Wikipedia's Cleanup of AI-Generated Content(spectrum.ieee.org)
484 points by rbanffy 1 day ago | 220 comments
tl;dr: Wikipedia's volunteer WikiProject AI Cleanup, launched in 2023, uses pattern recognition and AI-detection tools like Pangram to identify machine-generated content, which recently accounted for 15-20% of new articles and drafts. Beyond obvious tells like "stands as a testament," editors now hunt subtler structural patterns and fabricated citations, while deploying AI-powered source verifiers to fight back. Founding member Ellia Lebleu argues Wikipedia's human-written status has become core to its identity—both as "the last good place on the internet" and as valuable uncontaminated training data for AI companies.
HN Discussion:
  • ↑Wikipedia is a precious resource worth protecting from AI contamination
  • ↓Wikipedia's cleanup efforts are flawed - legitimate fixes get reverted by their own bots
  • ↑Appreciates Wikipedia's nuanced stance of fighting AI while still acknowledging its utility
  • ~The cleanup effort ironically trains AI to better evade detection, becoming a losing battle
  • ↑AI companies and executives profit while degrading shared resources like Wikipedia