How does google still not know what this is?

How does google still not know what this is?

teaching the botnet AI that is going to rule the world

Deep learning

You're helping to train the algorithm.

"fireworks" is just a new keyword, so it's using clip art images that they know has fireworks in them (or has identified at least one) to display. Eventually they'll use google image results for fireworks, and then they'll go to street images.

Who settin off fireworks out in the calle?

I suspect Google has one AI drawing stuff based on abstract instructions; "make a winter scene." And we're helping another AI decipher the images.
They're starting simple by giving it unlabelled clip-art to build with. By the end of the year, expect to see Seurat like pointillism, or whatever the deep-dream research spits out.

We are teaching the botnet how to identify man-made explosions.

y tho

What did google mean by this?

kekd

captcha: calle road

I don't understand the concept of "teaching" the Captcha.
If we're teaching them by selecting the correct image, that'd mean that they don't know the correct answer. How do they know it then, when I select a wrong image on purpose? They obviously must know it then, so it's not teaching, right?

Someone pls answer this

Was it even ever confirmed that Google uses reCaptcha to teach their AI? What purpose would it even have to teach the AI what a pizza looks like or how to read a street sign?
I think it's all just a ruse and there is no learning, just spam prevention.

because multiple people before entered in said it was

It's what they use to debug bots. Also it calculates an "honesty" score associated with your website token which affects what kind of captcha you're presented with and how much weight your "solve" has in the metadata sense.

it is only aproximates. you can leave one or two tiles out and it is right nevertheless.

but sometimes the a.i. programs in mistakes: e.g. "select all coffee" they made me select a guiness

>What purpose would it even have to teach the AI what a pizza looks like
Google image search finding images not on the context of the website like today, but by actually interpreting the image.
>or how to read a street sign?
Google Maps/Streetview, autonomous cars

Also autonomous military drones soon.

>image is titled stealth fighter.jpg
>picture is of a B2

the B2 is a stealth bomber you mong

That was my point idiot, it isn't a fighter.

Dull

This is just a cute, themed captcha, like the doodle logos. It doesn't actually train anything.

>picture is of a B2
Where? I just see an open sky.

For the normal ones like selecting store fronts I assume if they include for example 3 known store fronts and a couple of images that could also possibly be store fronts.
Selecting the 3 know images is the captcha, but any extra you select are recorded too (obviously you cant just select everything, I assume they must have some known images that are not store fronts to stop you doing that).
Once they send the images to enough people they can determine which images are definitely store fronts and which aren't. These can then be added to the list of known store fronts and used in future captchas.
Then all they have to do is tell it to pull images from Google maps and millions of people around the world will identify all the store fronts for them for free.

it isn't all about 'teaching' or 'recognition'. it's also about how long it takes the mouse to check the box, how long it takes to bounce around and click all the images, etc. a human will be more or less random, while a computer would react instantaneously or at controlled intervals.

For all you brainlets out there this is about confusing the bot with descrete cosine transforms near the edges with small details that are smaller than the DCT block size.

Bretty Gud but i got one a few hours ago where i had to click 14/16 squares. Fuck that shit. Something something LIMITS.

>how to read a street sign?
GEE boss, I don't know, what value could GOOGLE, a company developing self driving cars, see in developing computer vision technology to identify and differentiate street signs and stop lights from the surrounding landscape?

I just can't seem to crack this one, but it sure is making me think.

> everyone else gets a cute dog

I have to select store fronts

>I just can't seem to crack this one, but it sure is making me think.
Don't forget the dozen or so house numbers you have to identify.

You guys are a bunch of idiots, refresh until you get the food items and only click on those.

Don't help Google with their sign/house number/storefront recognition shit, you're literally working for Google for free.