The avalanche that hit math is coming for AI development
Reported by Hacker News1 source ↓
The essay’s author argues that rapid AI progress on verifiable math problems could lead to AI systems automating AI research, accelerating advances across other fields. The author estimates this could happen within months to a few years, while acknowledging uncertainty.
Key points
- The author says AI systems may match top human AI researchers within months or a few years
- METR attributed much of Anthropic’s reported twofold increase in progress on well-scoped tasks to AI use in R&D
- Anthropic said Claude led more than 26% of its AI R&D work in August, up from under 1% in February
- The author says current AI systems are better at solving defined problems than posing new research questions
Details
The essay argues that AI systems’ recent performance on verifiable math problems may signal progress toward automating parts of AI research. Its author says that matching top human AI researchers could be months to a few years away, and projects that broader expert-level capabilities could follow by 2030 or possibly sooner. These are the author’s estimates, not established outcomes.
To support the case for AI-assisted progress, the essay cites METR’s assessment that Anthropic’s rate of progress on well-scoped tasks was twice what it had been a year earlier, with most of the acceleration likely due to AI use in research and development. It also cites OpenAI’s report that its most intensive 10% of researcher users increased daily spending on internal coding agents from about $1,000 at the start of summer to more than $7,000 by the end of August.
The author describes AI systems as solving increasingly difficult math problems, including a Millennium Prize problem, while noting that current systems remain stronger at solving well-defined questions than at proposing powerful new ones. The essay argues that this distinction may matter less in AI research, where labs often value methods that improve training efficiency or model capabilities over theory for its own sake.
Written from coverage as of
Images
- Hacker News
- Hacker News
- Hacker News
- Hacker News
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh