• chevron_right

      Pirated Movies Flood YouTube, Millions of Views on Compromised Channels

      news.movim.eu / TorrentFreak · 5 days ago - 20:09 · 5 minutes

    youtube-movie-flood It’s Tuesday, April 24, 2007, and other than Beyoncé & Shakira singing Beautiful Liar on the radio, nothing much is happening.

    For someone called haroldlky , whose real identity is currently unknown, at least part of that day was spent opening a channel on YouTube, a video site that was yet to celebrate its second birthday.

    On that day more than 17 years ago, the fledgling YouTuber uploaded three videos that appear to have an engineering theme. The total running time for the trio, a modest 42 seconds.

    Whether ‘haroldlky’ was content with less than 400 views in total over the next 17 years, or whether he even visited YouTube ever again, is completely unknown. If he visited today, he might be a little surprised.

    After suddenly bursting back to life two weeks ago, three new videos were uploaded to his channel. All of these uploads were movies (Eragon, The King, and The Green Knight) dubbed in Hindi, and have since been viewed 376,000 times.

    No Isolated Incident

    After receiving a tip this morning that something unusual may be playing out on YouTube, we took a closer look. Similar events to that outlined above may have happened before but, roughly two weeks ago, the pace appears to have quickened and may have further increased during the past few days.

    Many of the channels currently offering pirated movies appear to be personal accounts that may have been compromised. That’s unlikely to mean an issue at YouTube but rather some kind of data breach on another platform, which typically lead to exploitation of users’ duplication of login credentials across various services.


    (Note: where channels relate to regular citizens and have been used recently, or contain home movies with families and friends, we have redacted information to limit identification of these potential victims)

    We have no specific knowledge of the mechanism through which credentials may have been obtained, if indeed that’s the case here. However, if we look at events from the opposite direction, it seems unlikely that a YouTuber uploading math tutorials receiving a few hundred views (above) would suddenly switch to Hollywood blockbusters overnight.

    In less than two weeks, these rogue uploads have already been viewed 917K times.

    Bigger Movies, Bigger Numbers

    It would be impossible for us to document every channel affected, but there are a few that catch the eye. Some channels may have been created in advance for use in nefarious activities. In most cases, however, it’s difficult to determine intent based on scant information. Out of caution, screenshots are partly redacted.

    The pair of images below show that channel creation dates can differ wildly. On the left is a channel with 117K subscribers and just seven videos, displaying a join date of September 17, 2023. The pirated films uploaded to the channel are responsible for most of the 7.4 million views generated in the last 10 days.

    On the right is a channel with just three videos, 6.1K subscribers, and an ancient join date of March 31, 2007. Just one pirated movie, Fate of the Furious, has already been viewed 582K times since it was uploaded yesterday.

    Who, Why, and How is It Possible?

    Who might be behind such a significant effort to spread so many movies is too early to say. We assume there’s no way of making money from these uploads, at least not on YouTube, so at least potentially, money may not be a factor*.

    Update: No surprises, money is the main factor, see explanation below

    Judging purely on outward impressions, various presentational factors, and the nature of those commenting on these movies, it seems likely that there’s at least some connection to India. Inbound links to the movies may be of interest to YouTube, but at least initially, a bigger question may be the focus.

    For reasons unknown and with no outward indication of video tampering, these movie uploads appear to have completely circumvented Content ID, YouTube’s anti-piracy fingerprinting system. While that can happen for older titles or those already in the system, one would assume every angle would be covered for new and recent movies.

    It’s feasible that fingerprints aren’t being supplied or maybe some glitch in the matrix is responsible. Interestingly, an upload of the Netflix series ‘The Gentlemen’ (every episode, running time 6h 23m) shows something that suggests it may have been subjected to scanning.

    As seen in the image below, a music track by ‘Zoxer’ titled ‘Forward’ is listed as making an appearance in the series.

    If this was indeed a content match and not something manually entered by the uploader, it seems unusual that the rest of the tracks in the series weren’t identified in the same way. Our searches found no record of ‘Forward’ by Zoxer appearing in the series; it may have done, we’ve just had no luck finding it.

