This week, Openai released His last model of generating video with artificial intelligence Sora 2, advertising him as a “big jump forward” for space. When the litter amazes the public, she will have to fight for the market share in a crowded market, including with a major competitor who is gaining momentum quickly: the Chinese company by BYTEDANCE, which owns TiktokField
Over the past few months, ByTedance has been released by SEEDANCE, AI video generator, which already many users have already Calling the best in the world and the new version of Seedream, elite Image model. Its LLM, Dubao, has 150 million active users per month, according to the Analytics Analytics website AICPB.comField
AI BYTEDANCE Achievements are a vivid example of how Chinese AI companies quickly catch up with American ones, despite the control of the chip export. Since their models of high quality, as well as cheaper, they defeat consumers around the world, including in America. But while these models captivate many users, they also represent many problems that many of the advanced models pursue: they allow anyone to create available deep -farming that are indistinguishable from reality, as well as freely reproduce the material protected by protected means.
Reaching the border
Over the past year, Baitedan has gathered a top talent, hiring The former Vice President of Google Deepmind, which leads the fundamental research of AI and luring other engineers and researchers from Alibaba and other startups, Financial times reported in December. He also invested Billions of dollars In the infrastructure, including advanced NVIDIA chips.
ByTedance released the first iteration of its video model in June, and in September a new image generator, SEEDream 4.0. Models can be available on US platforms through third -party platforms.
Jobrin Johnny, a designer based in Kerala, India, first discovered Seedream at the end of August and was especially impressed with how he represented someone from his region. “The generated face bore accurate functions and details of a real person from Kerala,” he says.
Johnny says that SEEDANCE is now his favorite video model, especially with how she captures physics and Natural movementsThe field will not hurt that the seedlings are much cheaper: on a third -party Freepik platform, it costs half as much loans than Google Veo 3. In social networks, several influential persons have Encouraged Their followers To switch to BYTEDANCE products depending on their price.
Tiezhen Wang, an engineer on the machine learning platform, hugging his face, tested the tools. He says that he developed an “amazing” poster with the beginning of the seeds, and this Seyan “shines on tasks with the image -Video, preserving the style and consistency of the character …
Eric Lou, co -founder of the program for editing the Kapwing video, offered his customers the generation of images of AI for several years, starting with stable diffusion. When SeeDance and Seedream were released, his team conducted internal testing to compare their rapid commitment, image quality, speed and cost to American competitors. “And it was not close – the models are better in every way,” he says.
Lou quickly switched the images of AI by default Kapwing to the seedling and seeds from American models. “It was almost not difficult, because we save money, but also give our users the best quality products,” he says.
“Unlimited” AI
But this growth in quality has many consequences. Firstly, this shows that Chinese companies have successfully moved through American export elements of chips management designed to slow down. Information reported In December, this Baitedance gained access to advanced NVIDIA chips, renting them outside of China. The company was It expands quickly Using the data processing center in Malaysia.
The price of ByTedance tools also allows many new users to contact AI to create realistic images. The availability and availability of these tools can facilitate the work processes of advertising, marketing and stock personnel industry. “Why buy clips when you can generate any shot, do you need instantly?” One user X wrote In the thread about the seedling.
As these hyperrealistic AI tools spread, threats of deep norms and disinformation are painted large. In June, Time found that the video model Google Veo 3 generated realistic clips, which contained misleading or inflammatory information about news events. After time, I contacted Google about these videos, the company said that it would begin to add a visible watermark to the video generated tool.
Read more: The new AI tool from Google generates convincing deep riots, conflict and election fraud
When the time was checked by many of the same clues that were used with the Google model with seedlings through the Dreamina's Capcut tool, he actually rejected many of them, on the grounds that they violated the leading principles of the community. Nevertheless, the model, like its competitors, created a decent realistic personnel, which, according to the visible, can be divided as misinformation on social networks, as this video created through seeders of American soldiers providing assistance to Palestinian refugees. The representative of ByTedance did not respond to a request about the comments.
BYTEDANCE realism also raises questions on copyright and similarity. Chinese scientists have Approved that China accepted the regulatory approach of “moderate condescension” from the point of view of teaching models on material protected by copyright. This is displayed in the output of the model: for example, one user X published an image of Seedream show Joker Hit Ledger, Margot Robbie Harley Kuinn and woman -Koshka Michelle Pfifer in Dive -Bare. The other created image With Spider-Man, Batman and Superman.
Lou, in Capping, says that seeds and seeds, it seems, are especially ready to recreate characters protected by copyright, whether Mickey Mouse or Minions. “I think that in the States there is much more attention in some of these large laboratories in terms where they supply the content on which they train,” he says. “I think that in China there is an unlimited ability of researchers to receive data that they need and teach it.”
Selina Suy, the policy of China and AI in the office of Erica Schmidt, says that “expected” that Baitedance and other Chinese companies teach their models for video surfaces created by the user from their social networks. She adds that video generation models are “a growing flow of income for artificial intelligence companies”.
Time was able to create an image of “Young Brad Pitt and Leonardo DiCaprio, shocking his hands” through the beginning of the beginning at the caping. Some members of the Congress, including Marsh Blackburn, Trying To accept legislation that will protect the voice and visual similarities of people and creators from digital replicas created without their consent. But such legislation is still far enough from passing.
Meanwhile, American companies begin to pay attention to these Chinese giants of AI, forcing them to more publicly grab copyright protection. In September Disney, Warner Bros. Discovery and Nbcuniversal Judging The Chinese company Minimax for “intentional and arrogant” violation of copyright.
“Heat the fire”
American laboratories have Argument that, since their Chinese colleagues have a weak relationship with copyright, they should also be able to train on a material protected by copyright, and that they create new images so that they are transformed and protected with fair use. Earlier this year, Openai announced that its rules related to the moderation of content will weaken, which will lead to a wave of Ghibli studio memes flooded on the Internet.
Read more: How these studio memes of died are a sign of a change of era of Trump Openai
“I am not convinced that this is managed by Chinese companies. Openai opened the gateways, to some extent in March, ”says Maribet Rauh, an artificial intelligence ethics researcher in the AI laboratory in Trinity -College Dublin. She says that the ability of Baitedans to create a semblance of protected copyrights and real people, “unfortunately, adds a heat to the fire of the fight to succeed in any price, and regardless of which laws or ethical consequences.”
Rauch has many concerns about the spread of Deepfake tools, including the fact that they can lead to an increase in harassment and disinformation and threaten user privacy. “People very open interaction with these models: such images that they are interested in creating, how they set them up, or if they introduce images of similarities with real people,” she says. “This is all the data that will be at risk.”
Katherine Trendacost, Director for Policy and Evil in the Electronic Frontier Foundation, claims that education is the key to reducing the risks of Deepfake. “We have reached this strange point where everything can be generated, but no one believes anything else,” she says. “But we never solve the main problem. We simply continue to focus on new technology, not media literacy or teaching analytical skills or how to evaluate sources. ”
