Tao: Open math problems being non-renewably mined by AI
- ID
- 22632
- Status
- summarized
- Published
- 09 Sep 2026, 5:00 AM
- Fetched
- 11 Sep 2026, 5:45 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://mathstodon.xyz/@tao/117237320796901560
- Source URL
- https://hnrss.org/best
Summary
- Score
- 6.0
- Created
- 11 Sep 2026, 6:52 AM
- Tags
- Audience
- ai-ml-learnersdeveloperssaas-startup-founders
What happened
Terence Tao argues that AI is mining good, fruitful open math problems in a non-renewable way, potentially making them scarce. He draws an analogy to a country surrounded by ocean but short on drinking water: you can generate infinite problems trivially (e.g., the 10^10^10th digit of pi), but most are worthless—too easy, too impossible, or lacking connections to deeper insights.
Why it matters
The same 'non-renewable mining' dynamic applies to AI benchmarks, coding challenges, and eval datasets that builders rely on—once AI saturates them, they lose signal. If you build evals or benchmarks, consider whether your test set is already contaminated or trivially solvable by current models, and design harder, more novel evaluations rather than reusing public problems.
Discussion angle
Does the 'non-renewable problem' framing apply to your own eval datasets or coding interview pipelines—are you testing on problems that current models have already seen or can brute-force, and what would a genuinely renewable eval look like?