@root I've been working on AGI for 35 years now. I don't think there's anything "special" about the human brain that can't be replicated algorithmically, but I still think we are a long way from emulating one with technology.

LLMs are not the way; that much is dead certain. There are so many things a human brain does that an LLM can't do, ever. LLMs learn to follow the statistical patterns in the *outward expression* of thought, not how to actually think. LLMs don't have emotions; they just imitate them. LLMs don't learn from experience; they just add stuff to a massive short-term memory (the context window) to compensate for that. LLMs can't generalize knowledge of familiar domains to unfamiliar ones; even asking an LLM to combine two familiar ideas in a novel way is too much novelty for them to handle, and they consistently break down when you do.