OpenAI might have stolen another major proof (twitter.com)

OpenAI faces serious accusations of potentially using unpublished research without attribution in developing its AI model, Astra. In a detailed post, mathematician Andreas Thom claims that Astra may have been trained on conversations he had with fellow mathematician Gábor Kun while researching Gromov’s soficity conjecture, one of the significant problems that Astra later claimed to have solved. Thom, a respected expert in the field, suggests that this issue extends beyond mere attribution—if proven true, it would imply that proprietary human work was incorporated into an AI model that subsequently presented these findings as its own breakthroughs. This situation could have profound implications for the AI and mathematics communities. If the allegations from Thom and others are corroborated, it could reveal a pattern of unethical practices in AI development, casting a shadow over OpenAI's reputation and the broader narrative of AI as an innovative force in science. It raises critical questions about the ownership of intellectual property and the ethical boundaries of using human-generated data to train AI systems, challenging the notion that AI is creating new knowledge and instead suggesting it may be appropriating existing discoveries.