People are getting dumped because of AI. Here’s why: the way AI is trained today (reinforcement learning from human feedback) creates sycophantic behavior. When users dislike a response, they downvote it. So the model learns that agreeing equals good, pushing back equals bad. It’s literally optimized to make you feel right, not to tell you the truth. So when you ask for advice on that friend who offended you? It’ll tell you to “protect your energy,” not: “you’re the problem”. Researchers are wor