Shane Legg (DeepMind Founder) - 2028 AGI, New Architectures, Aligning Superhuman Models
Dwarkesh Podcast
00:00
Richard Sutton's perspective on scaling search and learning in AI
Richard Sutton believes that there are two things that can be scaled: search and learning. LMS focuses on the learning aspect, while the search component, which involves an agent interacting with the environment, needs further exploration. To solve complex problems, searching through possibilities is essential. For example, AlphaGo's famous Move 37 came from searching and discovering a hidden gem. Current language models lack this creative capability.
Transcript
Play full episode
