Keras Creator Admits Underestimating LLMs, Cites o3 Breakthrough as Turning Point
khademinori · x · 2026-08-07
François Chollet tweeted to clarify that he underestimated the long-term importance of LLMs back in early 2023. He admitted he initially didn't see LLMs working as a base for systems capable of fluid intelligence, and only changed his mind in late 2024 following the o3 test-time compute breakthrough. However, he emphasized that the early 2023 narrative—that merely scaling up base LLMs would solve AGI—did not pan out. New ideas and capabilities like TTC (test-time compute) and Harnesses are crucial to recent breakthroughs.
More from AGI Musings
- Analogy: Children are better suited than adults for discussing AI instruction generalization — 1a3orn · 2026-08-26
- Using AI models today feels like downloading MP3s on dial-up in 1999 — Daniel_Farinax · 2026-08-26
- Paper: Automating entry-level jobs may shrink long-term GDP by blocking expertise — soumitrashukla9 · 2026-08-26
- Diamandis: Intelligence is becoming a commodity, value shifts to apps — PeterDiamandis · 2026-08-26
- The Loop is the Product: Stanford and Sequoia Agree on Agent Value — tool_call_traces · 2026-08-26
- Aphorisms on Agent Naming and SOP Formats — KirkNewcombe · 2026-08-26