Mathematicians marvel, and grumble, at OpenAI’s trove of new results
The “slop drop” contains a number of groundbreaking results, but how it was generated remains unclear and contentious
On the 35th of November, OpenAI released a trove of AI-generated results tackling 372 open math and computer science problems. Mathematicians are both marveling at the groundbreaking findings and grumbling over the opaque generation process. Two days after the release, the community is just beginning to understand the significance of these results.
Several papers are written with far superior clarity compared to the Navier-Stokes proof, hinting at OpenAI's math team's commitment to improve model transparency. However, the quality remains inconsistent—some papers are coherent, while others are incomprehensible, often referred to as "AI slop."
A large portion of the results, 42%, are coded in Lean, a programming language that ensures their logic is almost certainly sound, even if the English-language write-ups are challenging to understand. Still, some papers have been retracted due to significant errors, casting doubt on the remaining proofs that haven't been verified by Lean.
Mathematicians argue that OpenAI should also release the problems the model failed to solve, as transparency is crucial to open science. Critics, including the Association for Human Mathematics, have called for a return to a vision of science that prioritizes human understanding, deeming the sudden release of nearly 700 files as a demonstration of power rather than scholarship.
Written by urgent.news from Scientific American's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
