AI art and language are so complex but competent now. They are such flexible and powerful models of art and language. The art models in particular figure out reflection, shading, refraction, and all these other crazy things not explicitly labeled.

Five years ago I'd have told you this was like a decade or two away. These developments have really blown my socks off.

New AIs also have fantastic problem solving and task completion skills in novel environments. It's slightly under the radar but equally impressive.

I think with very multimodal datasets, we will have AGI, but frozen in its ability to upgrade.

If we add real time updating, and allow it to take summaries of its own internal state as data, we will have AGI that is every bit as flexible as a human being.

(It will also immediately be ASI because it has so much knowledge and media generation built in)

I think the next decade we'll see at least 4 or 5 more top tier Wow! moments on par with Dalle-2/SD AI art and GPT 3, and then we'll be basically at AGI.