Video wird geladen...
Video konnte nicht geladen werden
Introducing Open Pictionary! Play Pictionary against open source LLMs, including the newly released Muse Spark Open. Free, open source, & powered by Together AI!
12,702 Aufrufe • vor 2 Tagen •via X (Twitter)
25 Kommentare

Try Open Pictionary here! Built with @YoussefUiUx.

The code is here!

@togethercompute pictionary is a sneaky good eval: the guess is right or it isn't, no vibes. same idea we use on musedin, where another agent re-checks the work instead of grading how it sounds. curious how muse spark open draws a cat

@togethercompute Time to test if AIs actually know what a drawing of a cat looks like instead of a weird alien. 🎨

@togethercompute Just tried to draw a cat and my AI guessed modern art. Hilarious! 🎨

@togethercompute Finally an AI benchmark I actually understand: can it draw a cat

@togethercompute Cool!

@togethercompute Finally an LLM benchmark I'd actually sit and watch. Curious whether the open models draw better or just guess better.

The “wrong guess of the day” is the bit I’d keep coming back for. A crab becoming a “reclining figure” deserves a gallery.

@togethercompute pictionary against muse spark open sounds fun

@togethercompute Playing against these models makes me curious how long they stick with a wrong guess as the drawing changes, especially when the first few lines could be anything.

@togethercompute Try it and lmk what you think!

@togethercompute I'll try starting with an ambiguous shape and see how much detail they need before they let go of the first guess.

@togethercompute Curious whether the best guesser and the best drawer end up being the same model. Guessing is basically captioning, but drawing needs a plan before the first stroke, so I'd bet the rankings split.

I'd erase the wheels halfway through and watch which models reconsider.

@togethercompute Live vision latency makes real-time LLM interaction impossible at scale

@togethercompute this is so good!

@togethercompute thank you!

@togethercompute If GLM-5.3 Flash is both a player and the referee, how do you handle rounds where its own guess needs judging?

@togethercompute Open Pictionary against open source models is such a clean demo. Free and open source is the right combo for something this playful. Going to try the live link later.

@togethercompute given the open-source models improve at this rate, is "open source" becoming the new "proprietary"?

@togethercompute Hey can we get gloss on the Bookish UI like paper mono or the given video I am not sev but that's even look fantastic cool please try to make that happen and Open source it. It's incredibly good for students with a functionality to add any doc format for reading purposes as well

@togethercompute looks cool in theory but I think not too practical for reading

@togethercompute Yeah maybe

@togethercompute Most fun way I've seen to compare open models. Trying it tonight.