    Update and explanation: It appears that the audio content matches on these movies do not rely on the allegedly-infringed tracks being in the movie, only that they’re present in the video files uploaded to YouTube. The scheme apparently works something like this:

    1. Create or otherwise obtain tracks that are not detected by Content ID
    2. Register those tracks (illegally) with Distrokid, CD Baby, or Tunecore (who are oblivious to the fraud)
    3. Obtain a movie that will get lots of views
    4. Use software to stitch the audio track to the end of the movie
    5. Upload resulting video file to YouTube
    6. Wait for a content match on the music, monetize the entire video
    7. Get paid a small amount, or nothing at all if the royalty collection companies find out

    Those enjoying the uploaded movies appear to have no problem finding them as they’re racking up millions of views. It’s hard to say exactly how many millions of views overall, but it’s a significant number, especially when in normal circumstances the figure wouldn’t be worth reporting.

    There’s no requirement for YouTube to be proactive in these circumstances, but why nobody has reported the haul below, which represents just a few of the movies uploaded, is certainly interesting.

    From: TF , for the latest news on copyright battles, piracy and more.

    • chevron_right

      RadioWis | EP 6 | Looking at the past 7 months and some true colours

      comics.movim.eu / RandoWis · 6 days ago - 05:44

    Uh-oh, it’s been so long since the last episode! I explain things with terrible detail, and share my thoughts on recent events. Hope you enjoy the listen!

    The post RadioWis | EP 6 | Looking at the past 7 months and some true colours first appeared on RandoWis .

    • chevron_right

      I took part in YouTube’s Partner Program

      comics.movim.eu / RandoWis · 6 days ago - 05:34

    This video was meant to be released around December last year, but things got so out of hand with producing animations, that it’s only finally done in July 2024. Please enjoy the watch!

    The post I took part in YouTube’s Partner Program first appeared on RandoWis .

    • chevron_right

      Russia Slams Google’s ‘Censorship’ Yet Deleting 5.6m VPN Links is Just Fine

      news.movim.eu / TorrentFreak · Thursday, 18 July - 18:34 · 4 minutes

    putin-g Those who have taken an interest in the words of Dmitry Peskov since Russia’s invasion of Ukraine, will have noticed that Vladimir Putin’s press secretary is loyal, if nothing else.

    In respect of his commentary on matters important to the Kremlin, it can be difficult for foreigners to appreciate the finer details; is he always trying to mislead or just most of the time?

    Take for example a statement early February 2024, just a few weeks before new law came into effect that tightened the noose further still on the use of VPNs in Russia.

    These measures included a ban on advertising VPNs as a means to circumvent government blocking, which prevents access to other information currently blocked in Russia. Indeed, providing information on how to access blocked information using a VPN is now a criminal offense, with the unblocking information also subject to blocking.

    Peskov said this didn’t mean that VPNs were banned in Russia, and he offered reassurances that no decision like that had even been made.

    On one hand, that statement was truthful. On the other, Russia’s evisceration of VPNs since 2017 means that when Peskov referred to VPNs, he wasn’t even on the same page as many people listening. As the law stands, there can be no expectation of privacy when using a state approved VPN, i.e the only ones allowed under law.

    Censorship now applies to any mention of VPN unblocking capabilities and should those be discussed online and end up in search results, that information will be purged under the orders of telecoms watchdog Roscomnadzor.

    “YouTube Violates Key Principles of Free Dissemination of Information”

    This week Roscomnadzor informed news outlets including Interfax, TASS and RIA Novosti, that it had contacted Google CEO Sundar Pichai with complaints about YouTube. The regulator said that over 200 channels, operated by Russian media, various government agencies, politicians, and media personalities, are currently blocked by YouTube for supporting Russian authorities.

    Roscomnadzor said that it demanded action from Pichai to reverse what it described as “an act of censorship” carried out by YouTube under a “Russophobic” policy.

    “The victims are not only those who voice their position on various political issues, but also people who generally speak well of Russia,” Roscomnadzor said, inadvertently amusingly.

