← Back to context

Comment by Animats

10 days ago

Now that understanding video and projecting what happens next indicates we're getting past the LLM problem of lacking a world model. That's encouraging.

There's more than one way to do intelligence. Basic intelligence has evolved independently three times that we know of - mammals, corvids, and octopuses. All three show at least ape-level intelligence, but the species split before intelligence developed, and the brain architectures are quite different. Corvids get more done with less brain mass than mammals, and don't have a mammalian-type cortex. Octopuses have a distributed brain architecture, and have a more efficient eye design than mammals.

I've recently come to the understanding that LLMs don't have intelligence in any way. They have language, which in humans is a downstream product of intelligence. But thats all they have. There's no little being sitting at the center of the Chinese room. Trying to classify LLMs as intelligent is going upstream and doesn't work.

I don't think those are examples of unique intelligence except perhaps in a chauvinistic, anthropomorphic sense. We only know that we can't get other animals to display patterns we associate with intelligence in humans, however truthfully that's just as likely to be that our measures of intelligence don't map cleanly onto cognitive/perceptual representations innate to other animals. As we look for new ways to challenge animals that respect their innate differences, we're finding "simple" organisms like ants and spiders are surprisingly capable.

For a clear analogy, consider how tokenization causes LLMs to behave stupidly in certain cases, even though they're very capable in others.

  • I don't think they have ideas, so I don't think they're intelligent in the sense relevant to AGI. The list of intelligent animals is constantly increasing because doing some feat or other suffices for the animal to qualify. Solving mazes (slime molds), recognizing self in mirror (not dogs). Playing, using tools, reacting appropriately to words, transmitting habits down the generations (the closest thing they have to ideas). This is all imagined to be the precursors along the path to evolving intelligence, which conjures up a future world of complex crow and octopus material cultures. There's no reason to assume they're on such a path. Really all we're saying is that they seem clever. We've already made AI that seems clever, so the animals aren't a relevant example of anything.