3 ms·
> Although I would argue that the output AI makes still doesn’t match the complexity and nuance of an individual human being, though LLMs are very good at nuan
by ben_w 1y ago
> Although I would argue that the output AI makes still doesn’t match the complexity and nuance of an individual human being, though
LLMs are very good at nuance. Better than any human I've seen — so much so, I find it to be a tell.
> We have numerous ways of determining if a thing is human or not - biology, for one.
I don't care if the intelligence is human, I care if it's (1) (a) intelligent, (b) educated, and (2) has the ability to suffer or not so I know if it should have moral subject rights.
1a is slowly improving but we're guessing and experimenting: not really engineering intelligence, just scaling up the guesses that work OK. 1b was always easy, libraries fit "education" in isolation from the "intelligent" part of 1a. LLMs are good enough combination of (a) and (b) to be interesting, potentially even an economic threat, depending on how long the time-horizon between failures gets.
2 is pre-paradigmatic, we don't have enough understanding of the problem to ask the correct question — even ignoring AI for the moment, the same problem faces animal welfare (and why would the answer be the same for each of chimps, dogs, ravens, lobsters, and bees?) and even within humans on topics such as abortion, terminal stage of neurodegenerative conditions such as Alzheimer's, etc.