As of: 2026-11-22 13:32 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/p/2026-10-07-fchollet-jagged-frontier-math-code/ # Chollet: is the jagged frontier "mainly math + code" pushed by RLVR? François Chollet (@fchollet), X, 2026-10-07. Source: https://x.com/fchollet/status/2107625225768858076 ## Why it matters Prominent sceptic's framing (~68k views): verifiable domains vs everything else. ## Summary Posted 00:13 UTC: "What if the jagged frontier is mainly math + code (which you can push arbitrarily far with RLVR), and everything else starts to plateau because it is still bottlenecked by human generated data?" He asks whether steady gains in non-verifiable areas come from higher general ability or only from new human data, adding "A lot of things depend on the answer". Follow-up at 00:17 UTC (https://x.com/fchollet/status/2107626212613034064): does G "emerge" from math + code RLVR? ~68k views. Not explicitly about the release, but posted in its wake and listed with it on Techmeme. ## Archived text > What if the jagged frontier is mainly math + code (which you can push arbitrarily far with RLVR), and everything else starts to plateau because it is still bottlenecked by human generated data? > > Model performance in non-verifiable areas has kept improving steadily, albeit much slower than for math and code. But is that steady improvement a side effect of a higher G (itself driven by RLVR), or only a function of the amount of new human data getting injected into training (which is still continually happening on a massive scale)? > > A lot of things depend on the answer to this question Archived 2026-10-07 via fxtwitter (unofficial). Counts: 75,755 views, 1,027 likes, 73 reposts, 105 replies (at fetch time) ## Cited in - 2026-10-06: [OpenAI releases 722 AI-written math manuscripts claiming hundreds of open problems](https://postcutoff.com/e/2026-10-06-openai-math-release-722-manuscripts/)