Mathematician Andreas Thom asks OpenAI to prove it did not use his work — The Verge
Mathematician Andreas Thom said that OpenAI should provide evidence that it did not use users’ non-public mathematical work to improve its models. He linked his concerns to the company’s recent announcements of mathematical results, one of which concerns his specialization — nonsofic groups.
Questions about user data
As The Verge reports, Thom said in a series of Mastodon posts that he and his colleagues had communicated with ChatGPT before OpenAI’s announcement. One of the 10 results the company reported last month concerned nonsofic groups. OpenAI acknowledged that this result relied significantly on earlier work by Thom and his colleague Gábor Kun.
According to Thom, after the announcement he noticed OpenAI’s detailed command of methods that were neither the most obvious nor the most promising path to solving the problem. He wrote to OpenAI researchers Sébastien Bubeck and Mark Sellke asking whether his conversations with the chatbot could have been part of the training data or accessible during the model’s reasoning process.
More current news is available on the UA.News Telegram channel Telegram.
Demand for greater transparency
Thom said that the response he received addressed only direct access to chats, but did not explain whether the content of the conversations had entered the datasets the company uses to improve models. In his view, only OpenAI has the information that would make it possible to determine whether specific user materials were used. The mathematician believes the company should disclose the necessary datasets and explain the terms of their use if it denies using non-public research.
OpenAI previously stated that its researchers and agents had not seen other mathematicians’ work before its public disclosure, including that they had not accessed specific user data to solve the Navier–Stokes problem. At the same time, the company noted that it could not completely rule out that anonymized data obtained through the use of its products may have helped improve the models.
Thom stressed that anonymization may remove a name, but not the intellectual content of a mathematical idea. He called it ethically unacceptable if users’ non-public research could help improve models that later compete with those same researchers for priority in publication without their consent, disclosure, or proper recognition. OpenAI did not provide The Verge with a prompt comment.
