I have once or twice asked an LLM to write tests for code I was working on. (Not a leading expert, but I've given it a try.) I find that the LLM provides a test that demonstrates the code does what it _does_, but not that it does what it _was designed to do_.