When the generative AI conversations can't be more surprising to filmmakers or content creators, the Atlantic drops another bomb.
They exposed the pointless scraping of around 16 million YouTube videos to train the next generation of generation AI.
And the companies behind scraping aren't like startups. They are a huge company that uses what you put on YouTube to train programs that will replace you.
Let's dive in.
Unprecedented robbery
part of, part of The Atlantic Ocean The new AI Watchdog subsite reveals that over 15.8 million videos have been downloaded without permission from over 2 million YouTube channels.
You can use them Searchable database To see the videos used to train the generative model and which tech companies are using the material.
I searched and found some film school videos there, so it's fun.
Of course, this kind of thing is against YouTube's terms of service, but AI companies are finding other ways through third-party apps and other workarounds.
Not all YouTube videos are copyrighted, but many of the videos found in Atlantic exposures were.
Here are some of the articles that stood out to me: “Many major tech companies have trained AI using these datasets, according to research papers I read and research papers I spoke to. This group includes Microsoft, Meta, Amazon, Nvidia, Runway, Bytedance, Snap, and Tencent. They “respect” the creators of content and believe that their use of their work is legal under existing copyright laws. This focuses on developing ways to generate “persuasive, high quality ads from simple prompts” when video is involved.
How does this affect filmmakers?
I know that most of YouTube is not driven by filmmakers, but what do you speculate? It turns out that AI companies are pursuing filmmakers' works specifically.
Another excerpt from the article reads: “AI companies are more interested in some videos than others. 404 Media A former employee of Runway, who builds AI video generation tools, shows that the company has evaluated certain channels, including “high camera movements,” “beautiful film landscapes,” “high quality film scenes,” and “ultra-high quality sci-fi short films.” One channel was labelled “So far the Holy Grail of Automotive Movies.” The other was labelled “Only 4 videos” and is really well done. ”
In other words, these AI companies see them as workarounds for training their technology to replace those who make these channels. Imagine learning all these cool camera techniques and ideas and developing your original voice.
That's what's happening now.
This is our battle.
This is more than just a massive copyright violation. This is an existential threat to creative professions everywhere. And we need to fight back.
Every frame of your work they ingest is used to build a more effective tool to replace you. They are not enhancing human creativity. They automate it and reduce costs.
The WGA and SAG-AFTRA strikes have placed a bet on the ground, but this new front, the video data – is where the next battle will be fought. This is not to stop technology. It concerns demanding consent, compensation and management.
What can you do?
It's understandable that you feel overwhelmed. But Giving is not an option. Share Atlantic reports and search tools with all authors you know.
This issue is too big to be resolved by lawsuits alone. We need clear and powerful laws that protect creators' rights and enforce transparency from AI developers.
Add it all
This report proved to be phenomenal and infuriating. It feels like all these companies are stealing ours, waiting to be caught, and continuing to pay nominal fees.
That's a shame, but we're all involved in the fight together.
I'll keep you up to date.
