Keras Founder François Chollet: Future AI Will Shift from LLMs to Symbolic Learning

burny_tech · x · 2026-08-08

Keras founder François Chollet shared his updated views on the scaling of Large Language Models (LLMs). He previously predicted a capability plateau for base LLMs during the 2022-2024 era but changed his stance after seeing the o3 test-time compute demo in late 2024. The new models demonstrated genuine fluid intelligence, suggesting that this new line of work could enable unbounded capability scaling for LLMs—meaning "there will be no wall."

However, looking ahead 15 years, he still does not believe future AI will rely on the current LLM stack. Instead, it will necessarily evolve toward its optimal, final form: symbolic learning. Acknowledging this as a risky and contrarian belief, he argued that current techniques are 4 to 6 orders of magnitude away from optimal efficiency, with the core difference being efficiency rather than task-specific skill.

Related event: Keras Creator Admits Underestimating LLMs After o3 Breakthrough(5 posts)→

Original post →

More from AGI Musings

AGI Musings channel →