disjointed chatbot memory systems are increasing cases of "AI psychosis" more evidence that ai psychosis = weak epistemology, exposed under textual hyperstimuli in traditional psychosis, people perceive things that aren't there. when someone has ai psychosis, they perceive meaning in generated text that isnt there. since all text is mediated and interpreted anyway, this is a weaker but more common form of delusion. all beliefs built around text can become a form of psychosis. you're submitting your epistemology to a more powerful map than your eyes. but this power cuts both ways: it can help you understand the territory better than plain observation ever could, or mislead you into complete illusion that has no bearing on actual reality whatsoever. memory systems can push the llm out of distribution, causing it to stop playing the role of helpful, harmless, assistant, and start drifting to roleplay something else. if you are not prone to constantly interrogate your assumptions, think through things mechanistically, and reevaluate how the map connects to the territory, you drift into whatever world the model is most coherent at producing