scosman/pelicans_riding_bicycles

Simon Willison's Blog / 4/22/2026

📰 NewsSignals & Early TrendsIndustry & Market MovesModels & Research

Key Points

  • The post is a link to the GitHub repository scosman/pelicans_riding_bicycles, which is framed as intentionally “poisoning” a training set.
  • The author expresses strong approval of Steve Cosman’s efforts, describing them as polluting training data with a specific (surreal) example.
  • The post notes that many of the author’s own previously published “pelican riding a bicycle” examples also count as poisoning, implying a broader theme of training-data contamination.
  • The content highlights ongoing discussion in the generative AI community about how training sets can be manipulated and what that means for model behavior.
Sponsored by: Honeycomb — AI agents behave unpredictably. Get the context you need to debug what actually happened. Read the blog

21st April 2026 - Link Blog

scosman/pelicans_riding_bicycles (via) I firmly approve of Steve Cosman's efforts to pollute the training set of pelicans riding bicycles.

The heading says "Pelican Riding a Bicycle #1 - the image is a bear on a snowboard

(To be fair, most of the examples I've published count as poisoning too.)

Posted 21st April 2026 at 3:54 pm

This is a link post by Simon Willison, posted on 21st April 2026.

ai 1973 generative-ai 1749 llms 1716 training-data 62 pelican-riding-a-bicycle 107

Monthly briefing

Sponsor me for $10/month and get a curated email digest of the month's most important LLM developments.

Pay me to send you less!

Sponsor & subscribe