Several mathematicians have raised allegations of plagiarism regarding the mathematical and theoretical computer science achievements produced by OpenAI's next-generation model, "Astra."
Andreas Thom, a mathematician at the Dresden University of Technology in Germany, pointed out that the proof steps Astra performed regarding non-solvable groups are strikingly similar to his own research findings spanning over 20 years. Thom revealed that he had used ChatGPT to discuss unpublished research content prior to the publication of his findings, and he questioned the possibility that those conversation records were used for model training.
In response, OpenAI core researcher Mark Serki said that user conversations are not used for model training. Meanwhile, OpenAI has faced criticism from the academic community for being disingenuous, as the company has stated in a separate instance that it cannot rule out the possibility that anonymized data from customer products was used to improve models.
While OpenAI explains that Astra's achievements were developed through large-scale reinforcement learning, concerns are spreading among researchers regarding the impact of AI-driven scientific discovery on the existing research ecosystem and the handling of data.
Source: