Microsoft Research has similar studies.

@tante the boosters do not care. I had a coworker tell me to my face yesterday that we need to stop relying on "internal expertise" at our company, and instead hand that off to LLMs to write tests to validate everything. This was after I pointed out to him that the majority of tests the LLM wrote for him were fake, and tested nothing. You could write code that severely broke what the LLM has written for him and the tests would have continued to pass. How can we give up on the expertise we have, and give up on further building that expertise, if the tools aren't capable of doing the work? If you can't validate the output then you're guaranteed to create a catastrophic failure in the future.