4 ms·

> They can and do say "I don't know". Parent poster never claimed they couldn't generate the text of "I don't know." But when an LLM generates "I don't know",

• redlock 2y ago
  Parent poster did say they are aware they don’t know but can’t express it. I am guessing he is referring to mechanistic interpretability research like these: https://arxiv.org/abs/2405.16908 https://arxiv.org/abs/2405.16908 https://arxiv.org/abs/2407.03282 https://arxiv.org/abs/2407.03282 You are claiming they are statistical parrots, which I don’t think the parent poster meant. The “statistical parrots” argument might have been compelling with GPT-3, but not with today’s models and the results of mechanistic interpretability research, which show internal representations and rudimentary world models.
• elpocko 2y ago
  I pasted those examples from an earlier, less nuanced discussion; the last example was tongue-in-cheek and I probably shouldn't have pasted it here. Still: "AI don't know what they don't know" is not universally true and shouldn't be taken at face value. They clearly are able to tell when they're missing knowledge to answer a question. That is distinct from giving the wrong answer because the wrong tokens were sampled for whatever reason.