{"id":6810,"date":"2026-09-10T11:08:32","date_gmt":"2026-09-10T11:08:32","guid":{"rendered":"https:\/\/silvybrand.com\/?p=6810"},"modified":"2026-09-10T11:08:32","modified_gmt":"2026-09-10T11:08:32","slug":"where-does-openai-get-mathematics-training-data","status":"publish","type":"post","link":"https:\/\/silvybrand.com\/?p=6810","title":{"rendered":"Mathematicians want proof OpenAI didn\u2019t use their work\u00a0"},"content":{"rendered":"


\n<\/p>\n

\n
\n

Another researcher is challenging OpenAI about the data driving its increasingly impressive array of mathematical discoveries. Just days after a bitter row erupted over whether the company\u2019s models benefited from unpublished work, a second mathematician has come forward accusing the AI giant of unethical and \u201cdishonest\u201d behavior and a lack of transparency about the origins of its training data.<\/p>\n<\/div>\n

\n

In a series<\/a> of<\/a> posts<\/a> on Mastodon, mathematician Andreas Thom raised concerns that interactions he and his colleagues had had with the ChatGPT chatbot before OpenAI\u2019s triumphant announcement may have contributed to its success in the field. One of the 10 results<\/a> OpenAI announced with great fanfare<\/a> last month involved Thom\u2019s area of expertise, so-called non-sofic groups, and OpenAI acknowledged that their result built heavily on previous work by Thom and fellow mathematician G\u00e1bor Kun.<\/p>\n<\/div>\n

\n

Thom said he began reflecting on his own interactions with OpenAI after Tristan Buckmaster, a mathematics professor at New York University, publicly questioned<\/a> whether the company\u2019s AI models had benefited from his use of OpenAI\u2019s Codex. After OpenAI announced its non-sofic groups result, it was widely criticized in mathematical circles for failing to acknowledge recent contributions from Thom and Kun and the company quietly amended its writeup. Non-sofic groups are, roughly speaking, infinite mathematical structures that cannot be approximated by finite ones.<\/p>\n<\/div>\n

\n

Thom said he was also struck by \u201cOpenAI\u2019s detailed command of our techniques,\u201d which he said were neither the most obvious nor the most promising routes to a solution at the time. He said he wrote emails to OpenAI researchers S\u00e9bastien Bubeck and Mark Sellke, also a statistician at Harvard, to ask whether his interactions with ChatGPT were \u201cpart of the training data or accessible to the reasoning process\u201d and could therefore have contributed to the result.<\/p>\n<\/div>\n

\n

But the answer did not satisfy Thom, who said it only addressed whether his conversations with the chatbot could be accessed directly, not whether they had entered into the vast pools of training data the company uses to improve its models. \u201cNo such qualification, explanation, or evidence was given,\u201d he wrote. \u201cI take this as dishonesty to say the least.\u201d<\/p>\n<\/div>\n

\n

Thom said researchers aren\u2019t equipped to reverse-engineer OpenAI\u2019s training pipeline to figure out whether their work has been used or not. \u201cOnly OpenAI has the relevant data for that.\u201d If the company is going to deny doing this, he said the responsibility is on them to prove that by disclosing all necessary datasets and clarifying various settings and terms setting out how it uses data.<\/p>\n<\/div>\n

\n

OpenAI\u2019s reluctance to conclusively rule out any use of user data echoes the way it defended its recent Millennium Prize breakthrough, both in its public messaging and its communications with Buckmaster \u2014 who was working on the problems with Anthropic researcher Levent Alp\u00f6ge in a personal capacity. In the blog post announcing<\/a> the Navier-Stokes solution, which concerns the movement of fluids, OpenAI flatly denied using any specific<\/em> user data: \u201cWe (the researchers and the agents) did not see any of their work through any means until they released it publicly \u2014 in particular, no specific user data was accessed in order to solve this problem.\u201d<\/p>\n<\/div>\n

\n

But it would not conclusively rule out an indirect influence: \u201cWhile unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models\u2060.\u201d Thom said it is the same obfuscatory distinction the company drew in its communications with him. \u201cDe-identification may remove a name; it does not remove the intellectual content of a mathematical idea,\u201d he said.<\/p>\n<\/div>\n

\n

In light of recent events, Thom said \u201cSellke\u2019s categorical answer was, at minimum, unjustifiably broad and materially misleading; looking back it was plainly dishonest.\u201d<\/p>\n<\/div>\n

\n

Thom said it \u201cwould be ethically indefensible\u201d if nonpublic research supplied by users helped to improve models that the company then used to race those very same users to publication, without consent, proper disclosure, or credit.<\/p>\n<\/div>\n

\n

OpenAI did not immediately respond to The Verge<\/em>\u2019s request for comment.<\/p>\n<\/div>\n

\n

His comments add to mounting unease over OpenAI in mathematical circles at what should be a moment of triumph for the company. Its announced solution to one of mathematics\u2019 legendary Millennium Prize problems is an extraordinary achievement that, should it be verified, few would deny. But this was complicated by the unusual circumstances OpenAI said led it to pursue the problem in the first place: It heard rumors online that other researchers had made major progress and thought it would try too.<\/p>\n<\/div>\n

\n

The ongoing incident has left a sour taste in mathematicians\u2019 mouths<\/a>. Numerous researchers told The Verge<\/em> they worry behavior like this will push the field into a more secretive state if mathematicians know that even rumors they are close to a big breakthrough could ignite a race with a well-resourced tech giant eager for glory.<\/p>\n<\/div>\n

Follow topics and authors<\/strong> from this story to see more like this in your personalized homepage feed and to receive email updates.<\/span><\/p>\n