[georgeecollins]

OK I am bracing for the downvotes..

What about when the llm is smarter than the person? Sometimes I get material that is so bad I wish they had had AI do it. Then it would be poor to mediocre.

There was an episode of the podcast “Question Everything “ where they talked about how LLM s can sometimes talk people out of conspiracy theories by patiently refuting the arguments with facts. There have been academic studies on this.

I think people hate AI because it is often mediocre and flawed but sometimes it’s replacing humans that are inept.

[OptionOfT]

If the LLM is smarter than the person, the person cannot look at the output and judge to see whether it is correct.

[anigbrowl]

True, but that's a people problem.

[abalashov]

I thought it was pretty well-established that conspiracists aren't deterred with facts, and that their commitments to the conspiracy theory are essentially emotional and/or in-group identitarian?

[georgeecollins]

You should definitely listen to this podcast which is an interview with a Professor of Psychology at Columbia. He also thought what you thought-- that you can't convince a Q-Anon type-- and they were actually testing something else. When they found out that 25% of people who believed conspiracy theories actually changed their mind after three exchanges with an LLM.

They repeated the test and the results were replicable. (But still only about 25%, but that is something.)

This made me think: I grew up in a world where there was a flawed but consensus view of the world, its problems, institutions motivations. This came from a common mass media. Maybe getting our answers from AI will lead us to a new (inevitably flawed or even bad) consensus. Weird.

Source: https://play.cdnstream1.com/s/kcrw/question-everything/can-a...

[abalashov]

Very interesting. Thank you for that, and I say so with due recognition that my response wasn't really aimed at the central idea of your comment. Thank you for humouring it anyway!