    Rules For Thee, But Not For Me

    In the very same press statement criticizing YouTube for failing to provide unimpeded access to information, Roscomnadzor saw no hypocrisy in calling out YouTube again, this time for not censoring content at Russia’s behest. The content at issue, over 61,000 videos containing “fakes about a special military operation,” extremist material, and “promotion of non-traditional sexual preferences.”

    What these categories all have in common is obfuscation as a means to change and expand definition. The term “special military operation” was crafted to reassure Russians that there was no ‘invasion’ and certainly no ‘war’ in Ukraine. A new crime of ‘discrediting’ the military, which encompasses spreading ‘fake’ information, made the use of the words ‘invasion’ and ‘war’ in the context of Ukraine, punishable by years in prison.

    ‘Extremist material’ is a little like the shifting definition of VPN in Russia. It started out describing material promoting extremist ideology and then when it suited Russia, it meant something else. When Facebook, Instagram, and Meta were designated as extremist groups in 2022, the term could encompass almost anything. Promotion of non-traditional sexual preferences refers to anything that deviates from the Victorian standard of a man and a woman, period. Or at least it does today, the terminology seems broad enough to intrude even on that.

    Erasing the Existence of VPNs From Search Results

    As more overt wars of words attract attention, Roscomnadzor’s war on VPNs not only continues, but does so with significant assistance from Google. The company says it has no choice but to comply due to Russian law and Roscomnadzor certainly isn’t holding back.

    Russia’s campaign against VPNs and similar technology gathered pace in 2017 and hasn’t let off the gas since. In addition to targeting Tor , a massive search engine delisting campaign is ongoing.

    What began with takedown notices ordering Google to remove a few dozen links per month, soon escalated into orders to remove a few thousand, then tens of thousands . The invasion of Ukraine saw thousands of links crammed into single notices but nothing like the delisting campaign currently underway.

    A typical notice sent by Roscomnadzor to Google is shown below. This is just one of 18 notices sent to Google in June 2024 and orders the removal of a staggering 460,750 URLs from Google search, for being related in some way to VPNs.

    The exact nature of the URLs removed is not for public consumption. On top of the requirement to disappear the links, Google is compelled to deviate from its transparency policy due to restrictions under Russian law.

    What we can see are the numbers involved. In just 18 takedown notices sent to Google, Roscomnadzor demanded the removal of more than 5.6 million URLs, in June alone.

    Given the surge in volumes, these removals represent more than just links to VPN providers. Tutorials and discussion about VPNs have become a priority target and are likely to feature heavily here.

    As a result, it’s difficult not to conclude that this represents classic censorship and the denial of the right to freely receive and impart, knowledge and information. Russia knows that, and it knows that everyone knows that it knows. It just doesn’t care.

    Links to the takedown notices, courtesy of the Lumen Database, are available below

    #01 June 03, 2024 – 413,348
    #02 June 03, 2024 – 58,987
    #03 June 04, 2024 – 39,353
    #04 June 04, 2024 – 442,913
    #05 June 04, 2024 – 478,474
    #06 June 05, 2024 – 444,220
    #07 June 06, 2024 – 37,914
    #08 June 06, 2024 – 450,920
    #09 June 07, 2024 – 456,006
    #10 June 10, 2024 – 437,887
    #11 June 10, 2024 – 38,646
    #12 June 11, 2024 – 437,957
    #13 June 13, 2024 – 431,769
    #14 June 13, 2024 – 77,607
    #15 June 14, 2024 – 445,285
    #16 June 14, 2024 – 49,424
    #17 June 16, 2024 – 447,292
    #18 June 17, 2024 – 460,750

    From: TF , for the latest news on copyright battles, piracy and more.

    • chevron_right

      IA : pour entraîner ses modèles, la Big Tech est en train de piller YouTube

      news.movim.eu / JournalDuGeek · Wednesday, 17 July - 15:04

    Ia Youtube

    Plusieurs cadors de la tech comme Nvidia, Apple, Anthropic ou Salesforce font leur marché sur la célèbre plateforme vidéo pour entraîner des modèles IA à fort potentiel commercial, sans la moindre contrepartie. Une situation qui rappelle l'importance de la traçabilité des données.
    • chevron_right

      Former OpenAI researcher’s new company will teach you how to build an LLM

      news.movim.eu / ArsTechnica · Tuesday, 16 July - 21:40

    File photo of children in a classroom listening to a robot.

