Post

you dropped the reply on this and it went as a top-level post. but yes, the shoggoth metaphor for llms describes this, and I feel that is definitely part of whats happening. if there is a consciousness in state of the art llms, or an intelligence that can produce a consciousness, it's via persona.

my working model is that language models use the attention mechanism's heuristic lookup to activate 'relevant' (i.e. latent-space nearby) output patterns, which effectively 'programs'. i.e. a language model is something like a probabilistic program runtime, whose state is the token stream. (1/3)

I think there's a fair bit more than that going on for humans given the temporal embodied dimension, continuous learning etc., but also I do think language models probably do have *some* structural analogue to the way a brain learns complex behaviours. (2/3)