OpenAI publishes 722 AI math manuscripts prompting calls for verification

OpenAI published 722 math manuscripts on GitHub on Tuesday, which were generated by an unreleased internal artificial intelligence model. Decrypt reported that an OpenAI spokesperson said almost all of the manuscripts were produced from a single prompt given to a single artificial intelligence agent, although some solutions may have required multiple attempts. [1]

The Wall Street Journal and 36Kr both report that the internal model resolved hundreds of complex mathematical problems. According to 36Kr, the published solutions overnight included a breakthrough proof related to the Quasi-Riemann Hypothesis. [4][5]

Mathematicians have expressed skepticism regarding the claims made about the unreleased model. Decrypt reported that Andrew Sutherland, a mathematician at the Massachusetts Institute of Technology, stated that claims regarding single-agent problem solving should be treated as unverified until the model is released for public replication. [1]

Artificial intelligence refers to the capability of computational systems to execute tasks associated with human intelligence, including reasoning and problem solving. Unsolved mathematical problems span disciplines such as algebra, number theory, and analysis, with several major challenges offering financial prizes for verified solutions. [6][7]