5 days ago
Impressive. Did you also test other system one models, like D3? I tested internally, and the results show D3 is much better than Jev.
I haven't tried D3, however, because System 1 decision-making is highly task-specific, mathematically limited in output, and not reasoning-heavy by definition, I expect a fine-tuned, BERT-style model to outperform any off-the-shelf model if given enough LLM- and human-labeled data.
--
