Urgent.News

What's breaking now, across thousands of outlets.

AI

OpenAI just dropped 700 preprints of mathematical proofs and counterexamples

OpenAI has released a collection of mathematical manuscripts and supporting proof artifacts generated by an internal model. This repository consists of 722 manuscripts organized into 372 families, each categorized by mathematical discipline. The models were evaluated on open research problems, with performance on existing mathematical evaluations reaching a saturation point that prompted the expansion of these evaluations.

The model was posed approximately 4,000 problems during the evaluation process, with each result utilizing an average of three hours of ChatGPT Pro thinking compute. A significant number of results were obtained using the same procedure with the unreleased internal OpenAI model. However, some exceptional cases, such as the zero-free region for the Riemann zeta function and the proof of the Hodge Conjecture for CM abelian varieties, did not follow the fixed procedure.

These works were human-edited for readability, specifically the writeup for the Re(s) ≥ 11/12 zero-free region for the Riemann zeta function.

Not all manuscripts in this collection have accompanying Lean formalizations, which are formal proofs in the Lean theorem prover. The collection will continue to be updated with these formalizations as they become available. Some unformalized results might contain issues, which the team aims to address promptly. The repository will also explore community-hosted repositories for these materials.

The catalogue of manuscripts and families generated by the model was compiled by aggregating the output into result families and manuscripts, requiring an appropriate level of significance. Exceptions to the fixed procedure include the aforementioned zero-free region and Hodge Conjecture cases. The public release history of this collection will be preserved, with corrections and revisions recorded as new versions while retaining access to previously released versions.

To cite individual manuscripts, users can utilize the BibTeX block provided in their respective directories.

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

This story

This is one outlet's version. Read the fullest account.

Read the original at github.com →

More in AI

Your AI agent doesn't need fewer permissions. It needs a mission.

If you've been on dev Twitter or Hacker News this month, you've seen the stories. An agent asked to clean up a folder deletes tens of thousands of files.

  • AI agents misbehave due to lack of mission alignment, not permission issues.
  • Watchdog monitors agent actions against one-line mission statements.
  • Shield protects websites from abusive AI crawlers and bots without captcha.

More from Tuesday 6 October →