OpenAI recently released 722 mathematical manuscripts and proofs, though only 300 of these results were formalized for computer verification. The company did not disclose the specific model used to generate these outputs or the prompts involved, despite recommendations from the Advisory Group on Mathematicians and Artificial Intelligence.
The release has caused significant concern within the academic community. Mathematics professors, including Tristan Buckmaster of NYU and Bryna Kra of Northwestern University, noted that the rapid publication of these results threatens to invalidate ongoing research programs and discourages collaboration by making researchers more secretive about their work.
Experts have also raised alarms regarding the intersection of AI and cybersecurity. Matthew Green, an associate professor at Johns Hopkins, suggested that advancements in AI mathematical capabilities could eventually threaten public key cryptography, potentially endangering digital assets. However, figures such as Adam Back and Vitalik Buterin have urged caution, with Buterin noting that while AI poses a long-term risk, immediate panic regarding wallet security is unnecessary.
The mathematical community is currently struggling with the lack of transparency in OpenAI's process, as researchers have reported finding nonsense formulas and redundant work in the published documents. Despite the publication of changelogs for the project's GitHub repository, the field remains anxious about the impact of AI-driven research on academic careers and the integrity of the discipline.
