In all the fields I've tried (software/music/epidemiology/geosciences/mathematics) ChatGPT can mostly explain things at the lay level, but even there often gets things wrong that without more expert guidance, you will not notice.

Experts in each of these areas were consistently better at giving a correct explanation.

ChatGPT is great when occasionally being wrong doesn't matter. It's still pretty good when you can't be wrong, but you're an expert. But in any "no mistakes, I need to talk to an expert" situation, ChatGPT is setting you up for failure.

That's the key part - the information is unreliable/untrustworthy. As long as that is not a problem, it's awesome.

The problem is that it presents wrong data with absolute confidence in a very convincing voice. (Hence people calling it a con artist). It'll become a huge problem as it makes inroads in mis/disinfo campaigns.

(The usual tech problem. Don't let large scale interests touch it, and you'll be OK)