HackerTube / Video
An Observation on Generalization
Simons Institute for the Theory of Computing
Ilya Sutskever gives a Simons Institute talk on why neural networks generalize, framing learning as finding short descriptions, like compression. It was part of the Large Language Models and Transformers program.
Cited in discussions of learning as compression and of why neural networks generalize.
3 Hacker News mentions · 2023–2026