OpenAI’s Sora, a forthcoming service generating one-minute videos from text prompts, showcases remarkable vividness and imagination. While unlikely to replace Hollywood, it heralds a surge in short user-generated content, posing a threat to traditional TV. The impact on advertising, especially internet ads, is anticipated to be transformative, offering targeted and creative campaigns. Sora’s use of synthetic data could revolutionise AI learning, challenging Meta and Google’s data dominance. The acceleration of image-focused AI like Sora raises intriguing possibilities for scientific modelling and prompts contemplation about our reality being a simulation.
Sign up for your early morning brew of the BizNews Insider to keep you up to speed with the content that matters. The newsletter will land in your inbox at 5:30am weekdays. Register here.
By Tyler Cowen
Sora, a new service from OpenAI that produces one-minute videos in response to a textual prompt, isn’t yet available to the public. But the videos it has released are striking for their vividness, their detail and, if this is the correct term for an AI, their imagination. ___STEADY_PAYWALL___
It is worth considering the future economic consequences of this development.
First, Sora is unlikely to put Hollywood out of business. Eventually these videos will get much longer, but it remains to be seen how well AI can construct long story arcs and integrate them with images into a commercially appealing package. That still seems a long way off, and cost is an additional consideration.
The more clear and present danger to Hollywood is that would-be viewers might start making their own short videos rather than watching television. “Show my pet dog Fido flying to Mars and building a space colony there” is perhaps more fun than many a TV show.
Sora and comparable services will lead to a proliferation of short educational videos, internal corporate training videos, and just plain fooling around. Sora probably will be good for TikTok and other short video services. It is not hard to imagine services that splice your Sora-constructed videos into your TikTok productions. So if you’re doing BookTok, for example, maybe you put a battle reenactment in the background of your plug for your new book on the US Civil War.
Perhaps the most significant short-run use of these videos will be for advertising — especially internet advertising. Again, there is the question of how to integrate narrative, but the costs of creating new ads is likely to fall.
More advertising may sound like a mixed blessing. But ads will almost certainly be more fun and creative than they are now. Watching ads may become its own aesthetic avocation, as is already the case for Super Bowl ads. These ads also might be targeted, rather than serving a mass audience. If your internet history suggests you are interested in UAPs, for example, perhaps you will see ads with aliens telling you which soap to buy.
Some other implications of Sora are more subtle and more speculative. There is the notion that Sora has been trained on “synthetic data,” for example, which in this context means other videos created by AIs, rather than videos of real life. To the extent that is true, future AI progress will be to some extent liberated from constraints of data. AIs will be in essence able to teach themselves, which would accelerate AI progress even more.
To the extent synthetic data proves important, it may weaken the moats of Meta and Google, which have access to significant stores of data through Facebook, Instagram, YouTube and other services. That would give startups and smaller AI companies more of a competitive chance. On the other hand, the rising demand for computation — to produce and work with the synthetic data — would strengthen the market valuation of Nvidia and other companies selling the proper kinds of chips.
It’s also interesting to think of Sora as a new way of modeling physical systems. Its presentation of physics and geometry is by no means perfect (though likely to improve), but it is already good enough to fool the typical human eye. Whatever methods were used to model the physical systems of our world seem to apply quite generally, noting that OpenAI has not explained how it created Sora. In any case, those methods may someday contribute to building better models in physics, chemistry and engineering, and could accelerate scientific progress as well as economic productivity.
Finally, Sora is yet more evidence that, in AI, progress in images is proceeding more rapidly than progress in text. GPT-4 has now been out for well over a year, and it remains to be seen how much better its successors will be. In that same period of time, image and video production has made amazing strides. It was not so long ago that people used to complain that AI image services could not portray human hands convincingly. If progress in images continues to outpace progress in text, then maybe that old saying has it backwards: A word is worth 1,000 pictures. Or, to put another way, words are more truly “human” than images.
At the most speculative level, the success of Sora may increase the chance that we are living in a simulation — a computer-based world created by some high-powered being, whether a deity or aliens. Is that bullish or bearish for asset prices? It depends on how you assess the responsibility and ethics of the creator. At the very least, our planet Earth simulator seems to be able to generate videos that last longer than a single minute. Beyond that, I cannot say.
Read also:
- 🔒 Sam Altman’s rapid return to OpenAI strengthens Microsoft’s position in AI race: Parmy Olson
- Sam Altman returns as CEO after OpenAI’s roller-coaster ride
- From video games to AI dominance with ChatGPT – How Nvidia joined the magic $1 trillion club
© 2024 Bloomberg L.P.
