[nonethewiser]

In some way tests are the worst thing for LLMs to write. Because you have to verify what the LLMs generate. Are you going to write tests for tests? No.

I dont think this _necessarily_ makes them bad for writing tests but you really need to be careful and examine what it gives you and even tweak it to fail in a way that verifies it's working. Conversely, you could write the tests yourself and then literally copy-paste the code-to-be-tested without any significant examination.