AI can't be trusted in general, at least not for a long time. It gets basic facts wrong, constantly. The fear is that it will start eating its own dogfood and being more and more wrong since we are putting it in the hands of people that don't know any better and are going to use it to generate tons of online content that will later be used in the models.

It does make some queries much easier to find, for instance I had trouble finding out if the runner ups got the win in the Tour De France after the Armstrong doping scandal and it answered it instantly. The problem is that is offers answers with confidence, I think them adding citation is an improvement over ChatGPT, but it needs more.

Luckily, it's still a beta product and not in the hands of everyone. Unfortunately, ChatGPT is, which I find more problematic.