Hi！ How much can we rely on the answers GPT gives us?

I think it depends on the quality of the prompt. You can and should be qualifying the prompt such that there's some threshold of confidence that must be met. You could say something like "give me the answer to x provided that y condition is met, and if y condition is not met, then tell me z instead."