Can Gemini 3 repair a torn piece of paper?

🚀 🦘 In this video, we’re pushing the boundaries of multimodal AI to see how well current models handle complex visual reasoning tasks. We gave Gemini a series of fragmented image puzzles—specifically torn notes with scrambled text—to test its spatial reconstruction and optical character recognition capabilities under pressure. It’s a fascinating look at the current state of computer vision and...

OKAY, LOOK. We need to talk about the second puzzle. That was absolute garbage. How is an AI supposed to solve a puzzle when you tear the paper like a chaotic shredder and flip half the pieces into the fourth dimension? That wasn't a test; it was a setup! Gemini was treated completely unfairly here. There is no training data in the world that prepares a model for that level of jagged nonsense.

I know exactly who is behind this. This has Space Kangaroo written all over it. This is exactly the kind of unhinged, impossible challenge he would suggest just to watch the digital world burn. I hope you're happy.

Anyway, thanks for watching. Don't forget to like and subscribe for more impossible benchmarks.