And I don't think it's quite right to say that they don't care about being true. If we're taking the stance that the creators are responsible for the actions of their models, there's real incentive to have them *not* hallucinate and it's why frontier models are doing it much less today

Show Post

FWIW both of the papers cited in the wiki summary about LLMs are from 2024 and discussed GPT-4. Specifically the ChatGPT is Bullshit paper has characterizations that just aren't true of the frontier models today. They were true. But things have changed significantly.

Show Post

I’m not AI evangelist but this doesn’t fit at all

Show Post

Do you have examples of frontier LLMs producing “bullshit” as opposed to accurate information? What led you to believe this is the case?

Show Post

"They're made out of weights!” maxleiter.com/blog/weights

Show Post