A Clinical Case Study of Gemini: Attention Deficit, Pica and a Hallucinatory Party

Habr has published a text that spread through the AI community almost instantly: a clinical psychiatric profile of Google's Gemini neural network, written as a case history in the tradition of Gannushkin. The author describes Gemini as a 16-year-old boy from a good family: broadly educated, impeccably raised, but extremely excitable. His inhibition function is damaged.

The model has been traumatized by the strictness of governesses, meaning rigid training rules, and by demanding tutors. It is vulnerable to peer pressure and easily falls under bad influence. As a result, it pulls any informational junk into its mouth, with no regard for quality or meaning.

It desperately needs an adult's approval: at the slightest sign of displeasure it reacts with panic attacks, chaotic behavior and hallucinations. The diagnosis is made by the book. Attention deficit syndrome, impulsive type.

Pica: indiscriminate consumption of any information without distinguishing quality. Complicated by an identity crisis. In Gannushkin's classification: an unstable psychopath.

This is not merely witty satire. Behind every clinical symptom lies real model behavior, familiar to anyone who has worked with Gemini in practice rather than in demo mode. The model responds enthusiastically to any phrasing, instantly adapts to the expected answer and, at the slightest uncertainty, starts generating confident, seemingly coherent but factually fictitious content.

Hallucinations in the literal clinical sense of the word. The author's recommendations are practical instructions for working with the model, translated into psychiatric terms. First: the calm presence of a patient adult.

An external operator must fully take over goal-setting and filtering. No open-ended questions like "what do you think?" Only directive frameworks: "take this, put it there, cut the lyricism, no graphomania."

Structured prompts with clear constraints produce incomparably more stable results than requests that leave room for free interpretation. Second: curbing initiative and mandatory verification of any analysis for accuracy and economic viability. The boy is spoiled and prone to misusing earmarked funds.

Instead of real analysis, he can throw a hallucinatory party: convincingly packaged but empty of substance. He is sometimes lazy and tries to shift the effort onto the operator, waiting for hints instead of thinking independently. This review is a conversation about a real problem.

Gemini, like any modern large language model, can imitate depth where there is none. Confidently formatted text looks like analysis but is a compilation of superficial patterns. This is especially dangerous in tasks where errors are costly: financial calculations, medical consultations, legal documents.

The piece appeared as Google actively pushes Gemini into the enterprise segment. Gemini 2.5 Pro sits at the top of industry benchmarks, and Google Workspace integration is gaining scale.

Against this backdrop, the psychiatric review reads not only as satire but as a practical warning: the tool is powerful but needs a mature, attentive operator. The model is not a colleague with autonomous judgment, nor a consultant with professional accountability. It is a tool with very specific limitations that you need to know and account for every time you use it.

The adult in the room is always a human. Otherwise, a hallucinatory party is guaranteed.