François Chollet

François Chollet (born 1989) is a French software engineer and AI researcher who created Keras in 2015 — one of the most widely used deep-learning frameworks in the world — and spent nearly a decade at Google before leaving in late 2024 to co-found the research lab Ndea and the ARC Prize Foundation. His 2019 paper On the Measure of Intelligence, and the ARC benchmark it introduced, reframed how the field tests for genuine reasoning rather than memorization.

Against the hype — and the doom

Chollet occupies an unusual position in the AI debate: he built tools that millions of practitioners use, yet he is one of the sharpest critics of the claim that scaling language models leads to general intelligence. On the Dwarkesh Podcast in 2024 he argued that the LLM gold rush sucked the oxygen out of frontier research, saying OpenAI had set back progress towards AGI by five to ten years, and that benchmarks the industry celebrates mostly measure memorization, not intelligence.

He is just as dismissive of catastrophizing. His 2017 essay The Impossibility of Intelligence Explosion argued that a runaway self-improving superintelligence misunderstands what intelligence is — situational, embodied in civilization, and advancing on “a visibly linear ladder of progress.” The essay drew detailed rebuttals from Eliezer Yudkowsky and the AI-safety community, a debate that still frames the disagreement today.

With the million-dollar ARC Prize, co-founded with Zapier’s Mike Knoop, Chollet put money behind his definition of progress: systems that can adapt to genuinely novel problems. Notably, his skepticism has limits — as models began making real progress on ARC, he publicly shortened his own AGI timeline, saying he follows the evidence rather than a side.