Sora: What Educators Need to Know About The Video Version of ChatGPT
If you ever wondered what would happen if ChatGPT could create videos, you’re about to get your answer.
Sora is a new AI video-generation tool recently announced by OpenAI, the Microsoft-backed maker of ChatGPT. Though currently only available to cybersecurity experts who are testing it, Sora is expected to publicly launch soon.
While AI video generators already exist and are in use, the samples of Sora-created video released so far show a level of photorealism not previously seen. In addition, OpenAI has emerged as a leader in the AI field and its entrance into generative video could have big implications for society as well as for those of us tasked with guiding our students through this brave new world of computer-generated text, imagery, and now, movies.
As an educator, I can see many ways in which this technology can be helpful to my students in addition to the potential pitfalls and concerns. How this will all play out, of course, remains to be seen, but here’s what we know about Sora and its implications for teachers so far.
What is Sora?
Sora is a new text-to-video model from OpenAI currently being tested for safety by security experts before being made available to the general public. Educators, in particular, will need to remember not to confuse it with Overdrive’s K12 digital library tool of the same name.
“We’re teaching AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction,” OpenAI said in its post introducing Sora. “Sora can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt.”
Sora is powered by a combination of the diffusion model used by OpenAI’s Dalle-3 image generator as well elements of the GPT-4 model, Wired reports. The examples OpenAI has shared so far, are pretty stunning.
In one video created by a one-paragraph prompt, a woman walks down a rain-drenched Tokyo street awash in neon. In another, the prompt “historic footage of California during the gold rush,” led to a drone-style shot of a realistic-looking Western town with horses and people walking on the dusty streets.
“Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background,” notes OpenAI. “The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.”
What Are Sora’s Limitations?
Sora is far from perfect, however. Like all AI models, it can make mistakes or "hallucinations" that can be more uncanny and strange given how well they do other things. “It may struggle with accurately simulating the physics of a complex scene, and may not understand specific instances of cause and effect,” OpenAI notes. “For example, a person might take a bite out of a cookie, but afterward, the cookie may not have a bite mark.”
In another example shared by OpenAI, a video of man running on a treadmill is facing the wrong way. This is obviously a mistake, but it looks kind of cool. The clip reminds me of Maurits Cornelis Escher’s work and is one of the more interesting videos from Sora that OpenAI has shared so far.
How Much Will Sora Cost?
OpenAI has not released any details on pricing. But based on other AI tools it has released, it appears unlikely that full access will be free; however, that remains to be seen.
What Are Some Implications for Teachers?
Let’s focus on the positive first: There are many potential teaching applications for Sora.
As a writing professor, it could be fun to use this tool with students. I can see this as an engaging, if slightly gimmicky, way to help fiction-writing students visualize scenes. By requiring focused prompts, it can also help anyone learn to write with clear and precise language.
Teachers in other topics can find fun uses as well. For instance, I’d love to see how it models photosynthesis or the motion of the planets. Does it provide an accurate visual physics lesson or can your students point out the flaws?
Now about concerns. We all know AI has a tendency to “hallucinate” and can amplify the bias of those who created it in unintentional and unpredictable ways. This is concerning with text- and image-based AI but the stakes are even further heightened with video.
OpenAI seems concerned about some of these problems and says it is working to check for bias, and will provide an identifying tag on videos created with the tool to combat misinformation. Additionally, prompts that request extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others, will be rejected.
But given the challenges of existing AI technology, it’s hard to imagine some level of bias not slipping by these safeguards. Google, one of OpenAI’s rivals, for instance, recently announced it was pausing its AI portrait-generating tool after reports of historic inaccuracies, including reportedly depicting Nazis as people of color.
Even if OpenAI can navigate these obvious instances of inaccuracies and bias, there are types of bias that might be inherently unavoidable. What if a student asks Sora to create a beautiful person walking down a street -- what standard of beauty will the model use?
As with other AI tools, rather than ignore these challenges and concerns, I’d advise instructors to embrace them and discuss it all with their students. AI is a fascinating, fun, powerful and also often scary new technology. Let’s learn how to navigate it with our students.
