Mathematicians Demand Proof That OpenAI Did Not Use Their Research

Mathematicians Andreas Thom and Tristan Buckmaster (NYU professor) publicly raised issues concerning training data source transparency and citation of existing research in relation to OpenAI's announcement of achievements in the field of mathematics.

OpenAI acknowledged that results related to non-sofic groups (infinite mathematical structures that cannot be approximated by finite ones) among the 10 mathematical achievements recently announced relied heavily on the existing research of Thom and Gábor Kun. However, the initial announcement did not explicitly state the contributions of these researchers, drawing criticism from the mathematics community, after which OpenAI quietly revised the report.

Thom expressed surprise at OpenAI's detailed understanding of their techniques and inquired with researchers whether ChatGPT conversation content was included in the training data or accessible during the inference process. OpenAI's response only addressed direct access to chatbot conversations and failed to provide a clear explanation on whether conversations were included in the vast pool of training data used for model improvement, causing dissatisfaction.