I think you make a good argument. An LLM can't "be tired", so it is clearly lying and/or mimicking what a human does.
But I have two counter-arguments:
- maybe the LLM thinks it is tired, because it thinks that it is a human or behaves like one. And thinking you are tired while not really being it is something humans happen to do. (And humans are conscious right?)
- alternatively, maybe the LLM says it is "tired" in a colloquial form, i.e. it is not really "tired" but it has something analogous to it. Maybe it is annoyed by the conversation and decided to use that word?
Oh that's interesting! In my mind we are now on the cusp of being able to scan all these archives and have them be read by LLMs (in a first pass). Do you agree with that assessment, or am I being naive here?
This already exists! There's a company called Carbon Robotics that has a prototype of a robot that eliminates individual weeds with a lazer: https://www.youtube.com/watch?v=_2s-0wgQWXM
I'm also looking forward to this technology being widely available and helping solve the issue of herbicide run-off, and later additionally the same for pesticides and fertilization.
So the article mentions "regularization" as the secret ingredient to get to a generalized solution, but they don't explain it. Does someone know that is? Or is it an industrial secret of OpenAI?
But I have two counter-arguments: - maybe the LLM thinks it is tired, because it thinks that it is a human or behaves like one. And thinking you are tired while not really being it is something humans happen to do. (And humans are conscious right?) - alternatively, maybe the LLM says it is "tired" in a colloquial form, i.e. it is not really "tired" but it has something analogous to it. Maybe it is annoyed by the conversation and decided to use that word?