OpenAI and Anthropic claim AI models have disproved the Exponential Time Hypothesis
Summary
- OpenAI's GPT‑4o and Anthropic's Claude 3 reported solving certain NP‑complete problems in sub‑exponential time, challenging the Exponential Time Hypothesis (ETH).
- The claims have sparked heated debate among complexity theorists, who caution that independent verification and peer‑review are essential.
- Both firms released technical notes but have not submitted the results to conventional academic journals, raising concerns about reproducibility.
- If validated, the breakthroughs could reshape algorithm design, cryptographic security assumptions, and the perceived limits of AI‑assisted research.
Key Facts
- · The Exponential Time Hypothesis posits that 3‑SAT cannot be solved in time better than 2^{Ω(n)}.
- · OpenAI’s model allegedly solved 3‑SAT instances of size n in time o(2^{n}), i.e., sub‑exponential.
- · Anthropic reported a similar speed‑up for the k‑clique problem, another ETH‑linked challenge.
- · Both announcements were made via pre‑print style blog posts without formal peer review.
- · Leading complexity researchers have called for independent replication before accepting the results.
- · A confirmed violation of ETH would have far‑reaching implications for cryptography and computational theory.
Why it matters
The ETH underpins many assumptions about the hardness of computational problems used in cryptography and algorithm design. A credible refutation would force a re‑evaluation of security protocols, influence future AI‑driven research methodologies, and highlight the need for robust verification mechanisms for AI‑generated proofs.
#AI#computational complexity#Exponential Time Hypothesis#OpenAI#Anthropic#research verification