OpenAI claims AI models solve long-standing math and physics problems

OpenAI announced on Monday that its artificial intelligence models have produced three new scientific findings in mathematics, computer science, and physics. The company says the models proved the local smoothing conjecture in every spatial dimension for the first time. Researchers had previously established the conjecture only in three dimensions.

The local smoothing conjecture is a central proposition in harmonic analysis. This field studies how complex signals and functions break down into elementary waves. Alex Iosevich, a professor of mathematics at the University of Rochester, called the development "historic." Iosevich noted that the problem was formulated by his PhD adviser, Chris Sogge. He described his initial reaction as a shock but said he now feels joy and optimism about the potential for giant leaps in science.

Iosevich expects theoretical mathematics to change drastically because of AI. He admitted to a twinge of sadness but believes that feeling will give way to the realization that more can be accomplished. He noted that many pure math problems arise from concrete physical considerations and that those connections can now be explored rapidly.

OpenAI also said its models disproved the Exponential Time Hypothesis. This hypothesis relates to the time needed to solve complex computational problems. Additionally, the company stated that the models provided a mathematically rigorous proof of the existence of quantum chromodynamics. This theory studies the interaction of quarks and gluons.

The company reached out to outside scientists to turn raw research data into scientific understanding. It selected three expert groups to act as shepherds for the release process. The groups handling mathematics and computer science chose to make the AI-produced data public immediately. The physics group opted for a delay to write its own interpretation of the findings.

OpenAI acknowledged that these approaches involve tradeoffs. The collaborations allow experts to unpack results and place them in academic context. However, choosing initial shepherds shapes who gets involved first and creates an asymmetry of information. The startup, valued at $852 billion, said it welcomes feedback on how to broaden participation.

This announcement follows a similar claim from Anthropic last month. The creator of Claude said it discovered a new enzyme resembling CRISPR. That announcement drew mixed reactions from scientists. Some hailed the discovery as exciting, while others suggested the company had exaggerated its significance.