I think that's a pretty common opinion here. Here's what I thinks: LLMs are not AGI but are currently the best way to leverage everyone else's knowledge.
LLMs are fossilized intelligence. It takes a lot of energy to train them and have them learn new things (Calling RAG learning is like saying a human has memory because they have access to Google).
Therefore it is limited. It might do AGI in a universe where we could harness unlimited energy and create unlimited matmul compute. But if it needs to run on 3 square meals a day it doesn’t have much hope!
Therefore it is limited. It might do AGI in a universe where we could harness unlimited energy and create unlimited matmul compute. But if it needs to run on 3 square meals a day it doesn’t have much hope!