Researchers have found that advocacy organizations can influence what large language models say about their causes by strategically editing Wikipedia articles. According to a new study, Pro-Animal Wikipedians—a group focused on animal welfare—made a hundred and twenty-five edits to Wikipedia articles on related topics. These edits measurably shifted how language models responded to animal welfare questions, without affecting unrelated queries about the same entities. The researchers verified this using three different methods, confirming the effect was specific to the edited content. Since Wikipedia is weighted heavily in most large language model training datasets, they concluded that strategic editing could be a practical, low-cost tool for advocates trying to influence how AI systems discuss their issues.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton