I've found ChatGPT to almost never be wrong, can't think of an example ATM. Having said that, I have a sense for what it can and can't do, what sort of inputs will output a solid answer.

Where it goes hilariously sideways is if you talk to it like a person and keep following up. Hell no. You ask a question that can be answered objectively and stop.

No way the output went straight to, "Sure! Bromine's safe to eat." Either he asked a loaded question to get the answer he wanted or this came after some back and forth.