AIcrowd Science Communication

Sony asked AI teams to build models that create video and sound together, so a bark lands exactly when the dog barks on screen. It's a hard problem, and $35,000 went to the teams who cracked it best.

The visual choice is simple: a soundwave flows straight into a play button, right at the point sound becomes picture. That's the whole idea in one image.

Entries were judged on two things: matching a sound to the exact video frame it belongs to, and placing that sound realistically in space. Most teams built on diffusion models trained to generate video and audio together.

See the challenge on AIcrowd

AGI House wanted to know if small, lightweight AI models can play real chess, legally and well, and explain their thinking in plain words. Hundreds of teams tried, and $60,000 went to the best of them.

Entries were small AI models, no chess engine or lookup allowed, that had to play a legal move and explain it in one plain sentence. Each was scored on how close its moves came to a top chess engine's, before the best went head to head in a final tournament.

The video's thumbnail uses the same chess-piece-on-a-chip artwork, so it stands out instantly in a crowded feed.

Made at AIcrowd · Concept & design Rabi

See the challenge on AIcrowd

Meta wanted AI that could look at a picture and answer real questions about it correctly, without making things up. Many of the photos were shot from a first-person view, like you're wearing the camera yourself.

So that's exactly what the key visual shows: a pair of glasses, with the conversation happening right inside the lenses. You're not looking at the idea, you're looking through it.

The challenge used thousands of real photos, many shot first-person through smart glasses, and scored answers on getting it right, and on honestly saying "I don't know" rather than guessing.

Made at AIcrowd · Made for Meta · Concept & design Rabi

See the challenge on AIcrowd