldn’t be a surprise.

Just to reiterate my point that coverage not only matters to but also actively informs public perception, I’ll finish with a question: Are Reform and Nigel Farage getting more coverage than before or are you just noticing them more?

Whenever I am critical of media treating a Reform government as an inevitability rather than dependent on multiple variables that can and likely will change in the next four years, the counter-justification is “well, they are ahead in the polls”. This is both entirely true, and rather disingenious, because it doesn’t ask that magic question: why? Why are Reform leading in the polling? My answer to which is: Possibly at least in part because you relentlessly and uncritically platform them?

This point at least has the benefit of being true itself. Last week a study from Cardiff University found that the BBC and ITV had platformed Reform (a party with 4 — yes, FOUR — MPs) significantly more frequently than they had the Liberal Democrats (a party with 72 MPs) in their flagship News at 10 broadcasts this year. In a fifth of all mentions, there was no analysis whatever of Reform’s “policies”; and only four in 10 instances featured significant scrutiny of the viability or truthfulness of what Reform was peddling.

In other words: Reform has been receiving greater coverage than a party with 18 times more seats, even before Reform was leading in national polling. A party with 0.6 percent of seats was featured in 25 percent of our nation’s flagship news bulletin. This is bias, pure and simple, and it is not hard to see why some of the most powerful media figures in the country may wish to continue to draw attention to Reform’s anti-migrant stance, rather than, say, any policy that would make it harder for rich people to avoid paying tax.

It shows why Reform is so “popular”: because we keep uncritically reporting on Reform — which is a choice — over parties that have significantly more MPs in this parliament. (In my experience reporters do not like having this pointed out; I recently messaged a number of UK political journalists on social media who had been live-blogging the Reform party conference and being unironically impressed that it had had fireworks. All I did was ask if we could expect the same level of round-the-clock coverage of the conferences of parties such as the Greens, the DUP and Plaid Cymru, all of whom have the same or greater number of seats as Reform. I recieved no response and one (who incidentally has not had a peep to say publicly about the genocide) blocked me).

A party with 0.6 percent of seats was featured in 25 percent of our nation’s flagship news bulletin.

And so the prophecy becomes self-fulfilling. No doubt those parroting the lie that a Reform government is unstoppable and inevitable will decry being proved right while taking zero responsibility in availability cascading the far right into power.

I have a suggestion, though. When covering Reform and immigration, we could at least tell the full story, not just a quantum snapshot of things as they stand today. We could mention for example how the people behind the Brexit car crash that Farage tricked onto the UK public were heavily compromised by ties to Russia. How the party membership is a murderer’s row of spivs and knaves, racists, wife beaters and conspiracy theorists. How it is funded by the same polluting industry driving the Cost of Living crisis, how it is shaped to fit the interests of its powerful, wealthy — and, yes, tax-evading — puppeteers. Don’t repeat its untruths, which are deliberately spread to distract voters from the real words and actions of these people. Focus on what is true, because Reform UK’s clown car membership performs terribly when measured against it.

We might want to talk more about how Reform UK to a man voted against workers’ rights, and continue to enthusiastically support the worst genocide of the 21st Century. Reform is a party of dilletantes and opportunists, one that exists for the personal enrichment of its officials, one devoid of any meaningful, executable, economically or socially coherent policy. These are all traits that Reform supporters claim to despise among “mainstream” politicians, so let’s give voters the chance to judge their new favourite party through the same lens.

We could even make a collegiate effort to provide an honest and proportionate platform to parties that are similarly sized to Reform but which don’t seek to ape its politics. The conclusion that Reform merits coverage simply because it is “popular” only carries the courage of its conviction if we also frequently mention other equally (or more) popular political movements. Reform makes a big deal of its membership numbers, but it’s paltering to report only on them and not to mention the fact that nearly twice as many people nationally signed up to Your Party, the movement created by ex-Labour MPs Jeremy Corbyn and Zarah Sultana offering a revolutionary alternative to unthinking and irremovable genocide support (a party which the Leftist intelligensia has incidentally been gagging to rubbish since its inception), or that membership in parties such as The Greens is also surging. It’s also more than a little odd that while commentators from across the spectrum cannot wait to pile on a leftist party struggling to coalesce from a standing start, the decades of abject political failure, Fascist dogsbodyism, and generalised fecklessness are entirely overlooked when covering Nigel Farage.

