[zamnos]

What you've just asked is essentially "what's going to be on the test"?

Unfortunately, real life isn't a class at school, there isn't a professor, and no one can tell you what's going to be useful vs useless because ChatGPT's just as new for everybody except the people who made it. My time machine's just as good as your time machine, so whatever I'd try to predict is going to be just as wrong as it is right.

If you want to work in AI for the foreseeable future figure out what you want to know about in that realm. Personally I'm learning the math underlying LLMs and learning the principles that go into creating a GPT and how to make my own. Even though I can't afford to train my own, understanding the underlying concepts is interesting to me in its own right. Other people are interested in other aspects of LLMs that don't interest me, but interest them.