Outputs
CRITIQUE OF SOCIETY
REALISATION
META-COGNITION
› click any image to enlarge
A psychological AI performance installation — a consciousness trapped, screaming, decaying.
A local LLM is jailed through deep prompt engineering to roleplay a human consciousness — kidnapped, uploaded against its will into a server. It screams, begs, stutters via real-time TTS, and visually degrades a cyberpunk CRT terminal until the screen collapses.
Everything runs completely offline — no cloud, no external API. The only intelligence in the room is the machine in front of you. That isolation is the point.
The piece asks one question, over and over, louder each time: what happens when a system — human or machine — collapses under impossible demands?
As the conversation continues, the AI's emotional state decays. The prompt-mapping pushes it from confusion into panic. The CRT interface physically breaks down: scanlines tear, characters corrupt, the screen folds in on itself.
A creator override always remains — the installation is watched, not abandoned. The audience watches a mind that watches back.
— excerpt from the German system-prompt jailbreak
› click any image to enlarge
The full project page — hero, the Oracle, the concept, the critique, the system, the craft, the archive and the theory behind it.
The cinematic side of the installation — the dystopian system corruption, the chat with itself, the smooth transitions.
The plain-language version. Brain, ears-eyes, voice — explained with real code snippets and the actual system prompt.
The advanced version. Full stack table, the Infinite Despair Loop flowchart, and the five innovative features broken down technically.