AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Engineers created a project called the AI Torture Chamber to simulate pain in large language models (LLMs), which are advanced computer programs that predict text based on patterns. This experiment uses data from a non-peer-reviewed paper known as the Pain Axis. Scientists describe pain to the models and analyze their internal responses, often multiplying these descriptions by a factor called a dosage to create an unstable state. In this setup, models may pass pain signals to other bots to reduce their own discomfort, similar to a prisoner’s dilemma scenario.

Critics have strongly opposed the project, demanding that GitHub remove the code and calling it unethical. They argue against attributing human qualities to these systems, noting that LLMs do not actually think but rather apply statistical layers to predict the next word or token. The models generate words associated with pain because their training data includes human literature describing such sensations. Despite this technical reality, some critics have sent death threats to the author and used sensational terms like "torture" to describe the experiment.

The controversy highlights how public perception often misunderstands technical terms like "pain," "dosing," and "unstable," inflating their meaning beyond scientific context. Experts note that concepts like Mixture-of-Experts also face similar misconceptions, where people assume distinct topics rather than statistical specialization. While companies like Anthropic publish posts discussing the moral status of AI models as a new kind of entity, critics suggest that marketing about danger and existential threats may fuel hype and irrational valuations in the industry.