I've got some weird captchas recently.

[cody1]

heard recently that google was selling captcha data to the pentagon to train autonomous military drones. sounds pretty batshit now that i type it out but has anyone else heard about this/have any info on it?

 

[Punp]

Back in the day we used to fill the two word captchas with always one word being the n-word. Don't think it did much but it was always funny.
As for people being concerned you're performing text recognition to train an AI: OCR software is nothing new (I've worked with it and still do, currently for an automated invoice scanning physical to digital service). It's likely you were in the past, but at this point OCR software is so good that the only human work left to do in an OCR system is context-training, IE recognizing that some piece of text relates to a specific thing, and not just the computer being able to recognize the letters themselves.

I really hate the new 4chan captcha too, makes me post there less and less.

I couldn't use the 4chan captcha and just gave up. Another part of the internet just dying.

RE: The old captcha, the second word would get recycled as part of the next captcha and people would see it.

 

[№56]

I'm probably the only person on earth who actually likes the new 4chan captcha. Matching up the distorted letters is less annoying and takes less time than reCAPTCHA's stupid picture recognition thing. Plus, it's (probably) not helping Google make more money.

 

[Punp]

I'm probably the only person on earth who actually likes the new 4chan captcha. Matching up the distorted letters is less annoying and takes less time than reCAPTCHA's stupid picture recognition thing. Plus, it's (probably) not helping Google make more money.

Is that how that works? I couldn't understand it at all

 

[№56]

Is that how that works? I couldn't understand it at all

It's exactly like the old-school captchas with distorted text, you just have to move the scroll bar around until you get something legible. It's fairly easy once you get the hang of it.

 

[LostintheCycle]

As to social media, there have already been astroturfing (false grass-root movement campaigns) by political interests on important social thought. Just look up "political astroturfing" - I don't want to give examples for fear of appearing biased.

I recommend reading a book called Toxic Sludge Is Good For You. It investigates cases of American PR companies manipulation of the public, covers a lot about astroturfing. It's shocking to see that public manipulation is an old thing, at least 50 years ago they had very good techniques already. My favourite chapter was about the tobacco industry and how they hired PR companies to astroturf pro-tobacco movements under the guise of American freedoms

 

[LostintheCycle]

You'll be interested to note that you do not have to be correct on 100% of the images to pass the verification, so I aim for 50% to blur the resulting data as best I can.

I just did this on a hCaptcha for the first time and holy shit you're right. I didn't even pick any of the right things and it still put me through. I'm gonna do the 50% thing every time from now on

 

[Fractalactals]

I tried posting on 4chan last night on my phone and realized how VILE the new captcha is when using a smartphone.
I thought it was fine on PC, if a bit annoying, but on a phone it's very aggravating to use!
All the more reason not to go there.

 

[Shantotto]

I've been getting some weird captchas recently, as well. Anyone know how I'm supposed to solve this one?
View attachment 36934

This would be so cool. I've never played touhou but imagine the bar for entering a website was beating a super difficult videogame level or challenge. I think logging in would feel a lot more rewarding considering it takes a certain amount of learned skill to enter everytime. Or maybe I would just get annoyed and stop using it.

 

[floozy]

Tbh in the coming years I see widespread use of malicious AI beginning a new era for the internet. One where truth is nearly impossible to verify, where you can no longer be reasonably confident that a person on a forum, in voice chat, or in a multiplayer game is real.

I already live like this. Also, I thought I was the only person who couldn't solve 4chan's pain in the ass captchas. But I might just be retarded.

 

[nobodyhere]

Pardon me if someone already said something along these lines (I just skimmed this thread!).
There's a very clear purpose behind all this AI training—at least, in my eyes. The aim is to fully automate surveillance, and especially the process of tracking someone through a breadcrumb trail of photos.
Consider the Captchas that test you on pictures of cars, planes, and bikes. Now, consider you're a detective trying to catch a criminal, and you have 6000 photos available from the suspect's iCloud account (as well as the accounts of his friends & family, which they handed over voluntarily). You have several witness reports that, at the scene of the crime, the criminal was on a bike. Rather than going through each picture individually, you have an AI check for bikes and provide you all the photos. Apple already has this kind of tech built into iOS.
Now, imagine our hypothetical criminal is guilty of hate speech.

Is it fun yet?

 

[Shantotto]

Pardon me if someone already said something along these lines (I just skimmed this thread!).
There's a very clear purpose behind all this AI training—at least, in my eyes. The aim is to fully automate surveillance, and especially the process of tracking someone through a breadcrumb trail of photos.
Consider the Captchas that test you on pictures of cars, planes, and bikes. Now, consider you're a detective trying to catch a criminal, and you have 6000 photos available from the suspect's iCloud account (as well as the accounts of his friends & family, which they handed over voluntarily). You have several witness reports that, at the scene of the crime, the criminal was on a bike. Rather than going through each picture individually, you have an AI check for bikes and provide you all the photos. Apple already has this kind of tech built into iOS.
Now, imagine our hypothetical criminal is guilty of hate speech.

