Earlier quoted context omitted.
Perhaps a bad example, since ChatGPT consistently gets the steak temperature right (or at least gives the same values as Google). Internal temperature of about 130-135°F (54-57°C) for medium rare, etc.
“It tells me the right answer” and “I trust it to give me the right answer” are two different propositions.
It reminds me of when Wikipedia was new and we were told repeatedly that we couldn't trust it (by teachers, articles, etc.). It didn't matter if we could point to studies that found Wikipedia had similar accuracy to other encyclopedias. They objected on the grounds that anyone could edit it and anything on it _could_ be wrong and there's no publisher or paid editors to ensure accuracy.
ChatGPT tends to do well on common questions, where the answer is plastered in hundreds of articles across the internet. The internal cooking temperature of a steak is a great example of this. There are many other types of questions it fails at.
A better example of where you shouldn't trust ChatGPT is asking it the distance between two cities. It'll get it right for a few common pairs of cities (eg London to Paris), but it'll give you a wrong answer for most less common pairs (eg London to Cairo).