OpenAI Releases Results on Hundreds of Math Problems, Drawing Math Community Scrutiny
In short
OpenAI released findings on hundreds of mathematics problems spanning algebra, number theory, theoretical computer science, mathematical logic and topology. The New York Times and The Boston Globe reported 377 problems, while Engadget said OpenAI posted 722 manuscripts covering 372 problems to GitHub. The results came from an unreleased, more advanced AI model, and follow last month's disputed Navier-Stokes announcement.
Read the full story 2 min read
OpenAI released findings on hundreds of mathematics problems, a move The Boston Globe said deluged mathematicians with results spanning algebra, number theory, theoretical computer science, mathematical logic and topology. The outlets gave different counts: The New York Times and The Boston Globe reported 377 problems, while Engadget said OpenAI posted 722 manuscripts covering 372 major math problems on GitHub. [ 1 , 2 , 3 ]
Like last month's result on the Navier-Stokes equation, the new solutions used a more advanced AI model that has not been released publicly, according to The Boston Globe and Engadget. OpenAI said the average result took about three hours of computing, The New York Times and Engadget reported. The company included the model's reasoning, compute estimates and information about the number of problems attempted, according to Engadget, and for 10 of the findings it provided summaries of how the model came up with its solution, The New York Times reported. Engadget said OpenAI did not release specific compute times for problems or which prompts it used. [ 1 , 2 , 3 ]
The release follows OpenAI's announcement last month that it had cracked the Navier-Stokes equation, one of the Millennium Problems, The Boston Globe reported. That achievement, The Boston Globe said, raised questions over whether OpenAI's systems were employing creative thinking or simply completing the final steps of a proof after borrowing ideas from human mathematicians' work. Engadget said OpenAI stated last month that its models had resolved more than 100 long-standing open problems across most areas of mathematics. [ 2 , 3 ]
OpenAI has been working with an advisory board of mathematicians on how AI companies should communicate AI-produced results, The Boston Globe reported. The group is hosted by the Institute for Advanced Study in Princeton, New Jersey, and is independent from any AI company, according to The Boston Globe. The board recommended that all prompts to the AI agents and the agents' chains of thought be released, and on Tuesday night it released a statement calling the public release “the beginning, not the completion, of the process of human understanding and the incorporation of the work into mathematical knowledge,” the newspaper reported. Melanie Wood, a mathematician, said: “We want to create standards and practices so that results released from AI labs can be understood by mathematicians and can advance the field.” [ 2 ]
Engadget reported that OpenAI said it followed guidelines from its independent Advisory Group on Mathematics and Artificial Intelligence, including prompt release of results through traditional academic channels, the name of the model used, prompts and compute costs. OpenAI said it published the results in a GitHub repository with protocols for paper revisions and citations and that it is exploring other community-hosted alternatives, according to Engadget. Among the claimed results Engadget listed were a solution to the four-dimensional Kakeya conjecture, improvements to key computer algorithms and progress toward the Riemann hypothesis. [ 3 ]
Why it matters
The release deepens a debate in mathematics over how AI-produced results should be verified and credited, an issue the sources say is roiling the field. OpenAI published with guidelines from an independent advisory group of mathematicians, which the Boston Globe says was set up to advise AI companies on communicating such results. The documents do not give any other basis for significance beyond the mathematics community's reaction.
Key facts
- OpenAI released findings spanning algebra, number theory, theoretical computer science, mathematical logic and topology, according to The Boston Globe. [ 2 ]
- The New York Times and The Boston Globe reported 377 problems; Engadget reported 722 manuscripts covering 372 problems posted to GitHub. [ 1 , 2 , 3 ]
- For 10 of the findings, OpenAI provided summaries of how the AI model came up with its solution, The New York Times reported. [ 1 ]
- The average result took about three hours of computing, the company said. [ 1 , 3 ]
- The solutions used a more advanced AI model that has not been released publicly. [ 2 , 3 ]
- The advisory group of mathematicians is hosted by the Institute for Advanced Study in Princeton, New Jersey, and is independent from any AI company, The Boston Globe reported. [ 2 ]
- Engadget reported OpenAI did not release specific compute times for problems or which prompts it used. [ 3 ]
- Engadget said claimed results include a solution to the four-dimensional Kakeya conjecture, improvements to key computer algorithms and progress toward the Riemann hypothesis. [ 3 ]
Confirmed by several sources
- OpenAI released findings on hundreds of mathematics problems. [ 1 , 2 , 3 ]
- The results were produced by an unreleased, more advanced AI model. [ 2 , 3 ]
- OpenAI said the average result took about three hours of computing. [ 1 , 3 ]
- The release follows OpenAI's announcement last month that it had cracked the Navier-Stokes equation. [ 2 , 3 ]
- An independent advisory group of mathematicians advised OpenAI on communicating AI-produced results. [ 2 , 3 ]
Still unclear
- The number of problems covered by the release. The New York Times and The Boston Globe say 377, while Engadget says hundreds of solutions and progress toward 372 problems in 722 manuscripts.
- Whether OpenAI released all the prompts and chains of thought the advisory group recommended. Engadget reports OpenAI did not release which prompts it used or specific compute times, while the advisory group recommended all prompts to the AI agents and the agents' chains of thought be released, according to The Boston Globe.
- Whether the AI systems used creative thinking or completed the final steps of proofs after borrowing ideas from human mathematicians' work. The Boston Globe says the earlier achievement raised this question; the documents do not report it being resolved.
What local media are saying
Timeline, local time
- The New York Times reports OpenAI released findings on 377 math problems, characterizing the release as more responsive to concerns about AI disrupting research. [ 1 ]
- The Boston Globe reports the 377 results and that the advisory board issued a statement on Tuesday night calling the public release “the beginning, not the completion, of the process of human understanding.” [ 2 ]
- Engadget reports OpenAI posted 722 manuscripts covering 372 problems on GitHub, using an unreleased model. [ 3 ]