Sora is Mad Libs. Would a Mad Libs Festival make sense?

Is there a future for software on an "events" business model?

I’ve been spending a lot of time on Sora, a new app / network from OpenAI that lets users generate videos of themselves and others from text prompts. Nevertheless, I’m confident it has no future. There are good reasons for this, but I also have the advantage of having already seen how a product like this works, observing the extremely well-executed Can of Soup, created by my friend and former Substack designer Gabriel Birnbaum. Can of Soup was an app that worked much as Sora does:

1. Give the app your face (photos for CoS; video for Sora)

2. Connect with your friends

3. Prompt the app to create media of you and/or your friends in whatever kinds of scenarios

4. Scroll a feed of such media

In my opinion, Can of Soup did literally every part of this extremely well, and had a level of fit and finish that Sora certainly does not, yet. But both apps face the same challenges: first, that users only care about generative model content about themselves and/or their friends, and second, that the joke(s) get old, the possibility spaces gets exhausted, and people run out of reasons to keep at it.

This is hard for people who think very reductively about human behavior to understand. After all: wasn’t e.g. Instagram just photos of you and your friends? Why wouldn’t images of you and your friends in a literally limitless variety of scenes be similarly engaging? Isn’t it true that “once ‘AI’ is good enough, it’s all indistinguishable,” and won’t models then be better at generating “interesting images” than real life is? The answer is: no, of course not, and any regular people know it. But it’s beyond or perhaps beneath this essay to explain why reality matters, why humans do not consider the provenance of “content” to be immaterial, why some things simply aren’t fungible, the nature of life and culture, or what it is exactly that we’re interested in when we’re interested in other people (and photos and videos of them). It suffices to observe that this is so, and to wait for the eulogies for Sora, and Meta AI, and any other slop-centric feed apps to confirm it for good.

In fairness, this issue is misunderstood in part because there is a novelty effect of incredible intensity and durability for such apps. Sora is really, really, really funny, provided you and your friends are all on there, and I have been opening it a ton. Talking with David Cole and randa, I feel confident in a few things:

• Sora is like Mad Libs, the old game where you fill in blanks with words and then read back to your friends (or siblings) a resulting story. Especially for children, Mad Libs is insanely funny; there’s the absurdity of the sentences and narratives that result from random diction, of course, but also a strange psychological effect that comes from interfacing with your pal’s imagination, as well as from the asymmetric information arrangement and turn-taking. I might have never laughed harder than when playing Mad Libs, but I am confident I would never laugh at all reading the Mad Libs of strangers! Similarly, I watch my friends’ Sora videos and laugh until I cry, but never watch or share those of strangers, and cannot imagine doing so.

• Sora has at its core, again like Mad Libs, a non-negotiable randomness that means it cannot be “taken seriously.” This gives permission to players —users, but “players” may be the better term— to toy with norms, ignore sensitivities, abandon ego or vanity or insecurity, and just absolutely clown around without any sense of responsibility to e.g. “be good” or “present oneself in some way or another.” In the context of the ultra-boring wider Internet of 2025, it’s liberating. There is no such thing as a “serious” Sora video, as the process of making them is fundamentally unreliable (and incompetent). Ironically, improving the model will make this worse; if Sora is ever even remotely reliably good at allowing creators to make videos as they describe them, the one appealing thing about it will vanish and what remains will be of no enduring interest.

• Sadly, even Mad Libs eventually gets old: one tries all the categories of “being funny” or “not being funny” in supplying words; one gets accustomed to the various shapes of “ridiculous outcomes”; one grows bored with the mechanical result of the game. Sora is not presented as a game, while Mad Libs of course was, but they’re both games. They each have a game-feel, you might say, and once you have the game-feel down pat, only phenomena like competition or the desire for mastery could continue to engage you; and neither has any plausible chance at being competitive or rewarding mastery, I don’t think.

For OpenAI, this might be a problem. I don’t know this, but I suspect after hiring hundreds of top former Meta people, OpenAI is trying to use network effects and time spent mechanisms to build an advertising business: they hope that if their LLMs end every answer with a follow-up question, and their apps keep billions of people scrolling, they can insert ads, charge for placements, and hopefully justify their valuation before commodification in the generative model space leaves them exposed to margin-destroying competition from other companies and even open-source. Perhaps they can pull this off, but I doubt Sora will be part of them doing so; I think the shine, such as it is, will come off fast.

It makes me wonder about something I used to ponder when looking at apps made by Nikita Bier: is “building a sustainable business with software” really the right model for all social experience developers to pursue? Because of the venture-funding → acquisition pattern, it often works to cobble together something fun, novel, sticky, but empty, because if you flip it fast enough, founders can get rich and move on to other opportunities before the crash comes. But unless Apple buys them, OpenAI probably can’t get acquired anymore, so a fun, novel, sticky, empty experience will simply cost them a lot of money (unless they can trick more people into yet another round of investment!). Something like Sora seems less like a “new platform we can build an ad business on” than “a new movie everyone is going to see,” or perhaps “a big music festival many want to be a part of,” or some other ephemeral, but often lucrative, type of moment.

I imagine historically software was too expensive for this model, but with LLM-assisted coding, I wonder if in some number of years, software-as-event will be viable. If Sora had had minimal development costs —which it emphatically did not— it could probably turn a profit with a pop of usage that then declined, much as major blockbusters do. This is of course how video games worked for much of their history, although they often now seek to retain and monetize users indefinitely as well.

In any event, I could be wrong about literally all of this. Perhaps Sora will be part of the cultural landscape for decades! Maybe it’s all a ploy to “get more training data,” or some such strategic cope. But as with so many products in this category, it seems likelier to illustrate the limits of slop’s appeal than to displace any real cultural activity.¹ Mad Libs was big for a while, but I don’t recall it hurting book or newspaper sales, and even though I hang out with a lot of nerds, I can’t remember the last time someone suggested playing it.

Helping OpenAI train its models with the finest scenarios imaginable.

1. I should note that Sora does have some strange psychological effects. After being in a lot of videos with Randa, e.g., I think of us as having known one another for longer and in more contexts than we have; this has happened with a few people I’m in videos with, and I assume some number of people, especially young people, will get all screwed up by this technology in this and other ways.