So are different users served the same picture but with the squares aligned slightly differently? Or is it exactly the same always for each user?
This feels like a setup for an internet rabbit hole.
They fail you on the first one to get you to do at least one more.
I have noticed that the chances of success increase if one “forgets” the ever so slightly in squares or things like bicycle handlebars.
I get the impression that perfectionists have to purposefully make mistakes to count as humans in these things.
I now take a rapid fire who cares if it’s wrong approach to these. If I bother at all…
Man I’m glad I’m not the only one. And does the dude on the bike count as “the bicycle”?
deleted by creator
Audio captcha -> open-source* speech-to-text -> cursor inserts the captcha for me, w/o an extra dedicated CAPTCHA add on some of the corresponding potential hassles
(Closed-source superwhisper + Keyboard Maestro also make this a breeze on Mac :) )
* ggerganov/ whisper.cpp - Port of OpenAI’s Whisper model in C/C++
I suck at these, apparently. I especially hate the motorcycles, but this goes for almost anything. Do I select the tile with just some wheel in it? Is the wheel a motorcycle too? What about the rider, are they the motorcycle? I mean, when a bike passes you on the road, you don’t say “I got passed by a motorcycle and its rider”. What about the pole the traffic light is on, does that count as part of the traffic light? I end up doing them for like ten minutes until my wife comes over and gets it first try.
Highly recommend Buster extension. You click on it and it uses the audio version and solves it for you. Works like 95% of the time.
Ah, so we can now use an automated robotic script to prove we’re human? :)
do motorbikes and mopeds count as motorcycles?
They do but shouldn’t. I had a captcha scooter recently.
Note: They didn’t ask me to find the “scooter” squares.
Relevant: https://novehiclesinthepark.com/
That was interesting, and l liked the one it links to at the end, Staying Alive, even better.
If I protect the body it’s “why does the soul need a body” and if I don’t protect the body it’s “how does reconstructing the body make sure the soul comes with it” it’s a catch-22 is what it is.
Did it say anything about needing a soul to survive?
It’s just meant to be a thought exercise.
They’re certainly exercising thought
It’s weird, I followed the instructions too closely opposed to basing it on emotional response. Not sure whether i would answer differently responding emotionally. I’m only in 7% agreement with the sample population lol!
I find the statistics mildly disturbing on the last query.
Yeah the last question was the only one I deviated from the majority on.
To be fair, you also don’t get passed by a carbio and its head(s)
Why are we all out here doing unpaid image tagging so some tech company can have a dataset to sell to self-driving car companies? This is serfdom.
I always thought street lights ≠ traffic lights?
Yeah, that’s just the comic author
It says pick all squares that contain street light so it counts because it contains part of a street light. I get these wrong like 50 percent of the time so IDK.
But you know who doesn’t get these wrong? Bots!
(Source: experimented with Skyvern)
I used to do this but have learned (like a fucking machine) that its best not to as that gives me a better success rate.
Good thing self-driving cars were trained on your data!
(Flashback to that Will Smith movie…)
Flashback to that Will Smith movie…
I fail to see the connection between self-driving car data and Will Smith blasting aliens with a tiny gun.
It was oblique.
The robot deciding to save the protagonist vs the girl based on algorithms/statistics, compared with self-driving cars deciding how to react at traffic lights (and save you or not) because they learnt from your decisions in captchas.
Or… you got it right but they take the chance and ask again to grab info about how you solve some more.
Just try to do the worst posible just to pass it but at the same time fuck up the AI mind
I’ve found that if its only the tiniest bit then not to click. U also gotta account for the object hitboxes being rectangular. I have been trained by a machine in how to convince the machine I am human by lieing to it thus training the machine more wrong. A perfect case of misalignment lol.
It really does not matter is what I learned, I just click whatever I see in 1 second and that is it. I will have to redo some, but I always had to do that. So I stopped wasting any time.
Overthinking is a sign of not being a bot
sleep(10);
print(“You might be right.”)
Depending on the language, that might only sleep for 10 milliseconds, outing you as bot for sure.
It’s based on whatever the majority of humans using captcha say. They ask multiple question on the captcha. Some are testing you on. Some they don’t know and give to multiple people to figure out which boxes should be selected
Ok. So is there enough streetlight in the square?
Should be yes, but the human is only interested in getting passed the prompt, not with accuracy so it’s no. Which I’m pretty sure fucks with their AI training, but who cares.
If the majority of those given that specifc question say so, then yes
If the majority given it say no, then no
There’s no predefined answer here
One of creators of the captchas said in an interview that they themselves were never sure about the edges and if they should include it or not
These are made so the lowest denominator can pass. Both options usually work.
Gaaah! Wait! What’s the answer?? Sometimes I click that square, but other random times I don’t.
You’re determining the answer by training Google’s AI
If we’re supposed to be telling the AI what’s right, why do we so often get it “wrong?”
There are 5 lights.
The answer is the time and decision making, not correct choices.
