An Observation on Generalization and Unsupervised Compression
Connects Kolmogorov complexity and Solomonoff induction to next-token prediction, arguing that optimal sequence compression forces models to recover shared world structure.
Co-Founder of SSI & Former Chief Scientist of OpenAI
Self-contained public lectures, keynote addresses, and research seminars.
Connects Kolmogorov complexity and Solomonoff induction to next-token prediction, arguing that optimal sequence compression forces models to recover shared world structure.
Ilya Sutskever of OpenAI presents on advanced topics in meta-learning and self-play within the context of Artificial General Intelligence, delivered for an MIT course.