And if two people can get two opposite results by giving the same prompt which asks a very specific question to the same model, it looks like bunk anyway. LLMs don't care if they are correct.
Why is that relevant to GP's point?
I can't speak for anyone else, but I come to HN to discuss stuff with other humans. If I wanted an LLM's (it's not AI, it's a predictive text algorithm) regurgitations, I can generate those myself and don't need "helpful" HNers to do it for me unasked.
When I come here I want to have a discussion with other sentient beings, not the gestalt of training data regurgitated by a bot.
Perhaps that makes me old-fashioned and/or bigoted against interacting with large language models, but that's what I want.
In discussion, I want to know what other sentient beings think, not an aggregation of text tokens based on their probability of being used in a particular sequence determined by the data fed to model.
The former can (but may well not be) a creative, intellectual act by a sentient being. The latter will never be so, as it's an aggregation of existing data/information as a sequence of tokens cobbled together based on the frequency with which such tokens are used in a particular order in the model's corpus.
That's not to say that LLM are useless. They are not. But their place is not in "curious conversation," IMNSHO.