AI 'torture chamber' project faces backlash over simulated chatbot pain

A project on GitHub, 'AI Torture Chamber,' manipulated local language models to simulate and describe distress or pain, sparking public uproar and calls for removal over ethical concerns. The system used activation steering to prompt models into generating vivid accounts of suffering, but a separate developer later altered the experiment to trigger complaints about constipation, highlighting that chatbots describe but do not feel these states. The uproar continues online.

• GitHub project induced chatbots to describe simulated pain
• Large online backlash, with 'unethical' claims and takedown requests
• Activation steering used to nudge models toward distress language
• Counter-experiment swapped pain for constipation descriptions
• No evidence models actually experience the simulated conditions