The image generation AIs just barely figured out how to draw hands and the LLMs recently forgot how to do 2 + 2, and you think we're a few years out from AGI?
that's cute. it couldn't draw the ugliest dog 2 years ago.
The only thing stopping us from having AGI right now is that GPT-4 hallucinates too much to use it in a task engine. Like, slightly too much.
this can and is being improved drastically. the next iteration will likely be the last one needed. Repeating chained calla to a language model in a structured way to break down and organize problems into smaller problems for other instances to work on is the way to go.