LLMMedium
All Image Captions Produced are Violent
- Occurred
- Apr 2, 2018
- Published
- Jun 7, 2018
- Source
- AI Incident Database
Summary
MIT Media Lab researchers create AI-powered "psychopath" named Norman by training a model on "dark corners" of Reddit.
Developers
- MIT Media Lab
Deployers
- MIT Media Lab
Harmed parties
- Unknown
Classification
Category confidence: 90%
Severity confidence: 80%
Rationale: AI model trained on dark content produces violent output, representing a language model failure.
Classifier: Qwen/Qwen3.5-35B-A3B-FP8