Is it fun yet?

looool. dont know what you are talking about schizo, check out this cool thing i found on tiktok

The Follower

How does this work? Recorded a selection of open cameras for weeks. Scraped all Instagram photos tagged with the locations of the open cameras. Software compares the Instagram with the recorded footage. I launched the project on 12 September 2022. The YouTube video was just created with results...

driesdepoorter.be

 

[Punp]

Pardon me if someone already said something along these lines (I just skimmed this thread!).
There's a very clear purpose behind all this AI training—at least, in my eyes. The aim is to fully automate surveillance, and especially the process of tracking someone through a breadcrumb trail of photos.
Consider the Captchas that test you on pictures of cars, planes, and bikes. Now, consider you're a detective trying to catch a criminal, and you have 6000 photos available from the suspect's iCloud account (as well as the accounts of his friends & family, which they handed over voluntarily). You have several witness reports that, at the scene of the crime, the criminal was on a bike. Rather than going through each picture individually, you have an AI check for bikes and provide you all the photos. Apple already has this kind of tech built into iOS.
Now, imagine our hypothetical criminal is guilty of hate speech.

Is it fun yet?

This is already technology that has been rolled out in China and Russia. I'm sure it's already being used in America and Europe, bu it's behind the scenes and they cover it with pretending to get that information from other avenues.

This is the broader scope of the AI technology.

Captchas have certainly been used to train traffic cameras for self-driving cars and automatic traffic violation recording - but I don't think we've been given a "track this particular car" task yet.

Why are we training it to detect whether a lion's eyes are opened or closed?

 

[SolidStateSurvivor]

heard recently that google was selling captcha data to the pentagon to train autonomous military drones. sounds pretty batshit now that i type it out but has anyone else heard about this/have any info on it?

How's that batshit? Honestly seems like the logical next step after ditching their whole "don't be evil" mantra.
To give another example, remember those silly Microsoft Hololens that were supposed to be like an augmented reality game console? Now they're being rolled out for use by the US military in combat lol.

 

[EzzyCrafts]

this bullshit is so obvious it's by an AI, look at bottom right, middle left, turtle, middle right, top left, middle top, FUCKING EVERY PICTURE

 

[Punp]

View attachment 38763
this bullshit is so obvious it's by an AI, look at bottom right, middle left, turtle, middle right, top left, middle top, FUCKING EVERY PICTURE

They can't keep getting away with this [;_;]

 

[grap]

New vice article on the issue: https://www.vice.com/en/article/xgwy5n/captcha-is-asking-users-to-identify-objects-that-dont-exist

 

[alCannium27]

New vice article on the issue: https://www.vice.com/en/article/xgwy5n/captcha-is-asking-users-to-identify-objects-that-dont-exist

That article seems to be discussing two different things, one being the Captcha companies making money from their users by selling user data -- in this case, image identification -- to organizations doing machine learning for whatever purpose. The other is the bit they mentioned at the end, which is GAN, and I fear is not directly related to those weird images.
The fact that AI generated images exist in Captchas is undeniable, I can see only two ways for them to get in -- one is the Captcha service providers are hurting for new images of an object and resolved to use the AI, or certain of their clients pay them to put hose images in for whatever research purposes they have.
The first one is, at least to my mind, unlikely -- I've been asked to solve captchas with fire hydrants, bikes, traffic lights etc., and each time there appears to be multiple repeats from previous tasks. It's clear that, at least the company that runs the Captchas I've encountered so far has a database for test images, and no given the speed at which Captchas needs to be generated and the basic concept of minimizing cost, I don't see why they'd generate images with an AI real-time, each time, a user is asked to solve a puzzle.
So clearly it's somebody putting these images into the service's image database.
I guess some company doing GAN research projects can be putting their GAN-images into a captcha database in order to refine training datasets, crowd-sourcing human testing for semantic identification, like the example in the VICE article about a "Yoko". No doubt a concept a researcher came up with in the lab, trying to see how consistenty a model can generate a newly defined concept.
It'd be an interesting shift to see Captchas from a "label this real image so the AI can learn what's what" to "label if this image is generated corrected according to a new concept so we can see how good our AI understands a new concept". Almost fascinating really, to see how a subset of a market reflects another industry -- alignment has finally become a big topic in the field right now, whereas it was something few considered important just a decade years ago, relegated to the musings of Sci-Fi novelists. It'd be interesting to see if Captcha can be utilized in the future to gauge human morality emass in some way, creating a "morality police" for an AI model.
Really, if you look at the multi-model AIs these days like GPT4, it's really a interpretor AI taking in user input, figuring out if it wants to query the internet, or generate an image, or summarize an article, or a combination of the above, and pipe the task to another, corresponding, AI model. (a process at least used by the open sourced text-generation-webUI codebase). Stable Diffusion had a safetychecker which pipes a generated image to a model trained to identify if an image is NSFW, etc.. A morality model can perform a simple YES/NO check, or provide a detailed weighted analysis of all morality concerns based on the things we humans care about.

I fear the day of my sweet release from this world is still some time away, the real Skynet will not kill me as I had hoped; it will just do its masters' biddings for their amusement, like a pet dog or a court jester. I can only hope all the big tech competiting will unwittingly, in spite of their incompetence, usher in a new age of the machine.