Why OpenAI shut down the Sora App

I have a surprising amount of food-related analogies for generative AI, something that definitely says more about me than about AI…

One I come back to a lot is my experience at a Buddhist vegetarian restaurant in Beijing more than two decades ago. Many of their dishes were tofu recreations of meat, fish, or fowl. So they had this tofu salmon that looked and felt exactly like steamed salmon. For the first three chews, at least. Then it became chewed tofu. The same went for the sliced pork belly with fat, or the giant prawns, or the… you get the idea.

Everything went from "Wow, amazing!" to "Okay, it's tofu" in no time. A great party trick that drew a large crowd, but if you ever go, do yourself a favour and focus on their actual vegetarian dishes.

You can probably see where I'm going with this with respect to AI-generated videos. Great party tricks that quickly turn into a samey experience that leaves you hankering for the real deal.

I mean, do you even remember TCL's push into AI-generated content? They were planning to do a bunch of short videos that would play on their own TV channel, and then companies would buy ads to run alongside them. The videos were really really bad, just like the stuff posted on Sora.

But that's not why I think Sora didn't work as a standalone app. It definitely contributed, but slop content has been around since long before it could be easily generated straight in the app you were using to view the slop, and that hasn't stopped anyone.

If we disregard the obvious fact that Sora lost OpenAI way too much money, and pretend that they had a working ad business that could start paying for the astronomical cost of generating a video of a moonwalking tortoise that overtakes a sleeping hare on a hoverboard, I still don't think it could have become a success.

On every other content-pushing platform most people just consume. On these AI-driven platforms I think it's the reverse of the 90:10 rule. So rather than 10% of users making content for the platform, it becomes closer to 90%. And they all expect to make a buck or at the very least become famous. Which obviously can't work.

So we have a perfect mix of things working against the idea of an AI-driven content platform:

1. The skill and effort it takes to generate slop is closing in on zero.
2. But it's still really hard to make something worth watching.
3. 1 plus 2 means that if anyone succeeds in making something that catches on, the carbon copies will flood the platform in the blink of an eye.
4. Which in turn means that more and more people will be competing for the same attention, each getting a smaller and smaller share.
5. And all of that is just driving up the enormous costs of providing people with the platform.
6. The main reason to be on the platform is to make money somehow (because, as point 2 says, the content itself will be slop) rather than creating something worthwhile. As 3 says, spending time on creating content is for losers. Rapidly churning out slight variations of the top-trending slop is the smart play.

That is why the Sora app was doomed from the beginning.

What I'm NOT saying

I'm not saying there aren't uses for generative AI in video editing or special effects. I've seen Karen X's stuff , and I think it's great. But she is working her ass off, and knows how to do cool effects even without generative algorithms. She shares most (all?) of her processes so go ahead and see for yourself.

What I am saying is that a platform full of instantly generated slop has longevity on par with tofu's ability to resemble salmon. Sora user numbers agree.

We don't gather in huge crowds to watch people commute to work. We want something extraordinary. Which is why Karen X has 1.3 million followers on Instagram whilst the TCL slop channel on YouTube has less than one thousandth of that.