The "confident idiot" problem: Why AI needs hard rules, not vibe checks | Not Hacker News!