What I hate is the infuriating ones where you click a correct box, and it slowly disappears to be replaced with ANOTHER picture, which is probably also a fire hydrant, so you have to click that, too.
my guess is it's probably to avoid bots who are like insta-clicking the pictures and so the captcha is like "hey I didn't get time to load this next image, wth thats too fast". but idk /shrug
@@microman502 pick the image of the eldritch nightmare standing at the end of your 2nd floor hallway that is currently waiting for you recognize its presence.
I was once given a captcha that asked to select the images of parking meters. One of the images was a plastic mailbox that the captcha thought was a meter. I failed the captcha for knowing it wasn’t a meter.
These capture images go to computers worldwide, American mailbox's can look very similar to parking meters in the UK and other countries around the world and the ai learning will be messed up by this. And tbh, as a Brit, I originally thought those type of mailboxes were just a 'Hollywood movie' thing and didn't actually exist in real life (they do!). Cus why would anyone want potentially important letters placed in an easy to steal from box at the edge of the road... Just get a letterbox and have them pushed through your door directly into your houses. 🤣
@@Ryukai-san Unfortunately the United States is very large and there are many zip codes that mail needs to cover daily, and also people like to build houses separated from the road so letterboxes aren't thus fashionable. (Never lived in a city so I'm not sure about that, but I think for more cramped spaces in the US people either have a kind of have a wall/room of mail for their entire apartment/building where they go down to the desk and pickup their items (possibly get notified when some come in) or they have P.O. boxes) Tangentially related: my current belief as to why some things are like that in the UK is that the UK has been built up through literal centuries and therefore has housing (and other) conventions that last past their age, as opposed to the US where there's so much land people can kinda "waste" or "recycle" land on the newer conventions
@@richardpike8748 Also it probably has something to do with the fact that Americans seem to think they need a car to travel a few hundred metres. Hell the amount of times while I have been there I have simply walked to the destination and been there waiting for them to arrive when we left at the same time is silly. Over distances that short the time taken to get into a car is a significant portion of the time needed to simply walk.
@@Ryukai-san Security might also play a role here. Here in Venezuela we have so little security that we don't trust some random getting close to our doors to leave the mail behind, instead we prefer to get from a little distance away from our front door.
What makes us human? ❌ Love, affection, care ❌ Music, painting, lyrics and other art forms ❌ Reasoning, doubt, critical thinking ✅ Selecting all traffic lights
The best captcha is the one that records the imperfect nature of our physicality. Our hand forced to helplessly drag a mouse with all the meandering. It knows that bots are the only beings capable of moving from a to b in a perfect straight line since it's not bounded by our anti perfectionist universe.
In my opinion the most elegant one is the one where you simply click the "I'm not a robot" checkbox and all the analysis/evaluation is done behind the curtains.
Elegant and insidious. Elegant because it steals your habits without you realizing and insidious because those digital habits can be used for anything from placing ads where you can't ignore them to allowing hackers to bypass gateways by scripting mouse movements instead of other code more likely to be identified as a virus.
That's good because if filling out the form normally takes a human many seconds, a bot can't log into many sites per hour and the poor bot will get frustrated.
@@bob456fk6 -- Or you just run multiple instances of the bot set to type at human speed, using randomized variables to delay mouse movements and mimic human types of behavior, even occasionally making intentional "mistakes" to appear more human. People who code bots often have access to web logs and can see how real traffic behaves and, since the whole point of making a bot is to get money, they will implement those features.
Right now I'm regularly having to deal with a "choose the similar shapes" captcha. Problem is that I'm color blind, and the red-on-purple hexagons and green-on-yellow triangles are almost impossible for me to pick out. Especially given the teeny-tiny window it gives you.
@@Supreme_Lobster Not so easy to fake if some ML classification model is used. Human inputs are imprecise with some patterns, not totally random noise.
My favorite kind of Captcha is the one where you click the verify button and then just wait for a few seconds and then the Captcha is just done with no effort whatsoever.
it uses browser fingerprinting and analysing your mouse movements on the way to the checkbox, maybe checking cookies and whatnot too. if you use a privacy-first browser you'll probably never get just the checkbox and it will move on to having you solve a problem
fun fact: on some sus websites (pirated movies and stuff), if you're shown a captcha it might not be for the site you're on, instead it's getting you to solve it for a different site so it can bot *that* site don't need to make the AI better if you can just get a human to do it without knowing 😌
I remember this old audio test that still scares me to this day, it was one where you type the characters you hear but the audio sounded like a loud intercom in a parking lot with sirens on in the background, very unsettling
i remember one night i was making a new google account and the capchta was typing what you heard and the sounds were demonic growls, holy shit, i was scared shitless and just went to sleep
"Let's stop AI by making them read and at the same time train an AI to learn how to read" "We no longer can use that anymore as somehow AI's have learned how to read"
And once the AI could read and the captcha was useless, they said "Lets use object recognition as a test. And feed the data to our object recognition AI."
The problem with this one is that it assumes a stereotypical Hollywood AI which operates on total logic and zero cultural awareness. That's not how AI turned out to be in reality at all. An actual AI would easily choose the second option because it has seen that specific combination of text far more frequently in its training.
You might be interested in Chinese CAPCHA too, they have evolved to rely on different uses of the Chinese language. For example: 1. As Chinese characters are basically square-y pictures, like 架, 奇, 雯 etc., they ask you to rotate a random character upright in the correct way. People who knows Chinese can do that effortlessly. 2. Different Chinese characters have the same or similar pronunciation, e.g. 忍, 人, 認, 刃, 仁, 韌, 飪... are all "ren" in mandarin, spoken in different tones, and all commonly used. So there is a CAPCHA for you to pick "which word sounds the closest to the given word". And also, to match the word to the given pinyin (i.e. "ren") and vice versa. 3. Three random, uncommon Chinese characters are given, and the distorted version on these three characters are hidden in a colourful picture. You are asked to click on each, according to the order they are listed in the question.
In order to get a Japanese google account, I had to answer a captcha with distorted Japanese text. Certain characters in Japanese are difficult to distinguish even when not distorted. Guess which characters they used!
The one’s I’ve encountered ask you to click on four characters part of an idiom (成语) in the order they appear in that idiom. They’re usually rotated and on some kind of background, as per usual captcha tests. Sometimes I have to pull out my dictionary if I don’t recognize the idiom 😂
Just gonna drop my favorite quote which I always get reminded of when CAPTCHAs come up: "There is a considerable overlap between the intelligence of the smartest bears and the dumbest tourists." -Yosemite Park Ranger on why it's hard to design a bear-proof garbage can.
@@terig8974 they do that too if your a bot hoster you can buy a number of captcha solves from call centers usually in india where real people will just solve it for the bots i can sign up to solve captchas at a payscale of 1 dollar every 1000 solves aparently there economy makes this worth it sadly 1 dollars actually good pay too ive seen .40 per 1000 too
I remember a few years ago I was sobbing I couldn't get into my account because of that dumbazz captcha, it's a literal platform kids go to, why make it so hard even kids can't solve it?
Github had those back in 2021 and I legit could not solve it (Thankfully there was a request for a different one or otherwise I would not had got my account on there for Bug Reporting)
Gosh that brings me back to all the times I would get locked out of my Roblox account, because it would give me a 20 max level difficulty captchas, then every failure would make me restart all 20.
These captchas are almost like those free flash games we all used to play. I swear, if it keeps evolving like this, in 20 years we'll have to play a cs:go match to complete a single captcha.
@@Rilex037my boy playing dark soul reorganise every pattern to defeat every single boss and final boss and only to play flappy bird but boy it was worth it and you know i am human if i wrote a wrong letter can't wait to play flappy bird and get to play dark soul for free right?
Half the time the robot I am trying to convince I am a human doesn't even know the answers itself. I had one that said "Identify the mailboxes" and it counted it wrong because it thought a brown trashcan on the side of the road was a mailbox.
Back in the days, I used a very simple mechanism to block contact form spam on my website: I put a checkbox under the form, with a little text instruction over it that said "Don't check the checkbox below". Then I used CSS to hide this checkbox as well as the text. Normally a user with CSS styles on wouldn't see the checkbox and its description, so he would leave it unchecked. Similarly, a human user with CSS turned off would see the ckeckbox along with its description, and followed the advice to not check it, so it would remain unchecked as well. But a bot would most likely just see the raw HTML encoding the form and try filling up all its fields with some random garbage or spam. On the back end, I checked if the checkbox field contained anything, and if it did, the message didn't pass through. Only the unchecked messages did. Simple and efficient.
My favorite captcha was one that asked the user to click the animal that's walking backwards. The damn thing was so hard, I gave up trying to log into an account on an MMO's website. Many of my friends couldn't log in either. The problem was that the animals were walking in place on a white background - there's no sort of environment that's moving throughout the image to indicate which way theyre walking, so you have to think about the way an animals legs move. But that's hard to get correct four or five times in a row, which the captcha asked you to do
Have you had the "how long are the pants" captchas? After a few of those, I just wished that the internet would cease to be. It's bad enough already that apparently, most people think that anything(or part of anything) with 2 wheels is called a bicycle.
i remember the old recaptcha. it was such an asshole move to have us teach it, because it was clearly going to lead to captchas being more annoying in the future it was always fairly obvious which was the one we were teaching, so my method of coping was to always say the word was "butts"
At this point, we've gone from "how do we fool an AI that can't recognize text if it's a picture" to "how long before AIs learn how to recognize, behave in response to, and interpret abstract concepts better than humans?"
I think it was never about "How can we fool it" and always about "How long does it take until it's better than us". That's more often than not the reason for any program: Being faster than handish human work.
except the captchas are designed to train bots to be better - why do you think they're all "select all of a frog" or "horse made of clouds" or something - you're training ai to generate images, that other ais will recognise, it's kinda wild.
Captchas 10 years from now: - Do a photorealistic drawing of a dog to prove you're a human - Win a TF2 match to prove you're a human - Write a song about proposing to a supercomputer to prove you're a human - Bake lemon macarons and mail them to the company to prove you're human - Translate the Voynich Manuscript to prove you're human
I had the touch the mouse one once. Had 10s time for each of the 10 levels loading time included and they got progressively harder. When I finally did it it said I was too slow and had to try again. Gave up after 1 hour I was the mouse who couldn’t reach the cheese all along.
_Ah yes, let’s block the user for taking too long to solve our ridiculous nonsense CAPTCHA, because bots are well know for their lengthy deliberation on how to solve a problem._
I once got a CAPTCHA that drove me insane for almost an HOUR. It told me a certain number and showed me several images of dices (a few dices per image). I then had to calculate the sum of each image judging from the dots on the dices. But that alone wasn't annoying enough. It was TIMED. I was given around a minute to solve 9 fucking summations in dice-form. I either ran out of time before I could figure out which image was the right one, or made a mistake calculating due to the time pressure. And that wasn't even the worst. I had to go through that process more than once, and if I messed up at any point it would mean an instant restart and loss of all progress. It was truly infuriating, especially for someone like me who is a slow calculator.
slow calculator? its just stupid lol, i had to do one that was "click on the motorcycles" and i sat there screaming at a bunch of random people on motorcycles because i couldnt figure out if half a wheel in a different square was a part of the motorcycle. even the "which ones with bus" was dumb, i clicked all the buses, ALL OF THEM, and i guess it said that four crosswalks 2 cars and a fire hydrant were a BUS
@@Sarantis-107 I liked that one. It taught me that English speakers generally do not consider buses and trucks to be cars. Everything is a learning opportunity if you keep your mind open.
