OpenAI's Math Coup Shows Where AI Earns Its Keep

There is a useful diagnostic buried in this week's AI news cycle. On one side, OpenAI used an internal build of its next major model to crack ten mathematical problems that had resisted progress for at least a decade, each for under $2,000 in compute costs. On the other side, Sam Altman was publicly mocked for suggesting ChatGPT should narrate personalised podcasts about your child's activities during the morning school run, rather than, say, talking to your child. Gravity Falls creator Alex Hirsch's reply — essentially, "why not just talk to your kid?" — collected 122,000 likes against Altman's 9,600. The market has spoken.

The contrast is not coincidental. It maps almost precisely onto the fault line that Greg Brockman identified this week: people object when ChatGPT integrated into Slack reaches out to their colleagues on their behalf, even if those colleagues would happily help if asked directly. It is not the outcome they resist; it is the mediation. AI earns its keep when it handles work that humans find genuinely tedious, technically demanding, or beyond their capacity. It irritates people when it elbows into relationships that carry social and emotional meaning.

What the Maths Results Actually Demonstrate

The ten mathematics results deserve more attention than they have received. OpenAI published Lean 4 formalizations — machine-checkable proofs in a formal verification language — alongside conventional papers and LLM-generated PDFs reconstructing reasoning from model traces. The problems span geometry, cryptography, and complexity theory. These are not benchmark puzzles; they are open research questions that human mathematicians had not resolved in over ten years.

The framing matters. Mathematician Terence Tao has described the vision as