François Fleuret: autoregressive process and reasoning are two different conceptual levels
francoisfleuret · x · 2026-09-26
Reacting to the Nate Silver debate on whether LLMs 'just predict language', researcher François Fleuret argues people conflate two distinct conceptual levels: the autoregressive process and reasoning itself. The quoted thread (1/7) pushes back on reducing LLMs and human thinking to the same mechanism.
Related event: Meta's Fleuret Debunks 'LLMs Just Predict the Next Token' in 7-Tweet Thread(7 posts)→
More from AGI Musings
- Bindu Reddy backs Jensen Huang: accelerate AI, doomers are dead wrong, 0% extinction risk — bindureddy · 2026-09-26
- Physicists scooped by Anthropic AI: 'more low-hanging fruit than experts expect' — soumitrashukla9 · 2026-09-26
- "LLMs can't think" is the unscientific claim, argues Plinz in debate over burden of proof — AlexTensor · 2026-09-26
- AI researcher: it's a good thing when problems you've worked on for years get solved — rajammanabrolu · 2026-09-26
- Connectome's layered AI-companion memory goes viral, sparking AI life ethics debate — repligate · 2026-09-26
- Alignment take: today's weak misalignments mean we're not on the worst timeline — repligate · 2026-09-26