A couple of weeks ago, Twitter presented a "pick the shadow that matched the icons at the top of the image" captcha. My twitter account was in a head lock for a whole 2 days because the captcha was impossible for me to complete. I tried to get my dad to complete the captcha as well, and he couldn't do it either. The only thing that saved me was when the captcha changed to something else.
The ones that require you to pick the photos with certain images in them are actually meant for you to get wrong sometimes. The reason they sometimes have a square with a tiny piece of the traffic light on it or something is because the captcha knows that a human could miss those squares but that a robot would complete it perfectly without delay. This is why sometimes a captcha will let you into a website even if you don’t get it fully correct.
@@stm7810 actually yes it does. I've done so many recaptchas that I rarely read the the text anymore, I instantly start picking what seems "intuitive" to pick. Like if I see cars, then a picture of a bike I know that it's looking for bikes, if I see boats I know they want boats, see landscape I know they want hill side... some times I actually miss click images and still it accepts it, I'd say that 98% of the time I don't have to do another captcha even though I didn't read the text..... witch kinda makes me wonder, isn't the point of recaptcha to catch "bots" that don't read and just know the solution? also am I a bot? The new one with the animals is always throwing me off, but given enough time I'll probably get the intuitive feel of those challenges.... I guess that they could really just be testing intuitivety, as bots don't have intuitivety they just "know"
@@svampebob007 The problem might come from the fact that there's companies out there that would pay money in order to be able to ...I don't know... have bots advertising for adult sites write comments and chat messages (definitely an example that's completely unrelated to youtube). Which means that as long as human labor can be cheap enough, it becomes viable for other companies in low income countries to offer to have humans solve those captchas. And those humans will inevitably solve thousands of captchas per day; which means they stop bothering to read the prompt, and they develop their own subculture of what they consider a valid answer. And since each of those workers and offices solves so many captchas, they have much more influence on how future captchas are trained than any normal user.
The beauty of mini game CAPTCHAs is that it takes almost no time to develop a new variant of “add up to 12” but it takes days to develop (program/train) a bot-based solver for the new kind of CAPTCHA. Also, I noticed CAPTCHAs today are super time-consuming even if I can solve it and it seems it’s slowing me down intentionally to throttle traffic or limit number of accounts created per second.
@@melody3741 Usually, but once I encountered the sum dices to 14 captcha. You don't get just one sich image but like five of them. It is for some reason timed! I am pretty good at calculating but I am just not fast enough for that captcha. I got two of my friends to parallelize the captcha solvong. Each calculates one column and shouts if he sees the 14 sum. It still took us three attempts to get it done within the time limit. I think it took us at least 10 min to get past this captcha...
Actually of all the captchas mentioned, the one where you add up the numbers doesn't seem that hard. I've never done it but if the goal is to just pick a square that has a wheel of numbers, add them all up, see if that matches the correct value... that wouldn't be too hard to program in. You can just use OCR to extract every number from each wheel, and then run a sum total formula of all those numbers to see if it matches. Keep iterating through each wheel and viola. Not really all that hard.
@@boggless2771 Yeah, but it's not about where they can possibly lurk, but where they can lurk not having to care of any consequences because the company doesn't give a flying phück about it, or silently supports it :q
11:08 I solved quite a few of those math CAPTCHAs for fun (reloading the page), they look scarier than they are if you know basic algebra and calculus. And the answer to many of them is 0, including the example shown in the video!
Title: Worst CAPTCHA ever Video: The complete illustrated history of CAPTCHAs Bro, you anti-clickbaited me. I thought this was going to be some silly, vapid, time wasting horsesh*t I'm gonna watch, and you took me an entire emotional journey to Mordor and back. You had me questioning my humanity. You had me questioning Google's humanity (okay, that one's easy, but you still did it, so it counts). You are definitely getting subbed.
A really decent one was on a Japanese site where you not only had to slide a slider to line the hiragana symbols up but also answer what it told you to … in kanji this was used to teach a ai language model to translate between the two writing styles
@what the 4chan currently uses a slider with text based captcha. how it works is imagine those distorted text with lines captchas, and you cut some ovals into it. You’re supposed to line it up close enough to read the answer
Got the feeling that would be harder and harder for both human and computer to do it for Chinese and Japanese words are so blended together it's just hard to even tell which word is native and which is foreign in both languages nowadays, and that's not accounting everything that shows up before the 21st century that are by default considered all native (well, semi-native I mean, just like how people usually won't care that ketchup is a Chinese word).
I don't know why, but when it talked about how the 'select traffic lights' and those sorts of CAPTCHAs were from actual google maps images, my brain just exploded a little. I had no clue that the original text based CAPTCHAs were even used to teach AI how to read, and so with that knowledge, the second that it talked about traffic lights, I instantly realized what that actually meant. That's real cool, and extremely horrifying that I didn't even know what that was being used for.
It's interesting how as soon as self-driving cars were being more marketed and advertised as coming into existence suddenly captchas featured taxis and motorcycles, types of vans and cars, and especially traffic lights. so many traffic lights
Why are you saying that like it’s a unproven conspiracy lol. It’s well known they’re using them to train AI’s some of which drive cars. This video specifically brings that up like a few times at least 😂 Although I think it was more so for google street view than actual driving per se
@@t1Pz ehhhh, I mean AI recognition is generally pretty beneficial, but it *is* kind of crazy how so many people have unknowingly helped Google to do stuff like creating a digital globe where everyone knows what your house looks like if they find your address... I dunno, I'd just be careful to think that us being used to improve AI is always a good thing.
@@t1Pz I like to be paid for the work I do, even if it's a repetitious and brainless work. Let's just leave it at that. And especially when it comes to Tech Giants like Google, Meta and co.
I’ll be honest, finding Wally in a captcha is genius because it relies on pattern recognition rather than arithmetic and because it should be a bit of fun.
@@mikhail-cybertastic5349 Visually-impaired (fully or just partially blind) usually use screen-readers to travel the internet. It reads whatever is on the screen or what they are hovering over. How else do you think they made it to the website the captcha is on in the first place?
as a visually impaired person, I absolutely despise captchas. Due to my poor eyesight I can just never figure them out and spend way too long solving them
hCAPTCHA allows you to get an accessibility cookie that bypasses the CAPTCHAs. reCAPTCHA has an audio captcha button that 60% of the time refuses you and tells you to screw off because it takes clicking that as a red flag that you're a bot (???)
15:25 the most obnoxious captcha I've encountered was trying to sign up for a throwaway email on outlook recently and it hit me with a captcha like this except with a series of incredibly abstract and highly distorted image matches requiring you match 7 to 12 different images and then it wouldn't even explicitly state if you failed it would just load a new sequence of images to match or rotate. I assume that means I failed but I am not sure as it never told me. It started to feel like the site was just farming captcha input and never actually going to give me an email address. Ultimately I left and signed up with a different provider lol.
My question with the images has always been "How do they know I got it wrong if I am the one training their AI?" (I purposefully click wrong squares some times just to mess with the training)
@@Squant how can they, in good conscience, continue with their everyday work and hobbies, when they're basically training ai to read distorted words in scanned books and improve self-driving cars... the absolute NADIR of intelligence
OMG the math ones. Microsoft used those for a bit too. I was trying to help an older gentleman get back into his email account (I run an IT shop) and I COULD NOT answer them fast enough! I'm 37, not 5, and I took freaking calculus! The problem wasn't the math, it was the speed with which they thought you should be able to add all the numbers together in each box to pick the right one!
What's even worse nowadays is that you can cheaply buy a manual captcha solving service. So you have your bot interact with a website and when you're confronted with a captcha there's an actual person somewhere that solves it for you and you get back at botting...
The one with Roblox kind of reminds me of the copy protection on old PC games where the obscurity was the point, you had to prove you had access to the manual in order to get in because the manual was hard to copy as opposed to the software itself which could be copied easily. Carmen sandiego had you looking up specific words in specific sentences on specific pages of the big fat reference book you were supposed to use to solve the in-game mysteries.
Yeah. I remember Return to Zork let you play up until the school house, where suddenly the teacher will give you a pop quiz of random questions (answers were in the manual.) Failure to answer correctly meant death.
I remember Tie Fighter had that security feature as well. I think Sid Meier's pirates also had something similar to find which ports had the treasure laden ships docked so you could raid the big bucks... Ahh that was so long ago.
Jesus christ, I get flashbacks to the worst period of video gaming I've ever experienced. I could kill somebody to this very day for fcking up Pizza Connection. Neither the Pizza templates, nor many phone number actually worked. Makes it also incredibly frustrating when you want to play... like at all.
Some memorable ones I had were Eric the Unready where you had to tell your armor sizes to the armor craftsman so that he can make the armor that fits you. Was pretty cool, but extremely easy to copy, since you just had to copy one page. Another was really annoying (TMNT) where there was a dark dark purple sheet with many hundreds of very tiny numbers (4 digits IRC) printed in black, and you had to say the number on row y column x, but it was so hard to read due to the small size and dark-on-dark colors. I was young so it wasn't too terrible, but I remember always breaking out the super strong desk light and looking closely.
I have a screenshot of one of these weird captchas that says, "Pick the smiling dogs." Underneath it is a bunch of poorly photoshopped dogs that look horrifyingly like skinwalkers.
I can't believe it never occurred to me we were training AI every time we logged into something. If you're not paying for the product, you are the product, and all that.
don't worry, even if you are paying for the product, your data is still a secondary product. marketers would never give up that info just because you paid them.
It was a fantastic idea. Kinda like those apps that use your extra computer/phone RAM when you're not using them to add to a cloud based supercomputer.
Back in the day (at 3:11) I do believe they advertised that detail publicly that the service helped to authorize you + translate scans of old books and articles. And like in the example in the video, you could often determine which word it didn't know, since it would be the less wavy/warped one, and if it was too hard to read you could get past the captcha by just writing anything you want
i once got a RIDICULOUSLY hard captcha, you had to make like 20 math problems in one minute, i remember being at my friend's house that time and none of us could do it
One thing that I always found annoying about those early captchas is that, as a dyslexic person, I kinda struggled with them too, the occasional variation in fonts used and the distortions that got added later on made it so much harder for my brain to process what was in front of me when it already struggled with that enough on regular text. Sure it would normally get it eventually but it would take me quite a while, and sometimes it would take me a heap of captchas to get through that. As weird as some of them are, I'm glad they've moved on to imagine captchas, they're much more accessible and they don't make me feel terrible about not being able to read what's in front of me
@@trytoo5167 I'd realized there were accessibility issues, but you make a very good point. They're literally testing typical human abilities, so it's unavoidably non-accessible. Interestingly, the only way to do it well might be to have humans approve submissions, rather than computers.
