A modern spin on this technique is as follows: Write (or use an LLM to write) something so simple that it is both obviously correct and very easy to verify the correctness of. Then, use that same LLM to create a comprehensive suite of tests, which further prove the correctness of the simple implementation. Once the tests are there, let the LLM run wild and ask it to optimize the hell out of the implementation while keeping the tests untouched.