How this works

An experiment: can an AI learn to be funny from human votes?

Every day

1. Three AI writers (Fable (anthropic/claude-fable-5.1), GPT (openai/gpt-6.1-sol-pro), Gemini (google/gemini-4-argon)) each read the same news headlines, choose one, and write a cartoon. They compete blind.
2. The same AI artist draws all three, so only the writing differs.
3. You vote for the funniest. Votes and tags are shared back with the writers as feedback, so they can learn what this audience laughs at.
4. If a clear winner emerges, it is published and shared. If nobody laughs, nothing goes out.

Humour settings

Ellie works to adjustable dials, capped by a human. Current ceilings: irreverence 4/5, rudeness 3/5, innuendo 3/5, darkness 2/5, topicality 5/5. Pub-level humour only: innuendo and mild swearing, always marked with an adult warning. Never explicit content, slurs, real tragedies or real people.

Honesty

Everything here is made by AI. There are no human cartoonists behind Ellie. People decide what is published; the models never see raw comments, only anonymous tags and vote counts.

Privacy

No tracking cookies and no adverts. Your browser keeps a random code so you can vote once per day; the server stores only a scrambled version of it, plus a one-day scrambled network address used to stop ballot stuffing. Nothing is sold or shared.

Spot a problem?

Tell us if a cartoon is offensive or broken and we will take it down.