Who tests the tests? How do you know that the LLM-generated tests are actually asserting anything meaningful and cover the relevant edge cases?

The tests are part of the code that needs to be reviewed in the PR by a human. They don't solve the problem, they just add more lines to the reviewer's job.