– I had fun. Now it is about to tilt completely. So says Ole Petter Baugerød Stokke, editor of the online newspaper Kode24. He has been interested in artificial intelligence (AI) for a while now. – I honestly think it’s getting disgusting, he says. In a newsletter published on Friday, Stokke writes about how he is going crazy wondering if all the pictures he sees are fake. BEGINNING TO GO CRAZY: Kode24 editor Ole Petter Baugerød Stokke says he hates how good AI has become. Photo: Ole Petter Baugerød Stokke / Privat The new videos generated by Sora, the new service of OpenAI that was presented last night, did not ease the fears. – We have already talked a lot about the uncanny valley and that it is disgusting with artificial intelligence that looks a bit real, but now it is starting to get disgusting on an existential level, says Stokke. Sora generates videos that are up to a minute long, much longer than what has previously been generated from text to video. According to OpenAI, the videos can have complex camera movements and portray human expressions better and more complex than before. TOKYO DREAMING: This text command is the starting point for the video: “Reflections in the window of a train traveling through the Tokyo suburbs”. The video is generated by the AI tool Sora. For now, Sora is not available for testing among people, but OpenAI has released a number of examples of videos the AI service should have generated. Text-based video services are not new in themselves, but: – What is new about Sora is that it is so unbelievably good compared to what you have seen previously, says Stokke. Several people agree with him. Experts have commented that they did not expect video material on the level of Sora until two or three years from now. Now the problems that already exist in connection with AI will only intensify, Stokke believes. – The whole fake news thing can be taken to dizzying heights. The classics like fake pornography, scams and false information will only become bigger problems than they already are. “Prompts” used to create videos Open Ai has been open about which text commands, so-called prompts, it has fed Sora with to get the various results. Here we reproduce the text commands for the video clips used in this case. Video 1: Woman walking on the street in something that is supposed to represent Tokyo: “A stylish woman walks down a Tokyo street filled with warm glowing neon and animated city signage. She wears a black leather jacket, a long red dress, and black boots, and carries a black purse. She wears sunglasses and red lipstick. She walks confidently and casually. The street is damp and reflective, creating a mirror effect of the colorful lights. Many pedestrians walk about” Video 2: Woman “films” herself in the reflection from a train window: “Reflections in the window of a train traveling through the Tokyo suburbs” Video 3: A puppy that moves from window to window: “The camera directly faces colorful buildings in burano italy. An adorable dalmation looks through a window on a building on the ground floor. Many people are walking and cycling along the canal streets in front of the buildings” Video 4: Representation of the gold rush in California: “Historical footage of California during the gold rush” Video 5: Close-up of a man with glasses: “Reflections in the window of a train traveling through the Tokyo suburbs”.: An extreme close-up of an gray-haired man with a beard in his 60s, he is deep in thought pondering the history of the universe as he sits at a cafe in Paris, his eyes focus on people offscreen as they walk as he sits mostly motionless, he is dressed in a wool coat suit coat with a button-down shirt, he wears a brown beret and glasses and has a very professorial appearance, and the end he offers a subtle closed-mouth smile as if he found the answer to the mystery of life, the lighting is very cinematic with the golden light and the Parisian streets and city in the background, depth of field, cinematic 35mm film» Video 6: Cat in bed together with human: “A cat waking up its sleeping owner demanding breakfast. The owner tries to ignore the cat, but the cat tries new tactics and finally the owner pulls out a secret stash of treats from under the pillow to hold the cat off a little longer» Read more about the new service Sora from OpenAI here. A big step – This is the best I have seen so far of this type of video generation, says Morten Langfeldt Dahlback, who is head of innovation and technology at Faktisk. He believes that OpenAI has taken a big step with Sora. For fact-checkers like him, the technology will probably mean even more work. Here in Norway, videos taken out of context are still the type that make up the lion’s share of the fake videos Faktisk gets in its lap. AI-generated videos are still not very widespread, says Dahlback. But that could turn around if the development continues, he points out. INNOCENT: It’s hard to see how a cute but fake puppy could cause trouble. But the possibilities inherent in the technology are difficult to imagine, point out several of the experts news has spoken to. – The better this technology becomes, the more difficult it will be to see the difference. And the easier it becomes to create videos with AI, the greater the scope of fake videos can be. It can potentially become overwhelming for us and other fact-checkers. But he believes that traditional fact-checking methods still hold up. BUILDING: Sora is good at producing buildings and details, but Dahlback in Faktisk believes that the traditional fact-checking methods still stand up when it comes to revealing fake video material. – I think we have a good starting point. I’m more concerned about the volume of videos that will come. Some of the methods we use to check videos – details on buildings, light in relation to the time of day – these methods are difficult to bypass completely using AI. – But it will be much more resource-intensive to conduct fact-checking, he adds. Good on the face Previous AI generators have often had problems portraying people, especially when it comes to faces and body parts. But right here, Sora is impressively good, Dahlback believes. – It still struggles a bit, but compared to what I’ve seen before, there are much more realistic facial expressions and movements in these videos, he says. HUMAN-LIKE: OpenAI writes that Sora is better than before at portraying human expressions. Dahlback in Faktisk also thinks the video generator is impressive. Dahlback still believes that one must be open to the fact that the videos that have now been released by OpenAI are among the very best examples we will see. – We have to take into account that the videos that have been released now have been trimmed over a long period of time to look as good as possible. Among other things, he points out that the videos OpenAI CEO Sam Altman made live on X on Thursday evening looked significantly less lifelike than the promo videos. It is therefore difficult to say anything about how good it is in practice, Dahlback points out. – Does it give good results all the time or do you have to work a lot with the text commands to get good results? We won’t know this until it is actually tested outside of OpenAI, which has developed it, he says. NSM: – Will be abused New forms of AI were among the threats the National Security Authority (NSM) pointed to in this year’s risk report, which was presented this week. – This technology is guaranteed to be misused, says director of security culture, Roar Thon, in NSM. He has long worked with how people use technology and how this affects safety in society. – We are already seeing a professionalization of crime related to AI and our concern is that new technology will affect even more people and be significantly more targeted, he says. THERE IS SOMETHING WRONG HERE: The video, generated by the AI tool Sora, shows a cat and a person. But if you pay close attention to the arm of the person in bed, you will see that there is something that is not quite right. The fact that video is generated from text also means that the costs of making video go down drastically, Thon points out. – The order does not have to be so advanced, compared to a few years ago where you needed to learn an entire software to create deepfakes. OpenAI writes that they will now test Sora for malicious intent with the help of a so-called Red Team. – But history shows time and time again that despite good intentions, there are forces out there that will constantly find ways around this, says Thon.
ttn-69
– Now it is about to tilt completely – news Norway – Overview of news from various parts of the country
