[krupan]

What does this even have to do with the parent? Your capabilities have nothing to do with LLM capabilities. The two work in completely different ways. The reason LLMs work is because they are huge and have been trained on vast amounts of data, full stop. Sure, there's potential someday to get something useful using less data, but we aren't there.

[avadodin]

You are right on the limitations of the architecture but I wouldn't call LLMs huge. Flagship models maybe but that's just because they don't scale very well.

A universal translator with image and voice recognition and a decent breadth of encyclopedic knowledge in only a small fraction of an English Wikipedia dump(6GB/20+GB) is not "huge".

It is probably closer to the theoretical limit than anyone could have expected.