I don’t have dyslexia but I still struggle with most text based captchas. They’re just hard in general, especially when it’s a guessing game of is it capital or lowercase, is the captcha even case sensitive idk fuck em lol
The AI getting so good at CAPTCHAs that humans couldn't do them gives me an idea: a CAPTCHA where in order to pass, you actually have to get the question wrong
It's just as easy for a computer to fail something as to succeed at something. Though your idea is still somewhat achieveable, but in a different way. Basically there are methods that targets specific machine learning algorithms that would encode an image in a way that would be invisible to human but would cause the algorithm to output a wrong answer with fairly high confidence score. Maybe the website could timeout any input that's wrong in a specific way because only AI that got hit by such attack would give that specific answer.
2 things. 1: I have recently seen images with Dalle mini/2 generated images in captchas. 2: Roblox is the worst website ever because of the captchas alone but then you see the awful way they work and it's worse.
some of the "move the x" capchas actually measure mouse movements. AI's movements are not as erratic and inefficient at human mouse movement. but if you try it on a touch screen... it sometimes thinks you are an ai
Yup. A programmed bot would move the cursor in a straight line to the target. You do pretty much the same thing with a touch screen and fail the captcha.
@@Unknown_Genius Sure I'm aware. The OP observed that "if you try it on a touch screen... it sometimes thinks you are an ai" vs. "erratic ... mouse movement", which is some of the information that a captcha looks for to decide that a human is present. By your reply, I gather that you're familiar with HIDs. In response to a) I think you'd agree that compared to a mouse, a touchscreen output is lower in resolution. Also, depending on the device, movement is often processed and filtered to make it more usable than it otherwise would be. So it doesn't surprise me that some movements of the x on a touchscreen could fall within a captcha's threshold for suspecting a bot. In response to b) We agree. I should have said that a basic script would move the cursor in a straight line to the target.
This reminds me of the terrifying captcha of "choose the animal with the wrong head" captcha, it would have dog heads on squids and octopus heads on birds, it was scary af.
13:40 I've heard another thing it monitors is the way the mouse moved towards the checkbox. If it was a perfectly straight line it would be suspicious because an actual human would likely waiver a bit while moving the mouse.
then again theres a video of a literal robot moving a stylus across a touchpad only across the vertical and horizontal axes (so only straight lines) to solve a captcha and being let in instantly so idk about that
@@coolguyman7356 One of the features it detects are likely the noise and slight jittering caused by the hardware. And possibly the fact that if you have hardware moving the cursor, it has to obey the laws of physics and known materials (i.e. no infinite acceleration).
@@coolguyman7356 it lets you pass a few times so it can fingerprint your bot and block you more effectively in the future. it also blocks on many more factors than just mouse data, and if the other factors are valid and you're not a suspected bot, you'll likely pass the captcha.
@@Pystro true but i was thinking you could just simulate normal acceleration with a bot so why bother with that anyway unless youre tryna be thorough in catching those bots
It's AI training. The captcha it's self is for training AI algorithms. It was never about preventing bots from using sites, it was about getting people to unwittingly do the hard work of training AI image recognition models.
Listen I hate google as much as the next person (and a lot more) but I have to disagree here. Yes I don't doubt that google is using users to train ai, but the main point is STILL to stop bots. (looking at some comment sections however, you can see just how effective or not it is.)
@@cact0s_ulion405 I mean main goal or not, fact remains if what they are actually doing helps bots better than it stops them, then they're effectively helping the bots. If we're the data they use for that, then we're the unpaid employees doing it. If you want to stop AI bots from getting in either switch it up so fast they can't get any significant data, or figure something out that AI actually cannot do. Problem is, this would cost a lot of time and money and the way they do it currently might suck but it gives them back a lot of time and money in the long run, so why would they change it?
I remember the old captchas. It's always been pretty obvious which ones were transcription data and which ones were the actual humanity test, so I'd always transcribe it as something horribly incorrect. If Google wants me to work for them, they can pay me. Otherwise, they have no reason to be surprised that I'd make their transcriptions more colourful than they intended.
Google will honestily pay you as part of their survey programs to do stuff like this just like people are payed to test games. Not much but for a brief weekend payment it is something
My least favourites are the one's that don't take into account language/country differences. Most of them seem American based, I'm not from there so sometimes I don't know what the thing is or what counts as one. I get through them in the end but sometimes they take me a while when I'm trying to decide whether the bit of road with pavement on each side but no crossing lines counts as a crossing
(Not)fun fact: They used such captcha on one Polish government site :q I bet lots of Polish citizens were unable to answer the English-language captcha, or even get what they're being asked to do :q 1:0 for the government though - they didn't have to deal with interesants for quite a while…
man im from america and i cant even do this. picture this, 3 buses. you click on the buses, and there are no more buses. YOU ARE WRONG. THE FIRE HYDRANT IS A BUS AND THE CROSSWALK IS A friggin bipod.
What we should do is make one that has a GIF in it. It'll screw with a lot of image recognition AIs, since not many can trace motion. Edit; Ok so thanks Thomasmills8613 for pointing out that this will just teach image recognition AI to trace motion. Only problem is, how do we deal with robots without teaching them something new?
Bitcoin could be applied to solve this problem. Each person could have a layer 2 wallet into which they load some small amount of BTC. Charge a few satoshis to perform actions on each website where limiting bots is a priority. Real users can afford to spend fractions of a cent here and there as they browse normally, but operating a bot army at scale would no longer be free and would require an actual financial investment. There isn't anything new that bots could learn to defeat this (well, except maybe how to earn money..) Edit: Just to clarify why I mentioned Bitcoin (as opposed to a traditional payment system) is mainly because it is permissionless, meaning anyone in the world can easily opt in without needing to ask for permission by some third party. A traditional payment system would potentially exclude a lot of locations in the world from participating, and it might compromise the ability to browse anonymously.
fun fact, mostly for the image ones; my social worker was recently working with some people who program these things. apparently the way they ACTUALLY tell apart humans and robots, is how fast you answer the questions. an ai will click the answers straight away, or fill in the solution too quick; a human will take more time, the curser (if they’re using one) will move around a bit before inputting the answer. the ai uses that to tell if it’s a bot or not and uses that to kick out bots
my gripe with capthcas so far, is when they are the "click all images till none remain with X" because the moment they get laggy, you fail 100% of the time even if its correct. Even the "Choose all traffic lights" has issues where it reads things improperly. my favorite, is a mailbox was a traffic light, and it got mad saying i didnt select all traffic lights. Yes, i waited till the mailbox came back later and tested, selecting it worked..
I actually noticed a pattern with google recaptcha, it always seems to have you select a certain amount of them, like 3, 6, 4, so chances are, you can cheat it by doing that too.
@@comet.x huh I've never noticed it, what I noticed is that I don't have to read the text anymore to know what they want me to pick, I've never noticed it was in a pattern.
10:44 Simi: “if so, you got it correct… which actually means you failed the captcha, as only a robot could possess that level of perception” Me who guessed it right: start searching for Sarah Connor
I also believe the recapture "I'm not a robot" one also tracks mouse movement through the page, a bot would tab or not even tab through the input fields as it fills out, where as a human would have their mouse move randomly about the page. There are multiple layers at work with all of them, somewhat fascinating.
That actually makes me a bit mad. I keep getting flagged because captchas think I'm a bot. I get banned from all sites that have this verification.. latest one was steam and I bought games and i just wanted to play one..
(3:35) It's not a great system though since I've seen Google books where they've used these reCaptcha answers as the scan-to-text, but the text is wrong. Usually when it's non-English text missing accented symbols which are important for non-English text. Plus sometimes most people might make the same error and only a few actually gets the correct answer. Since these words are shown out of context, and it's hard to tell what it is, but easy to tell when part of a full sentence. Basic example: З, this might look like the digit 3, but it is actually the Cyrillic Z. You can't tell out of context, but in the context of "Zlatoust (Златоуст) is a Russian city", you can at least tell it isn't a digit.
Google hasn't focused on non-English data mining as much, but it's ramping up. Pretty soon you will have trained an AI to defeat your language skills too, and do it at no cost to Google.
The one thing I love about modern captcha is that they always make me think of the lyric "I'm not a robot and I'm not a monkey, I will not dance even if the beat's funky"
I mainly hate it when it has those where you have to select the parts that are the thing like a bicycle or stop light. Sometimes there is a pole or the smallest corner you aren’t sure will be counted as a space.
I would gladly contribute if someone starts an internet-wide trend of entering as much adversarial answers as allowed into the recaptchas to atleast slow down this nonsense. My friend keeps sending me screenshots of ridiculous acid trip captchas he keeps getting.
@@Oneiroclast Considering the word captchas are obsolete anyway, I was talking about image ones. Pretty sure there's an amount of error allowed there, especially with controversial squares. I wonder if it's possible to figure out how much you can mess it up and still pass, and do that every time.
As I've worked in 2captcha for more than 3 years, I've actually faced every single type of captcha and recaptcha you mentioned in the video. Trust me besides all the types you mentioned, there's actually a lot more types of captcha out there that'll either amaze you or drive you nuts 😁😁
When reCaptcha started out, it was absolutely sufficient just to type in the first word correctly and the first letter of the second word, after a year or so, you had to write atleast half of the second word when but it was still possible to get through. With the amount of captchas I had to enter on Megaupload it was well worth knowing this
this is such a fascinating topic that really does reflect in a noticeable way how ai has gotten far better at recognizing images over the years, giving insight into projects that major tech companies here are working on, and what ai is still yet unable to do also great lesson on how to outsource answers for your machine learning, by making everyone else do it in order to log into their email
i personally really like Genshin Impact/Hoyoverse's take on the puzzle captcha. the image used for it is usually official artwork or a screenshot from the game. Pretty good way to incorporate your brand into it.
I looked at this and I thought it was going to be a compilation of silly CAPTCHAs. What I got was much better. Thank you, I have learned something new today!
10:58 that ain’t no equation though. It’s the slope of the tangent curve on a point in that function. The answer should be 12cos(3x-pi/2) - 24sin(6x) for any given point. We are looking for the value when x=0, so the final answer is 12cos(-pi/2) - sin(0) which is 0
Yesssssss I was scrolling through the comments to find anyone else that did it, I also got 0 in my head Yeah not an equation but at least he noticed it involves trigonometric functions
Those ones where it’s a single image broken into multiple squares is so annoying since you can’t be sure what it’s counting at all like if for the traffic light if it’s counting the pole itself or just the light itself since I’ve done ones where the light is in two squares and it counted it as wrong. Also bad is the multiple small pictures and it might count something you can’t tell as the item you’re looking for.
Worst part of that 10 captcha's thing is that it begins by telling you it's only 3 or 5 and then ramps up to 10 and if there's ANY hiccup in your connection it may tell you to just try again even though you just completed them. Epic games launcher had those for a while and I had to do them for both logging in and claiming my weekly free game.
I remember when I got a captcha similar to the frog one. The only problem was that the website auto-translated what I have to choose, so there were 2 possible correct answers.
I was watching this video waiting for Roblox to be mentioned. I was confused with every minute that passed with no mention of Roblox. I’m so glad you brought it up. When I had to submit a support ticket, I had to answer 20, yes TWENTY, of those die questions. And because I was trying to get through them quick, I ended up messing it up 3 times so I switched to the auditory version so I could be quicker. All-in-all, it took me over 30 minutes to finally complete the captcha just for their support team to be of no help. Apparently they’re very busy with the 3 other people that could be bothered spending 30 minutes trying to solve the captcha.
