Who said what
On October 10, IT Home relayed a Business Insider report from that morning. OpenAI’s large release of mathematics results drew different reactions in academia. Some scholars were glad to see a research advance. Others worried about researchers still working on the same problems.
Tristan Buckmaster, a mathematics professor at New York University, told CNBC’s Squawk Box that the work of some young mathematicians had been hit hard. He said the other side released so many results at once that an entire research plan was wrecked. He raised a second worry: researchers had submitted unpublished material to AI tools, and he asked whether that material helped OpenAI produce the results it released. He said the company can see their grant proposals. After a proposal is submitted, reviewers have to read it and write a summary, and while writing the summary they often enter the proposal into a model.
[1]
What was published
IT Home writes that on the 6th, local time, OpenAI published more than 700 mathematics manuscripts on GitHub. The announcement says the results came from a frontier model inside the company. Many of the proofs also come with a version a computer can check.
For some mathematicians, the issue is not only that a problem they had been working on was solved first. Bryna Kra, a mathematics professor at Northwestern University, says AI can help move mathematics research forward, but releasing a large batch at once may damage the long-standing collaborative habit of the field, and that collaboration is an important basis for progress. She says mathematics is a highly collaborative discipline. Nobody at OpenAI, she says, can give a talk on any one of these results or answer technical questions about it.
Mathematicians regularly share unfinished lines of thought in talks and conversations. Kra worries that if people fear an idea will be absorbed before it is formally published, they may stop speaking about it in public.
[1]The company, and a different view
IT Home writes that OpenAI’s announcement says this release set rules for revising papers and for citation. The company also plans to improve how the papers are presented and to fund workshops that help researchers understand these AI-generated mathematics results.
Alex Kontorovich, a distinguished mathematics professor at Rutgers University, is more optimistic. He says AI is pushing academia to reconsider hiring and prizes, and that the value of mathematical training may rise. He says he would not want to give a prize to someone who only presses a button. Learning mathematics, he says, is above all about thinking clearly, deeply, and for a long time about hard problems. He thinks demand for that ability will grow, because it is hard to acquire and takes long training.
[1]The limit of this account
All of the above comes from IT Home’s retelling of Business Insider and a television interview. This article did not check the CNBC tape or open the GitHub repository. A separate story this morning reported that some mathematicians called for an end to cooperation with OpenAI. This one records only the named comments from October 10. It does not fold the glad, worried, and optimistic views into one conclusion.
[1]