Anthropic and OpenAI each say an AI model refuted the Exponential Time Hypothesis, a 1999 rule that says some yes-or-no puzzles have no real shortcut. Both say a proof-checking program (Lean) accepted their steps. No person has finished checking them yet.
Does it make computers faster? Anthropic’s own paper says its proven speed-up only beats trying everything once a puzzle has more than 10^10,000 switches. OpenAI’s paper gives no number, and its announcement says practical gains “will require significant further research”.
Is this P vs NP? Not quite. ETH is a bolder version of P ≠ NP: if ETH were true, P ≠ NP would follow. Breaking ETH does not settle P vs NP; the claimed speed is still far from fast.
The 1 followed by 10,000 zeros, and 10^80 atoms in the universe, are our arithmetic and a standard estimate, not numbers from either paper.
Full Video:
Sources
OpenAI, “The Exponential Time Hypothesis is False”
Anthropic, “A 2^o(n) Time Algorithm for 3-SAT”:
Music: “Dreamer” by Kevin MacLeod (incompetech.com), CC BY 4.0,
#AI #Math #Shorts #OpenAI #Anthropic
source

For the experts: ETH is a stronger form of P ≠ NP, so breaking it would not settle P vs NP. The claimed speed is still far from polynomial.
Do you think the proofs will survive the human check? We'll post what the reviewers find.