"We're almost at AGI" is still an unsupported inference.
A GI must be able to extract valid strings from raw environmental data streams—video, audio, etc.— decompose those streams into distinct fragments, and correlate them. And it has to be able to update its model on the fly.
Anyway, intelligence is only the first step. Creativity is a different paradigm. What we've seen so far is equivalent to a million hacks who have never experienced the worlds they are trying to describe.