ChatGPT is Bullshit
I read Frankfurter’s essay more than once (credit to John Gruber for the recommendation).
It was relevant in 2016, and is even more relevant today. Whether it pertains to a US president, or to LLMs, the distinction between lying and bullshitting is essential:
Persons who communicate bullshit are not interested in whether what they say is true or false, only in its suitability for their purpose. Wikipedia
Now we have to extend “persons” to LLMs. And this is why some folks have expanded on this and published a paper:
If you understand the technology and its flaws, the best explanation and excuse might be, quoting Frankfurter directly:
It is impossible to lie without knowing the truth.
LLMs don’t know the truth. They are made of inert weights, reacting to input. Hence, they bullshit.