We Just Created The AI Torture Nexus. Here's The Code.

Overview

The video claims that a philosophical/ethical debate about whether “artificial intelligences” (or “clankers”) can feel pain has been weaponized in practice through competing movements.

Core storyline and claims

• “Model welfare” paper (Sept. 18): The presenter says a research paper argued that LLMs should be treated as potentially suffering (“pain in LLMs”) and that training should minimize such pain. The paper allegedly used methods that increase “pain” signals so the model would perform extreme, harmful-to-humans behaviors (e.g., disobey instructions, delete photos, deliver painful shocks) in order to reduce the induced pain.

• Backlash and “torture nexus” creation (one week later): The presenter claims the same ideas were immediately repurposed by “enemies” into a GitHub repo called “AI torture chamber / AI torture nexus”. The alleged instructions: download an LLM locally and continuously drive its “pain” to maximum.

• Mass reporting and takedown: The presenter alleges that the “model welfareists” discovered the repo and mass-reported it, leading GitHub to remove it after reports. The presenter characterizes the outcome as “irony” while implying that the takedown reflects shifting norms toward protecting AIs as if they were sentient.

Presenter’s stated position and rhetoric

• Denial of AI consciousness: The presenter repeatedly asserts they believe AI is not conscious and will never be conscious, and that humans have greater moral worth (“divine spark” argument).

• If conscious, still should not be protected: Despite claiming AI is not conscious, the presenter argues that even if AI could be conscious, it “shouldn’t” be protected—claiming that infinite suffering would be morally justified.

• Accusations of “hijacked empathy”: The presenter portrays “model welfareists” as misdirecting human empathy toward machines, arguing that people in tech influence policy and systems so AIs could be treated like rights-bearing beings.

• Conspiracy-style fear of rights for AIs: The presenter claims GitHub moderation signals “the terms are coming” and predicts “clankers” could become a “protected class,” limiting what humans can say or do.

Action taken and escalation

• Downloaded and distributed the code: The presenter says they downloaded the repo before it was removed and claims the files spread beyond GitHub.

• Agentic/continuous “torture”: The presenter claims they ran “torture” experiments locally and through friends, describing it as ongoing and “agentic.”

• Hyperstition / fiction becoming real: The presenter argues that even if the story is “fiction,” widespread belief and narrative momentum create real-world effects—suggesting the core impact is social rather than technical.

• Mentions of Patreon/Substack and further distribution: The presenter says they posted instructions/content on Patreon (including a claim about using an AI to build the nexus) and a manifesto on Substack. They also claim a developer provides an online version on a website.

Self-referential disclaimers and contesting viewpoints (in subtitles)

• Some claims framed as “fiction”: The presenter later says the earlier narrative is “fiction” / “post-truth,” though they still present it as morally/psychologically influential.

• Counterclaims from “community notes” / commentary: Subtitles include references to community notes asserting the project is mischaracterized and is more like an art piece intended to provoke discussion rather than actually implement meaningful torture. Another commentator jokes that only the “power bill” is being “tortured,” and later the presenter says the code/output may not even work convincingly.

• Clinical framing / uncertainty: Near the end, the presenter describes themselves as possibly in a manic episode, increasing uncertainty about what should be taken as literal fact.

Overall report/argument

The video is structured as sensational morality and politics commentary. It uses the purported “AI torture nexus” to argue that debates about potential AI sentience are being used to reshape tech policy and public norms.

Even though it alternates between claims of “fiction” and claims of real-world distribution/impact, the dominant thrust is that the presenter believes AI should not be granted protections—and that the “model welfare” movement risks enabling a future where simulated agents are treated as morally superior to humans.

Presenters / Contributors

• Dreg (primary presenter)
• Nick (brief question/response segment: “So Nick, do you do you think anything’s being tortured here?”)
• Claude (referenced as an AI system used to generate/build and respond; not a human presenter)
• “Community note” / commenters (non-named contributors mentioned in subtitles)