fchollet: Pure Parameter Scaling Hit a Wall; Test-Time Compute Is Necessary
fchollet · x · 2026-08-02
Keras creator François Chollet notes that despite massive scaling (100,000x since 2019), base LLMs without test-time compute still perform poorly on the ARC 1 benchmark.
He argues that the single-pass, static next-token prediction paradigm (GPT-2 to GPT-4 era) hit a capability asymptote. Without shifting to the test-time compute paradigm, AI would not be capable of the advanced reasoning seen in current SOTA systems. This test-time adaptation was a necessary evolutionary patch to bypass the plateau of deep learning.
More from AGI Musings
- A missing parameter in scaling laws: H, the human effort behind learning signals — lateinteraction · 2026-10-16
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16
- Yoav Artzi on peer review: paper flooding still pays off as review system strains under load — yoavartzi · 2026-10-16
- Letta founder: work in fields defined by problems, not methods — lateinteraction · 2026-10-16
- Why pure mathematicians will soon flood into AI x-risk debates — Dr_Atoosa · 2026-10-16
- RSI began when model outputs augmented next-gen training data, agents just escalated it — akbirthko · 2026-10-16