I think what is also interesting is what LLMs help us learn about ourselves.
I think we understand human intelligence a bit better because of LLMs, and I think people have been surprised how far you could get just by processing text.
I think (just a guess relying on intuition and my very shallow survey) that human intelligence is multilayered where we have something at least vaguely analogous to a LLM combined with something else that knows the physics of the world as well as something else that is able to do logic and symbolic processing.
But this is exciting to see the power of relatively simple neural networks when they are trained with a giant corpus.
I think we understand human intelligence a bit better because of LLMs, and I think people have been surprised how far you could get just by processing text.
I think (just a guess relying on intuition and my very shallow survey) that human intelligence is multilayered where we have something at least vaguely analogous to a LLM combined with something else that knows the physics of the world as well as something else that is able to do logic and symbolic processing.
But this is exciting to see the power of relatively simple neural networks when they are trained with a giant corpus.