The Most Hyped Bot Since ChatGPT

For more than two years, every new AI announcement has lived in the shadow of ChatGPT. No model from any company has eclipsed or matched that initial fever. But perhaps the closest any firm has come to replicating the buzz was this past February, when OpenAI first teased its video-generating AI model, Sora. Tantalizing clips—woolly mammoths kicking up clouds of snow, Pixar-esque animations of adorable fluffy critters—promised a stunning future, one in which anyone can whip up high-quality clips by typing simple text prompts into a computer program.

But Sora, which was not immediately available to the public, remained just that: a teaser. Pressure on OpenAI has mounted. In the intervening months, several other major tech companies, including Meta, Google, and Amazon, have showcased video-generating models of their own. Today, OpenAI finally responded. “This is a launch we’ve been excited for for a long time,” the start-up’s CEO, Sam Altman, said in an announcement video. “We’re going to launch Sora, our video product.”

In the announcement, the company said that paid subscribers to ChatGPT in the United States and several other countries will be able to use Sora to generate videos of their own. Unlike other tech companies’ video-generating models, which remain previews or are available solely through enterprise cloud platforms, Sora is the first video-generating product that a major tech company is placing directly in users’ hands. Chatbots and image generators such as OpenAI’s DALL-E have already made it effortless for anybody to create and share detailed content in just a few seconds—threatening entire industries and precipitating deep changes in communication online. Now the era of video-generating AI models will make those shifts only more profound, rapid, and bizarre.

OpenAI’s key word this afternoon was product. The company is billing Sora not as a research breakthrough but as a consumer experience—part of the company’s ongoing commercial lurch. At its founding, in 2015, OpenAI was a nonprofit with a mission to build digital intelligence “to benefit humanity as a whole, unconstrained by a need to generate financial return.” Today, it pumps out products and business deals like any other tech company chasing revenue. OpenAI added a for-profit arm in 2019, and as of September, it is reportedly considering revoking the control of its nonprofit board entirely. Sora’s marketing is even a change from February, when OpenAI presented the video-generating model as a step toward the company’s lofty mission of creating technology more intelligent than humans. Bill Peebles, one of Sora’s lead researchers, told me in May that video would enable “a couple of avenues to AGI,” or artificial general intelligence, by allowing the company’s programs to simulate physics and even human thoughts. To generate a video of a football game, Sora might need to model both aerodynamics and players’ psychology.

Today’s announcement, meanwhile, was preceded by a review by Marques Brownlee, a YouTuber famous for his reviews of gadgets such as iPhones and virtual-reality headsets. Altman wore a hoodie emblazoned with the word Sora. Altman and the Sora product team spoke for more than 17 minutes; Peebles and another researcher spoke for one minute and 45 seconds, mostly lauding how the company is launching a “turbo” version of Sora that is “way faster and cheaper” in order to launch a “new product experience.”

The Sora release comes on the third of “12 Days of OpenAI,” a stretch of releasing or demoing a new product to users every day. What the company has announced certainly resembles a product more than a computer-science breakthrough: a sleek interface for creating and editing videos, with features such as “Remix,” “Loop,” and “Blend.” So far, many of Sora’s outputs have been impressive, even wonder-inducing. The company hasn’t built a new, more intelligent bot so much as an interface in the style of iMovie and Premiere Pro.

Already, videos that OpenAI staff and early-access users generated with Sora are trickling onto social media, and a deluge from users the world over will follow. For more than two years, cheap and easy-to-use generative-AI models have turned everybody into a potential illustrator; soon, anybody might become an animator as well. That poses an obvious threat for human illustrators and animators, many of whom have long been sounding the alarm against generative AI taking their livelihood. Sora and similar programs also raise the specter of disinformation campaigns. (Sora videos come with a visual watermark, but with OpenAI’s highest tier of subscription, which costs $200 a month, customers can create clips without one.)

But job displacement and disinformation may not be the most immediate or significant consequences of the Third Day of OpenAI. Both were happening without Sora, even if the program accelerates each problem: Production studios were already experimenting with enterprise AI products to generate videos, such as a recent Coca-Cola holiday commercial. And cheap, lower-tech methods of creating and disseminating false information have been extremely successful on their own.

What the mass adoption of video-generating AI products could meaningfully change is how people express themselves online. Over the past year, AI-generated memes, cartoons, caricatures, and other images, sometimes called “slop,” have saturated the internet. This content, much of it clearly generated by AI rather than intended to deceive—a medium of crude self-expression, not sophisticated subterfuge—may have been the technology’s biggest impact on the 2024 presidential election. That anybody can generate such images provides a way to immediately express inchoate feelings about an inchoate world through an immediately digestible image. As my colleague Charlie Warzel has written, such content is meant to be consumed “fleetingly, and with little or no thought beyond the initial limbic-system response.”

A flood of AI-generated videos might provide still more powerful ways to visually communicate confusion, charged feelings, or persuasive propaganda—perhaps a much more lifelike version of the recent, low-quality AI-generated video of Donald Trump and Jill Biden in a fistfight, for instance. Sora might take over TikTok and similar short-form-video platforms just as AI image-generating models have warped Facebook and altered how people show support on X for political candidates.

Sora’s takeover of the web is not guaranteed. Back in May, Tim Brooks, another Sora researcher who has since joined Google, likened the program’s current state to GPT-1, the earliest version of the programs underlying ChatGPT, which are currently in their fourth generation. OpenAI repeated the analogy today. That comparison has broken down as the company has become more and more profit-driven: GPT-1 was highly preliminary research, a concept before a proof of concept, and four years removed from the release of ChatGPT. Sora might be just as undeveloped as an avenue for AGI, but it has become a full-fledged product nearly 10 months after OpenAI teased the model. Such early-stage technology might not mark significant progress toward curing cancer, solving the climate crisis, or other ways the start-up has claimed AI might benefit humanity as a whole. But it might be all that OpenAI needs to boost its bottom line.