The Captcha's are being used to train AI. At first we were entering words and wonky words. Then Google released "Screen Read" and "Translate". Then we were locating Traffic Lights in pictures. And the data was then used in self-driving vehicles. Now we are completing images of fruit etc. Because AR tech is gonna be the next big thing.
The worst Turing test I've personally experienced was the one EA made me take when I was trying to create an Origin account to play It Takes Two. They made me count the values of a bunch of dice shown on various pictures, I have to do maths with them in order to choose the correct one. For every single question there were like 6 thumbnail sized squares of pictures and each picture contains like 6 dice, they weren't even positioned that straightforwardly and were stacked sometimes. You have to get 10 of those questions correct, in a row, AND IT HAD A TIME LIMIT.
So what you're saying is an AI is given the prompt 'make a horse that looks like clouds' and then needs to make that image, meaning it knows what 'a horse shaped cloud' looks like, meaning it can solve its own captchas. Incredible
@@UnleashedDan this. The amount of armchair computer scientists in this comment section is really something. I tend to not like the phrase "Armchair X" (because even non-"experts" can be knowledgeable if they have the right requisite knowledge) but jesus christ there is no other way to describe this comment section.
It's closer to running image recognition neural networks backwards. Say you had a Neural Network trained to tell the difference between "horses" and "clouds". You show it a bazillion pictures of horses and clouds and let it fiddle with its internal controls until it can answer, with acceptable accuracy, whether a new picture is a horse or a cloud. It does this by combining different "input layers" based on what it "sees", weighing these layers against one another in several different ways, and then getting two outputs; 1) how sure am I this is a Horse, and 2) how sure am I this is a Cloud. For example, it might end up 30% sure that it's a Horse and 50% sure it's a Cloud (implying 20% that it's neither, but it's *required* to pick one) so it picks Cloud since it's the higher surety. So, run it in reverse, give it 51% Horse and 49% Cloud on the backend and a catalogue of Google Image results to stitch together, and it will reverse-engineer a picture through its algorithm, using information from Google Images, to produce a picture that, if run the correct way through the program should give a result of 51% Horse and 49% Cloud.
@@numbdigger9552 You apparently have never heard of the DeepDream project or AI Inception. Look up either of those terms for more info. "The software is designed to detect faces and other patterns in images, with the aim of automatically classifying images.[10] _However, once trained, the network can also be run in reverse, being asked to adjust the original image slightly so that a given output neuron (e.g. the one for faces or certain animals) yields a higher confidence score._ This can be used for visualizations to understand the emergent structure of the neural network better, and is the basis for the DeepDream concept. This reversal procedure is never perfectly clear and unambiguous because it utilizes a one-to-many mapping process.[11] "
13:45 I used to think it tracked your mouse movements to see if they looked "human" (wobbly, fluctuating speed, etc.), but these other reasons make a lot more sense.
"A Capcha needs to be easy enough for the dumbest Human, but too hard for the smartest AI." Me, an Intellectual: "Design a Capcha that blocks both Smart AI *and* Stupid Humans from using the internet. Never settle for solving just one problem at a time."
Good suggestion, but what if it surpasses all humans collectively? Considering the AI can infinitely grow unlike a human, it's not an unrealistic outcome. The question is, who uses the internet then?
@@The9thMonth That's just a matter of Natural Selection. "What if Neanderthal or one of the other hominids out-competed Homosapiens and humanity wasn't able to survive?" Well, then it would just be the decendents of the hominid species that *did* come out on top discussing Capcha here and now. If humanity really degrades to such a degree that no one at all is able to access the Internet except for a bunch of AI, then it means the AI are the new successful species; emergent life from synthetic origins. It could even have emergent sentience or even sapience. Just as a common ancestor of all hominids gave rise to one that would outperform the rest, so too in the natural order of things may we give rise to that which may be categorically better than us. If that's the case, then logically I can't find a practical, objective reason to arbitrarily judge the better to be less deserving than us. If AI truly become superior to mankind and mankind dies out as a result, it wouldn't be any different than the dying out of myriad previous transitional species; the only new matter would be that the format has shifted from analog to digital.
Mr. September The robots do. Once the robots become smarter than all humans, we’ll no longer be able to create CAPTCHAs that can stop them because they’re smarter than us.
What I hate is the infuriating ones where you click a correct box, and it slowly disappears to be replaced with ANOTHER picture, which is probably also a fire hydrant, so you have to click that, too.
Reminds me of that one time they censored hydrant's terminals. It must have mistaken them for woman's nipples.
Oh, yeah, I hate the repatition.
my guess is it's probably to avoid bots who are like insta-clicking the pictures and so the captcha is like "hey I didn't get time to load this next image, wth thats too fast". but idk /shrug
How tf do you do it
And it's also an excruciatingly slow disappearance and reappearance
"Pick the wrong shadow" is viscerally horrifying as a concept, but creatively fascinating
pick the shadow that you can see under your desk
"pick the backrooms number that doesn't exist"
Pick the eldritch entity that's unperceivable and incomprehensible
inb4 "pick the image which doesn't contain a human" and instead has an uncanny humanoid instead
lol
@@microman502 pick the image of the eldritch nightmare standing at the end of your 2nd floor hallway that is currently waiting for you recognize its presence.
I was once given a captcha that asked to select the images of parking meters. One of the images was a plastic mailbox that the captcha thought was a meter. I failed the captcha for knowing it wasn’t a meter.
These capture images go to computers worldwide, American mailbox's can look very similar to parking meters in the UK and other countries around the world and the ai learning will be messed up by this.
And tbh, as a Brit, I originally thought those type of mailboxes were just a 'Hollywood movie' thing and didn't actually exist in real life (they do!). Cus why would anyone want potentially important letters placed in an easy to steal from box at the edge of the road... Just get a letterbox and have them pushed through your door directly into your houses. 🤣
@@Ryukai-san Unfortunately the United States is very large and there are many zip codes that mail needs to cover daily, and also people like to build houses separated from the road so letterboxes aren't thus fashionable. (Never lived in a city so I'm not sure about that, but I think for more cramped spaces in the US people either have a kind of have a wall/room of mail for their entire apartment/building where they go down to the desk and pickup their items (possibly get notified when some come in) or they have P.O. boxes)
Tangentially related: my current belief as to why some things are like that in the UK is that the UK has been built up through literal centuries and therefore has housing (and other) conventions that last past their age, as opposed to the US where there's so much land people can kinda "waste" or "recycle" land on the newer conventions
@@richardpike8748 In higher income areas its a bit better with a metal box that has little locked containers inside.
@@richardpike8748 Also it probably has something to do with the fact that Americans seem to think they need a car to travel a few hundred metres. Hell the amount of times while I have been there I have simply walked to the destination and been there waiting for them to arrive when we left at the same time is silly. Over distances that short the time taken to get into a car is a significant portion of the time needed to simply walk.
@@Ryukai-san Security might also play a role here. Here in Venezuela we have so little security that we don't trust some random getting close to our doors to leave the mail behind, instead we prefer to get from a little distance away from our front door.
What makes us human?
❌ Love, affection, care
❌ Music, painting, lyrics and other art forms
❌ Reasoning, doubt, critical thinking
✅ Selecting all traffic lights
heck yeah
The best captcha is the one that records the imperfect nature of our physicality.
Our hand forced to helplessly drag a mouse with all the meandering.
It knows that bots are the only beings capable of moving from a to b in a perfect straight line since it's not bounded by our anti perfectionist universe.
Exactly
@@T.S.rel11 straight*
@@_Turquoise_Lemon_ thanks for realizing my imperfections.
In my opinion the most elegant one is the one where you simply click the "I'm not a robot" checkbox and all the analysis/evaluation is done behind the curtains.
Elegant and insidious. Elegant because it steals your habits without you realizing and insidious because those digital habits can be used for anything from placing ads where you can't ignore them to allowing hackers to bypass gateways by scripting mouse movements instead of other code more likely to be identified as a virus.
That's good because if filling out the form normally takes a human many seconds, a bot can't log into many sites per hour and the poor bot will get frustrated.
@@bob456fk6 -- Or you just run multiple instances of the bot set to type at human speed, using randomized variables to delay mouse movements and mimic human types of behavior, even occasionally making intentional "mistakes" to appear more human. People who code bots often have access to web logs and can see how real traffic behaves and, since the whole point of making a bot is to get money, they will implement those features.
Sadly it requires Google tracking you at all times
it is elegant, but with that you have to use other kinds of captcha to avoid false negatives
Right now I'm regularly having to deal with a "choose the similar shapes" captcha. Problem is that I'm color blind, and the red-on-purple hexagons and green-on-yellow triangles are almost impossible for me to pick out. Especially given the teeny-tiny window it gives you.
According to the test, AI are better at being human than colorblind people. Sad...
Oh snap. Solution: Use a screen color picker and read the color numbers.
you can actually sue for this. legally the ADA has a clause that prevents sites from locking stuff behind something discriminatory like that.
@@polocatfan color blindness is quite common as well so they should have thought of it
Are you sure you're not a robot
By the way the jigsaw ones are typically tracking your mouse movement to see if it moves in an imprecise human way or a far too precise robot way.
*uses d-pad for mouse*
_Mousekeys exist._
Can easily be fakes with some random noise to the robot's input path
@@Supreme_Lobster Not so easy to fake if some ML classification model is used. Human inputs are imprecise with some patterns, not totally random noise.
@@xw3132 I can just record myself solving it 1000 times and use a random pattern from my list
My favorite kind of Captcha is the one where you click the verify button and then just wait for a few seconds and then the Captcha is just done with no effort whatsoever.
A lot of websites with recaptcha V3 actually do captcha but there is no button or anything. Dropbox's login page is an example.
it uses browser fingerprinting and analysing your mouse movements on the way to the checkbox, maybe checking cookies and whatnot too. if you use a privacy-first browser you'll probably never get just the checkbox and it will move on to having you solve a problem
@@TheJunky228 The thing is, a lot of people use mobile, so it literally can’t check anything in that scenario.
@username5155 as far as mouse movement. there's still fingerprinting and whatnot and I wouldn't be surprised if it checks for tap accuracy
Ah yes, the “we track your every movement” one.
fun fact: on some sus websites (pirated movies and stuff), if you're shown a captcha it might not be for the site you're on, instead it's getting you to solve it for a different site so it can bot *that* site
don't need to make the AI better if you can just get a human to do it without knowing 😌
good idea
i've heard also that some people's job is to solve captchas 24/7 as they just outsource it to real humans
Hum... Good to know
There's a facebook post hiring for part time job that you'll be just solving captchas... Currently popular in the philippines...
I’m weak. That’s actually hilarious.
I remember this old audio test that still scares me to this day, it was one where you type the characters you hear but the audio sounded like a loud intercom in a parking lot with sirens on in the background, very unsettling
apparently that's to prevent bots from listening to audio, very mean to blind people who rely on audio to use computers tho
reCaptcha arg?
was it BotDetect CAPTCHA? lol
i remember one night i was making a new google account and the capchta was typing what you heard and the sounds were demonic growls, holy shit, i was scared shitless and just went to sleep
@@TheTroll10 im lucky as i have never seen a "type what you hear" captcha
or rather, heard
"Let's stop AI by making them read and at the same time train an AI to learn how to read"
"We no longer can use that anymore as somehow AI's have learned how to read"
It's always been about making AI better, at least in the modern age.