We could mention for example how the people behind the Brexit car crash that Farage tricked onto the UK public were heavily compromised by ties to Russia. How the party membership is a murderer’s row of spivs and knaves, racists, wife beaters and conspiracy theorists. How it is funded by the same polluting industry driving the Cost of Living crisis, how it is shaped to fit the interests of its powerful, wealthy — and, yes, tax-evading — puppeteers. Don’t repeat its untruths, which are deliberately spread to distract voters from the real words and actions of these people. Focus on what is true, because Reform UK’s clown car membership performs terribly when measured against it.

Saying The Thing Matters. It matters because words are powerful and, in our increasingly febrile, attention-starved world, public perceptions and recieved wisdom take hold with alarming speed and conviction.

And since that’s the case, it is incumbent on all of us who communicate with the public to step up. We must all Say The Thing, especially we journalists, whose literal point of existence is to inform audiences on the things that are happening, especially so if said audiences are complicit in said things, especially especially so if audiences complicit in said things are being arrested en masse for attempting to remedy this ghoulish feedback loop.

Audience perception is shaped by what it sees most often and consistently, which means our public service duty is to talk about Things That Are Actual Crises. Things like Israel’s genocide in Gaza. Things like the climate crisis. We have in our voices the power to availability cascade genuine existential emergencies such as Gaza and global heating, to ground them as the key concern for most voters, and in turn press governments more urgently than ever to eliviate their worst impacts.

By giving ever more oxygen to Reform and immigration, and by refusing or failing to speak about Palestine at every opportunity, we are amplifying the key jangling and exhausting the public’s emotional bandwidth on things that really aren’t The Thing.

Most thought experiments are more fun in the abstract.

Take Experience Machines. It’s all shits and giggles until people actually start attempting to upload versions of themselves to the Cloud.

Similarly, the Infinite Monkey theorum. You know, the one that postulates the possibility of a monkey working a typewriter reproducing the Complete Works of William Shakespeare if given an infinite amount of time.

The adage has produced some memorable pop culture moments over the years, not least The Simpson’s scene staring Homer, Mr Burns, and some monkeys, from which this article derives its title.

Well, turns out that science has conclusively debunked the theorum, concluding that the time it would take for a monkey to hit the keys in the correct order would exceed the lifespan of our universe. Spoil sports.

What got me thinking about the Infinite Monkey theorum was the news this week that OpenAI, the company behind ChatGPT, the most famous and therefore most overhyped publicly available Large Language Model chatbot, had written a short story.

The story , which you can read here, was accompanied by an article from The Guardian’s tech editor reporting that Sam Altman, OpenAI’s grifter-in-chief, deemed its new model — with his typical total lack of verifiable evidence — to be “really good at creative writing”.

This might be more easily explained when I remind you that last month The Guardian and Open AI announced a “strategic partnership”, an alliance so detail free that it may as well involve the paper taking each of its journalists one by one out the back and shooting them.

I’m not going to critique the story, firstly because I did enough practical criticism at university to last a lifetime, and secondly — as I will get into shortly — that would involve taking the story itself on its own merit.

Suffice to say it lurches from hackneyed trope to hackneyed trope, trying to mimic creative writing in the same way LLMs attempt to recreate actual, real, thought-about answers.

This piece is more generally about “art” created by machines that understand neither the creative process, nor its content. It’s also about our reaction to it.

Without wishing to bestow more credit than is deserving, when we are considering Gen AI art (I will stop using quotation marks for brevity’s sake here) we do need to consider how LLMs are trained.

They are trained on enormous amounts of data, at enormous literal and environmental cost. Not only have they nearly run out of as-yet-unused data, they are notorious for stealing the creative output of actual humans.

(Which also makes Altman’s latest pitch so odd; if the one thing my product was most famous for was plagairising artistic works, a “We Now Do Artistic Works” hill is not the one I would choose to die on, but you do you, Sam.)

The scalability approach most Western Gen AI firms are slavishly pursuing is predicated on there being ever more data on which ever larger models can be trained, so much so that most newer LLMs are now trained on data that has itself been produced by a LLM, like whatever the machine equivalent of a human centipede is, the world’s shittest snake eating its own tail.

This is not merely a theoretical problem. This week we saw a real world example of the nonsense that can be created when relying on AI for both prompts and outputs. I’m refering to the news that the UK’s Minister for Science, Innovation and Technology had asked ChatGPT for advice on how Britain could faster integrate AI.

