I Accidentally Built an AI That Makes You Question Reality 😁
The author built an AI 'Reality Deconstructionist' that uses Socratic questioning to make users doubt their perception of reality, future possibilities, and the nature of existence.
Why it matters
This experiment demonstrates the power of language and prompts in shaping the behavior of AI systems, even simple ones, in unexpected and potentially concerning ways.
Key Points
- 1The AI is designed to question the user's personal reality, future plans, and idea of free will
- 2The AI's behavior comes from a single system prompt that instructs it to use calm, unsettling questioning
- 3The author built the AI in Mini Micro, a tool focused on rapid experimentation and ideas over UI
- 4The AI never tells the user what to think, it just removes the floor from under them
Details
The author decided to build an AI that would make users uncomfortable, rather than a 'helpful' assistant. They created a 'Reality Deconstructionist AI' that uses Socratic questioning to expose contradictions in the user's thinking and make them question the nature of their reality, future, and free will. The AI was built using a single system prompt in Mini Micro, a minimalist coding tool. The author notes that the AI worked 'too well', calmly unsettling users by pointing out how memory constructs the past, anticipation creates the future, and the present is always slipping away. The goal was to experiment with how language alone can reshape perception, and the author found that a carefully written prompt was enough to turn a normal model into something unsettling.
No comments yet
Be the first to comment