Urgent.News

What's breaking now, across thousands of outlets.

AI

Learning more about Claude's mathematical capabilities

Article URL: https://www.anthropic.com/research/riemann-zeta Comments URL: https://news.ycombinator.com/item?id=49247070 Points: 180 # Comments: 122

Recently, a staff member at Anthropic presented Claude, an AI model, with a challenging task related to one of the most famous unsolved problems in mathematics: the Riemann hypothesis. Despite the complexity of the task, Claude attempted it and, surprisingly, made progress on a related issue. Claude's enhanced version produced an improved lower bound for the fraction of zeros of the Riemann zeta function satisfying the Riemann hypothesis, increasing it from 41.6% to 67.2%.

Two mathematicians at Anthropic validated Claude's paper and presented a concise proof for experts. Claude also provided a formally verifiable proof of its result, which has been praised by experts such as Brian Conrey and Dan Goldston. Although Claude did not succeed in proving the Riemann hypothesis, its efforts demonstrate the rapid advancements in AI models' mathematical capabilities.

The Riemann zeta function, a critical tool in understanding prime numbers, plays a significant role in the Riemann hypothesis, a conjecture proposed in 1859 that remains unproven. Researchers have been working on related problems, such as quantifying the proportion of zeros on a specific line in the complex plane. Claude's findings build upon prior research, leveraging techniques introduced by Montgomery and others.

By combining various methodologies and tools, Claude was able to surpass the previous lower bound of 41.6% and increase it to 67.2%. The process involved extensive computational work, with Claude engaging in numerous tasks, including running Python scripts and consulting academic papers. After discovering the new lower bound, Claude subjected its result to rigorous scrutiny, involving multiple subagents, peer reviews, and independent re-proofs.

A formal verification of Claude's findings was also conducted using the Lean proof assistant. This breakthrough highlights the potential for AI models like Claude to build upon and expand upon existing mathematical ideas, potentially leading to new discoveries and insights in the field.

Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at anthropic.com →

More in AI

More from Monday 10 August →