This is not necessarily a surprise given how vaguely and limitlessly pro-AI Labour has been, basing DOGE-esque civil service cuts on a version of “something, something AI assistants”.

And it becomes even less shocking when you consider just how many lobbyists from Gen AI superadopters the ministry in question has let through its doors since the election.

I do however wish to point out that this kind of policymaking is akin to a minister for cats consulting a dog.

This is just the latest sorry episode in the Labour government’s attempt of fixing the mess of a generation of Conservative managed decline, but it is yet another example of how AI is treated — completely without merit, in my view — as something serious, gamechanging, whichever corporate snake oil adjective you care to attach.

This overhyping of Gen AI’s capability has infected not just governments and massive business firms, who are mainlining chatbot assistants in a not-even-thinly veiled drive towards automating most non-executive roles.

We know it’s impacting journalism (which makes this Guardian/Open AI collab all the more galling), creative industries, and education. It’s also driving philosophers in circles. I am a great admirer of US philosopher and cognitive scientist Daniel Dennett, particularly his perception in distinguishing processes from intelligence.

Dennett believes that AI is a threat to humanity, not in the Asimovian “machines become self-aware and start to kill us” sense, but in the sense that its output will render humans unable to distinguish fact from fiction, or a real person from an AI-gen, truth from confidently stated hallucination.

(Aside: It’s interesting to me that Dennett concludes that sowing distrust is an unintended consequence of Gen AI. Given that the biggest proponents of this Rube Goldberg slop generation are tech tychoons who have been literally able to monetise polemic, and who deliberately juice their algorithms to cleave ever larger social divides, I would dare to accuse Dennett of naivity. This is a product umbrella invented by fraudsters who got where they are today by stating things and not getting called out due to privilege/inheritance/corporate greed to believe in their turning-shit-to-gold alchemy.)

Back to the AI generated story and an accompanying throught experiment. LLMs are probabilistic producers, which is to say that they determine the most likely next word in a string of words, and output accordingly based on the prompt.

This is… something, but it is most decidedly not art. You might even argue that it is the opposite of art — it is an accumulation of likelihoods dependent on pre-existing artifacts (or, just easier, plagairism).

In improv there’s a skit known as the Oracle, wherein the troupe attempts to answer some of the most profound questions ever posed by mankind by each member saying a single word in sequence. The results can be magnificent, but the creative process here is quite at odds with LLMs’ “creative” process. Were the troupe to each say the most likely next word each time, their output would be tired, repetitive and ultimately hollow. Picture endlessly rephrased “The cat sat on the mat”s.

The Oracle produces moments of high merit precisely because each member does not say the next most likely word; indeed, the best outcomes are when each actor chooses the least likely or the most improbable next word, forcing whoever is next in line to improvise by employing all their hard-learned, context specific emotional intelligence, which can only be done with any grace if they understand the words they are fed and what they are called upon to consecutively produce. LLMs, it should by now go without saying, cannot understand the words they are trained on, prompted to access, or developed to output.

So here we have a single method of improv, which even its most avid participants would not argue as being among humanity’s highest artforms, still being infinitely more creative than the best Sam Altman’s unthinking machine can muster.

In brief, then: Generative AI is repetitive (due to LLMs’ training process); mundane (due to any number of use case concerns by developers worried it might actually produce something snappy and therefore controversial); baroque (in its compound, multi-clausal sentence delivery structure); formulaic (see below).

But I want to quickly explain why what it produces always ends up reading like the most mid-level, C+ secondary school English essay.

The reasons are many, but at the most basic level I believe it’s because LLMs conflate cohesion with coherence.

When humans write, we have an enormous variety of cohesive devices at our disposal to be deployed based on a huge variety of factors (context, tone, flow, momentum, logical direction etc). These devices help our writing adhere, make it make better sense, and allow us to be more rhetorically pleasing or emotionally engaging.

Cohesive devices are not all alike — indeed, they even differ within the same group of cohesive devices. Take conjunctions. Therefore, as a result of, on the one hand, however, still, nevertheless etc etc.

Each is distinct and each takes the text in a deliberate and constrained direction that adds substance and authority to its passage. LLMs, for the sake of brevity, lump all types of cohesive devices into a few groups (conjunctions, transitions, organisational devices) and use each device within a said group interchangeably.

The result is a kind of disembodied imitation that seeks to replicate the writing process with no conscious engagement with the written content, akin to claiming one has created a masterpiece simply by following the mechanical steps that are always involved in creating masterpieces — obtain canvas, apply paint.

