OpenAI has unexpectedly published an archive of 722 scientific papers on GitHub, containing solutions to 372 complex mathematical problems. This archive covers fundamental mathematics, theoretical computer science, and mathematical physics, demonstrating an unprecedented level of autonomy for modern neural networks. According to the developers, all of these extensive works were generated by a new, unreleased advanced AI model in response to just a single prompt with hints.

Technical Generation Details and Automated Verification

According to the technical documentation, the model tested about 4,000 tasks, with the creation of one successful solution taking an average of about three hours of ChatGPT Pro-level computing power. To allow the scientific community to objectively assess the correctness of such complex calculations, the developers included formal proofs written in the specialized programming language Lean. The use of Lean allows third-party algorithms and mathematical programs to automatically verify the logic and correctness of each individual step of the proof without manual sorting.

Controversial Data

Despite technical innovations and formal proofs in Lean, the academic and scientific community met OpenAI's claims with a heavy dose of skepticism. The main stumbling block was that the company still has not provided external researchers with direct access to the model itself to independently reproduce the results. Mathematicians note that without open testing, claims of solving hundreds of complex problems with a single prompt remain unverified. Additionally, critics point to the lack of detailed data on the exact computation time for each individual case and the complete concealment of the exact text of the source prompts.

Consequences for Science and Prospects

The situation surrounding the publication of the OpenAI archive has intensified long-standing discussions about the transparency of commercial developments in artificial intelligence. On the one hand, demonstrating the capabilities of automated problem-solving opens up grand prospects for fundamental science and accelerating technological progress. On the other hand, scientists rightly fear the secrecy of major tech players who demand that their achievements be taken on faith. Experts agree that a real breakthrough in mathematics using AI requires an open ecosystem where independent researchers can audit and recheck every stage of data generation.