If you can use it, it's useful. That being said, not every useful thing is safe. My kitchen knife is useful, but if I operate it absent-mindedly then I get cut

If you can use it, it's useful. That being said, not every useful thing is safe. My kitchen knife is useful, but if I operate it absent-mindedly then I get cut. AI is largely the same way; you can trust it to do it's job, but you can never trust it to work safely without supervision. So, AI really just needs to be applied right. People are right to squabble over how it generates it's answers, because largely that decides whether the output is fact or fiction. In this case, ChatGPT is trained on a combination of facts and fiction, with no way for it's model to differentiate them.