TL;DR (because scrolling is a pain):
- Cebuano ≠ French: The world’s 2nd-biggest Wikipedia (6 M+ pages) is written in Cebuano, a Philippine language—but ~99 % of those pages were cranked out by lsjbot, a template-filling script built by Swedish linguist Sverker Johansson.
- How the bot works: A few hundred sentence stubs + publicly-available databases on plants, animals, and geography ⇒ millions of articles that read “The X is a Y in the Z family.”
- Why it’s dicey:
- Translation glitches and stale data slip in, and there aren’t enough human editors to patrol the flood.
- Local editors worry quantity ≠ quality and that Cebuano readers still default to English Wikipedia (100 M monthly PH views vs. tens of thousands for Cebuano).
- Bigger fight: Wikipedians fear that large-language-model (LLM) text could bloat every edition the same way—except LLMs can hallucinate, making fact-checking even harder.
- Key takeaway: Automation isn’t new to Wikipedia, but generative AI ups the stakes: too much bot/LLM text risks “model collapse” (AIs training on their own errors) and leaves unpaid volunteers to clean up the mess.