OpenAI Says Its Model Has Solved More Than 100 Math Problems, in Collaboration With a Panel of Math Advisors
On September 21, OpenAI announced it will work with an independent advisory panel of mathematicians, which the mathematicians formed themselves, to provide guidance on how AI-related mathematical results should be reviewed and released. According to an OpenAI statement, this “Mathematics and Artificial Intelligence Advisory Panel” has nine members in total. The members are unpaid and may publicly share their views. The announcement also said that the company’s internal model, trained beginning August 28, has solved more than 100 long-standing mathematical problems that had remained unsolved, in addition to the Navier-Stokes problem. The panel’s nine members are unpaid and may proactively offer advice that OpenAI has not asked for. OpenAI said the panel will help assess the importance of new mathematical results, suggest how to coordinate the release of findings, and provide input on the academic and professional standards of mathematical research. The panel will also offer advice on how OpenAI’s tools can support mathematical research and learning.
The announcement emphasized that the panel operates independently of OpenAI. It can offer advice that OpenAI has not been asked for, comment on OpenAI’s impact on the math community, and publish its recommendations. The members are not paid by OpenAI, and the list can be adjusted by the panel itself. However, OpenAI also stated that the panel is not responsible for the pace at which OpenAI advances its internal mathematical research.
The nine members are:
François Charles (École Normale Supérieure PSL—Paris)
Camillo De Lellis (Institute for Advanced Study, Princeton)
Timothy Gowers (École Normale Supérieure de Lyon, Cambridge University)
Martin Hairer (Swiss Federal Institute of Technology Lausanne, Imperial College London)
Nikhil Srivastava (University of California, Berkeley)
Ulrike Tillmann (University of Oxford)
Ravi Vakil (Stanford University)
Edward Witten (Institute for Advanced Study, Princeton)
Melanie Matchett Wood (Harvard University)
OpenAI: The internal model trained beginning August 28 has solved more than 100 unresolved problems
At the beginning of its announcement, OpenAI wrote that the company began training a new internal model on August 28. In addition to the Navier-Stokes Millennium Prize problem, the model has solved more than 100 long-unsolved problems spanning most areas of mathematics. The pace of progress surprised even OpenAI’s internal mathematicians, and also sparked internal discussions about how to ensure the broader mathematical community becomes aware of this rapid advancement so the field can prepare and adapt.
When OpenAI released its Navier-Stokes results on September 9, it sparked debate over research ethics and credit after a New York University mathematician, Tristan Buckmaster, questioned whether OpenAI had gotten ahead of his work after learning about his research progress.
In response to an open letter signed by Fields Medal winners, Martin Hairer—who is also a signatory—OpenAI directly quoted from the letter published by the mathematicians on mathandai.org. OpenAI said the mathematicians expressed concern about the negative externalities created by “using the resolution of unsolved problems as a benchmark for evaluating a new AI system,” and said these criticisms highlight the need for AI companies to interact carefully with the mathematics community.
The statement, signed by multiple Fields Medal winners and reported on September 12, criticized that answers produced by AI are often announced hastily, without time to properly write up and refine new methods and ideas, and without citing relevant prior research. This has raised concerns about credit and plagiarism. Martin Hairer, one of the panel members and a 2014 Fields Medal winner, is also listed as a signatory to the statement.
OpenAI described this collaboration as “a first step,” and said it hopes to place mathematicians at the center—deciding how AI can support understanding in mathematics and how the results can benefit a wider community.
This article, in which OpenAI says the model has solved more than a hundred math problems and that it worked with a panel of mathematician advisors to review the results, first appeared on .
On September 21, OpenAI announced it will work with an independent advisory panel of mathematicians, which the mathematicians formed themselves, to provide guidance on how AI-related mathematical results should be reviewed and released. According to an OpenAI statement, this “Mathematics and Artificial Intelligence Advisory Panel” has nine members in total. The members are unpaid and may publicly share their views. The announcement also said that the company’s internal model, trained beginning August 28, has solved more than 100 long-standing mathematical problems that had remained unsolved, in addition to the Navier-Stokes problem. The panel’s nine members are unpaid and may proactively offer advice that OpenAI has not asked for. OpenAI said the panel will help assess the importance of new mathematical results, suggest how to coordinate the release of findings, and provide input on the academic and professional standards of mathematical research. The panel will also offer advice on how OpenAI’s tools can support mathematical research and learning.
The announcement emphasized that the panel operates independently of OpenAI. It can offer advice that OpenAI has not been asked for, comment on OpenAI’s impact on the math community, and publish its recommendations. The members are not paid by OpenAI, and the list can be adjusted by the panel itself. However, OpenAI also stated that the panel is not responsible for the pace at which OpenAI advances its internal mathematical research.
The nine members are:
François Charles (École Normale Supérieure PSL—Paris)
Camillo De Lellis (Institute for Advanced Study, Princeton)
Timothy Gowers (École Normale Supérieure de Lyon, Cambridge University)
Martin Hairer (Swiss Federal Institute of Technology Lausanne, Imperial College London)
Nikhil Srivastava (University of California, Berkeley)
Ulrike Tillmann (University of Oxford)
Ravi Vakil (Stanford University)
Edward Witten (Institute for Advanced Study, Princeton)
Melanie Matchett Wood (Harvard University)
OpenAI: The internal model trained beginning August 28 has solved more than 100 unresolved problems
At the beginning of its announcement, OpenAI wrote that the company began training a new internal model on August 28. In addition to the Navier-Stokes Millennium Prize problem, the model has solved more than 100 long-unsolved problems spanning most areas of mathematics. The pace of progress surprised even OpenAI’s internal mathematicians, and also sparked internal discussions about how to ensure the broader mathematical community becomes aware of this rapid advancement so the field can prepare and adapt.
When OpenAI released its Navier-Stokes results on September 9, it sparked debate over research ethics and credit after a New York University mathematician, Tristan Buckmaster, questioned whether OpenAI had gotten ahead of his work after learning about his research progress.
In response to an open letter signed by Fields Medal winners, Martin Hairer—who is also a signatory—OpenAI directly quoted from the letter published by the mathematicians on mathandai.org. OpenAI said the mathematicians expressed concern about the negative externalities created by “using the resolution of unsolved problems as a benchmark for evaluating a new AI system,” and said these criticisms highlight the need for AI companies to interact carefully with the mathematics community.
The statement, signed by multiple Fields Medal winners and reported on September 12, criticized that answers produced by AI are often announced hastily, without time to properly write up and refine new methods and ideas, and without citing relevant prior research. This has raised concerns about credit and plagiarism. Martin Hairer, one of the panel members and a 2014 Fields Medal winner, is also listed as a signatory to the statement.
OpenAI described this collaboration as “a first step,” and said it hopes to place mathematicians at the center—deciding how AI can support understanding in mathematics and how the results can benefit a wider community.
This article, in which OpenAI says the model has solved more than a hundred math problems and that it worked with a panel of mathematician advisors to review the results, first appeared on .