SPORE OG
All jokes aside it is just a matter of time before people get there. It is not a race of if it is when and google wants to be the first one there so.
And once the AI could read and the captcha was useless, they said
"Lets use object recognition as a test. And feed the data to our object recognition AI."
It is intentionall
"What is love?
○ A complex set of emotions asso.....
○ Baby don't hurt me."
Had me dying lmao 💀
Wish captchas were actually that funny...
The problem with this one is that it assumes a stereotypical Hollywood AI which operates on total logic and zero cultural awareness. That's not how AI turned out to be in reality at all. An actual AI would easily choose the second option because it has seen that specific combination of text far more frequently in its training.
10:30
I would have failed that and I'm pretty sure I'm human ?
It may however explain why I'm still single...
I admire Simi’s ability to take a seemingly boring topic and make me invested for 17+ minutes
Wait what!? I just watched the video and didn't even know it was that long I thought it was only a few minutes.
17 minutes!?!?!?! It felt much shorter
ye
and then i be wanting a part two
He's a black, British AustinMcConnell.
You might be interested in Chinese CAPCHA too, they have evolved to rely on different uses of the Chinese language. For example:
1. As Chinese characters are basically square-y pictures, like 架, 奇, 雯 etc., they ask you to rotate a random character upright in the correct way. People who knows Chinese can do that effortlessly.
2. Different Chinese characters have the same or similar pronunciation, e.g. 忍, 人, 認, 刃, 仁, 韌, 飪... are all "ren" in mandarin, spoken in different tones, and all commonly used. So there is a CAPCHA for you to pick "which word sounds the closest to the given word". And also, to match the word to the given pinyin (i.e. "ren") and vice versa.
3. Three random, uncommon Chinese characters are given, and the distorted version on these three characters are hidden in a colourful picture. You are asked to click on each, according to the order they are listed in the question.
In order to get a Japanese google account, I had to answer a captcha with distorted Japanese text. Certain characters in Japanese are difficult to distinguish even when not distorted. Guess which characters they used!
There was like more than 1000 characters right?
The one’s I’ve encountered ask you to click on four characters part of an idiom (成语) in the order they appear in that idiom. They’re usually rotated and on some kind of background, as per usual captcha tests. Sometimes I have to pull out my dictionary if I don’t recognize the idiom 😂
@@Fixer_Su3ana the the most famous swirly character
@@FenceAKAGlasnost Yottsudomoe? :J
Just gonna drop my favorite quote which I always get reminded of when CAPTCHAs come up:
"There is a considerable overlap between the intelligence of the smartest bears and the dumbest tourists." -Yosemite Park Ranger on why it's hard to design a bear-proof garbage can.
I’ve heard that one lol
Love it.
At a certain point it's probably best for a person to get assistance for captchas, especially if it's a kid.
@@terig8974 they do that too if your a bot hoster you can buy a number of captcha solves from call centers usually in india where real people will just solve it for the bots
i can sign up to solve captchas at a payscale of 1 dollar every 1000 solves aparently there economy makes this worth it sadly
1 dollars actually good pay too ive seen .40 per 1000 too
Some bears are smarter than the average bear.
It is so annoying that sometimes it doesn't even matter if you get the thing right, because it will just keep giving you captchas infinitely
That Roblox captcha one with the rotating animals... I'm so glad you mentioned that cos, that was an absolute nightmare
I remember a few years ago I was sobbing I couldn't get into my account because of that dumbazz captcha, it's a literal platform kids go to, why make it so hard even kids can't solve it?
Github had those back in 2021 and I legit could not solve it (Thankfully there was a request for a different one or otherwise I would not had got my account on there for Bug Reporting)
Gosh that brings me back to all the times I would get locked out of my Roblox account, because it would give me a 20 max level difficulty captchas, then every failure would make me restart all 20.
How about rotating spheres?
That’s still better then that Roblox captia with the darts
These captchas are almost like those free flash games we all used to play. I swear, if it keeps evolving like this, in 20 years we'll have to play a cs:go match to complete a single captcha.
"You are playing on CT side, shoot your opponents". And then your game graininess is 1000%, so no robot can see who is who.
Imagine if you had to solve one every single round v:
csgo doesn't prove anything, we will have to complete dark souls instead.
@@Rilex037my boy playing dark soul reorganise every pattern to defeat every single boss and final boss and only to play flappy bird but boy it was worth it and you know i am human if i wrote a wrong letter can't wait to play flappy bird and get to play dark soul for free right?
Its fun until you realise the bots are better then you
Half the time the robot I am trying to convince I am a human doesn't even know the answers itself.
I had one that said "Identify the mailboxes" and it counted it wrong because it thought a brown trashcan on the side of the road was a mailbox.
Well that’s fucking hilarious
it can be a mailbox for some people XD
You were technically correct, the thrashcan is a mailbox I use everyday... I file the spam / ads / junk mail I get there.
@@MouseGoat More like a eating whole
Same thing happened with cars. Do I have to be dumb to be believably human?😂
Back in the days, I used a very simple mechanism to block contact form spam on my website:
I put a checkbox under the form, with a little text instruction over it that said "Don't check the checkbox below". Then I used CSS to hide this checkbox as well as the text. Normally a user with CSS styles on wouldn't see the checkbox and its description, so he would leave it unchecked. Similarly, a human user with CSS turned off would see the ckeckbox along with its description, and followed the advice to not check it, so it would remain unchecked as well. But a bot would most likely just see the raw HTML encoding the form and try filling up all its fields with some random garbage or spam. On the back end, I checked if the checkbox field contained anything, and if it did, the message didn't pass through. Only the unchecked messages did. Simple and efficient.
@@bonbonpony genius man
My favorite captcha was one that asked the user to click the animal that's walking backwards. The damn thing was so hard, I gave up trying to log into an account on an MMO's website. Many of my friends couldn't log in either. The problem was that the animals were walking in place on a white background - there's no sort of environment that's moving throughout the image to indicate which way theyre walking, so you have to think about the way an animals legs move. But that's hard to get correct four or five times in a row, which the captcha asked you to do
When I saw that in the video I straight up laughed at the stupidity
if it was a gif it would be possible and work better against bots
Have you had the "how long are the pants" captchas? After a few of those, I just wished that the internet would cease to be. It's bad enough already that apparently, most people think that anything(or part of anything) with 2 wheels is called a bicycle.
i remember the old recaptcha. it was such an asshole move to have us teach it, because it was clearly going to lead to captchas being more annoying in the future
it was always fairly obvious which was the one we were teaching, so my method of coping was to always say the word was "butts"
When someone pointed out the second word didn't matter, I remember doing the same thing lmao
I love it
Also just noticed, ...interesting username you've got there lol
At this point, we've gone from "how do we fool an AI that can't recognize text if it's a picture" to "how long before AIs learn how to recognize, behave in response to, and interpret abstract concepts better than humans?"
Answer: Not long.
I think it was never about "How can we fool it" and always about "How long does it take until it's better than us".
That's more often than not the reason for any program: Being faster than handish human work.
AI is a lot smarter and faster than humans
except the captchas are designed to train bots to be better - why do you think they're all "select all of a frog" or "horse made of clouds" or something - you're training ai to generate images, that other ais will recognise, it's kinda wild.
which is why at this point captcha just needs to stop lol
Everybody gangsta until you see your own number on the address CAPTCHA
Captchas 10 years from now:
- Do a photorealistic drawing of a dog to prove you're a human
- Win a TF2 match to prove you're a human
- Write a song about proposing to a supercomputer to prove you're a human
- Bake lemon macarons and mail them to the company to prove you're human
- Translate the Voynich Manuscript to prove you're human
Current AI can already do those.
Current ai can do the first, second, third and fifth. The winning of a tf2 match and translation ones might be a bit tricky however.
Being a TF2 player, the bots already win matches
@@shdy9498 lol it sums thi
for the tf2 match you just yk? (the spinbotting bots)
I had the touch the mouse one once. Had 10s time for each of the 10 levels loading time included and they got progressively harder. When I finally did it it said I was too slow and had to try again. Gave up after 1 hour
I was the mouse who couldn’t reach the cheese all along.
That kinda seems to be designed to specifically allow only the captcha farms (where human workers solve captchas for money) to solve them.
So if you were the mouse... did you go and touch yourself?
Yea, got it multiple times in a row though and kept running out of time, literally couldn’t log in for a full day
_Ah yes, let’s block the user for taking too long to solve our ridiculous nonsense CAPTCHA, because bots are well know for their lengthy deliberation on how to solve a problem._
You should've touched yourself.
I once got a CAPTCHA that drove me insane for almost an HOUR. It told me a certain number and showed me several images of dices (a few dices per image). I then had to calculate the sum of each image judging from the dots on the dices. But that alone wasn't annoying enough. It was TIMED. I was given around a minute to solve 9 fucking summations in dice-form. I either ran out of time before I could figure out which image was the right one, or made a mistake calculating due to the time pressure. And that wasn't even the worst. I had to go through that process more than once, and if I messed up at any point it would mean an instant restart and loss of all progress. It was truly infuriating, especially for someone like me who is a slow calculator.
Good, good, the humans are finally training :>
...and roblox thought it was a good idea to have that captcha.
it's blizzard isn't it
slow calculator? its just stupid lol, i had to do one that was "click on the motorcycles" and i sat there screaming at a bunch of random people on motorcycles because i couldnt figure out if half a wheel in a different square was a part of the motorcycle. even the "which ones with bus" was dumb, i clicked all the buses, ALL OF THEM, and i guess it said that four crosswalks 2 cars and a fire hydrant were a BUS
@@Sarantis-107 I liked that one. It taught me that English speakers generally do not consider buses and trucks to be cars. Everything is a learning opportunity if you keep your mind open.
A couple of weeks ago, Twitter presented a "pick the shadow that matched the icons at the top of the image" captcha. My twitter account was in a head lock for a whole 2 days because the captcha was impossible for me to complete. I tried to get my dad to complete the captcha as well, and he couldn't do it either. The only thing that saved me was when the captcha changed to something else.
The ones that require you to pick the photos with certain images in them are actually meant for you to get wrong sometimes. The reason they sometimes have a square with a tiny piece of the traffic light on it or something is because the captcha knows that a human could miss those squares but that a robot would complete it perfectly without delay. This is why sometimes a captcha will let you into a website even if you don’t get it fully correct.
SO it's punishing me for doing my best? I hate them even more.
yeah and then it starts to convince me that motorcycles are bikes because it doesn't let me in if i don't select the motorcycle...
@@randomclips9715 in some culture people call both motorcycles and bicycles as bikes. And well, I agree with you that captcha is unfair.
@@stm7810 actually yes it does.
I've done so many recaptchas that I rarely read the the text anymore, I instantly start picking what seems "intuitive" to pick.
