What was already known
Two levels of Strassen’s method use 49 scalar multiplications for this setting.
What changed and why it matters
A concrete improvement to a long-studied small-matrix computation, discovered through reinforcement learning.
The AI contribution
AlphaTensor uses reinforcement learning to search for tensor decompositions representing algorithms.
Limitations
This claim is specific to arithmetic over F₂ and fixed dimensions. It is not a new bound on the general asymptotic exponent, nor a universal wall-clock speedup.
Verification evidence
Evidence labels show what supports the claim. Importance rates its impact if correct.
Author claim
Claim reported by the authors.
Accessible artifact / proof
Authors’ public code and supporting material; not independently reproduced here.
Peer review
Published in Nature; peer review does not establish infallibility.
Correction and publication history
Historical seed, researched for the initial catalog. Initial alerting disabled.
Read this published revision