I think this is terrible, but I suppose I think it is less bad if ChatGPT didn’t get the phone number from the bank’s files/data.
If ChatGPT just provided it from public info/the training set, am I wrong to think that isn’t as bad?
Just for clarity, I think it’s not a good idea to use LLMs if you care whether the answer is right or wrong. So this is a terrible use case.