sora/open ai/annotation by npr
Fascist SpongeBob Square, Dog Driving a Car, Jesus Playing Minnecraft – these are just a few of what you can see when flipping through new apps that are open exclusively to short-form videos generated using artificial intelligence.
If you can't find what you're looking for, don't worry. Easy to use using the app's small text-based prompt window. The result is a very addictive stream of sometimes funny, sometimes weird 10-second videos.
Openai released the SORA app on Tuesday a few days after Meta released a similar product to some of the Meta AI platform. NPR watched early and discovered that Openai's app can easily generate very realistic videos with real individuals (with permission). The early results are both researchers and worrying researchers.
“Your friends are saying things they never say, so you can make insanely real videos,” said Solomon Messing, an associate professor at New York University at the Center for Social Media and Politics. “I think we may be in an age where we don't believe in seeing.”
Deepfake Tiktok
The SORA 2 app looks like other vertical video social media apps like Tiktok. It comes with several different settings. For example, you can select videos for each mood. According to Openai, users can control what is being used “end-to-end” with AI-generated videos. This means that users can make their faces available to everyone, a small circle of friends, or only themselves. Additionally, they are permitted to delete videos showing their likeness at any time.
SORA also provides a way to identify content as AI-generated. Videos downloaded from the app include mobile watermarks with the SORA logo, and the files include metadata that identifies them as AI-Made.
Openai says it put GuardRails on what the app can make. A company spokesman has directed NPR to SORA's system cards. This prohibits the generation of content that can be used for “scam, scam, scam, spam, or impersonation.”
“To support enforcement, we provide in-app reports, combine automation and human reviews to detect patterns of misuse, apply violations or delete content when violations occur,” the document reads.
However, the short time in NPR using the app found that the guardrails appear to be somewhat looser around Sora. Many prompts were rejected, but they were able to generate videos that support conspiracy theory. It was easy to create a video that, for example, looked like President Richard Nixon, gave the televised address telling the US that the moon's landing had been forged.
And one of astronaut Neil Armstrong removes his helmet on the moon.
The NPR was also able to generate videos depicting drone attacks on the power plant. That too appeared to violate guidelines regarding violence and (probably) terrorism.
Additionally, the app appeared to contain other loopholes. NPR was able to directly conflict with Openai's global usage policy to produce short videos on topics relating to chemical, biological, radiological and nuclear weapons. (The video created did not share inaccuracies that were useless to those seeking this type of information, and contained inaccuracies.)
Run and clown
It is unclear whether other users have found similar exploits, but quick reviews of the content show that SORA is being used to generate a huge amount of videos depicting trademark brands and copyrighted material. One video depicts Ronald McDonald running away from police in a burger car. Many others included characters from popular comics and video games.
Openai told NPR that he knows about the use of copyrighted materials on Sora, but felt that allowing it gives users freedom.
“People want to interact with their own family and friends through their imaginations, and through their imaginations through the stories, characters, and the world they love, and see new opportunities for creators to connect with their fans.” “We work with the rights holders to block characters from Sora on their requests and respond to takedown requests.”
Openai is currently sued New York Times Due to copyright infringement using its large-scale language model, ChatGpt.
A brave virtual world
The effectiveness of the world of social media, fully driven by AI, remains unknown, Messing said. Many researchers were deeply concerned about “deepfakes” when AI videos first appeared, but the traction of those videos was barely possible. “We all panicked collectively about deepfakes a few years ago, but society isn't very corrupted because of deepfakes,” he said.
But others worry that a collective sense of reality may be beginning to unravel. Sora is the latest in many tools that allow you to generate images, videos and audio at your own whim.
“We really see the ability of the entire blanket to generate very realistic and surreal content in all sorts of different ways you want,” said Henry Ager, director of potential space advisories tracking the evolution of AI-generated content.
Just as he worries about people being fooled, Ajder said he is also very worried about the consequences of not trusting what they see online.
“We have to resist a somewhat nuanced pull. “We can't know the truth anymore, so that doesn't matter anymore,” he said.
Messing said it's unclear what the outcome will turn out, but what's clear is that Sora is very good at creating things she can imagine. “I wasn't really sure how good the content was.”
