Experts find flaws in tests that check AI safety and effectiveness | Not Hacker News!