Goodhart's Law Hits the Professions: "AI is breaking our proxies for expertise"
Sean Goedecke's essay "AI is breaking our proxies for expertise" argued that AI output is overwhelming the signals professionals use to distinguish expertise from its imitation, opening from the mathematicians' declaration A Severe Misalignment of AI in Mathematics. The October front page turned that abstraction into a concrete, still-developing story: OpenAI dropped a batch of competition-level mathematical results, and the reaction escalated from shock to technical debunking within days.
The pieces:
- The Verge: "'Pure insanity': Mathematicians react to OpenAI's math results drop" — mathematicians blindsided by a corporate press-drop of theorem-level results.
- The Conversation: "Is this the mathocalypse? Why OpenAI's latest results dump has left mathematicians in shock". The thread's load-bearing comment quotes the piece on what happens when a mathematician actually reads one of the accompanying papers:
A senior colleague of mine found one of his favourite problems among those solved and attempted to read the accompanying paper. He told me it was so unintelligible that, had he received it as an editor at a mathematics journal, "it would have gone straight into the bin". (ColinWright)
- New Scientist: "OpenAI mistranslated mathematics into code for its Navier-Stokes proof" (57 points) — the dump's flagship result is already reported to have a formal-math-to-code gap.
- The community is doing post-hoc auditing in public: "Show HN: Reducing an OpenAI proof by 25% in a few hours".
What this means: the verification bottleneck is now the story. Results arrive via press release, the papers fail journal norms on readability, the formal math mistranslates into the code that produced it, and the burden of checking falls on an ad-hoc volunteer network working in GitHub repos. The expertise proxy doesn't just break — it inverts, with the audience doing more rigor than the publisher. Adjacent governance data point in the same window: OpenAI fired 3 safety researchers in a dispute over AI risks (AP), and the fired researchers asked the company to preserve visibility into AI reasoning (WSJ) — the same visibility-and-verification fight, moved inside the lab.