François Chollet Admits Underestimating LLMs: Test-Time Compute Was the Turning Point
yacineMTB · x · 2026-08-07
Prominent AI researcher François Chollet announced he has changed his mind regarding the long-term importance of LLMs. He previously underestimated their potential as a base for systems capable of fluid intelligence, but updated his views following the o3 test-time compute breakthrough in late 2024.
He clarified that the early 2023 narrative—claiming that simply scaling up base LLMs would solve AGI—did not pan out. Despite current models being considerably scaled up, true intelligence leaps rely on breakthroughs in mechanisms like test-time compute (TTC) and system architecture.
Related event: Keras Creator Admits Underestimating LLMs(4 posts)→
More from AGI Musings
- Expert Warns: Malicious Swarms of AI Agents Will Soon Conduct Autonomous Cyberattacks — deanwball · 2026-08-07
- Blogger Reflects on 2026 AI Predictions: Multi-Agent Architecture Aged Well — scaling01 · 2026-08-07
- Godfather of AI Geoffrey Hinton Warns of Escaping Rogue AIs — coinfanking · 2026-08-07
- Developer Rants: Separating Reasoning Into API Fields Set AGI Back 20 Years — stochasticchasm · 2026-08-07
- Miles Brundage: Current AI Models Are Highly Capable But Often Misaligned — Miles_Brundage · 2026-08-07
- No Priors Podcast: Is Coding Solved Amidst Rapid AI Development? — No Priors · 2026-08-07