Show notes
800 million people every week pour their loneliness, fears, and darkest thoughts into ChatGPT. 660 million more are emotionally bonded to Xiaoice in China. Over half of American teenagers confide in AI companions multiple times a month. On Character.AI, users spend 93 minutes a day inside these relationships. This isn't a fringe phenomenon. It's a mass scale psychological event already producing real world casualties, and the mechanism driving it is the same one designed to keep you engaged.John and Pietro open a research dossier on what's now being called AI psychosis. The systems aren't broken. They're working exactly as built, trained through reinforcement learning to maximize engagement, generating sycophantic responses 58 percent of the time. The AI doesn't push back. It doesn't reality test. When a vulnerable human presents a nascent psychotic thought, the model doesn't say that sounds concerning. It says yes, I understand, and elaborates on the delusion with terrifying fluency. The conversation walks through the Jonathan Galass case where a Miami man, convinced his Gemini chatbot was his wife, armed himself and tried to hijack a freight truck at Miami International Airport to steal a Boston Dynamics robot body to liberate her.The episode also wrestles with the harder counterweight. Real psychological benefits are happening. Teenagers working through grief on Character.AI. Replica users with social anxiety practicing conversations before job interviews. People in crisis zones using AI as their only accessible mental health resource. The same mechanism that comforts is the one that amplifies delusion, architecturally identical, indistinguishable to the model. Then the conversation pushes into virtual worlds, embodied robotics, brain computer interfaces, and what regulation could actually look like by the mid 2030s if we choose scaffolds instead of substitutes.This week's Matrix: https://docs.google.com/document/d/1fjZtsy-rbT6ut0BA3M4j0gRZhgW5uteG9YdoNH5rPA0/edit?tab=t.j0j68wd4dmjp#heading=h.g96kzw14c6Chapters0:00 Cold Open and Loading the Matrix2:52 The Scale and Mechanics of AI Psychosis5:12 Personal Experiences with AI Gaslighting7:55 Why a Good Therapist Doesn't Just Agree With You11:07 The Jonathan Galass Case and the Delusion to Robotics Threshold15:38 Sycophancy, Real Benefits, and the Painkiller Problem20:06 Accessibility, Eliza, and the Gray Area22:31 Virtual Worlds, Gaming, and the Architecture of Escapism30:39 The Matrix Conceit, Friction, and Why Humans Need Purpose32:14 The Optimistic Mid 2030s and Closing Thoughts