25 Fields Medalists Challenge AI Labs Over Rushed Proofs, Worldcoin (WLD) in Focus

25 Fields Medalists Push Back on AI Labs

Twenty-five winners of the Fields Medal, mathematics' highest honor, have published a manifesto accusing AI laboratories of treating problem-solving as a benchmark-chasing exercise divorced from the discipline's real goal. The statement, hosted at mathandai.org under the title “Mathematics and AI,” carries the signatures of Terence Tao, Peter Scholze, Maryna Viazovska, June Huh, Pierre Deligne and 2026 laureate Yu Deng, with award years spanning 1978 through 2026. The signatories concede that large language models have improved sharply in recent months and can now resolve major open problems across several fields. Their objection is procedural: AI-generated solutions, they write, are frequently “announced in haste,” without the careful write-up, the distillation of new methods or the citation of prior related work that converts a result into shared knowledge. Celebrated open problems, the manifesto argues, function as signposts of progress rather than finish lines, and credit assigned without community process cannot be integrated into the canon. The text names no company, but its timing is pointed. It follows OpenAI's September 9 announcement that GPT-6 Astra had produced a breakthrough on the Navier-Stokes equations — a claim several mathematicians disputed after evidence emerged that researchers involved had relied on Codex while OpenAI moved to publish first. The signatories widen the argument beyond mathematics, noting that professional training across creative fields builds understanding rather than merely delivering products, and that the two aims diverge once AI supplies the answer directly. They close by granting that AI “offers the potential to enhance and accelerate genuine mathematical research,” while stressing that the outcome depends on decisions made by whoever controls the technology. For traders, the statement lands squarely in the ecosystem behind Worldcoin's identity-token design, the Altman-linked asset most exposed to OpenAI's reputation.

Tibo Details Three Astra Fixes and a Quota Reset

OpenAI faced its own reckoning the same day, as Codex product lead Thibault Sottiaux — known in the community as Tibo — used a September 12 post to unpack the quality degradation Astra users had documented for days. He listed three root causes, all found and fixed. Skills written for earlier models were triggering too often, in some cases stopping the model from checking its own work. An opt-in context-management experiment caused the model to halt early or reply to stale messages; Sottiaux estimates it touched roughly 4,000 to 5,000 users and it has been disabled. And a set of misconfigured engines, which produced measurable quality drops on long-tail traffic, has been removed. The post closes with a reset: paid users' hourly and weekly usage limits return to full before midnight local time — a gesture the community has nicknamed the “Tibo reset.” A community-run tracker counts 29 verified resets to date, the most recent on September 3, and notes none has ever landed between 1 a.m. and 8 a.m. Pacific. The quota gesture arrives at a sensitive moment: OpenAI paused new signups for its $200 Pro plan on September 11 after Astra demand overwhelmed capacity. Measured against the 20 million active users Sottiaux cited for Codex and ChatGPT Work on August 21, the experiment's affected cohort is roughly 0.025% — the skills and engine fixes, which carry no user counts, are the ones that touch the majority. Sottiaux also promised more consistent execution, better adherence to each user's latest message and stronger in-task checks, thanking users who shared failing cases and helped the team locate the problems privately. Readers tracking the market in real time can follow live spot and futures prices on Bitget.

September 12 posthttps://x.com/thsottiaux/status/2098612714704891959?ref_src=twsrc%5Etfw

WLD as the OpenAI Sentiment Proxy

The through-line is credibility under load. In one document, 25 Fields Medalists question whether frontier AI claims survive the norms of the discipline they touch; in the other, OpenAI's own Codex chief concedes in writing that configuration errors degraded Astra's output and offers quota resets as compensation. Worldcoin (WLD) sits downstream of that credibility — our Worldcoin topic hub tracks the full thread. The Altman-co-founded identity token has historically traded as a sentiment proxy for OpenAI headlines: it slipped 3.2% when Altman signaled openness to slowing AI, and it has stayed in focus through congressional antitrust questions and a 70% extinction-risk estimate from an OpenAI safety researcher. Unlike a conventional initial coin offering, WLD's launch keyed distribution to verified humans, with positions held in a standard crypto wallet — and the Google RSI model leak rumor shows how quickly OpenAI-adjacent narratives move the token.