Fleuret: Generating Intermediate Steps Is 'Very Close to Actual Reasoning'
francoisfleuret · x · 2026-09-26
Closing tweets (6-7) of François Fleuret's thread: a model may struggle to generate the answer directly, but it can generate intermediate sequences covering intermediate arguments and results — small steps that can be produced one at a time.
That reframes 'it just predicts the next word' as 'it generates intermediate results that progressively lead to a valid answer', which he says sounds very close to actual reasoning and intelligence.
Related event: François Fleuret's Thread: LLMs Are More Than 'Next-Token Predictors'(6 posts)→
More from AGI Musings
- AI-drafted 166-page Navier-Stokes proof is correct but nearly unreadable for humans — Pascallisch · 2026-09-26
- Debate: LLMs are made of human writing but experience time, identity and death differently — qualadder · 2026-09-26
- Leo Gao on AI views: boosting productivity and doomerism aren't mutually exclusive — nabla_theta · 2026-09-26
- Two years in, this practitioner says truly unattended AI content is a myth — Cold_Hall_5384 · 2026-09-26
- Dean Ball mocks 'AI can't think' posters as Bluesky debate fatigue spreads — moultano · 2026-09-26
- Neuroscientist Anil Seth's TED talk: today's AI can't and won't become conscious — anilkseth · 2026-09-26