The thing is AI is already "good" for a lot of things. It all depends on your definition of "good" and what you require of an AI model.

It can do a lot of things that are generally very effective. High reliability semantic parsing from images is just one thing that modern LLM's are very reliable at.

You're right. I use it for api documentation and showing use cases, especially in languages i don't use often.

but this other attribution people are doing- that it's going to achieve (the marketing term) AGI and everything will be awesome is clearly bullshit.