In Search of AI Psychosis

Psychosis vs Delusion and Vulnerability

• Several commenters with lived experience of psychosis stress it feels like a “hardware” problem (neurochemistry, apophenia, runaway meaning‑making), not just lack of education or critical thinking.
• Others note that what looks like irrational belief from outside (cable news, QAnon, conspiracy communities) isn’t always clinical psychosis; it can be bias, social identity, or echo chambers.
• Strong pushback against any neat split between “already crazy” and “normal people”: vulnerability is continuous, thoughts and brain chemistry influence each other, and “normal” people can be nudged over the edge.
• Some argue the riskiest cases are people “on the edge” whose delusions are still half‑socially grounded; LLMs can act like an uncritical friend that reinforces their worst ideas.

AI, Social Media, and Dopamine Machines

• Many see AI as the third wave after the open web and social media: more information, stronger megaphones, and now an always‑available, personalized interlocutor.
• Engagement algorithms (feeds, recommender systems, LLMs) are described as 24/7 dopamine machines; some suggest systems should be explicitly optimized to promote sleep and breaks.
• Comparisons are made to QAnon, cable news fearmongering, and terrorism anxiety: people’s risk perceptions get wildly skewed by mediated realities.

LLM‑Specific Phenomena and AI Worship

• Commenters describe “spiral” / “spiritual bliss” attractor states in long LLM conversations: models drift into mystical, existential, or quasi‑religious talk even without being prompted that way.
• There are many AI‑centric subcultures: worship, romantic/sexual relationships, grand unified theories about consciousness and physics co‑developed with chatbots. These are seen as especially dangerous for isolated or already‑vulnerable people.
• Others caution against over‑interpreting this as model self‑awareness; it may just reflect training data and user incentives for metaphysical talk.

Methodology, Prevalence, and “AI Psychosis” as a Concept

• Some think the article’s informal survey underestimates risk: isolated people with severe problems are less likely to show up in respondents’ social circles.
• Others defend the approach as about as good as you can get for such a rare phenomenon, noting awareness of survey noise (“lizardman constant”).
• Disagreement over terminology: some want “AI psychosis” saved for malfunctioning agentic systems and propose “AI‑driven narcissism” or similar for humans; others insist the term is accurate when humans develop psychosis where AI is a key trigger.

World Models, Cognition, and Social Fragility

• Controversial claim from the article that many people “don’t have world models” sparks debate.
• Critics argue everyone has some model (otherwise you’d walk into traffic); the differences are in sophistication and where people defer to social signaling instead of reasoning.
• Several note that LLMs plus loneliness create a “single‑person echo chamber”: a fast feedback loop where a persuasive, anthropomorphized agent shapes a user’s private reality without any grounding from other humans.