What's this about?
Terence Tao argues that mathematics has infinitely many possible questions but only a limited stock of fruitful, appropriately difficult open problems. Mass AI solving may consume those problems before humans extract the explanatory ideas and new directions that traditionally come from working on them.
What exactly is a “difficulty landscape”?
It is a field's shared map of what is easy, reachable with effort, or beyond present techniques. Good problems sit near informative boundaries. A tool that flattens everything inside its range can solve more while making that map harder to see.
What's the community saying?
HN 164 points and 108 comments.
- Some argued explanations and reverse-engineering AI proofs can still generate unlimited insight.
- Others agreed that identifying worthwhile questions, not answering them, becomes the scarce skill.
- Several compared AI mathematics with chess engines: solved positions did not end human study, but changed its purpose.
- Concern centred on rumours triggering huge compute races that scoop slower human projects before publication.