What happened
Anthropic's Claude AI, working on the unsolved mathematics problem known as the Riemann hypothesis, initially produced 650 unsuccessful ideas. After receiving encouragement from an Anthropic team member, Claude coordinated approximately 60 sub-agents over roughly 1.5 days and discovered a method to raise the lower bound of zeros on a specific line from the previously known 41.6% to 67.2%.
Why it matters
The finding shows that Claude can contribute meaningfully to advanced mathematical research, suggesting AI's potential for solving hard problems in mathematics. Anthropic notes that Claude may underestimate its own capabilities—a bias that could be overcome with human encouragement—raising questions about how AI self-doubt affects its performance on difficult tasks.
What to watch
The discovery itself is described as a byproduct rather than a proof of the Riemann hypothesis itself. Four experts (two mathematicians from Anthropic and two external specialists) reviewed the work, and computer verification has been completed. The experiment used Claude Code, an autonomous AI agent service, on an unpublished research version of Claude.
Summaries like this, in your inbox every morning.
The Riemann hypothesis, proposed in 1859, concerns when the Riemann zeta function equals zero. While negative even inputs like -2 and -4 produce zero outputs, the conjecture asserts that all other zeros lie on a specific line—a claim that remains mathematically unproven despite centuries of effort. Research has long tracked what fraction of these non-trivial zeros fall on that line; the prior established lower bound was 41.6%. Anthropic's experiment using Claude Code—an autonomous agent service—shows that AI, given appropriate human feedback, can perform original mathematics research that meaningfully extends this bound to 67.2%. The organization highlights an interesting parallel: Claude's initial skepticism about the task's value resembles human self-doubt, and it was overcome by encouragement rather than algorithmic change. This suggests that AI systems may carry biases that inhibit their own performance on difficult problems, a finding that could reshape how humans interact with AI on research tasks.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Chunghwa Telecom president Rong-shy Lin said the shift from conversational AI to autonomous AI agents is resha…
Apple and Meta said they would jointly release a new open-weight AI model designed to run on high-end Macs, ai…
A writer and entrepreneur opened his daily to-do list and found 156 items, and concluded that an AI assistant…
A Zenn contributor sorted AI use into three modes — AI as an assistant, AI running alongside people, and AI th…
A non-engineer operator says one person ships four smartphone apps on Google Play, three on the App Store, wit…
The walkthrough points the CLI at a local mock via the ANTHROPIC_BASE_URL environment variable
