TL;DR: OpenAI’s video generation app Sora is shutting down, the first casualty of the company’s reported plan to cull its “side quests.” This also means the end of Disney’s landmark $1 billion equity investment with the AI startup, which had always been a controversial coupling. Sora’s shuttering is a surprising twist after OpenAI spent the past couple years making Hollywood sweat, but it shows the company is serious about focusing its resources on enterprise and coding—for better or worse.

What happened: OpenAI announced this week that it would put an end to its video generation ambitions with the shuttering of Sora and its API. Also on the chopping block is Disney's historic $1 billion deal with OpenAI—in which the AI company could license its characters and other IP to appear in Sora videos (though sources say no money ever changed hands to begin with).

“We respect OpenAI’s decision to exit the video generation business and to shift its priorities elsewhere,” a Disney spokesperson said in an emailed statement to Tech Brew.

This frees up OpenAI’s resources as it reportedly focuses on a forthcoming model codenamed “Spud.” And the Sora team is pivoting to world simulation models for robotics, a common strategic move for video AI companies as this area of AI research heats up.

Sora’s 15 minutes: OpenAI launched Sora as a video generation tool for filmmakers and studios, then later expanded it into a TikTok-like social feed. The AI startup reportedly spent months courting Hollywood studios—despite lots of resistance—trying to get them to use the tool and even holding a film festival to show off Sora’s creations.

So where did it go wrong? Last week, the Wall Street Journal reported that OpenAI is narrowing its focus to enterprise and coding in the face of increasingly heated competition from Anthropic and Google. That means freeing up limited compute resources from “side quests” that don’t contribute to those comparatively lucrative areas—all those clips of cats playing musical instruments or deepfakes of Sam Altman require a huge amount of computing resources, more so than even large language models.

A love-hate relationship: Video AI remains a hot space—it reportedly drew $3 billion in venture funding last year—and companies regularly claim that Hollywood studios do tap their tools without drawing attention to it. But it’s also particularly fraught with legal landmines, which has limited how much studios and other companies are willing to use it.

The Disney deal drew criticism from creative professional unions, especially as it and other entertainment giants remain embroiled in various lawsuits against other AI companies for copyright infringement.

The Sora app social feed also brought content moderation headaches and legal issues that may have created more trouble for OpenAI.

Still, OpenAI competitors like Google and a host of well-funded startups remain focused on selling video AI to creators, filmmakers, marketers, and others.

Bottom line: OpenAI may have decided this “side quest” was not worth the trouble, despite what could have been $1 billion in guaranteed investment from Disney. To put that number in perspective, though, OpenAI just closed a $110 billion round last month, plus an extra $10 billion this week (after the internal announcement of its pivot). The company seems to be seriously betting that enterprise and coding are where the real revenue is as it scrambles to keep up with Anthropic and Google. —PK

Apple Screen Time’s loophole problem

As the parent of a toddler, I’m already dreading the day my daughter asks for a smartphone. The internet has changed so much since millennials like me made our first Myspace pages, and thanks to on-demand streaming services, there are now endless ways to spend time staring at a screen. Today’s Bug Report about Apple Screen Time from Tech Brew reader Gabriel from Seattle, Washington, isn’t exactly calming my fears.

Apple's Screen Time feature is awful. Parents trying to use it to help limit kids' screen time are thwarted regularly, both by kids' abilities to break/glitch it, but even worse, by its lack of granularity. Ideal would be to be able to put actual limits on specific apps, including time of day use, amount of time per day used, etc. While it ostensibly has those, kids seem to jailbreak them with ease.

Further, they're not granular. For example, it'd be great to be able to limit Instagram to X hours per day and only in use between the hours of Y & Z, again A & B, etc. This way parents could, say, remove it during school hours and at a certain time in the evening for the rest of the night but allow kids more free use in other hours. The same should be true for all apps. The fact you can't do that is irresponsible of Apple.

In the early 2000s, dial-up internet essentially imposed time restrictions on my computer use (couldn’t tie up the phone line for too long). And with just cable TV, it was hard to binge-watch a whole season of a series in one sitting (unless a marathon of America’s Next Top Model was airing, of course). So if Apple can’t outsmart a kid with wi-fi 20-plus years later, then I might have to time-travel back to 2003 and raise my daughter there instead. —CM

Sometimes AI slop is obvious—like when you see a video of a golden retriever playing chess or an anthropomorphized talking strawberry going through a bitter domestic conflict.

But there's a subtler strain that’s probably all over your feed: AI-generated recipes that might pass the smell test on a quick skim until they call for half a cup of cinnamon in your stir-fry. Yesterday, The Cut published the results of a weeklong taste test of some of the AI Frankenrecipes that have taken over social media—and the results mostly ranged from bad to inedible. Many of the meals looked nothing like the pictures (though, to be fair, that happens with human-crafted recipes too) and also didn’t have the consistency you’d expect—like garlic breadsticks that felt more like a rubbery sheet of eggs. (That same garlic breadsticks recipe claimed garlic was “optional.”) The only dish that actually tasted good, receiving four out of five thumbs, was the spicy buffalo chickpea wrap—which The Cut suspects was plagiarized almost verbatim from human-created recipe blog Minimalist Baker.

It’s no mystery why AI slopcipes have taken over the internet. Recipes are among the most voraciously consumed content online, and AI makes it trivially simple to churn them out for easy views. Meanwhile, some independent food bloggers have reported traffic drops of up to 80% (and believe it’s due to AI), according to a Fortune report. AI models can now write a fully functioning app based on a quick back-of-the-napkin sketch, but bot-made recipes are a reminder that they don’t have the taste buds to know if a dish could use more butter (the answer is yes)—or the impulse to include a sentimental, 800-word preamble on how this meal reminds them of grandma. —WK

Chaos Brewing Meter: /5