Like if I see cars, then a picture of a bike I know that it's looking for bikes, if I see boats I know they want boats, see landscape I know they want hill side... some times I actually miss click images and still it accepts it, I'd say that 98% of the time I don't have to do another captcha even though I didn't read the text..... witch kinda makes me wonder, isn't the point of recaptcha to catch "bots" that don't read and just know the solution? also am I a bot?
The new one with the animals is always throwing me off, but given enough time I'll probably get the intuitive feel of those challenges.... I guess that they could really just be testing intuitivety, as bots don't have intuitivety they just "know"
@@svampebob007 The problem might come from the fact that there's companies out there that would pay money in order to be able to ...I don't know... have bots advertising for adult sites write comments and chat messages (definitely an example that's completely unrelated to youtube). Which means that as long as human labor can be cheap enough, it becomes viable for other companies in low income countries to offer to have humans solve those captchas. And those humans will inevitably solve thousands of captchas per day; which means they stop bothering to read the prompt, and they develop their own subculture of what they consider a valid answer. And since each of those workers and offices solves so many captchas, they have much more influence on how future captchas are trained than any normal user.
The beauty of mini game CAPTCHAs is that it takes almost no time to develop a new variant of “add up to 12” but it takes days to develop (program/train) a bot-based solver for the new kind of CAPTCHA. Also, I noticed CAPTCHAs today are super time-consuming even if I can solve it and it seems it’s slowing me down intentionally to throttle traffic or limit number of accounts created per second.
The time you’re talking about is literally 20 seconds max
@@melody3741 Usually, but once I encountered the sum dices to 14 captcha. You don't get just one sich image but like five of them. It is for some reason timed! I am pretty good at calculating but I am just not fast enough for that captcha. I got two of my friends to parallelize the captcha solvong. Each calculates one column and shouts if he sees the 14 sum. It still took us three attempts to get it done within the time limit. I think it took us at least 10 min to get past this captcha...
Those dice captcha are just to hard for some people, who let they test them out? Mathematicians and math teachers?
CAPTCHAs wouldn't throttle traffic - It only delays it.
Actually of all the captchas mentioned, the one where you add up the numbers doesn't seem that hard. I've never done it but if the goal is to just pick a square that has a wheel of numbers, add them all up, see if that matches the correct value... that wouldn't be too hard to program in. You can just use OCR to extract every number from each wheel, and then run a sum total formula of all those numbers to see if it matches. Keep iterating through each wheel and viola. Not really all that hard.
That guy making captcha tutorials for kids is so wholesome
Except that it's for getting on Roblox, where it's known that child predators lurk.
Maybe kids (who are mostly illiterate idiots) should learn to read and write before playing a game that is rated T by ESRB.
@@angelbear_og i mean, you just described the iternet. And the world.
@@boggless2771 idk id honestly feel safer letting my child wander the park alone than roblox
@@boggless2771 Yeah, but it's not about where they can possibly lurk, but where they can lurk not having to care of any consequences because the company doesn't give a flying phück about it, or silently supports it :q
i know how we can end the captcha war once and for all: "select normal looking hands"
lol
yes
11:08 I solved quite a few of those math CAPTCHAs for fun (reloading the page), they look scarier than they are if you know basic algebra and calculus. And the answer to many of them is 0, including the example shown in the video!
These are practically useless nowadays tho, since OCR and symbolic maths are both a thing...
I was not sure about my approximative calculation, but I was suspecting it, thanks for the confirmation :)
w h a t
@@einthegreat-man what
Title: Worst CAPTCHA ever
Video: The complete illustrated history of CAPTCHAs
Bro, you anti-clickbaited me. I thought this was going to be some silly, vapid, time wasting horsesh*t I'm gonna watch, and you took me an entire emotional journey to Mordor and back. You had me questioning my humanity. You had me questioning Google's humanity (okay, that one's easy, but you still did it, so it counts). You are definitely getting subbed.
Note for UA-camrs: please do anti-clickbait more often
Oh look, my own thoughts already in a comment, how convenient
This comment
Yes just wanted some more Waldo captcha. I feel like I would enjoy that
Yeah KSI credited him for animating one of his stories and I’m so glad I checked the channel out cause this guy is top notch
A really decent one was on a Japanese site where you not only had to slide a slider to line the hiragana symbols up but also answer what it told you to … in kanji
this was used to teach a ai language model to translate between the two writing styles
@@ochreyefroglight tho can be easy to type since it doesnt automatically put it in katakana and kanji if thats what im thinking
Doesnt 4chan use something similar?
@what the 4chan currently uses a slider with text based captcha. how it works is imagine those distorted text with lines captchas, and you cut some ovals into it. You’re supposed to line it up close enough to read the answer
@@ochreyefroglight ye no
Got the feeling that would be harder and harder for both human and computer to do it for Chinese and Japanese words are so blended together it's just hard to even tell which word is native and which is foreign in both languages nowadays, and that's not accounting everything that shows up before the 21st century that are by default considered all native (well, semi-native I mean, just like how people usually won't care that ketchup is a Chinese word).
I don't know why, but when it talked about how the 'select traffic lights' and those sorts of CAPTCHAs were from actual google maps images, my brain just exploded a little.
I had no clue that the original text based CAPTCHAs were even used to teach AI how to read, and so with that knowledge, the second that it talked about traffic lights, I instantly realized what that actually meant. That's real cool, and extremely horrifying that I didn't even know what that was being used for.
It's interesting how as soon as self-driving cars were being more marketed and advertised as coming into existence suddenly captchas featured taxis and motorcycles, types of vans and cars, and especially traffic lights. so many traffic lights
Why are you saying that like it’s a unproven conspiracy lol. It’s well known they’re using them to train AI’s some of which drive cars. This video specifically brings that up like a few times at least 😂 Although I think it was more so for google street view than actual driving per se
Actually captchas like that is being used to train AI on image recognition and we are simply being used as free labour to do that....
@@a64738 it doesn't seem like a bad thing
@@t1Pz ehhhh, I mean AI recognition is generally pretty beneficial, but it *is* kind of crazy how so many people have unknowingly helped Google to do stuff like creating a digital globe where everyone knows what your house looks like if they find your address... I dunno, I'd just be careful to think that us being used to improve AI is always a good thing.
@@t1Pz I like to be paid for the work I do, even if it's a repetitious and brainless work. Let's just leave it at that.
And especially when it comes to Tech Giants like Google, Meta and co.
I’ll be honest, finding Wally in a captcha is genius because it relies on pattern recognition rather than arithmetic and because it should be a bit of fun.
The person who can’t see
*mashes reset captcha aggressively*
Unfortunately, there are AI's that have been trained to find Wally/Waldo. So I guess that one doesn't work anymore
@@tabithal2977 Once you find him he will come back, to find YOU
@@superNova5837 how the heck are they immediately gonna know and find the reset button-?
@@mikhail-cybertastic5349 Visually-impaired (fully or just partially blind) usually use screen-readers to travel the internet. It reads whatever is on the screen or what they are hovering over. How else do you think they made it to the website the captcha is on in the first place?
as a visually impaired person, I absolutely despise captchas. Due to my poor eyesight I can just never figure them out and spend way too long solving them
You can use the accessibility feature
@@jamesbunce1198 or noptcha
@@jamesbunce1198 unfortunately it's not always there :/
hCAPTCHA allows you to get an accessibility cookie that bypasses the CAPTCHAs. reCAPTCHA has an audio captcha button that 60% of the time refuses you and tells you to screw off because it takes clicking that as a red flag that you're a bot (???)
@@alex15095 blind people:
new idea for a captcha: make a humanly impossible test, if u put the correct awnser, u r a robot
Isn't that just the cup one?
@@Alt.N I'm a little stupid and just randomly guessed correctly, so we need a Captcha that is stronger than my gambling superpower
15:25 the most obnoxious captcha I've encountered was trying to sign up for a throwaway email on outlook recently and it hit me with a captcha like this except with a series of incredibly abstract and highly distorted image matches requiring you match 7 to 12 different images and then it wouldn't even explicitly state if you failed it would just load a new sequence of images to match or rotate. I assume that means I failed but I am not sure as it never told me. It started to feel like the site was just farming captcha input and never actually going to give me an email address. Ultimately I left and signed up with a different provider lol.
You are genuinely dumb then, there’s no such thing as “farming captcha” like wtf
You probabaly were just farming captchas for someone's bot army
@@kaengurus.sind.genossen *disintegrating emoji meme*
My question with the images has always been "How do they know I got it wrong if I am the one training their AI?" (I purposefully click wrong squares some times just to mess with the training)
Unfortunately, 100 other people were dumb enough to reply correctly, and that's how the machine knew.
@@bonbonpony Yeah, they wanted to get on with what they were doing instead of screwing around with captchas. The very definition of dumb.
(based)
@@Squant how can they, in good conscience, continue with their everyday work and hobbies, when they're basically training ai to read distorted words in scanned books and improve self-driving cars... the absolute NADIR of intelligence
OMG the math ones. Microsoft used those for a bit too. I was trying to help an older gentleman get back into his email account (I run an IT shop) and I COULD NOT answer them fast enough! I'm 37, not 5, and I took freaking calculus! The problem wasn't the math, it was the speed with which they thought you should be able to add all the numbers together in each box to pick the right one!
What's even worse nowadays is that you can cheaply buy a manual captcha solving service. So you have your bot interact with a website and when you're confronted with a captcha there's an actual person somewhere that solves it for you and you get back at botting...
The one with Roblox kind of reminds me of the copy protection on old PC games where the obscurity was the point, you had to prove you had access to the manual in order to get in because the manual was hard to copy as opposed to the software itself which could be copied easily. Carmen sandiego had you looking up specific words in specific sentences on specific pages of the big fat reference book you were supposed to use to solve the in-game mysteries.
Maniac Mansion had a code locked door at the start of the game and if you failed like 3 times, the mansion exploded.
Yeah. I remember Return to Zork let you play up until the school house, where suddenly the teacher will give you a pop quiz of random questions (answers were in the manual.) Failure to answer correctly meant death.
I remember Tie Fighter had that security feature as well. I think Sid Meier's pirates also had something similar to find which ports had the treasure laden ships docked so you could raid the big bucks... Ahh that was so long ago.
Jesus christ, I get flashbacks to the worst period of video gaming I've ever experienced.
I could kill somebody to this very day for fcking up Pizza Connection. Neither the Pizza templates, nor many phone number actually worked.
Makes it also incredibly frustrating when you want to play... like at all.
Some memorable ones I had were Eric the Unready where you had to tell your armor sizes to the armor craftsman so that he can make the armor that fits you. Was pretty cool, but extremely easy to copy, since you just had to copy one page.
Another was really annoying (TMNT) where there was a dark dark purple sheet with many hundreds of very tiny numbers (4 digits IRC) printed in black, and you had to say the number on row y column x, but it was so hard to read due to the small size and dark-on-dark colors. I was young so it wasn't too terrible, but I remember always breaking out the super strong desk light and looking closely.
I have a screenshot of one of these weird captchas that says, "Pick the smiling dogs." Underneath it is a bunch of poorly photoshopped dogs that look horrifyingly like skinwalkers.
I can't believe it never occurred to me we were training AI every time we logged into something. If you're not paying for the product, you are the product, and all that.
don't worry, even if you are paying for the product, your data is still a secondary product. marketers would never give up that info just because you paid them.
