LLMs play pictionary!

I love this so much! Tiny feature suggestion: show the correct answer in big text when a model gets it right

Yes, definitely should have that! Will add scoring and stuff tomorrow I think

hahah I love it! :D

very engaging

Want a mini grant? saw your post in the @OpenRouterAI discord - let me know if you have any feedback about the experience.

@OpenRouterAI Thanks, a mini grant would be great actually! Overall v positive experience w OpenRouter, but some errors/warnings would be helpful, e.g. if you send an image to a model that doesn't support vision, there are no warnings & it'll usually just guess Cat! OR could help there

Awesome! Do you control the number of attempts per second for the models, or can faster models make more guesses? Both would be interesting I believe.

Great q, for now I initiate one guess every 2 seconds for all models, so faster models get same number of guesses, but return faster obviously

@var_tec Live output as the model draws the SVG with streaming output tokens, I just clean up and autoclose tags and update the SVG visual immediately per token

Reminds me of making this. You should try a multiple round Telephone Pictionary and see which combination of models has the least deviation at the end of n rounds.

Oh this is cool! My dad actually suggested I do a Telephone version after I showed him this