Andreas Thom is a mathematician at Technische Universität Dresden. His research spans geometric group theory, real algebraic geometry, algebraic topology, ergodic theory, and operator algebras. He also participates in public discussions about mathematical research and scientific practice.
DI has documented statements by Andreas Thom on Artificial Intelligence and Scientific Research. Read the quotations, original sources and related cases below.
Mathematician Andreas Thom, in a guest post on Terence Tao's blog, said he asked OpenAI researchers Mark Sellke and Sébastien Bubeck whether conversations he and a colleague had with ChatGPT about his research were part of the training data or accessible to the model's reasoning process, and that Sellke denied it. Thom calls that denial materially misleading in light of OpenAI's later statement in the Buckmaster-Alpöge case that it could not rule out that de-identified usage data helped improve its models. He also writes that Bubeck's acknowledged career-related remark to Buckmaster and his objection to including Alpöge in a proposed paper presenting OpenAI's proof "deepen my concern about their commitment to academic standards."
“deepen my concern about their commitment to academic standards”
Thom calls Sellke's denial materially misleading and says Bubeck's acknowledged career-related remark to Buckmaster and his objection to including Alpöge in a proposed paper "deepen my concern about their commitment to academic standards," adding that these episodes together raise serious questions about their judgment and personal integrity. That is a direct condemnation relevant to the case's pattern-of-conduct question.
Role at the time: Mathematician describing his own experience and ethical concerns in a personal public post
Condemned
Statement summary
Thom challenged OpenAI's explanation of research-related user data and requested evidence supporting its denials. Drawing on his own experience, he argued that de-identification would not eliminate the ethical issue if unpublished mathematical ideas were used to compete with their authors.
“De-identification may remove a name; it does not remove the intellectual content of a mathematical idea.”
Read the original text (opens in a new tab)Andreas Thom questions OpenAI's explanation of unpublished research and user dataThom's numbered 2/3 post replying in the Buckmaster discussion
Why we used this label
The full thread calls potential competitive use ethically indefensible and criticizes the current explanation as missing the moral issue. Factual uncertainty does not erase the direct ethical condemnation.
Tristan Buckmaster says OpenAI pressured him over authorship and publication; Sebastien Bubeck disputes his account. Read their statements and the timeline of competing mathematics work, including what remains disputed.
