- Published on
OpenAI's Next-Gen AI Model Astra Accused of Plagiarism in Math Results
- Authors
- Name
- DeepHorus
Recently, a report released by OpenAI on the mathematical research results of its next-generation AI model Astra was accused of academic misconduct by several mathematicians for failing to adequately cite previous work. This incident has sparked heated discussion about AI research ethics and academic norms.
It is reported that OpenAI released a 249-page technical document on July 31, detailing the "significant progress" made by the Astra model in multiple mathematical fields such as high-dimensional sphere packing, quantum game theory, group theory, and extremal combinatorics. OpenAI claimed that these results demonstrate AI's potential in solving complex mathematical problems, and pointed out that if calculated at Sol API's rates, the token cost required to solve these problems would be approximately $2,000, equivalent to about 13,519 RMB.
However, shortly after the paper was released, multiple scholars raised severe doubts after careful review. Steven Miller, a mathematician at Yeshiva University, pointed out that the core argument of Astra's work on sphere packing in spaces of dimension 1000 or higher actually first appeared in a 2016 paper he published with collaborators, but Astra presented it as its own discovery without giving due citation.
Another controversy involves a long-standing problem about soficity in group theory. Astra claimed to have proven that there exists at least one group lacking this property. However, Cambridge mathematician Francesco Fournier-Facio and Andreas Thom, a mathematician at Dresden University of Technology, found that Astra's key mathematical steps combined ideas from papers published in 2016 and 2019, again lacking proper attribution and citation.
In response to these allegations, an OpenAI spokesperson said that the company is responsible for the correctness of the results and claimed that these achievements meet the general standards of human mathematicians, while also stating that they plan to make minor updates to the paper this week. However, this response did not completely allay the concerns of the academic community. Fournier-Facio emphasized that OpenAI has fully participated in high-level research and must be subject to the same academic standards as human scholars.
This incident reflects the new challenges brought by AI's increasingly deep involvement in scientific research. AI models can quickly integrate existing literature and generate new results, but without a transparent citation mechanism, it is easy to cause intellectual property and academic integrity issues. At present, OpenAI has not yet announced specific modification details, and mathematicians are calling for stricter review procedures.
Looking ahead, as AI's applications in basic sciences such as mathematics and physics continue to expand, it is urgent to establish academic norms applicable to AI. How to ensure respect for previous work while encouraging innovation will be a common challenge for the scientific community and AI developers.
