Current LLMs are limited by the fact that they can’t train during inference. i.e. They can’t learn

Learning is much more than just training during inference.

True unsupervised learning is still in its infancy, and requires a completely different architecture than the current fancy auto-suggest boom. Unsupervised learning has been the dream since at least the 1950s when AI was an aspiration, even before transistors existed. Nobody has figured out how to do it yet.

If pigs learn to fly, then things are going to actually be different.