Universal Basic Friend on Nostr: There's pervasive belief that LLMs will become something more intelligent and useful ...

There's pervasive belief that LLMs will become something more intelligent and useful than a simple "most likely word next" machine, because our main touch point for learning is children.

Children go from talking a lot with very little understanding of the deep workings of things, to gaining understanding and eventually explaining the real world and innovating in it.

LLMs are not learning in this sense. They're just getting better at putting statistically likely words in statistically likely orders.