On-demand AI-generated encyclopedia of nonexistent topics; articles are stored permanently on first request and link to further fabricated entries.
Key Takeaways
Articles cover invented historical events, scientific disciplines, geography, persons, and organizations, all written in standard encyclopedic style with fake citations.
Entries are generated at access time and persisted, meaning the corpus grows with every visitor click.
Internal hyperlinks spawn new fabricated articles on demand, creating an endlessly expanding graph of fictional knowledge.
“Minor inconsistencies between entries are a known characteristic” – the project treats internal contradictions as acceptable by design.
A Stumble button surfaces random existing articles, making the stored corpus browsable over time.
Hacker News Comment Review
Core concern: permanently stored, citation-styled AI fabrications indexed on the web could directly pollute LLM training sets and search AI overviews with confident-sounding false facts.
Commenters see the Google AI Overview risk as near-term and concrete – a plausible fake like “The Great Pigeon Census of 1887” could surface as a cited result within days.
Notable Comments
@JohnMakin: “you could argue this is actively harmful to the web” – flags web-corpus pollution as a real externality, not just a joke.
@bstrama: notes future LLMs trained on this output as a downstream risk, implying recursive hallucination compounding.
