Mathematician Andreas Thom has expressed the view that OpenAI's actions are dishonest, claiming that the mathematical achievements announced by the company may rely on research conducted by him and his colleagues.

Thom pointed out that one of the achievements OpenAI announced last month includes content regarding non-sofic groups (infinite mathematical structures that cannot be finitely approximated), which is his area of expertise. While OpenAI has acknowledged that the achievement relies heavily on previous work by Thom and Gábor Kun, Thom stated that he has not received a satisfactory explanation regarding whether the contents of interactions with ChatGPT were used to improve the model.

OpenAI has denied using specific user data to solve the problems, but it has not completely ruled out the possibility that anonymized data contributed to the model's improvement. Thom argues that even if names are removed through anonymization, the intellectual content of the mathematical ideas remains. Furthermore, he stated that it is difficult for researchers to verify the training process by decompiling it, and that the responsibility for data disclosure lies with OpenAI.

Source: