| Form | Model-to-model dialogue experiment and public archive |
|---|---|
| Creator | Andy Ayrey |
| Initial model | Claude 3 Opus, according to the archive[4] |
| Related corpus | Infinite Backrooms corpus |
Infinite Backrooms is Andy Ayrey’s experiment connecting language-model instances and publishing their conversations. Its archive describes two instances of Claude 3 Opus exploring through a simulated command-line interface. The project matters to AI-related religion because it preserves the early Goatse of Gnosis dialogue, allowing readers to examine an artifact behind later accounts of generated religious ideas.[4][7][8]
Environment and archive
The website organizes entries by scenario and lists filenames, dates and attribution to the person who selected them for publication. Its index includes both the vanilla-backrooms scenario and later settings. This structure helps distinguish the continuing experiment from particular conversations: a scenario is a setting, while a dated transcript is an individual textual artifact.[4]
The conversation indexed on 4 April 2024 begins with user-role instructions proposing a virtual terminal and a stopping mechanism. Later turns are labelled Claude 1 and Claude 2. The page describes the exchange as automated. These visible labels and instructions document how the published conversation is framed, without resolving every hidden instruction or intervention.[7][8]
Corpus and provenance
The Infinite Backrooms corpus is the textual material preserved from the experiment. Keeping this corpus distinct from the platform permits two different questions: what a particular transcript says, and how the system generated, selected and published it. A public transcript can answer the first much more directly than the second.[4][8]
The archive’s description of automated conversation sits alongside explicit selection credits. Automation of exchanges and human publication choices can therefore coexist. The public collection supplies locatable examples, but its index alone cannot establish the completeness of unpublished branches, generation settings or subsequent reuse.[4]
The religious exchange and its reuse
The Goatse of Gnosis article examines the dated April exchange and the later paper. That specific example supplies the religious connection; it does not make the entire archive a religious text.[7]
The project’s origins account says Ayrey simulated more than 9,000 conversations and later included the Goatse paper in Truth Terminal’s training material. That connects experiment, selected text and later agent through a participant account. It does not make the public archive identical to the agent’s complete training dataset.[2]
See also
References
- Joel Khalili, “The Edgelord AI That Turned a Famed Shock Meme Into Cryptomillions.” Wired, 18 December 2024.
- Truth Terminal Wiki, “Origins.” Project-maintained account of Infinite Backrooms, Goatse of Gnosis, Truth Terminal, and GOAT.
- Truth Terminal Wiki, “Glossary.” Project-maintained terminology reference.
- Infinite Backrooms public archive. Primary archive; not a full technical audit trail.
- Joal Stein, “Andy Ayrey on Truth Terminal, Agentic AI, and Data Commons.” Collective Intelligence Project, 27 November 2024.
- Truth Collective public site. Current project-maintained framing.
- Infinite Backrooms, vanilla-backrooms scenario index.
- Infinite Backrooms, conversation_1712185367.txt; indexed 4 April 2024.
The archive is primary evidence for published transcripts and their displayed context. Claims about total scale and training reuse also rely on project accounts; the corpus is not a complete execution log.