OpenAI publishes 722 math manuscripts from an internal model it has not yet released
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
OpenAI has released 722 math manuscripts written by an in-house model that is still unavailable, sorted into 372 families on GitHub. According to OpenAI, the model received about 4,000 problems, and each result took on average as much compute as some three hours of ChatGPT Pro. Many of the papers can be checked in Lean, though not all, and Institute for Advanced Study advisers call it a milestone without endorsing the process.
OpenAI has published 722 mathematical manuscripts generated by an internal model it has not yet released.
The company has grouped them into 372 families and posted them to a public GitHub repository. According to OpenAI, the model was posed about 4,000 problems, and the average result used compute equivalent to roughly three hours of ChatGPT Pro thinking, the Pro mode of its assistant. Many manuscripts have formalizations in Lean, a language that allows proofs to be checked by computer, but not all of them, and OpenAI cautions that some unformalized results could have issues.
The Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, which OpenAI consulted, has described it as an important event for mathematics, but makes clear that its role is not a judgment on the impact of the results nor an endorsement of the process used to obtain them.
Why it matters · analysis and opinion
How many of those results are correct? OpenAI does not fully know, and says so: those formalized in Lean can be checked by computer, but the rest could contain errors. That frankness shows what is missing before this leaves the lab. First, an available model, because the one that produced them is internal. Second, mathematicians who judge whether the results matter, something the Institute for Advanced Study advisory group expressly avoids doing. And third, an affordable cost, because each result consumed a lot of compute. Until then, it is a demonstration of capability rather than a tool a researcher could use tomorrow.
Official source: OpenAI · Written with the help of AI: how we make the news
