Will make time to watch this. His older deep dive videos made me realize llms are "just" very lossy compression algorithms which, by maximizing chance, spew out better content than what was originally ingested when inspired by the right prompt. Great for those of us who need to "get it"