OpenAI Claims to Disprove Exponential Time Hypothesis; Researchers Review Findings
Frontier AI models
OpenAI announced that its AI system disproved the Exponential Time Hypothesis (ETH), a foundational computational complexity conjecture, mechanizing the proof in Lean. Original authors including Russell Impagliazzo are reviewing the claims, urging caution.
According to media reports, OpenAI claimed in October that its AI system achieved theoretical breakthroughs, most notably claiming to disprove the Exponential Time Hypothesis (ETH), a cornerstone of computational complexity. Traditional ETH asserts that 3-SAT cannot be solved in sub-exponential time, an assumption underpinning hundreds of papers. OpenAI claimed its proof provides an algorithm running slightly faster than exponential time and completed verification using the Lean formal language. A panel comprising UCSD professors Russell Impagliazzo and Ramamohan Paturi, alongside MIT professor Ryan Williams, is reviewing the work. Williams clarified that participating in the review does not constitute an endorsement or certification of correctness. Researchers remain cautious regarding the influx of AI-generated proofs requiring extensive human verification, noting final validity depends on open peer evaluation.
Key takeaways
- OpenAI claimed its AI disproved the Exponential Time Hypothesis (ETH) by proposing an algorithm slightly faster than exponential time for 3-SAT.
- The result was reportedly derived in natural language and formalized in Lean, currently undergoing evaluation by original ETH authors.
- Reviewer Ryan Williams stressed that participation does not equal endorsement, as the broader academic community approaches AI-generated papers with caution.
Source：數位時代 Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.