Mathematicians need to act - The blog of Tasmin Chu

The blog of Tasmin Chu

SubscribeSign in

Mathematicians need to act<br>Do not collaborate with AI companies for the foreseeable future.

Tasmin Chu<br>Aug 02, 2026

21

Share

This morning, I woke up to a text from a postdoc (soon-to-be-professor) friend, saying that OpenAI’s new model, Astra, had proved that a nonsofic group existed. At first, I was shocked. Then I became angry.<br>Let me begin with a few technical notes. I first learned about the problem of nonsofic groups in undergrad; it is one of Gromov’s famous open problems, who in 1999 essentially asked if every group was sofic (though the reader may notice he did not use the word sofic in that paper, which was coined by Weiss a year later). For the last 15 years, the mathematical community widely believed that nonsofic groups existed, but it is quite hard to show that a given group is nonsofic. This was known to be a very important problem in group theory, with real implications; sofic groups are known to satisfy interesting properties, such as Gottschalk’s conjecture and Kaplansky’s direct finiteness conjecture. As recently as March of 2026, I went to a conference at UT Austin which was solely about sofic groups and the Connes embedding problem, although as a percolation theorist I have never worked directly on this problem or even adjacent problems myself.1<br>Nine other results were announced by OpenAI in the same press release, including new bounds on sphere packing and a superexponential lower bound on multicolour triangular Ramsey numbers. Of all the facts accompanying this press release, the following stood out the most to me: the cost of these results was roughly $2,000 USD of tokens at Sol API rates.<br>In lieu of continuing with the technical details of this quite impressive result, let me now speak as a human being. This is a polemic, not a press release. This ruined my morning. I know that others feel the same way; I have spent today crashing out and learning that top mathematicians I know are crashing out. We in the mathematical community are all grappling with the implications of these highly effective large language models. What does this mean for our field? How should the practice of mathematics continue in light of increasing capable artificial intelligence systems? How should we interact with AI companies? Should we partner with them and work for them? Should we use these technologies to accelerate mathematical progress as quickly as possible and initiate scientific breakthroughs? What role do human mathematicians have to play in a world where artificial intelligence is increasingly capable of autonomously proving important theorems?<br>Many prominent mathematicians have articulated visions of what the future could hold, such as Terence Tao, Tim Gowers, Jacob Tsimerman, Daniel Litt, and others. Some, like Terence Tao, predict that we may have to adapt to an “abundance” environment where mathematical insights are plentiful rather than rare. Others, like Jacob Tsimerman, are broadly pessimistic about the implications of artificial intelligence for mathematics. Let me also draw attention to the Leiden Declaration on Artificial Intelligence, which offers a good number of intelligent and unobjectionable statements, and simultaneously assert that it does not go far enough.<br>I have talked to mathematicians who believe that in the future, we will simply be like artists; is this so different from what we usually do, which is already useless? One talented PhD student told me that while photography may have the obliterated the industry of portraiture when it made the production of images extremely cheap, at least photography still exists. Of course, artists and photographers are famously well-respected and paid in this society of ours.<br>I am sure that many obsolete careers of yesteryear were quite wonderful. Perhaps it was a great gig being the court astrologer. Pick up Silas Marner and you will realize that weaving used to be a well-paid and respected profession; while “Luddite” may be a pejorative today, the textile workers who destroyed mechanized looms were not behaving so irrationally from an economic standpoint.<br>I believe that today, mathematicians may be in the same position as these bygone professions. While I believe mathematical activity will continue and that some version of the profession of “mathematician” will continue, I think that LLMs pose an existential threat to the mathematical community and the values of that community. I believe our lives and livelihoods are important. I believe in chasing fantastic scientific and mathematical progress with the help of LLMs, we risk imperiling our institutions, our professional norms, and human mathematical understanding. Moreover, I believe that we are capitulating and surrendering in advance to AI companies that are broadly building a world that we do not want to live in. Mathematicians are not the only victims of these disruptive technologies, and...