Few technology products are as broadly controversial as video-generating artificial intelligence.
These complex algorithms that combine millions of data points into a video gob over a few seconds are well-known for their unique material, leading to a wide range of ethics. and legal concerns. (That's before even mentioning how much energy it takes to integrate AI video.)
Tech billionaires tend to argue that this is simply the way things are needed. If you need AI, you need to feed copyrighted books, music and videos.
However, a group of AI researchers are working to prove that argument is wrong.
Moonvalley is a Los Angeles-based AI startup offering a “3D-Aware” video integrated model that claims to be 100% trained in public domain films.
The startup's flagship product, Marey made his debut in March with limited running. It's now publicly available, TechCrunch Report, It uses a credit-based system typical of most AI video software.
The company has attracted the attention of big names in the cinema world, like Ed Wolbritch, a VFX artist and producer who worked on films such as “Titanic,” “Benjamin Button,” and “Top Gun: Maverick.” Moonvalley hired Ulbrich in June to get in touch with Film Studios. This is the role he was drawn to the company's “clean model” as he calls it.
Ulbric used to fall into generative AI, but says that Moon Valley's approach has helped him change his mind.
“I think I am a game changer and why I was forced is at the heart of that. [Moonvalley’s] Ethically sourced, ethically trained, clean model of what is legal and appropriate,” Ulbric said. deadline In an interview. “There's no stolen pixels, no internet scraping. It's done in a great way. It's very important that it happened.”
Similar projects have been launched in other media. In June, a team of over 20 AI researchers openly trained large language models (LLMs) on licensed or public domain data, proving that they don't need millions of stolen books to build AI chatbots.
It was certainly a ton of work, and the team formatted everything and adjusted more than 8 terabytes of data, equivalent to around 1,685,461 Bibles, to reaffirm the copyright status of all the materials.
The result was a more or less stacked LLM on Meta's Llama 1 and 2 7b.
As they argue, it remains to be seen whether Moon Valley data is really public, but it could provide a strong rebuttal to the story of Big Tech's lack of data and necessary looting.
Details of Generated AI: “Indie Rock Band,” which uses AI, claims that “we don't use AI.”
