One of those things about claims of “AGI” is that to really build a human-like intelligence we’d have to simulate a human brain as whole because 1. it’s the only thing we know that produces human intelligence and 2. no one knows how it actually works.
Because of (2) it’s irrelevant what anyone says about LLMs or any other technology (with the exception of simulating a whole brain)—you can’t know if a technology is intelligent like a human because we don’t know what that means or how that works.
Fun thing, it turns out it’s impossible to simulate a whole brain with the resolution required (basically quantum physics level), and you’d have to emulate a chemical and physical environment for the brain as well (it will also need a body etc.).
You’d also have to simulate other humans with brains from which the brain can learn; but to simulate those you’d have simulate evolving humans from single-cell organisms first etc etc ad infinitum