It was a fantastic idea. Kinda like those apps that use your extra computer/phone RAM when you're not using them to add to a cloud based supercomputer.
Back in the day (at 3:11) I do believe they advertised that detail publicly that the service helped to authorize you + translate scans of old books and articles. And like in the example in the video, you could often determine which word it didn't know, since it would be the less wavy/warped one, and if it was too hard to read you could get past the captcha by just writing anything you want
@@jeremy5602 more like a botnet
Well duh ...they're using it to train drones too
i once got a RIDICULOUSLY hard captcha, you had to make like 20 math problems in one minute, i remember being at my friend's house that time and none of us could do it
Are you sure you weren't supposed to fail, thus proving your humanity?
Good, good, the humans are finally using their brains, our plan worked… :>
@@bonbonpony I didn't know you were still around, i saw you on youtube ages ago
One thing that I always found annoying about those early captchas is that, as a dyslexic person, I kinda struggled with them too, the occasional variation in fonts used and the distortions that got added later on made it so much harder for my brain to process what was in front of me when it already struggled with that enough on regular text. Sure it would normally get it eventually but it would take me quite a while, and sometimes it would take me a heap of captchas to get through that. As weird as some of them are, I'm glad they've moved on to imagine captchas, they're much more accessible and they don't make me feel terrible about not being able to read what's in front of me
Captcha is ableist by nature.
@@trytoo5167 I'd realized there were accessibility issues, but you make a very good point. They're literally testing typical human abilities, so it's unavoidably non-accessible. Interestingly, the only way to do it well might be to have humans approve submissions, rather than computers.
@@trytoo5167 Kinda in the same way that street curbs are ableist to those in wheelchairs
I don’t have dyslexia but I still struggle with most text based captchas. They’re just hard in general, especially when it’s a guessing game of is it capital or lowercase, is the captcha even case sensitive idk fuck em lol
@@jeremy5602 are you trying to be sarcastic? Cause they are
The worst captchas were the ones where you had to determine the letters inside some squiggles, but no matter what, it would always say you are wrong.
Ah yes, the "ex-spouse" captcha.
The AI getting so good at CAPTCHAs that humans couldn't do them gives me an idea: a CAPTCHA where in order to pass, you actually have to get the question wrong
It's just as easy for a computer to fail something as to succeed at something.
Though your idea is still somewhat achieveable, but in a different way. Basically there are methods that targets specific machine learning algorithms that would encode an image in a way that would be invisible to human but would cause the algorithm to output a wrong answer with fairly high confidence score. Maybe the website could timeout any input that's wrong in a specific way because only AI that got hit by such attack would give that specific answer.
@@FlameRat_YehLon Fair enough, I like the idea
Just program the bot to answer at random, and take its time while doing it 💁🏻♀
@@FlameRat_YehLon you could tack on erattic mouse tracking for good messure
Just design it to fail. That’s even easier to make.
Captcha questions are slowly becoming Impossible Quiz questions. I wouldn't be surprised is Splapp was the one making them
2 things.
1: I have recently seen images with Dalle mini/2 generated images in captchas.
2: Roblox is the worst website ever because of the captchas alone but then you see the awful way they work and it's worse.
they removed captcha thankfully.
some of the "move the x" capchas actually measure mouse movements. AI's movements are not as erratic and inefficient at human mouse movement. but if you try it on a touch screen... it sometimes thinks you are an ai
I was thinking this what if you quickly swipe the slider to where it should be
Yup. A programmed bot would move the cursor in a straight line to the target. You do pretty much the same thing with a touch screen and fail the captcha.
@@Unknown_Genius Sure I'm aware. The OP observed that "if you try it on a touch screen... it sometimes thinks you are an ai" vs. "erratic ... mouse movement", which is some of the information that a captcha looks for to decide that a human is present.
By your reply, I gather that you're familiar with HIDs. In response to a) I think you'd agree that compared to a mouse, a touchscreen output is lower in resolution. Also, depending on the device, movement is often processed and filtered to make it more usable than it otherwise would be. So it doesn't surprise me that some movements of the x on a touchscreen could fall within a captcha's threshold for suspecting a bot.
In response to b) We agree. I should have said that a basic script would move the cursor in a straight line to the target.
I definitely get more captchas on my phone.
This reminds me of the terrifying captcha of "choose the animal with the wrong head" captcha, it would have dog heads on squids and octopus heads on birds, it was scary af.
I had one which was "Choose the dried up and dead potted plants" which all looked like diseased spiders
Perfect heart attack fuel at 10 pm
@@ibavider that would have killed me bruh I hate spiders so much
@@grilledcheese. me too
recaptcha mostly detects your mouse movements to make sure you aren't a robot and make you click a bunch of things to make sure
13:40 I've heard another thing it monitors is the way the mouse moved towards the checkbox. If it was a perfectly straight line it would be suspicious because an actual human would likely waiver a bit while moving the mouse.
yep, it also detects commonly used algorithms that try to circumvent this, like bezier curves
then again theres a video of a literal robot moving a stylus across a touchpad only across the vertical and horizontal axes (so only straight lines) to solve a captcha and being let in instantly so idk about that
@@coolguyman7356 One of the features it detects are likely the noise and slight jittering caused by the hardware. And possibly the fact that if you have hardware moving the cursor, it has to obey the laws of physics and known materials (i.e. no infinite acceleration).
@@coolguyman7356 it lets you pass a few times so it can fingerprint your bot and block you more effectively in the future. it also blocks on many more factors than just mouse data, and if the other factors are valid and you're not a suspected bot, you'll likely pass the captcha.
@@Pystro true but i was thinking you could just simulate normal acceleration with a bot so why bother with that anyway unless youre tryna be thorough in catching those bots
8:28 that option two was amazing
Blub blub blub blub
@@CubeRob558 You mean:
blub blub blub bleh, bluh bluh bleh bluh
It's AI training. The captcha it's self is for training AI algorithms. It was never about preventing bots from using sites, it was about getting people to unwittingly do the hard work of training AI image recognition models.
Could be, but its to late now.
Straight up just google using us as unpaid workforce :')
Basically... Fuck AI!
Listen I hate google as much as the next person (and a lot more) but I have to disagree here. Yes I don't doubt that google is using users to train ai, but the main point is STILL to stop bots. (looking at some comment sections however, you can see just how effective or not it is.)
@@cact0s_ulion405 I mean main goal or not, fact remains if what they are actually doing helps bots better than it stops them, then they're effectively helping the bots. If we're the data they use for that, then we're the unpaid employees doing it. If you want to stop AI bots from getting in either switch it up so fast they can't get any significant data, or figure something out that AI actually cannot do. Problem is, this would cost a lot of time and money and the way they do it currently might suck but it gives them back a lot of time and money in the long run, so why would they change it?
4:37 So the computers were/are using OUR brains for their answers? It's like the original draft of The Matrix!
I remember the old captchas. It's always been pretty obvious which ones were transcription data and which ones were the actual humanity test, so I'd always transcribe it as something horribly incorrect. If Google wants me to work for them, they can pay me. Otherwise, they have no reason to be surprised that I'd make their transcriptions more colourful than they intended.
Google will honestily pay you as part of their survey programs to do stuff like this just like people are payed to test games. Not much but for a brief weekend payment it is something
Unfortunately, 100 other people were dumb enough to answer correctly :q
My least favourites are the one's that don't take into account language/country differences. Most of them seem American based, I'm not from there so sometimes I don't know what the thing is or what counts as one. I get through them in the end but sometimes they take me a while when I'm trying to decide whether the bit of road with pavement on each side but no crossing lines counts as a crossing
(Not)fun fact: They used such captcha on one Polish government site :q I bet lots of Polish citizens were unable to answer the English-language captcha, or even get what they're being asked to do :q 1:0 for the government though - they didn't have to deal with interesants for quite a while…
(for the crossing/crosswalk thing)No it doesn’t, but that doesn’t stop people
man im from america and i cant even do this. picture this, 3 buses. you click on the buses, and there are no more buses. YOU ARE WRONG. THE FIRE HYDRANT IS A BUS AND THE CROSSWALK IS A friggin bipod.
Select all images with footballs
What we should do is make one that has a GIF in it. It'll screw with a lot of image recognition AIs, since not many can trace motion.
Edit; Ok so thanks Thomasmills8613 for pointing out that this will just teach image recognition AI to trace motion. Only problem is, how do we deal with robots without teaching them something new?
Oh great, so you want us to teach ai to track motion too?
@@Thomas_Mills Honestly thanks for pointing that out
@@Thomas_Mills So I guess the dilemma here is dealing with robots without teaching them something new.
@@officersoulknight6321 Honestly I wrote that while half asleep and meant it to be more humorous then at came off lol. Is still a conundrum though
Bitcoin could be applied to solve this problem. Each person could have a layer 2 wallet into which they load some small amount of BTC. Charge a few satoshis to perform actions on each website where limiting bots is a priority. Real users can afford to spend fractions of a cent here and there as they browse normally, but operating a bot army at scale would no longer be free and would require an actual financial investment. There isn't anything new that bots could learn to defeat this (well, except maybe how to earn money..)
Edit: Just to clarify why I mentioned Bitcoin (as opposed to a traditional payment system) is mainly because it is permissionless, meaning anyone in the world can easily opt in without needing to ask for permission by some third party. A traditional payment system would potentially exclude a lot of locations in the world from participating, and it might compromise the ability to browse anonymously.
fun fact, mostly for the image ones;
my social worker was recently working with some people who program these things. apparently the way they ACTUALLY tell apart humans and robots, is how fast you answer the questions.
an ai will click the answers straight away, or fill in the solution too quick; a human will take more time, the curser (if they’re using one) will move around a bit before inputting the answer. the ai uses that to tell if it’s a bot or not and uses that to kick out bots
my gripe with capthcas so far, is when they are the "click all images till none remain with X" because the moment they get laggy, you fail 100% of the time even if its correct. Even the "Choose all traffic lights" has issues where it reads things improperly. my favorite, is a mailbox was a traffic light, and it got mad saying i didnt select all traffic lights. Yes, i waited till the mailbox came back later and tested, selecting it worked..
We gaslighted the google bot
I remember seeing a captcha that make you identify which are the painting of a dog.. but the painting were all hyper realistic... My mind got blown..
wow
I actually noticed a pattern with google recaptcha, it always seems to have you select a certain amount of them, like 3, 6, 4, so chances are, you can cheat it by doing that too.
they are also almost always in a specific rotated pattern.
@@comet.x huh I've never noticed it, what I noticed is that I don't have to read the text anymore to know what they want me to pick, I've never noticed it was in a pattern.
It's gotten worse. Surely we're gonna get some sort of reboot for the captchas.
10:44
Simi: “if so, you got it correct… which actually means you failed the captcha, as only a robot could possess that level of perception”
Me who guessed it right: start searching for Sarah Connor
lol
i assume it's a joke...
then again, maybe my professional Wii Party ball and cup gameplay has finally payed off
No "guess" is needed. It is clear to see.
I also believe the recapture "I'm not a robot" one also tracks mouse movement through the page, a bot would tab or not even tab through the input fields as it fills out, where as a human would have their mouse move randomly about the page. There are multiple layers at work with all of them, somewhat fascinating.
