It’s just like with programming: The people who are scared of AI taking their jobs are usually bad at them.
AI is incredibly good at regurgitating information and translation, but not at understanding. Programming can be viewed as translation, so they are good at it. LLMs on their own won’t become much better in terms of understanding, we’re at a point where they are already trained on all the good data from the internet. Now we’re starting to let AIs collect data directly from the world (chatGPT being public is just a play to collect more data), but that’s much slower.