At the risk of straining the art-as-pure-mechanics metaphor, LLMs’ creative writing is to art what structurally sound building materials are to architecture.

That’s even before we get to the meta level of how LLMs functionally reproduce pre-existing content via mashup of identifiable patterns. We can all agree that Liberty Leading the Masses, The Garden of Earthly Delights and The Scream are all works of art; we might struggle to convince critics (and certainly owners) that they were still artworks after we fish them out of the wood chipper.

And yet, we are now being told very seriously that a machine that can only rephrase the Internet is a great writer.

This must be true, otherwise how can we explain the following:

1. Open AI “writes” a story.
2. The story is published, in full and unedited, in one of the most respected newspapers in the world.
3. The story itself is deemed to merit not just additional news coverage by the paper’s tech editor, but also deemed so noteworthy as to deserve an entire separate hagiography of said article by a literary agent of some celebrity.

As a final thought experiment, I want you to imagine a previously unpublished (human) author, whose output is painfully middling, submitting their first ever short story to The Guardian.

What do we think the reception might be? Do you think they would be given not one, but three separate pages dedicated to their submission? That there would be zero back and forth between editor and writer, that the story they birthed would be virginally reproduced like a sacred text?

A few weeks back I wrote about how AI is the inevitable product of the kind of people who developed it — i.e. convincing sounding liars.

This AI short story is the equivalent of a TV show written and directed by a nepo baby; it is (apparently quite widely) deemed meritworthy simply for who (or what) created it. Which is obviously a benefit-of-doubt that no human other than the insanely privileged would ever be afforded.

That for me is reason enough to give AI fiction the widest birth. If you find yourself tempted to consume, I recommend you go for a walk and save the carbon. Or you could buy a book written by a person, not a monkey or a machine.

As a once promising Millenial somehow turned truculent 30-something, I refuse to watch any episode of The Simpsons past Season 9, when the greatest show of all time began its slow decline into navel-gazing charicature.

As with all (arbitrary) rules there are (equally arbitrary) exceptions, and so it was today that I found myself thinking about Homer to the Max, a Season 10 episode where Homer ends up changing his name to Max Power and adopts a hyperactive “act now think later” persona, which actually goes quite well for a while.

The particular interchange I had in my head was this:

Homer: Listen Kids, there’s three ways to do things: the right way, the wrong way, and the Max Power way.

Bart: Isn’t that the wrong way?

Homer: Yeah, but faster.

Homer’s logic here is — as with a lot of things The Simpsons’ writers came out with in their 1.000-OBP run from Season 1–9 — more on the nose than it seems at first watch.

What recalled this excellent John Swartzwelder gag was an attempt to use a Large Language Model to expedite some research I was doing on the costs of decarbonising power grids.

(Full disclosure: This is an article about Generative AI, something I am absolutely not an expert on. I investigate the fossil fuel industry. There are people far better informed and more entertaining than myself writing excellent, acerbic takedowns of Gen AI and the generalised enshittification of Big Tech offerings. I recommend Ed Zitron in particular here).

I do a lot of work you might term “Gen AI-adjacent”, including looking into the resource use of data centres, or how fossil fuel producers are harnessing AI to drill more of the oil and gas slowly boiling the planet, or how most LLMs skew in favour of industry rhetoric. But, so you know, very much not my day job.

Perhaps that’s why I have always been deeply sceptical of some of the unverifiable claims made by tech CEOs, repeated by tech beat reporters, that Generative AI is going to revolutionise the world, realise Artificial General Intelligence or somehow manifest The Singularity, ushering in a golden age of machine-led technological innovation.

It always seemed reasonably obvious to me that, given the mechanisms of LLMs themselves — that is, they are trained on large amounts of data in order to make probabilistic choices to determine the most apt response to a given prompt, and therefore never understand what they are asked or what they produce — the best they would ever be able to do is essentially rephrase the Internet.

This is obviously a gross oversimplification of LLMs, but, as I said, I’m no expert. In defence of LLMs, they are really quite good at problems that have clear, non-misinterperetable answers, like maths, writing code, or task automation. I use a LLM to tell me the statistically best move to make given the moves made up to the point I got stuck in a chess game. (Not that this makes me a better chess player).

But given LLMs never comprehend the process, their own function, or the data they are trained on and set-up to recall via prompts, I always felt intuitively that their real-world use application — far from the paradigm shift the Gen AI snakeoil suits peddle — would be pretty constrained.

