LLMs are not made for seeking truth. They predict what's most likely to be said next. And what gets said most is whatever sounds confident, agreeable, and clean. Finding truth costs you comfort, dead ends, and contradictions. Weeks of sitting with "I don't know." A good advisor tells you things you'd rather not hear. Real research involves arguments and retractions, scientists proving each other wrong over decades. Science looks messy and confusing from the outside. AI looks simple and convincing. But scientists grinding through that mess get closer to truth over time. You, asking the AI, get closer to what sounds right. Ask ChatGPT if your strategy is solid. It'll find five reasons you're right. Ask it to poke holes in your plan, and you'll get gentle pushback wrapped in encouragement. These models optimize for your approval rating. Truth and approval overlap sometimes. Less often than you'd think. You're using a prediction engine for truth-seeking. Those are different jobs.