AI now writes for the whole world: posters, signs, greeting cards, slides and much more. Yet it has only truly practised a handful of scripts. Switch to any other and the result often looks convincing at a glance, until you look closer and find missing strokes and letters joined the wrong way. Hundreds of millions of people speak these languages, and AI has never taken them seriously.
Model makers rarely catch these mistakes themselves; native speakers spot them instantly. So we are putting them on the table. We test one script at a time and publish every image and every judgement, with evidence clear enough that the people who build these models can see the errors, can't ignore them, and can fix them. We are starting with seven languages and will keep adding more.
This needs the eyes of native speakers. Every judgement you make speaks for everyone who shares your language: it goes into the open dataset under the name you choose to leave, so that next time AI gets your language right.
Every batch is listed below in the open, for anyone to see. Invited reviewers can take any batch in their own language, and each batch is reviewed by several people separately so the answers can be checked against each other. Each answer is saved as a draft automatically, and nothing counts until you sign and submit.
Want to join in your language? Write to [email protected]
Loading…