OpenAI’s internal model Astra is facing another round of scrutiny over a claimed mathematical breakthrough. Mathematician Andreas Thom said in a recent post that OpenAI may have absorbed conversation records tied to his work with mathematician Gábor Kun on the Gromov soficity conjecture while training Astra, after which the model went on to solve a closely related problem. The conjecture is a long-unsolved question in group theory, and Thom said he has worked on it for 20 years. When OpenAI published Astra’s “top 10 mathematical breakthroughs” in August, the third item was the construction of a non-sofic group. The company had earlier described the result as having been completed by an internal version of Astra, but later revised that language after criticism from mathematicians, acknowledging that the proof relied on recent mathematical work, including research by Kun and Thom. There is currently no evidence showing that OpenAI actually used the private conversations in Astra’s training. Even so, the accusation has revived debate over AI training data, ownership of unpublished research, and whether some purported AI discoveries rest on human work that had not been made public.
OpenAI’s internal model Astra is again under scrutiny over a mathematical result. Mathematician Andreas Thom said in a recent post that OpenAI may have absorbed conversation content generated during his research with mathematician Gábor Kun on the Gromov soficity conjecture while training Astra, and that Astra later solved a highly related problem.
The Gromov soficity conjecture is a major long-unsolved question in group theory. Thom said he has spent 20 years working on the problem.
Claim centers on one of Astra’s published math results
When OpenAI published Astra’s “top 10 mathematical breakthroughs” in August this year, the third item was the construction of a non-sofic group. OpenAI had previously said the related result was completed by an internal version of Astra.
After drawing criticism from the mathematics community, the company revised its wording and acknowledged that the proof relied on recent mathematical work, including research by Kun and Thom.
No evidence so far that private conversations were used in training
At present, there is no evidence showing that OpenAI in fact used the private conversations to train Astra. Still, Thom’s accusation has pushed several issues back into focus: the source of AI training data, ownership of unpublished research results, and whether claimed AI discoveries may be built on human work that had not been made public.
If the allegation is ultimately confirmed, the dispute would extend beyond academic attribution and could develop into a broader controversy involving AI companies’ data use, research ethics, and intellectual property.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
