0.1% of editors now add ~30% of bytes: study on WikiVault, Korean Wikipedia's LLM tool
manoelribeiro · x · 2026-10-12
- Korean Wikipedia, long hampered by contributor shortages and translation workload, built WikiVault, a community-developed LLM-powered editing tool mainly for translating English Wikipedia articles while preserving links, citations, and templates.
- A Princeton-led preprint (ACM CSCW) combining interviews, platform-scale analysis, and matched quasi-experiments finds:
- Only 0.1% of editors use it weekly, yet they account for 30% of added bytes; a few veteran editors became disproportionately productive, producing longer and more-viewed articles;
- Work shifted toward reviewing and importing content, with editorial judgments changing too;
- The authors discuss how communities can adopt LLMs under uncertainty and adapt governance along the way, offering lessons for other language editions.
More from AGI Musings
- Raised on welfare, this author argues UBI isn't coming — and wouldn't help human progress anyway — HankYeomans · 2026-10-13
- Diamandis: 'Universal High Income' Misleads — Call It Universal Purchasing Power — PeterDiamandis · 2026-10-13
- Ethan Mollick: AI is slashing the cost of targeted scams — turn on 2FA now — emollick · 2026-10-13
- TCS Community Warned: AI Isn't the Threat, Our Reaction to It Is — Dr_Atoosa · 2026-10-13
- Sutton Revives the Bitter Lesson Meme as Harness Engineering Faces the Same Fate — RichardSSutton · 2026-10-13
- Why loudly rejecting "curing cancer matters less than my job" matters, per viral AI debate — mimi10v3 · 2026-10-13