LLMs are just the latest in a very long line of disparate attempts at making AI, and is arguably the most successful.

That doesn't mean the approach isn't an evolutionary dead end, like every other so far, in the search for AGI. In fact, I suspect that is the most likely case.