Countering the 'slow math takeoff' claim: without preregistration the AI progress curve misleads
Jsevillamol · x · 2026-10-07
Responding to claims that AI's math takeoff looks slow, Jsevillamol argues the issue is a lack of preregistration plus an abundance of unsolved problems of similar difficulty, where the count is mediated by how much effort OpenAI puts into solving them. With a preregistered problem set, the curve looks much smoother.
He adds that claims "90 problems haven't been solved" are plainly false — the takeoff is slow, but the data holds. The thread, citing slatestarcodex, is a substantive methodological debate on measuring AI research progress.
Related event: OpenAI's Math Breakthrough Sparks Awe, Skepticism and Career Anxiety(46 posts)→
More from AGI Musings
- Dario Amodei's 2016 AI safety paper resurfaces: thinking about safety pre-Transformer — bookwormengr · 2026-10-07
- Mathematician littmath: establishing what is true is itself a substantial contribution — joshgans · 2026-10-07
- Delivering personal agents will be far harder than any prior AI product category — nickbaumann_ · 2026-10-07
- The scaling debate hinges on two readings of "predictably better with scale" — Diyi_Yang · 2026-10-07
- Coinbase Institute paper: AI agents are the internet's second customer, and stablecoins fit micro-payments — kleffew94 · 2026-10-07
- Does RSI help with data? Frontier model progress may hinge on data, not compute — maxsloef · 2026-10-07