Representation of Facts in AI Systems and the Role of Interpretability

(Overtime segment available to paid subscribers below the paywall.)

0:00 Sam’s Bloggingheads bona fides 2:39 Is AI “interpretability” worth the trouble? 13:38 Natural selection and artificial intelligence 19:47 Are large language models absorbing human nature? 29:53 What Sam thinks singularitarian AI doomers get wrong 36:36 Is it time for an AI-safety Manhattan Project? 51:12 Sam: Let a thousand GPT-4 (but not GPT-5) plug-ins bloom. 57:09 Where are our AI safety blindspots?

Robert Wright (Bloggingheads.tv, The Evolution of God, Nonzero, Why Buddhism Is True) and Samuel Hammond (Foundation for American Innovation, Second Best). Recorded May 30, 2023.

Comments on BhTV: http://bloggingheads.tv/videos/66231 Twitter: https://twitter.com/NonzeroPods