Back to feed
arXiv cs.CL
arXiv cs.CL
6/25/2026
Small edits, large models: How Wikipedia advocacy shapes LLM values

Small edits, large models: How Wikipedia advocacy shapes LLM values

Short summary

Research demonstrates that coordinated Wikipedia editing campaigns measurably shape LLM outputs on targeted topics. Using gradient-based attribution methods (MAGIC, TrackStar) on Llama models, researchers found Pro-Animal Wikipedians' edits account for 68% of highest-attributed documents on animal welfare queries. This proves small groups can influence model behavior through Wikipedia—a major LLM training source.

  • •Wikipedia content is heavily weighted in LLM training and measurably influences model behavior
  • •Small, coordinated editing campaigns can shape model responses on specific topics with statistical significance
  • •Data attribution methods prove causal links between training data edits and model outputs

Generated with AI, which can make mistakes.

#research-breakthrough#ai-tools
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Explore more

Assert, don't describe: Linguistic features that shift LLM reasoning about animal welfare
arXiv cs.CL
Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training
arXiv cs.CL
Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints
arXiv cs.CL
Perfect Detection, Failed Control: The Geometry of Knowing vs. Steering in Language Models
arXiv cs.CL