OpenAI's internal AI model solves over 100 open math problems, prompting new advisory group
Advisory Group on Mathematics and Artificial Intelligence
OpenAI trained an internal model that resolved more than 100 long-standing open problems across mathematics, surprising even its own mathematicians. In response to concerns from the math community about the negative externalities of using open problems as benchmarks, OpenAI is working with an independent advisory group of mathematicians. The group will advise on reviewing and communicating results, academic standards, and how AI tools can support mathematical research and learning.
The pace of its progress in mathematics has surprised the mathematicians within OpenAI.
- 6gvONxR4sf7o
> The group will operate independently from OpenAI. The group will have the freedom to offer advice we have not requested, comment on OpenAI’s impact on mathematics, and make its advice public.
That seems so empty. I also operate independently from them and have the freedom to tell them things they haven't requested and comment on their impact and do so publicly. But I doesn't have any impact on OpenAI. If they're trying to say how this will actually have any impact on anything at all, they're doing a shit job.
- kingstnap
There are three big things in this announcement it seems.
> In addition to resolving the Navier–Stokes Millennium Prize problem , this model has now resolved more than 100 long-standing open problems across most areas of mathematics.
Rumor was they were sitting on the solution to the Hodge conjecture. I wonder if thats one of these.
> The group will advise on the review and communication of emerging results.
With hundreds of results, and the seeming exponential vertical from earlier in summer (10 major results from Astra, now hundreds...) I do wonder how exactly such a group decides to disseminate results. If it will instantly bottleneck. I mean I checked their site and it seems to be a google form asking for opinions.
> Importantly, the group will not be responsible for advising us on how to pace our internal progress on mathematics.
Spicy. I think the super intelligence benchmark for me is P vs NP. That one I know is proper hard and feel it in my bones.
In related notes I watched a panel of mathematicians discuss AI recently on YouTubes. Specifically the hlf26 one. [0]. I also watched Terrence Taos recent one on SAIR [1]. In addition to a few others like Tom Rocks and Martin Hairer [2]. Whats abundantly clear is that no one has any answers to what this means for the future of mathematics. They all waffle about what anyone is going to be doing in a few years.
[0] https://youtu.be/H7_d_sgui6o
- Certhas
Mathematics has enough cultural capital within the AI companies that they get to have this "advisory board".
Nobody else is getting that. When/If the severe impacts on labor materialize, workers won't get an advisory board. They will just get fired, and the companies will celebrate it as efficiency wins.
I think it's time to get very very real about regulation/taxation. Token/Compute sales tax that gets redistributed as UBI?
- mwkaufma
"Advisory Group"
Importantly, the group will not be responsible for advising us on how to pace our internal progress on mathematics.
- anothermathbozo
> internal progress on mathematics
This alone feels like a horrible new paradigm