OpenAI's Three AI Proofs Get Academic Shepherds, Credit Questions Linger
OpenAI's three AI-generated results, including a disproof of the Exponential Time Hypothesis, come with academic shepherds and Lean formalizations, but the release model leaves credit and data-provenance questions unsett
OpenAI announced three AI-generated results: a proof of the local smoothing conjecture in every dimension, a disproof of the Exponential Time Hypothesis, and a rigorous existence proof for four-dimensional interacting quantum field theories including QCD and a Standard Model completion. Each was formalized in Lean, the company says, and released through different academic-shepherded collaborations.
The math output is public with its Lean proof. A group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams will explain the ETH refutation while raw data is public.
David Gross leads a physics group writing up the Yang–Mills/QCD result and deciding on artifact release. OpenAI says it seeks no academic credit and is funding a $5 million Simons Foundation initiative.
Key takeaways
- The mathematics result proves Sogge's 1991 local smoothing conjecture in every dimension, which OpenAI says also resolves the Kakeya conjecture and Stein's restriction conjecture. The proof rests on a novel auxiliary partial differential equation; the natural-language output appeared Oct. 18, 2026, with Lean formalization the next day.
- The computer science result refutes the Exponential Time Hypothesis, a conjecture used in hundreds of papers to derive conditional hardness results. OpenAI says the natural-language disproof was produced Oct. 16, 2026, and formalized in Lean on Oct. 17; the shepherd group requested public raw data and a funded conference.
- The physics result gives a rigorous existence proof for four-dimensional interacting quantum field theories with gauge fields and fermions, including QCD and a unified Standard Model completion. The proof includes pure Yang–Mills existence, part of the Clay Millennium Prize problem, building on Balaban's constructive approach.
- OpenAI is testing three release strategies: public AI output for math, an external explainer group plus public artifacts for computer science, and deferred artifact release for physics. It says participants sign no NDAs, receive no academic credit, and work voluntarily; it is funding a $5 million Simons Foundation initiative.
What people think
Analysis details
21 comments analyzed. Analyzed .
The sample selects active discussion branches and includes available parent comments. It may omit parts of the full thread. The analysis included 21 of 95 usable stored comments because of the input limit. Selected themes do not measure community opinion or how common a view is.
Credit and data provenance
The release model draws objections that OpenAI's models are trained on researchers' inputs, then the company announces results and invites selected experts to explain them. Critics say this lets OpenAI claim credit for work built on the sum of researchers' efforts, including failed and unpublished attempts, and that shepherding does not resolve the asymmetry. Counterpoints stress that attribution matters but knowledge progress matters more, and that AI has not yet affected mathematicians.
ETH refutation and P vs NP priors
The disproof of the Exponential Time Hypothesis prompts debate over whether it shifts priors toward P=NP. One view treats it as significant hope; another warns P=NP would be catastrophic; a third frames ETH as a promising formalism now shown to be a dead end. A joke about sending SAT instances to a model for a few hours captures the uncertainty.
Formal verification favors math over biology
Progress in Lean-verifiable domains may reflect cheap correctness checks and reinforcement-learning reward signals, whereas biology requires wet-lab experiments that slow iterations by a factor of 100 or more and computational approximations with weak predictive power. A counterargument holds that knowledge transfers across domains, so frontier models will need less data than expected to become competent in harder-to-verify fields.
Alternative collaboration models proposed
A proposed alternative is for OpenAI to collaborate with scientists during research and let them publish with acknowledgment, rather than announcing results and inviting selected experts to explain them. This would address concerns about credit and data provenance, though whether such a model is feasible or acceptable to the company remains untested.