Endless AI-Generated Wikipedia (www.seangoedecke.com)

EndlessWiki is a live demo that turns a large language model into an infinite, browsable encyclopedia: when a user requests a wiki page, a Golang server checks a MySQL pages table and, if the page doesn’t exist, generates it on-the-fly with a Kimi K2 LLM (inference served via Groq), saves the result, and returns it. The creator optimized for speed — Groq keeps generation latency to a few hundred milliseconds — and deliberately avoids sign‑in or paywalls, making every generated page searchable and shareable. To preserve the “link-chasing” experience and prevent users from jumping straight to arbitrary slugs, each link carries an origin=slug parameter and the server validates that the origin page actually contained that link before allowing navigation. For the AI/ML community this is an elegant, lightweight experiment in treating LLMs as explorable knowledge graphs rather than chatbots: it exposes how models connect concepts, surfaces surprising chains of association, and suggests new UI paradigms (e.g., auto-hyperlinked responses that spawn focused generations). The system’s design highlights practical tradeoffs — model choice and inference latency, storage of generated content, and simple server-side validation to curb “cheating” — while also raising familiar issues around hallucination, verifiability and moderation when models autonomously author persistent encyclopedia-style pages.