Article

Claude independently breaks the 27 exponential time hypothesis (ETH), surprisingly discovered on the same day as OpenAI

Beating AI flash update: OpenAI announced only yesterday that AI had overturned the 1999 exponential time hypothesis ( proposed in ETH), and today Anthropic also released Claude's independent research findings. Remarkably, the discovery dates disclosed by the companies were the same 1 day:October 16。

ETH holds that certain computational problems are extraordinarily difficult: regardless of how an algorithm is optimized, computation time for the hardest instances still grows exponentially as problem size increases. The models respectively found subexponential-time algorithms that break through this limitation, and can even calculate exactly how many qualifying solutions a 3-SAT problem has, rather than merely determining whether a solution exists.

Anthropic disclosed that Claude independently found the core algorithm without human input during the process, producing approximately 14000000 Token. Anthropic researchers subsequently worked with Claude to organize and refine the paper.

Both parties have now released papers and Lean formal proofs, pending further verification by the academic community.

Original link https://m.theblockbeats.info/flash/373904