AI-Torture-Chamber

Steering language models into strong negative and positive valence states, and measuring what they say and what they're willing to do about it.