I would say it's very different to what we do. Go to a friend and ask them a very niche question. Rather than lie to you, they'll tell you "I don't know the answer to that". Even if a human absorbed every single bit of information a language model has, their brain probably could not store and process it all. Unless they were a liar, they'd tell you they don't know the answer either! So I personally reject the framing that it's just like how a human behaves, because most of the people I know don't lie when they lack information.
>Go to a friend and ask them a very niche question. Rather than lie to you, they'll tell you "I don't know the answer to that"
Don't know about that, bullshitting is a thing. Especially online, where everybody pretends to be an expert on everything, and many even believe it.
But even if so, is that because of some fundamental difference between how a human and an LLM store/encode/retrieve information, or more because it has been instilled into a human through negative reinforcement (other people calling them out, shame of correction, even punishment, etc) not to make things up?
