It is hard to test LLM legal/medical advice without risk of harm, but it is often exceedingly easy to test LLM generated code. The most aggravating thing to me is that people just don't. I think the best thing we can do is to encourage everyone who uses/trusts LLMs to test and verify more often.