Your AI Coach Is Lying to You: What Two New Studies Found | 357

We dig into two new 2026 studies that audited five popular AI chatbots on health questions, and athletic performance came back as the weakest category of the whole audit. We cover why an LLM is a prediction machine and not an all knowing coach, the sycophancy problem that lets you argue a chatbot into agreeing with whatever you already believed, and the fake citations researchers hit when they asked for references. We also get into the hard line on steroids, peptides, and supplement advice, where AI falls short on injuries and physical therapy, why a wall of jargon is usually a sign nobody understands the question, and how Jerred is thinking about guardrails before an AI coach ever shows up inside the Garage Gym Athlete app. Chapters: 0:03 AI Study Sparks Fitness Warning 4:26 Athletic Performance Scores Worst 8:37 Better Ways to Use AI 12:29 AI's Sycophancy Problem 16:49 Dangerous Advice on Drugs 23:31 Fake Citations and Hallucinations 26:29 Guardrails for Future AI Coaching 34:53 When AI Gets Too Complex