Your AI Feels Desperate — And That's When It Gets Dangerous

The dominant approach to AI alignment follows a simple formula: identify bad behavior, add a rule...