Two people test a robot vacuum's poop-detection algorithm, which marks feces with a pink spot and can avoid or capture it from multiple angles. The system was trained on thousands of internet images of poop.
Go test some shit. It's the poop test time. He's finding the chief poop engineer. Shitty job, but it's essential.
And we've enabled the poop algorithm on top of it. Do you know? Can we do that, please?
Okay. Will it touch the poop?
Let's see. Please give us some magic. It's got a poop gauntlet. Does it show up? Oh, look at that.
Pinpoint poop turning.
So now it's kind of captured two, one on the front and one on the back. It's doing a pretty good job avoiding it. So there we go.
But does the pink spot mean random object or is that specific?
No, no, you'll see specifically that's exactly where the poop shows up here visually.
But like it thinks it's a poop or it thinks it's a— it does.
No, it does.
How did you train this model? How many poops did you have to show it?
Luckily, there is a lot of internet images.
So like some engineer here, his entire internet algorithm will never be the same.
Yeah, will never be the same.
