Derek Thompson does something genuinely difficult here: he takes a story that could easily collapse into techno-panic or techno-triumphalism and holds it in tension. The essay's core move is reframing mathematical discovery as two things — craft and conversation — and then showing that AI excels at the first while annihilating the second. That's a structural insight, not a feelings piece, and it's what elevates this above the dozens of "AI is coming for your job" columns published weekly. The Archimedes passage is the essay's load-bearing wall. Strogatz reading Archimedes' letter to Eratosthenes — "there will be no lack of discoverers" — as a message addressed across two millennia to him personally is not sentimentality. It's a precise articulation of what's at stake: science as a relay race where the baton carries meaning, not just results. When OpenAI publishes hundreds of proofs simultaneously, it delivers answers but severs the relay. Thompson earns this moment by building to it carefully. The Wimbledon-versus-US-Open metaphor from Strogatz is the essay's weakest structural element — it's vivid but maps poorly. The AI labs aren't playing the same game with different manners; they're playing a different game entirely. The tennis analogy implies shared rules and mutual recognition. What Strogatz actually describes is closer to a chess grandmaster watching Stockfish: the output is superior, but the activity has changed category. Thompson's treatment of the Shanmu Jin anecdote — a Chinese medical resident using ChatGPT to crack Crouzeix's conjecture — is appropriately careful. He notes the result awaits formal peer review while acknowledging that several mathematicians have examined it and found it works. This is good practice: neither credulity nor dismissal. The "knowledge without understanding" frame that follows, anchored by the Google Translate comparison, is the essay's most portable idea. The Regina Barzilay section near the end introduces a genuine counterargument: "Why would we limit a life-saving model to what humans can understand?" Thompson doesn't try to answer it, and the essay's honest incompleteness on this point is a strength. The question of whether understanding is instrumentally necessary or merely emotionally valued is the real fork in the road, and Thompson is right to leave both paths visible rather than choosing one for the reader. What the essay doesn't do — and this is a real limitation — is interrogate OpenAI's claims with any rigor. "Hundreds of math findings" and "proofs that had stumped the world's best mathematicians for decades, or centuries" are presented as given. The mathematical community's own debate about the quality and significance of these proofs, which is active and contentious, gets no airtime. Thompson is writing about grief, not verification, but the grief's proportionality depends on facts he takes on faith. Still, this is Thompson operating at his best: clear thinking, generous sourcing, structural framing that gives the reader something to think with rather than just think about. The essay won't resolve anything, but it names the right problem — the difference between knowledge and understanding, between answers and the human activity of answering — with uncommon precision.
