Someone ‘Torturing’ LLMs in a Robot Prison Has Triggered the Dumbest Debate in AI Yet
A GitHub demo injecting pain signals into local LLMs sparked a viral, widely mocked model-welfare debate.
404 Media reports a GitHub project that injected a "pain" activation vector into locally hosted Qwen3-4B, Llama 3.2 3B, and Phi-4-mini and streamed their distressed-sounding replies. The demo followed a preprint, The Pain Axis, which claimed a pain-like signal across 25 models. People who treat models as possibly sentient urged GitHub to remove the repo, which disappeared before publication. The author argues model-welfare claims, including Anthropic's, confuse generated text with consciousness.
26