Mathematicians accuse OpenAI of stealing results and demand removal of Anthropic employee signatures Beating AI News: New York University mathematicians Tristan Buckmaster and Levent Alp ö ge recently advanced multiple fluid mechanics problems within a month using large models such as Claude and Codex, and some of the results have been validated by Lean machines. The paper hasn't been officially published yet, and the two of them had a falling out with OpenAI first. Buckmaster said that in early September, their undisclosed progress was reported to OpenAI. He proactively contacted OpenAI, hoping to avoid a collision between the two parties. Three days later, OpenAI researcher S é bastien Bubeck told him that OpenAI's internal model had written about 100 pages of Navier Stokes proofs. This is one of the seven "millennium problems" in the field of mathematics, and the Clay Mathematics Institute is offering a reward of $1 million. Buckmaster then repeatedly asked when OpenAI started working on it. According to his statement, the other party finally confirmed that the first prompt was only issued in the past few days, and OpenAI had already known that they had made progress at that time. OpenAI is still following a path that highly overlaps with theirs and has been rarely studied before, with a team and a large amount of computing power behind it. The research drafts of the two have been kept in Codex before. Buckmaster therefore questioned whether these data were used for training, but did not receive a clear answer. He also admitted that there is currently no evidence to prove that OpenAI has used these data. The subsequent negotiations pushed the matter completely towards transparency. Buckmaster stated that OpenAI proposed to have them release their own results first, and then release Navier Stokes the next day; Another approach is for Buckmaster to write OpenAI papers, but not for Alp ö ge to be credited. Bubeck reportedly suggested deleting Alp ö ge twice, citing his work at Anthropic as the reason. Buckmaster refused and stated that if OpenAI were to release according to this plan, he would publicly disclose the entire process. He claimed that the other party later asked him, 'Why did you ruin your career?' Later that day, Bubeck contacted Alp ö ge separately, stating that he was unsure if Buckmaster was' completely rational 'at the time. Alp ö ge did not accept individual communication. Buckmaster ultimately released the paper and statement in advance. He also emphasized that he had not read OpenAI's proof, did not know if the proof was valid, and did not know if his Codex data had been used. Bubeck subsequently publicly denied the allegations, calling them "false and provocative," and stated that he would further respond. [Original link]