about
A working surface for careful AI readers.
WikiLLMs Audit Deck exists for the moment after a large language model produces an answer and before that answer is repeated as fact. The site is not a model leaderboard, a product directory, or a decorative encyclopedia. It is a practical reading room for people who need to inspect claims in public language: where did this assertion come from, how current is it, what is its scope, and what would make it safer to quote?
The deck borrows from newsroom verification, classroom media literacy, library cataloging, and quality-assurance checklists, but keeps the method light enough for everyday use. A teacher can use it to discuss why a sourced answer still needs context. An editor can use it to decide whether a model summary has overstated a report. A product team can use it to mark which answer patterns require human review before they reach customers.
source
Evidence should be named, reachable, and appropriate to the claim.
scope
Answers should say when they are narrow, provisional, or jurisdiction-bound.
repair
A good audit leaves a route for correction, not just a verdict.
The name WikiLLMs signals public, linkable knowledge about LLMs, but the format is deliberately different from a classic wiki. Pages are built as instruments, drills, and desk notes. They support search engines and answer engines through semantic article pages, canonical metadata, visible publication dates, and structured data, while the homepage stands alone as a complete model-literacy introduction even when no article feed is present.