Terence Tao's "Math 2.0" lecture: AI has ended proof scarcity, and optimizing for solved problems is now harming mathematics

Why it matters

【Key points】

Terence Tao gave a public lecture, "Math 2.0," at Caltech on the evening of Oct 9 (Pacific) and posted the 27 slides on his blog the next day.

His argument: mathematics ran for a century on proof scarcity; because proofs are objectively verifiable, digital and data-rich, frontier labs prioritized math, and "we are now entering an era of proof abundance." Social media shows only the problems AI solves, while most frontier problems stay unsolved even with massive compute, so he calls for assessments that report resource use and negative results. Problem solving was always a proxy for deeper goals; it is now misaligned, and "further blind optimization of problem-solving alone is now actively harmful." His alternatives include AI-found explanations of opaque proofs, living networks of results, Mathlib, and interpretable open math models, some "to be released very soon."

【Verifiable】

Blog post and slide PDF are linked here. The most debated slide is a thought experiment: an AI-found cancer cocktail whose model is verified in Lean and passes a phase 3 trial; would you want one human expert to understand it first? Andrew Curran's rebuttal drew about 1.2M views and 579 quotes; related X posts run about 280 in 24 hours, still 17 to 27 an hour.

【Our read】

This is more actionable than the AHM boycott statement. DeepMind reported 9 of 353 Erdős problems; OpenAI reported about 4,000 problems for 722 manuscripts at ~3 hours of Pro compute each. Denominators exist; failed-attempt detail does not. Whoever publishes failures and costs earns credibility. Critics read the cancer slide too narrowly: Lean checks a proof against its formal statement, not whether the statement matches reality, so an exploited trial flaw passes every machine check. Human understanding is the audit layer. Watch who ships the open math models, and whether the next lab math release includes failure rates.