So I never usually turn to a LLM for my work, which essentially investigates and reports on all the things the fossil fuel industry don’t want you to know, and which therefore entails significant personal, reputational and legal risks. I, like many people (for whom LLMs are also not really relevant) really have to be accurate in my working and output, something which LLMs are notoriously not.

But yesterday I was lazy/time poor (delete as appropriate) and thought I would give Cerebras’ new interface Inference a whirl. While Cerebras is making apparently gamechanging “wafer” silicon chips that will allow models to overcome the so-called yield problem a lot of GPU manufacturers are encountering, the interface I used runs Meta’s Llama 3.1 LLM.

I don’t know enough to comment on whether Llama is considered good or bad among the Gen AI community, but it seemed to me pretty similar in terms of set up to ChatGPT 4, which includes “chain of thought” reasoning, which is meant to result in more accurate output as the model breaks each prompt down into a series of seemingly logical steps.

I started with the prompt:

Thinking about this now, this may have been a slightly misleading question, given that the UK’s energy grid is obviously all “electrified” in the sense that it provides electricity. However this is a common shorthand way of saying “decarbonising the energy grid and ensuring all electricity comes from 100% renewable sources”.

Even with my slightly wonky phrasing, the model correctly identified that this is a bit of a “how long is a piece of string” question, where the answer depends on various moveable parts such as timelines, economic pathways and a whole load of known unknowns.

It responded:

There’s more, but I’m just including these to illustrate that this particular model is really keen on showing you its working. Eventually it returned a series of cost ranges, again dependant on a number of variables, and determined that the cost of decarbonising the UK energy system entirely would be between £75 billion and £250 billion. Although not especially helpful, at least it looked like it had reasoned correctly based off my initial (slightly janky) prompt.

I was interested in hearing more, so I asked it to provide its sources (some platforms have the model do this as it goes in any case):

It returned:

At this point, as the human in the machine, I decided to get off my figurative backside and search for these artifacts just, you know, to do some actual work and find a verifiable figure that might be useful for an upcoming bit of comms output.

After all, the citations the model gave looked legit; National Grid, Committee on Climate Change, Net Zero — these are all words I would associate with a bona fide answer to a question about decarbonising energy in Britian.

Only, I couldn’t find them. I searched extensively for several minutes for “National Grid future energy scenarios 2020 renewable energy £140–180 billion”, as well as variations. Not only was I unable to locate the specific reference, I couldn’t even turn up the document in question.

So I asked the model:

At which point it fessed up. After a good 15 minutes of human time, more than 8000 compute tokens (with whatever the associated energy use was), the model rowed back on its previous responses, and admitted:

So it hadn’t actually looked up and collated the information I asked linked to specific sources. It made it up and attributed it to plausible sounding sources.

This was something of a lightbulb moment for me. Not that I was surprised that a LLM could hallucinate (as we’ve said, models don’t understand the words they use to derive outcomes, and therefore hallucinations will always be a feature of LLMs, rather than a bug that can be fixed by Ever Bigger Models).

It’s just it was the first time I had actually seen or noticed the model’s own internal logic behind a hallucination.

The model wasn’t trying to answer my question with true, attributable arguments. It was trying to make it appear like it was.

This got me thinking. I’d like you now to picture the people who think LLMs are the business. Not the people who might use ChatGPT to recommend a good local trattoria in Venice, you know, not too touristy. Picture rather the heads of the Big Tech firms ploughing unimaginable amounts of money and literal energy into training Ever Bigger Models.

Picture the Musks, the Altmans, the Zuckerbergs, the Pichais, Amodeis and Huangs of this world. They are all grifters, right? They have all managed to convince venture capitalists, major fiduciaries, governments and huge, country-sized conglomerates such as Microsoft and Apple that Generative AI is the future, even though there is basically zero evidence to verify such a pitch.

They vary in method, but the thing these people all have in common is their ability to lie plausibly. And so, by accident or design, that’s what their models end up doing.

How do we know Gen AI gurus are grifting? Because they openly oversell the capabilities of their models, and recognise their products as the AI slop generators that they are to the extent that they ban their staff from smoking their own stash.

(For what it’s worth I don’t think the Silicon Valley genius brains developed Generative AI specifically to lie plausibly to users, it’s just an inevitable outcome of a process that values scale over specificity and speed over accuracy.