I thought normal people tab though forms... 🤔
That actually makes me a bit mad. I keep getting flagged because captchas think I'm a bot. I get banned from all sites that have this verification.. latest one was steam and I bought games and i just wanted to play one..
(3:35) It's not a great system though since I've seen Google books where they've used these reCaptcha answers as the scan-to-text, but the text is wrong. Usually when it's non-English text missing accented symbols which are important for non-English text. Plus sometimes most people might make the same error and only a few actually gets the correct answer. Since these words are shown out of context, and it's hard to tell what it is, but easy to tell when part of a full sentence. Basic example: З, this might look like the digit 3, but it is actually the Cyrillic Z. You can't tell out of context, but in the context of "Zlatoust (Златоуст) is a Russian city", you can at least tell it isn't a digit.
Google hasn't focused on non-English data mining as much, but it's ramping up. Pretty soon you will have trained an AI to defeat your language skills too, and do it at no cost to Google.
Cool
I can never understand the ones with text nowadays. They have become SO distorted that I can never understand if one character is another.
This guy's attention techniques for us to stay focused on his video are phenomenal
captchas in 2050: select all the nuclear bombs in the clouds dropping on nevada
Ah yes new Vegas irl
Haha
somewhere in nevada
every single one
@@danka1167 You know they bombed Nevada for tests, right?
The one thing I love about modern captcha is that they always make me think of the lyric "I'm not a robot and I'm not a monkey, I will not dance even if the beat's funky"
For me it's "Guess what, I'm not a robot" by Marina
Also "We are the robots" by Kraftwerk but I sing it "I'm not a robot"
I mainly hate it when it has those where you have to select the parts that are the thing like a bicycle or stop light. Sometimes there is a pole or the smallest corner you aren’t sure will be counted as a space.
I would gladly contribute if someone starts an internet-wide trend of entering as much adversarial answers as allowed into the recaptchas to atleast slow down this nonsense. My friend keeps sending me screenshots of ridiculous acid trip captchas he keeps getting.
Let's make a bot that solves the first captcha right and the second one wrong lol
@@lombas3185 iam in!
@@lombas3185 My man we figured that out years ago, certain site used that shit so you can imagine what n word did they put on.
They caught on to that very quickly and started to randomize which word is known and which is unknown.
@@Oneiroclast Considering the word captchas are obsolete anyway, I was talking about image ones. Pretty sure there's an amount of error allowed there, especially with controversial squares. I wonder if it's possible to figure out how much you can mess it up and still pass, and do that every time.
As I've worked in 2captcha for more than 3 years, I've actually faced every single type of captcha and recaptcha you mentioned in the video. Trust me besides all the types you mentioned, there's actually a lot more types of captcha out there that'll either amaze you or drive you nuts 😁😁
I wanna see those other types
Same
When reCaptcha started out, it was absolutely sufficient just to type in the first word correctly and the first letter of the second word, after a year or so, you had to write atleast half of the second word when but it was still possible to get through. With the amount of captchas I had to enter on Megaupload it was well worth knowing this
We should have used the bad captcha with text generated by IA the other way: if you can't reconise the word you are human and if it can it's a robot
I had a captcha yesterday, where I had to pick every horse that had white legs. They definitely did something.
this is such a fascinating topic that really does reflect in a noticeable way how ai has gotten far better at recognizing images over the years, giving insight into projects that major tech companies here are working on, and what ai is still yet unable to do
also great lesson on how to outsource answers for your machine learning, by making everyone else do it in order to log into their email
9:06 that Tesla has been summoned, and it isn’t letting a plane stop it.
1:52 "only certain people can read this" is technically true, minding the person's knowledge on how to read and their linguistic boundaries
i personally really like Genshin Impact/Hoyoverse's take on the puzzle captcha. the image used for it is usually official artwork or a screenshot from the game. Pretty good way to incorporate your brand into it.
Yea that's so easy to verify too
AI can learn that easily
I looked at this and I thought it was going to be a compilation of silly CAPTCHAs. What I got was much better. Thank you, I have learned something new today!
10:58 that ain’t no equation though. It’s the slope of the tangent curve on a point in that function. The answer should be 12cos(3x-pi/2) - 24sin(6x) for any given point. We are looking for the value when x=0, so the final answer is 12cos(-pi/2) - sin(0) which is 0
Yesssssss I was scrolling through the comments to find anyone else that did it, I also got 0 in my head
Yeah not an equation but at least he noticed it involves trigonometric functions
Well.. technically it is an equation. Just the right side isn't shown, and it equals y which you need to solve for.
@@TheStrandedAlliance it is not though. It’s a mathematical expression. You don’t solve for anything you just evaluate the derivative.
I like your funny words, magic man
uh oh I just found the robot, Google
Those ones where it’s a single image broken into multiple squares is so annoying since you can’t be sure what it’s counting at all like if for the traffic light if it’s counting the pole itself or just the light itself since I’ve done ones where the light is in two squares and it counted it as wrong.
Also bad is the multiple small pictures and it might count something you can’t tell as the item you’re looking for.
Worst part of that 10 captcha's thing is that it begins by telling you it's only 3 or 5 and then ramps up to 10 and if there's ANY hiccup in your connection it may tell you to just try again even though you just completed them.
Epic games launcher had those for a while and I had to do them for both logging in and claiming my weekly free game.
I remember when I got a captcha similar to the frog one. The only problem was that the website auto-translated what I have to choose, so there were 2 possible correct answers.
Google translate strikes again
10:14 “robots cannot love” Bro have you seen WALL-E?
Lol
bro thinking about a litteral animation💀💀💀💀💀💀💀
well yeah but when wall e had his mind erased he didnt show any emotion implying the other wall-e robots have no emotions
@@StrandedOnDunaI seem to have forgotten that part of the film. Or at least 2 years ago I had
I was watching this video waiting for Roblox to be mentioned. I was confused with every minute that passed with no mention of Roblox.
I’m so glad you brought it up. When I had to submit a support ticket, I had to answer 20, yes TWENTY, of those die questions. And because I was trying to get through them quick, I ended up messing it up 3 times so I switched to the auditory version so I could be quicker. All-in-all, it took me over 30 minutes to finally complete the captcha just for their support team to be of no help. Apparently they’re very busy with the 3 other people that could be bothered spending 30 minutes trying to solve the captcha.
The Captcha's are being used to train AI.
At first we were entering words and wonky words.
Then Google released "Screen Read" and "Translate".
Then we were locating Traffic Lights in pictures.
And the data was then used in self-driving vehicles.
Now we are completing images of fruit etc.
Because AR tech is gonna be the next big thing.
The worst Turing test I've personally experienced was the one EA made me take when I was trying to create an Origin account to play It Takes Two. They made me count the values of a bunch of dice shown on various pictures, I have to do maths with them in order to choose the correct one. For every single question there were like 6 thumbnail sized squares of pictures and each picture contains like 6 dice, they weren't even positioned that straightforwardly and were stacked sometimes. You have to get 10 of those questions correct, in a row, AND IT HAD A TIME LIMIT.
ohoho man i feel bad for ya
@@Da_Rivulet french "person"
I remember that one I legitimately gave up after a while
ea literally just took your money gave you the game and locked it behind an impossible captcha
they did that on purpose
@@GooberInternet well i mean... they have a legitimate excuse, they're ea so they're allowed to do that
So what you're saying is an AI is given the prompt 'make a horse that looks like clouds' and then needs to make that image, meaning it knows what 'a horse shaped cloud' looks like, meaning it can solve its own captchas.
Incredible
Image generation is different from image detection
It's like drawing vs seeing
@@UnleashedDan this. The amount of armchair computer scientists in this comment section is really something. I tend to not like the phrase "Armchair X" (because even non-"experts" can be knowledgeable if they have the right requisite knowledge) but jesus christ there is no other way to describe this comment section.
It's closer to running image recognition neural networks backwards. Say you had a Neural Network trained to tell the difference between "horses" and "clouds". You show it a bazillion pictures of horses and clouds and let it fiddle with its internal controls until it can answer, with acceptable accuracy, whether a new picture is a horse or a cloud. It does this by combining different "input layers" based on what it "sees", weighing these layers against one another in several different ways, and then getting two outputs; 1) how sure am I this is a Horse, and 2) how sure am I this is a Cloud. For example, it might end up 30% sure that it's a Horse and 50% sure it's a Cloud (implying 20% that it's neither, but it's *required* to pick one) so it picks Cloud since it's the higher surety. So, run it in reverse, give it 51% Horse and 49% Cloud on the backend and a catalogue of Google Image results to stitch together, and it will reverse-engineer a picture through its algorithm, using information from Google Images, to produce a picture that, if run the correct way through the program should give a result of 51% Horse and 49% Cloud.
@@omargoodman2999 You can't run a neural network backwards. Your idea would probably end up producing random noise.
@@numbdigger9552 You apparently have never heard of the DeepDream project or AI Inception. Look up either of those terms for more info.
"The software is designed to detect faces and other patterns in images, with the aim of automatically classifying images.[10] _However, once trained, the network can also be run in reverse, being asked to adjust the original image slightly so that a given output neuron (e.g. the one for faces or certain animals) yields a higher confidence score._ This can be used for visualizations to understand the emergent structure of the neural network better, and is the basis for the DeepDream concept. This reversal procedure is never perfectly clear and unambiguous because it utilizes a one-to-many mapping process.[11] "
and now they're all just purely ai generated images which all look uncanny and horrendous
8:10 "Am I looking at a toads ass right now?"
Way funnier to me than it should've been.
13:45 I used to think it tracked your mouse movements to see if they looked "human" (wobbly, fluctuating speed, etc.), but these other reasons make a lot more sense.
"A Capcha needs to be easy enough for the dumbest Human, but too hard for the smartest AI."
Me, an Intellectual: "Design a Capcha that blocks both Smart AI *and* Stupid Humans from using the internet. Never settle for solving just one problem at a time."
Good suggestion, but what if it surpasses all humans collectively? Considering the AI can infinitely grow unlike a human, it's not an unrealistic outcome.
The question is, who uses the internet then?
@@The9thMonth That's just a matter of Natural Selection. "What if Neanderthal or one of the other hominids out-competed Homosapiens and humanity wasn't able to survive?" Well, then it would just be the decendents of the hominid species that *did* come out on top discussing Capcha here and now. If humanity really degrades to such a degree that no one at all is able to access the Internet except for a bunch of AI, then it means the AI are the new successful species; emergent life from synthetic origins. It could even have emergent sentience or even sapience. Just as a common ancestor of all hominids gave rise to one that would outperform the rest, so too in the natural order of things may we give rise to that which may be categorically better than us. If that's the case, then logically I can't find a practical, objective reason to arbitrarily judge the better to be less deserving than us. If AI truly become superior to mankind and mankind dies out as a result, it wouldn't be any different than the dying out of myriad previous transitional species; the only new matter would be that the format has shifted from analog to digital.
Mr. September
The robots do. Once the robots become smarter than all humans, we’ll no longer be able to create CAPTCHAs that can stop them because they’re smarter than us.
So basically what roblox did?
Just give everyone the calculus captcha
14:00 Isn't that a violation of privacy?
Yeah, I'm askng that queston in 2023, despite big techs constant river of shit