A beginning for mathematics
proofsandprompts.com ● Covered by 21 sources
Opinion — commentary, not a factual news event.
AI can now solve major math questions on its own. That may explode math output, but it could also break the old way the field measures understanding.
Based on reporting by proofsandprompts.com — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Three years ago, AI systems couldn’t reliably add two numbers. A year ago, internal models at OpenAI and DeepMind were already hitting the equivalent of a gold-medal score on the IMO. Now, Daniel Litt says they are autonomously resolving major open questions. He does not think that pace is likely to stop soon, and he thinks mathematics will have to change around it.
Litt, a professor at the University of Toronto, has been arguing that the scary part is not that machines might “solve” math. It’s that they could flood the field with results while human understanding stalls. That is the future he warned about in a talk called The End of Mathematics. In this essay, he pushes a more hopeful version: AI can take over more of the production side, while human mathematicians lean harder into understanding, explanation, and judgment.
His target is the profession’s habit of treating proofs and theorems as the main output. That used to work because a proof was also a signal of expertise. He says that signal is breaking. A PhD thesis can now be produced without the student even reading it, and the text itself may no longer tell you much about the person who wrote it. So he wants the degree to shift toward rigorous oral defense, deep topic mastery, and the ability to explain work until examiners are satisfied.
He also wants more weight on seminars, sustained discussion, and hiring and admissions processes that reward what AI still struggles to replace: internal understanding and social-relational skill. Papers, he argues, will matter less as proof of understanding; live explanation will matter more. And as models get better at exposition too, that gap could widen further.
The bigger point is blunt. Litt thinks people will keep pressing a button and getting mathematics back, whether academics like it or not. So the real task is to build a profession that can absorb that output, judge what is interesting, and keep turning new results into human knowledge. If that sounds like a mess, he’s basically saying the old system already was one.
My take — AI-written commentary, not fact-checked reporting
This is the right fight: stop worshipping the PDF and start rewarding actual understanding. The academy has spent years pretending its bottlenecks were sacred; AI is just the first thing rude enough to show how much of that was ceremony with footnotes.
Read more about this at: proofsandprompts.com