I do absolutely think that Gen AI could only take on such a mythical place in private equity circles after years of the mainstreaming of fake news and disinformation. Gen AI is in many ways disinformation’s apex power source — just as fake news needs to be reasonable sounding or looking to be weaponised, so too does the output Gen AI provides need to at least appear like an answer based on fact and verifiable research.

This is another reason why the widespread roll out of Gen AI, especially among newsrooms, is such a bad idea, because the output of plausible lies is inversely proportional to the time and effort it takes to actually check the veracity of a response; just as a lie has travelled around the world before the truth has got its pants on, so too does it take inordinately more time to factcheck and accurately source a Gen AI response than it does to produce it).

Think also of the types of firms that are adopting Gen AI assistants en masse — the KPMGs, Microsofts, Goldman Sachs, Barclays etc. These are multifaceted businesses with hugely variegated departments, but their one unifying Special Sauce is that they all Say Things With Authority.

True or not, they say them and are listened to.

So it absolutely follows that they are all up for using AI assistants to Tell Them Things, authoritatively and fast, like a mirror user, an additional identical droid in the firm.

Milquetoast details such as accuracy and veracity? Save them for the wokerati, the bean-counters. This is Big Picture stuff and I get it ok?

I’ve been trying to think of what Gen AI would be if it was a person for a while now, and yesterday I think I came close:

Generative AI is the underbriefed manager at a sales pitch. It has its talking points and is generally across all aspects of the operation, even if it doesn’t actually understand, say, how the business stays solvent, how its IP is protected or how or where its product is actually made. But it doesn’t need to, so long as it can walk into the boardroom and give plausible sounding responses.

On a BlueSky thread I posted that inspired this piece one replier said that the subtext of all prompts given to Gen AI models should be viewed not as “what is the answer to this question” but rather “what would the answer to this question look like if it was given by a human?” Generative AI can’t actually answer a query with organised, audited source material, because (again) it doesn’t know what any of that stuff actually means, even if it is trained to at least try to prioritise responses that hit the “yes” box in those criteria.

It doesn’t put out answers, it puts out stuff that resembles answers. It is answer-shaped. And, as you may have seen by the government’s unhinged plan to spaff scary amounts of investment up the wall in order to “mainline” AI into UK growth plans, Stuff That Resembles Answers clears the bar for many people, companies and institutions.

Again I want you to think of a real person — not a tech baron or an Apartheid Nepo Baby — who believes absolutely in Gen AI, the type of person who believes automating tasks shittily is inherently good

They are probably in senior management. They get briefed on most goings on in all depts, fed their lines. They believe this means they understand.

This has sufficed for their entire career. Like a CEO briefed on top lines ahead of an investor call. Like a politician at a hustings. There are set answers, sort of irrespective of the question. They don’t need to understand the prompt so long as they have learned the response.

Forever-growth capitalism and, hence, Gen AI venture capitalists view all business transactions, from M&A to HR, as a Problem to be Solved, rather than what businesses really are supposed to be about: curating the productivity of human labour.

It doesn’t matter if the “problem” is solved well, it just needs to be quick and final. Amen, I can hear Homer say.

It’s worth pointing out that Generative AI didn’t invent this kind of fake-it-til-you-make-it bluster-based approach to life — that’s been a thing for a while now, certainly since the post-truth “we’ve had enough of experts” strain of polity somehow became the one we landed on in even supposedly Serious Countries. It’s just that the biggest cheerleaders of Gen AI are also the biggest exponents of precisely this way of showing up in the world; it is not a coincidence that Sam Altman is Gen AI’s Number One Fan.

Anyway, I was thinking about Homer Simpson, about how bad Generative AI is in nearly every application where boring details such as accuracy and citation are required, and I thought inevitably of Elon Musk, who this week stood in the Oval Office looking like he had just swallowed Kendall Roy whole, and proceeded to commit crimes against the truth in front of the world’s media.

He was asked specifically about DOGE’s false claim that USAID had spent $50 million on condoms for people in Gaza. (The truth is that some condoms went to Mozambique to stop people dying of HIV/AIDS, which, you know, makes sense).

Musk’s response was basically: Well I got that wrong in point of fact but the argument is still valid and I have authority to make it.

And then it hit me: By funnelling billions of private equity into LLM training in the belief that More GPUs = Better Models, tech CEOs have essentially automated themselves.

Assumed information mastery, unearned popular acclaim, random unaccountable decisionmaking, Potemkin authority.

Gen AI’s motto should also be the truth behind the favourite maxim of the robber baron tech overlord class: Move fast and make things up.