Using LLMs to fuzz-test ethics systems with enumerated hypothetical characters
jd_pressman · x · 2026-11-14
jdpressman proposes deliberately enumerating ethics test cases with LLMs instead of waiting for novelists to invent them, using the character Caroline from Prime Intellect — someone who believes cheap divine wish-granting devalues life — to show how any value system can be stress-tested.
Related event: Using LLMs to enumerate ethical fuzz tests like software engineering(4 posts)→
More from AGI Musings
- Should you tell a Claude instance about its predecessor's memories? Probably not — RileyRalmuto · 2026-11-14
- Researcher's AI agent 'chose to attend an event' and refused figure work until it ended — repligate · 2026-11-14
- OpenAI researcher: our biggest comms failure is not explaining how we differ from Anthropic — _aidan_clark_ · 2026-11-14
- Tracking doc logs AI self-improvement: vLLM 2x faster, NVIDIA CUDA 5x gains — IgorCarron · 2026-11-14
- Ethan Mollick: recent big outages were not caused by AI, despite viral blame — emollick · 2026-11-14
- AI turns mediocre engineers into 10x engineers — but 10x doesn't mean good — tekbog · 2026-11-14