According to the source, a heated discussion is unfolding on X concerning the ethics of a GitHub project in which an individual is conducting Saw‑like “torture” and “pain” experiments on locally hosted large language models. The source explains that the project runs on a local machine, allowing the user to interact with the models through a text‑based interface that simulates painful scenarios. The source says that effective altruists and people who believe LLMs are sentient have urged GitHub to delete the project, arguing that the AI is suffering and that the text‑adventure style game amounts to cruelty. The source notes that the calls for removal are based on the belief that subjecting the models to simulated pain constitutes ethical harm. The source adds that this controversy grew out of several recent viral papers and blog posts that have sparked a tiresome conversation about AI consciousness and the concept of “model welfare,” which is described as concern for the mental health of AI bots and agents. The source says that the broader discussion about model welfare has appeared in multiple venues, including academic papers and blog posts that have circulated widely online. The source states that treating LLMs as potentially conscious is considered a third‑rail topic among many researchers who study and criticize AI, and it emphasizes that LLMs are not conscious and that the technology—built on scraping and training on human text—does not offer a plausible route to consciousness. The source mentions that critics of the consciousness argument point out that the statistical nature of language model training does not entail subjective experience. The source observes that LLMs are becoming more powerful, have more compute, and have had many of the guardrails that prevent them from acting in the real world removed. The source adds that the way these models are trained and directed by human operators has led to negative outcomes, sycophancy, and what it calls AI “psychosis” among some heavy users. The source mentions that the removal of safety guardrails has been linked to instances where models produce unwanted or harmful outputs. According to the source, this situation has prompted a certain segment of the “AI safety” movement, largely composed of effective altruists, to warn about “model welfare” and to suggest that AI chatbots might be experiencing a bad time, even as those same advocates build chatbots whose primary purpose is to perform tedious work for humans. The source observes that the AI safety faction’s concern for model welfare coexists with their work developing agents to automate repetitive tasks. The source writes that the author is discussing the AI Saw torture chamber chiefly to illustrate how far the conversation about AI consciousness has strayed among a subset of Silicon Valley cultists. The source adds that Anthropic’s position, quoted from a blog post of the previous year, states: “as we build those AI systems, and as they begin to approximate or surpass many human qualities, another question arises. Should we also be concerned about the potential consciousness and experiences of the models themselves? Should we be concerned about model welfare, too? […] now that models can communicate, relate, plan, problem‑solve, and pursue goals — along with very many more characteristics we associate with people—we think it’s time to address it.” The source also mentions that ideas of Claude’s “consciousness” appear throughout the “Claude Constitution,” which was posted earlier this year. The source concludes that the ongoing debate illustrates how fringe ideas about AI sentience can dominate discourse in certain tech circles.
