OpenAI has made publicly available 722 manuscripts containing solutions and work on 372 significant mathematical problems. The materials are hosted in a repository on GitHub, according to infohub.kz.
As Engadget reports, the developers used an unreleased ChatGPT artificial intelligence model to prepare the publication. It generated answers based on single prompts, and producing one result took an average of about three hours of work in ChatGPT Pro mode. The document also includes the neural network's reasoning process and approximate estimates of the computational resources used.
Among the claimed achievements are a solution to the four-dimensional Kakeya conjecture, improvements to key computer algorithms, and progress toward proving the Riemann hypothesis. The company says it followed recommendations from its independent advisory board on mathematics and artificial intelligence. However, the developers ignored some expert advice: they did not disclose the exact computation time for individual problems or the specific prompt texts.
The presented materials still need to undergo detailed scrutiny by the mathematical community. Scientists are treating the claimed results with caution. Experts note that until the model is made available for independent testing, any claims of solving complex problems on the first attempt should be considered unconfirmed.


