OpenAI Announces AI Independently Achieves Three Scientific Breakthroughs, ETH Conjecture Disproved
nashnova research
OpenAI announced its AI model independently produced major discoveries in mathematics, computer science, and theoretical physics — most strikingly, disproving the Exponential Time Hypothesis (ETH) that underpinned algorithm complexity theory for nearly 30 years. If confirmed, hundreds of papers built on that assumption face systematic re-examination.
ETH disproved — what does that actually mean?
The Exponential Time Hypothesis held that NP-complete problems like 3-SAT — a class of problems widely regarded as extremely hard — cannot be solved in sub-exponential time. For nearly 30 years, hundreds of papers in graph theory, databases, and computational geometry used ETH as the foundation for setting algorithmic speed limits.
OpenAI released a 67-page manuscript titled, plainly: *The Exponential Time Hypothesis Is False*. This means → if the proof holds up, the theoretical foundation underneath three decades of algorithm research has been pulled out.
In plain terms = everyone assumed "these problems can never be solved fast." The AI says they can — and wrote a proof. If it stands, the entire field must go back and check which conclusions still hold.
Who is verifying? How has academia responded?
ETH's two original authors — Russell Impagliazzo and Ramamohan Paturi — along with MIT complexity theorist Ryan Williams, have publicly joined the verification effort. The people who proposed the hypothesis are reviewing the disproof themselves — the highest possible standard.
Nobel physicist David Gross commented: "For those unfamiliar, the disproof of ETH is extremely surprising — the implications are too far-reaching."
OpenAI theoretical computer scientist Sebastien Bubeck called it the biggest surprise in his 20 years of following the field. This reflects the fact that even senior researchers did not expect ETH to fall this way.
Mathematics — a century-old "parent conjecture" solved in one day?
The AI extended the local smoothing conjecture — a mathematical statement about how smooth the solutions to wave equations are, posed by Sogge in 1991 — to all spatial dimensions (n ≥ 2), using a novel auxiliary PDE approach.
This means → the local smoothing conjecture, as a "parent conjecture," implies both the Kakeya conjecture and the Stein restriction conjecture. The Kakeya conjecture alone has challenged mathematicians for over a century and is the primary focus of Fields Medalist Hong Wang.
From natural-language proof to completed Lean formal verification — a computer checking every logical step — the AI took one day.
Theoretical physics — touching a Millennium Prize Problem, but not solving it
Following the constructive quantum field theory approach of Balaban, the AI produced a rigorous mathematical existence proof for four-dimensional quantum chromodynamics (QCD) and pure Yang-Mills theory. In plain terms = the AI proved, in strict mathematical language, that this physical theory is internally consistent.
The verification panel is led by Nobel laureate Gross and includes Douglas, Jaffe, Klebanov, and Nekrasov — all top-tier scholars in the field.
OpenAI explicitly stated: the positive mass gap problem remains unsolved, and a complete answer to the Millennium Prize Problem has not been reached. This reflects a deliberate restraint in framing how far the result goes.
Why did OpenAI give up authorship credit?
OpenAI classified the three manuscripts as "raw scientific data" and explicitly declined to seek academic authorship for the AI's original discoveries. It also contributed $5 million to the Simons Foundation.
The Simons Foundation, together with OpenAI and Anthropic, committed a combined $15 million in funding and compute resources to support independent review and workshops. This means → the verification process is not OpenAI grading its own homework — external, independent funding and academic expertise have been brought in.
OpenAI's math repository now holds 721 manuscripts, grouped into 373 result families; roughly 42% of top-level conclusions have completed Lean formal verification.
What is the real thing to watch here?
If the ETH disproof is ultimately confirmed by the independent expert panel, it will directly undermine hundreds of papers that relied on the hypothesis. The foundational framework of algorithm complexity theory faces systematic re-examination.
The three discoveries span mathematics, computer science, and theoretical physics, and all have launched external expert verification — this is not an isolated event in one field but a concentrated stress test of AI as a research tool.
In plain terms = the key question is not just "did AI make discoveries" but "can these discoveries survive line-by-line scrutiny from the world's top human experts." That is the real test of whether this new research paradigm is sustainable.
市场有风险，内容仅供研究参考，不构成投资建议。