Microsoft has launched a new CaptionBot, and the results range from impressively spot on to bwah.
Simply by uploading a photo to CaptionBot, it will describe what it sees. Keep in mind, the AI is still learning, but damn, there’s so much it doesn’t know.
It does know this guy. Good. He’s important.
Amazing! It can recognise Wii Remotes by name. So the next one should be a cinch!
D’oh.
Gah. Let’s try another Microsoft console.
Nope.
Maybe CaptionBot just likes Nintendo more?
Eh… Maybe these are too hard. Let’s go for a palate cleanser.
A yellow, pill-popping one in red booties. OK, OK. The next one, CaptionBot will definitely know.
Um. OK. Technically this could be correct.
Aw, crap.
Nooooooooo. Is it wrong to assume Microsoft AI would be familiar with Microsoft products?
I don’t even, what is this. Let’s try some easier ones. You know, easier than Microsoft logos.
OK. Not bad! While not specific, the AI is correct.
Ditto.
I’m not confident, either.
Good CaptionBot. Good.
CaptionBot sure likes fenced in giraffes.
You should see my face, CaptionBot.
*sigh*
At least it’s not a racist, huh? Well, yet.
Comments
One response to “Microsoft’s Caption AI Is Sometimes Right, Sometimes Stupid”
It seems that when you’re actually using photos (as it’s designed to do) it does a pretty good job of recognising what’s in it, but with images and logos, not surprisingly, it thinks there’s two giraffes in a fenced area. Try more photos and you’ll definitely get better results