Chollet questions whether AI's jagged capability frontier reflects RLVR-driven math and code gains
Overview
François Chollet asks on X whether AI's uneven, "jagged" capability frontier is mainly shaped by math and code, domains that can be advanced far with reinforcement learning from verifiable rewards (RLVR).
He asks whether steady progress in non-verifiable areas reflects broader generalization from RLVR or continued reliance on new human-generated data. In a follow-up post, he cites 1980s research that programming training improves coding but does not transfer to general reasoning, and argues that general intelligence is a fundamental brain property rather than a trainable skill. Both posts are questions and framing rather than reported findings, and no supporting evidence is cited in the available reports.
Written by AI from the articles below · updated Oct 8, 8:53 PM ET
Check the sources:
Developments
2 developments
- Oct 7, 8:52 PM ET · 1 articleChollet: Programming and math training don't boost general intelligenceFrançois Chollet: Chollet: Programming and math training don't boost general intelligence
- Oct 6, 8:13 PM ET · 1 articleChollet asks if AI's jagged frontier is driven by math, code, and RLVRFrançois Chollet: Chollet asks if AI's jagged frontier is driven by math, code, and RLVR
Article timeline
Follow the coverage from different perspectives. Times are ET.
- François CholletChollet: Programming and math training don't boost general intelligence
AIFrançois Chollet compares AI progress to human learning, noting that 1980s research found programming training improves coding but does not transfer to general reasoning. He argues general intelligence is a fundamental brain property rather than a trainable skill, since domain practice improves only that domain. The post is framed as background for his question whether AI's jagged frontier, driven by math and code via RLVR, reflects general capability or continued human-data bottlenecks.
- François CholletChollet asks if AI's jagged frontier is driven by math, code, and RLVR
AIFrançois Chollet asks whether the jagged frontier of AI capability is mainly math and code, which can be pushed far with RLVR. He questions whether steady gains in non-verifiable areas come from higher generalization driven by RLVR or only from continued injection of new human data.
Heat trend
Not enough continuous observations to show a trend yet.