Evidence item

Google DeepMind reports AlphaTensor discovering novel, provably correct matrix-multiplication algorithms from reinforcement learning, including methods more efficient than prior human-designed algorithms; this is evidence against the blanket claim that AI merely recombines training data. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 9899

Source: Discovering novel algorithms with AlphaTensor · locator {} · excerpt hash sha256:829d4ee10f16a8eea3a81bde8feb3ca1735f6378cd6faadb6cae49dab97e7462

We introduce AlphaTensor, the first artificial intelligence (AI) system for discovering novel, efficient, and provably correct algorithms ... AlphaTensor discovered algorithms that are more efficient than the state of the art for many matrix sizes. ... We then trained an AlphaTensor agent using reinforcement learning ... starting without any knowledge about existing matrix multiplication algorithms ... discovering algorithms faster than previously known.