I really appreciate the fact that the Google search summaries happily will make up false information about me or claim that I have said things that I have never even written in the first place.
@atax1a I got it to leak confidence intervals once by just rapidly refreshing the page.
Good technology.
@atax1a I’m still of the belief that language models are not the way forward for real intelligence, but only are a poor simulacra at best.
I honestly suspect that the success of these things is tied to the fact that they output in natural language, which provides unconscious incentives to anthropomorphize the machine, while also riding on top of wishful ideas from science fiction.