"Isaul Vargas, a New York-based software tester, spotted the videos in a post on BoingBoing and recognised them from an automation conference he had been at a year ago. They were being shown by a European firm that made streaming software for set-top boxes, the kit that sits under a TV and connects to services such as Sky or Netflix.
The company needed to be able to quickly and reliably upload digital video, a capability which it tested by uploading short, randomly generated snippets to its YouTube channel and running image-recognition software on it. "Considering the volume of videos and the fact they use YouTube, it tells me that this is a large company testing their video encoding software and measuring how Youtube compresses the videos," says Vargas."
I'm surprised nobody noticed that the names of the videos are from mktemp(3).
>>> tempfile.mktemp()
'/tmp/tmplOcKyZ'
>>> tempfile.mktemp()
'/tmp/tmpW0dJUR'
tmpwxm2CP
3 weeks ago - 31,842 views
tmpElnFwp
3 weeks ago - 112 views
Betting aqua.flv is the source file, the name of the video is the temporary output file (probably from a transcoder), and the article's conclusion about testing is accurate.
aqua.flv would probably match the Aqua Teen Hunger Force clip behind the paywall?
To do some math (that may be helpful or not):
77000 YouTube videos * 10 frames per YouTube video / 24 frames per second / 3600 seconds per hour = 8.91 hours of video
See also: A man who has uploaded 6,300 videos and counting of him doing nothing but smoking pipes and grumbling unintelligible streams of broken English that are conveniently transcribed in the descriptions: https://www.youtube.com/user/SMOKERSOFCIGARSPIPES/
At a former work place, we were gathering data about where entrances, lifts, passageways were in different transit stations. These videos (or someone on the same kick) helped us get an inside view of many of these stations.
I bet his viewers would be willing to pitch together some money and get him some sort of steadycam rig; the random video I clicked on was very hard to watch.
YouTube actually offer to fix videos that the algos find to be shaky; I wonder if they are choosing not to run the fix or if it's failing (the vid I viewed has some really fast transient wobble that doesn't seem like normal camera shake).
The youtube video stabilization makes for wobbly video. I take video inside of race cars, and they're unwatchable if you use the youtube stabilization. They turn out much more watchable without it.
Certainly is. During uni I did some work in a bar in a hotel, and one day the manager came in and said he'd been googling the name of the hotel and found a video on youtube. It was a video of a toilet in one of the rooms being flushed, and the dude who uploaded it had thousands of videos of toilets all around the world being flushed. Nothing else, just a single flush of the toilet.
This is his second account. A while back his first account "ConradsCigars" or something like that was posted on somethingawful and it got quite popular. At the time it was implied that these were some sort of smoking fetish videos. He didn't like the attention and deleted them all. Looks like he's going strong again now.
Brilliant. Andy Warhol recorded 8 hours long movie of his friend sleeping. In the youtube era he would make even greater fame than he managed in his time.
alantutorial is an alternate reality "game" starring a fictional character named "alan", made by Alan Resnick[1]. It seems all those videos are narrating a story!!! Now, I need to see them all :)
BBC has a different take on this [1] and denies Isaul Vargas's explanation. The part with hidden Aqua Teen Hunger Force clip and the fact that it's named aqua.flv is curious.
If it's a set-top box, it's not hard to imagine some set of circumstances where it uploads a stored or live clip instead of the random one. Or possibly the ATHF could have been uploaded from the company itself as a test of automatically uploading copyrighted content. It's behind a french paywall and the other video is a vid from a french balcony, so that could be a test of user-uploaded video.
This was a machine learning experiment and has since been shut down, due to the agent's unexpected capabilities. We hope to re-enable it once we've managed to control the psychosis inducing effects.
Also, please don't watch more than 8 of these in a row.
500+ hours of similar video content, most clearly computer-generated, with junk titles (from mkfile), and a new video being uploaded every 20 seconds. No "normal" videos other than the one behind a paywall and the one of the Eiffel Tower.
I don't know what the YouTube TOS says about automated bulk uploads, but I think such a user would have raised many flags on the system, regarding abuse, and the account would have been deactivated by now, right?... I'm betting that whoever is behind this has a special permit from Google.
If only there was a way to read the YouTube TOS and see what it says about automated bulk uploads (hint, it doesn't say anything) I don't see anything in the TOS that says doing something like this isn't allowed.
This ran as a piece on BBC Radio 4's today programme (quintessentially English), er, today.
"Do you think it's a machine producing these? Or is somebody sitting there all day producing these?", and other such questions for the non-technical.
As mentioned in the article they compared it to the phenomena of the 'Numbers Station' and suggested it might be governments communicating with their overseas spies. Priceless.
One interesting point they raised is that out of the 77,000 uploads there are two non-rectangle videos, both of scenes in Paris, France - so I bet the 'European' company is French.
A couple of years ago I was researching the distribution of filenames people uploaded onto flickr. One user there had multiple accounts and was uploading thousands of versions of an identical image with the same filename.
People see things and recognise things that aren't really there in clouds, as in internet clouds.
If you are building some type of API or interface then this stuff happens. The behaviour - when a machine does it - is inexplicable to the Radio 4 listening types that get other people to 'write their API's'.
I was hoping this might be an internet version of the old shortwave "numbers" stations [1] but now I see that there is likely a more mundane explanation.
OTPs and number stations can't be cracked mathematically and the channel is secure as long as the receiving end doesn't let the pad get into bad hands (this is achieved by compartmental communications and code words on top).
I would have thought that recent OpenSSL, SecureTransport and GnuTLS news would disprove your point. Not only that, we don't know what bad actors with unlimited resources can do with our communications if tapped.
So number stations and OTP are still viable and always will be.
With respect to numbers stations the transmission is centralised as well which means it's hard to break the channel infrastructure without crossing a border physically.
How does an operator reliably, repeatedly and securely use "asymmetric public-key cryptography" in an environment where the possession of a laptop is cause for suspicion?
The audio reminds me of some of the slower digital modes used in HF ham radio... Would be interesting to see if there is any pattern to the frequencies.
Just make something random, mysterious, use strange name and make it look like a cryptography thing...and enjoy watching the internet goes into a storm.
"Isaul Vargas, a New York-based software tester, spotted the videos in a post on BoingBoing and recognised them from an automation conference he had been at a year ago. They were being shown by a European firm that made streaming software for set-top boxes, the kit that sits under a TV and connects to services such as Sky or Netflix.
The company needed to be able to quickly and reliably upload digital video, a capability which it tested by uploading short, randomly generated snippets to its YouTube channel and running image-recognition software on it. "Considering the volume of videos and the fact they use YouTube, it tells me that this is a large company testing their video encoding software and measuring how Youtube compresses the videos," says Vargas."