Simon Willison Endorses pelicans_riding_bicycles — Deliberate Training Data Poisoning as Creative Resistance

Simon Willison·low signal

Simon Willison highlighted Steve Cosman's GitHub project that deliberately mislabels images (e.g., a bear on a snowboard tagged as 'pelicans riding bicycles') to poison AI training data scraped from the web. Willison 'firmly approves' and notes his own published examples similarly contaminate datasets. The project represents a growing grassroots movement of deliberate data poisoning as protest against unconsented web scraping for model training.