    Enlarge (credit: Getty Images )

    On Tuesday, former OpenAI researcher Andrej Karpathy announced the formation of a new AI learning platform called Eureka Labs . The venture aims to create an "AI native" educational experience, with its first offering focused on teaching students how to build their own large language model (LLM).

    "It's still early days but I wanted to announce the company so that I can build publicly instead of keeping a secret that isn't," Karpathy wrote on X.

    While the idea of using AI in education isn't particularly new, Karpathy's approach hopes to pair expert-designed course materials with an AI-powered teaching assistant based on an LLM, aiming to provide personalized guidance at scale. This combination seeks to make high-quality education more accessible to a global audience.

    Read 5 remaining paragraphs | Comments

    • chevron_right

      YouTube creators surprised to find Apple and others trained AI on their videos

      news.movim.eu / ArsTechnica · Tuesday, 16 July - 21:23 · 1 minute

    A tech commentator holds an iPhone while speaking to the camera

    Enlarge / YouTuber Marques Brownlee discusses iOS 18 in a new video. This specific video wasn't part of the large dataset that was used to train AI models, but many of his others were. (credit: Marques Brownlee )

    AI models at Apple, Salesforce, Anthropic, and other major technology players were trained on tens of thousands of YouTube videos without the creators' consent and potentially in violation of YouTube's terms, according to a new report appearing in both Proof News and Wired .

    The companies trained their models in part by using "the Pile," a collection by nonprofit EleutherAI that was put together as a way to offer a useful dataset to individuals or companies that don't have the resources to compete with Big Tech, though it has also since been used by those bigger companies.

    The Pile includes books, Wikipedia articles, and much more. That includes YouTube captions collected by YouTube's captions API, scraped from 173,536 YouTube videos across more than 48,000 channels. That includes videos from big YouTubers like MrBeast, PewDiePie, and popular tech commentator Marques Brownlee . On X, Brownlee called out Apple's usage of the dataset, but acknowledged that assigning blame is complex when Apple did not collect the data itself. He wrote:

    Read 13 remaining paragraphs | Comments

    • chevron_right

      Tibo InShape dévoile son vote : pourquoi la position politique est-elle si scrutée sur les réseaux sociaux ?

      news.movim.eu / Numerama · Monday, 8 July - 09:09

    Durant ces législatives anticipées de 2024, les prises de position politiques des influenceurs et influenceuses étaient particulièrement observées. Au centre de l'attention, le refus de Tibo InShape, premier influenceur de France, de se prononcer contre l'extrême droite. Jusqu'à ce qu'il en dise plus sur son vote, après les résultats.

    • chevron_right

      AI trains on kids’ photos even when parents use strict privacy settings

      news.movim.eu / ArsTechnica · Tuesday, 2 July - 19:37

    AI trains on kids’ photos even when parents use strict privacy settings

    Enlarge (credit: Aitor Diago | Moment )

    Human Rights Watch (HRW) continues to reveal how photos of real children casually posted online years ago are being used to train AI models powering image generators—even when platforms prohibit scraping and families use strict privacy settings.

    Last month, HRW researcher Hye Jung Han found 170 photos of Brazilian kids that were linked in LAION-5B , a popular AI dataset built from Common Crawl snapshots of the public web. Now, she has released a second report , flagging 190 photos of children from all of Australia’s states and territories, including indigenous children who may be particularly vulnerable to harms.

    These photos are linked in the dataset "without the knowledge or consent of the children or their families." They span the entirety of childhood, making it possible for AI image generators to generate realistic deepfakes of real Australian children, Han's report said. Perhaps even more concerning, the URLs in the dataset sometimes reveal identifying information about children, including their names and locations where photos were shot, making it easy to track down children whose images might not otherwise be discoverable online.

    Read 24 remaining paragraphs | Comments