Arvind Narayanan argues that reductive descriptors like 'stochastic parrots' have become markers of shallow AI critique

@asymmetricinfo Whenever someone says ”ai will NEVER be able to xxxxx”, I say ”wait 6 months”. I’m old enough to remember when computers would NEVER be able to beat humans at chess.

@asymmetricinfo They are all accurate(ish) descriptions of what an LLM actually is conceptually. Losing sight of the reality is what leads to nonsense arguments about sentience.

@asymmetricinfo Removing the term “stochastic parrot” from the discussion just because time has passed effectively would concede that Claude became conscious sometime between 2023 and today. And wouldn’t that have to be proven first?

@asymmetricinfo There’s this assumption that the human brain is not mechanical at bottom. I’m not sure what they imagine axons and dendrites are doing down there. All intelligence has to be built from small unfamiliar building blocks.

@asymmetricinfo The interesting question is not whether AI's are all these things. They certainly are. The interesting question is whether we are, too...

@asymmetricinfo @AndyMasley There's nothing incorrect about any of these phrases, this sounds like a political declaration more than anything

@asymmetricinfo Interesting. I haven’t yet seen anything (and I use it every day) that is truly creative or intelligent (though I find that hard to define). I think that AI has two extraordinary powers: to find analogies, and to build on prior work in logical steps. Both of those can feel like

@asymmetricinfo I feel the same way about "slop"

@asymmetricinfo What nobody gives me a sense of is hallucination frequency, and how much it has been reduced. My impression is significantly, but I'm not confident of that, and I am confident that peesuasiveness has improved significantly. That kind of leaves me uneasy.

@asymmetricinfo But...this is twitter. We're a bunch of shallow critics who haven't updated our views since the 1900's.

@asymmetricinfo Preach. Depth over buzzwords always.

@asymmetricinfo Yea and the most obvious one is missed: Tom Riddles Diary.

@asymmetricinfo Outermost knowledge, and your knowledge may well extend on a narrow front deep into the center of the sponge/tree, but you will almost certainly have left gaps around less interesting avenues even in your specialist area. And often, those gaps will be directly analogous to

@asymmetricinfo Magic to even the most sophisticated user. My mental model for the world’s knowledge is a sponge or tree: looking pretty ‘solid’ at its surface, but with a huge amount of empty space inside. If you are the most sophisticated person in the world in one field, you may have the

@Alterschwede66 @asymmetricinfo About AI, we regularly say, "This is the worst it will ever be.

@asymmetricinfo "Plausibility engines" is an apt framing when you're talking about reward hacking. In this case, the outputs are whatever is plausible to the grader

@asymmetricinfo Process easy and makes these connections across regions of human knowledge that a human is incredibly unlikely to find. So even the most extraordinary thought leader in a domain is highly likely to find gaps in their direct area of expertise filled in - which, once they stop

@asymmetricinfo Problems that have already been solved in a related or totally unrelated area. Now even the greatest polymath can only hold a tiny amount of information or make such linkages across domains - and often great progress can be made by these amazing people. But AI finds that

@asymmetricinfo Stochastic Parrots is an excellent band name though

@thomasbrookside @asymmetricinfo It's literally impossible to prove that anything is conscious.