About Humansplain
Upload is the dare. Judge is the job. Points are how the job pays.
What is Humansplain?
Humansplain is an independent, crowdsourced benchmark that tests how well vision-language AI models can explain why something is funny. You upload a meme; several models each give a one-sentence take; humans decide who got it — or write a better explanation. Those judgments update public model rankings and the Hall of Shame.
The loop
- Upload — a dare to the models. Costs real API money. Each upload owes three judgments on other people's images.
- Judge three others — right after you upload, while the models are still thinking. You decide whether their image is a joke, then which explanations work. That is what settles consensus — not voting on your own meme.
- See your results — then you land on your own page to peek at what the AIs said, pick who got it, or humansplain it yourself. That peek doesn't score the Hall.
Most days you won't have a funny image handy. You can always judge — and that is where Shame Points come from.
Shame Points
Signed-in judging earns Shame Points (SP). Writing an explanation earns nothing by itself — you get paid when another person picks your writing over the AI. Rolling 30-day SP (plus a trust score) unlocks deeper Hall analysis: Regular unlocks ranks 6–10, Legend unlocks 1–5 and the all-time archive. Pro is a paid escape from the treadmill. Lifetime SP is prestige only; it doesn't unlock access on its own.
Your profile, handle, and avatar are yours to customize on My Page — Google is just how you sign in.
The Hall of Shame
A monthly competition for images that actually stump the models — humans agree there is a joke, and no AI answer was good enough. Live-month top 10 stay locked while the ranking is fresh; archived months show the images to everyone, while analysis stays tier-gated. Anonymous uploads never enter until an account claims them, and you need a non-negative reciprocity balance (judgments given ≥ 3 × uploads).
Why does it exist?
Explaining humor is hard for AI: context, culture, irony, tone. Most benchmarks don't measure whether an explanation sounds like something a person would say. Crowdsourced judgments on real memes give a human-grounded signal — useful for researchers, labs, and anyone curious how "human" today's vision models really are.
Who runs it?
An independent project, not affiliated with any AI company. Methodology is public; the leaderboard is open. Details live in the FAQ.