Researchers Debate Counterintuitive Foundations of AI Paradigms

AI researcher jdpressman argues that the pretraining-then-finetuning paradigm was counterintuitive before GPT-2, and questions whether five years of mechinterp research could be 'negative knowledge' if the polysemanticity assumption fails.

2026-10-16 ~ 2026-10-16 · 2 related posts

• Researcher: the foundation-model paradigm is still counterintuitive outside AI — jd_pressman · 2026-10-16
• Researchers debate whether five years of mech interp work amounted to negative knowledge — jd_pressman · 2026-10-16