275
This actually works. Why didn't anyone tell me about this before? (lemmy.today)
submitted 4 months ago by to c/fuck_ai@lemmy.world
45 comments fedilink hide all child comments
 

all 45 comments

sorted by: hot top controversial new old
[–] 66 points 4 months ago (10 children)

I hate those where they ask you to select all the squares containing something. How much of the thing must be in the square for it to count? What if it just touches a corner of a square? I almost never get it right on the first try.

Also those stupid ones where you have to click all the pictures containing a certain thing, which then gets replaced by a different picture and you have to keep clicking until they're all gone. Those pictures often take FOREVER to appear, which causes you to time out.

Fuck Google.

  • permalink
  • fedilink
  • source
  • hideshow 10 child comments
    •
  • [–] 35 points 4 months ago (1 child)

    It's intentional. Recapacha will outright lie to you about getting a challenge wrong if it thinks you're protecting yourself too much. Same with [Google] outright refusing logins if it doesn't like your IP address (but it does confirm the username and password were correct!)

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 11 points 4 months ago*

    If it contains one pixel, it contains it.

    But I don't know if a saddle bag on a motorcycle counts as a motorcycle. Or if the handrail on the stairs is part of the stairs.

    It really pisses me off when it tells me I failed it, when I got it right..

  • permalink
  • fedilink
  • source
  • parent
  • [–] 10 points 4 months ago (3 children)

    Yeah I feel like recently they've gotten much pickier and I've had a lot of failures. Super annoying.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 3 points 4 months ago (1 child)

    Because one thing "AI" is good at is faking it's human. Now we have a machine that's good at faking humanity and thus actual humans have to jump through even more hoops to prove they are not "AI". Too bad "AI" is not useful anywhere else.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 4 points 4 months ago

    Just rapidly select a large square and delete what absolutely doesn't have part of the object, cause there seems to be a really heavy weight on "they corrected themselves" to accept the first time

  • permalink
  • fedilink
  • source
  • parent
  • [–] 3 points 4 months ago

    Those were designed with the lowest common denominator in mind. Either option is fine.

    Doesn't matter if it has a pixel inside the square or not. It has multiple answers.

    I don't bother when a fractions of a square is filled. Most of the time, selecting 3 squares is more than enough.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 41 points 4 months ago (5 children)

    https://addons.mozilla.org/en-US/firefox/addon/buster-captcha-solver/ Works like a charm

  • permalink
  • fedilink
  • source
  • hideshow 5 child comments
    •
  • [–] 15 points 4 months ago (3 children)

    As they say, the best tips are always in the comments. Thank you!

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 11 points 4 months ago (2 children)

    As a someone who works behind a corporate proxy, where dozens are sharing a single external ip, I can tell that the puzzle is really 20% of what actually goes on.

    1. Mouse tracking
    2. Identifiable browsing tracking data extraction
    3. Automatic rejection for unusual high traffic

    As a rule of thumb, any thing that requires Google CAPTCHA, I do it out of the corporate network.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
    • [–] 136 points 4 months ago (9 children)

  • permalink
  • fedilink
  • source
  • hideshow 9 child comments
    •
  • [–] 88 points 4 months ago (7 children)

    Ahhhh! An uncensored swear word!

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 28 points 4 months ago (2 children)

    I instantly died of instant death. Fatally.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
    • [–] 32 points 4 months ago (1 child)

    But if you do that, google will block you ಠ_ಠ

    Great that I'm not blind... Thx google

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
    • [–] 12 points 4 months ago

    *fuck

  • permalink
  • fedilink
  • source
    •
    [–] 9 points 4 months ago

    Then came this

    https://cybernews.com/privacy/google-captcha-hand-gesture-camera-privacy-fears/

  • permalink
  • fedilink
  • source
    •
    [–] 7 points 4 months ago (10 children)

    this gets trained on AI??? i thought they used it for like teaching people how to drive.

  • permalink
  • fedilink
  • source
  • hideshow 10 child comments
    •
  • [–] 27 points 4 months ago (7 children)

    Captcha verification has always been to train neural networks. That is why it started with "what letters are here?" And has progressed to the images we see today. I even got one where the images seemed "deep fried" to try to train thn nerual nets on worse input.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 1 point 4 months ago (4 children)

    Sorry, but captcha image is obviously generated AND google knows what text is on it, because it won't let you in if your answer is wrong.

    Nice try, but no. Google doesn't need your input to train NN for text recognition.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 6 points 4 months ago

    As others have said, yes the server already knows what it means, or which images are correct. However it is based on confidence, that that's how neural networks work. It's a statistical model to determine the right answer. So even though its already confident that "pictures 1, 3, and 8 have a bicycle", when you say that those images have bikes, you are reinforcing and training it further.

    https://www.geeksforgeeks.org/machine-learning/what-is-reinforcement-learning/

  • permalink
  • fedilink
  • source
  • parent
  • [–] 3 points 4 months ago (1 child)

    There's usually a couple of pictures that the captcha knows what it is, and some that it's uncertain about. So you have both verification ones to check you're actually doing it right and some used for training.

    It was a lot more obvious with the book/text recaptcha where one word was a sully stylised one, and the other was scan from a book. You could usually put in whatever you wanted on the book text.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 4 months ago

    I agree with the point it can be used for training on images, or text scanned from the book.

    I doubt it is useful on generated image of text, and I can only recall those on text based captchas, can't remember any scanned text.

    Anyways, smart move from Google.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 1 point 4 months ago (2 children)

    https://github.com/dessant/buster

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
  • [–] 1 point 4 months ago (1 child)

    But what if they use a buster-buster?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 3 points 4 months ago

    Anyway, these recaptchas are not longer used, currently there are more and more "I'm not a Bot" tests where you have to slide an block to the right or direct an marking field where the test ist based on the cursor move (humans curser move is never so exact and straight forward as one of an bot), since recaptchas and also traditional captchas nowadays, more with corresponding AI apps which solve it better than humans, useless. In accounts, to avoid spambots you even don't need an captcha, its enough to wait half an hour before sending the activation mail, due that spambots use an temporary disposable mail, they never receive it. This methode has shown as the most effective in several forums I know.

  • permalink
  • fedilink
  • source
  • parent
    • 6 comments in c/degoogle@lemmy.ml view that post
    [–] 16 points 4 months ago (4 children)

    Technically it doesn't fuck up the AI training; it just sidesteps it (still good and definitely easier)

  • permalink
  • fedilink
  • source
  • hideshow 4 child comments
    •
  • [–] [S] 3 points 4 months ago (3 children)

    If you input text that doesn't match the audio and they use this for training AIs, surely that will harm the training, no?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 4 points 4 months ago (2 children)

    I don't think the audio challenge trains AI since it only plays a single clip. The visual challenge works because it mixes in answers it knows (to confirm you're a human) with answers it doesn't know (to train the AI)

    My theory is that the audio challenge is purely for verification and they're okay with it not doing any training since a small minority of users will opt for it anyway.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] [S] 3 points 4 months ago (1 child)

    But if the audio challenge is not for training and they know the right answer why are you able to pass it by typing something completely different but with some syllables in common? It seems to me either they only have a vague idea of what the right answer is or they are open to the AI being wrong and in future it will be adjusted by input they think is reliable because it's sort of similar.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
    • [–] 29 points 4 months ago

  • permalink
  • fedilink
  • source
    •