Gabriel Petersson is 23 years old, has no university degree, and works as a research scientist at OpenAI. On the Sora team, specifically, the video generation project that's been getting all the attention. He joined in December 2024.
The story behind how he got there is either inspiring or concerning, depending on how you feel about the traditional education system.
The unconventional path
Petersson dropped out of high school in Sweden in 2019 to join a YC-backed startup called Depict.ai. He was 17. The startup was building product recommendation systems using computer vision and NLP. He didn't know how to code when he started. Learned on the job.
After that came Dataland (another YC company), then Midjourney, where he built internal tooling and what he describes as the world's most performant web image grid. There's a demo on his GitHub if you're curious about the technical claims.
And then OpenAI. No PhD, no master's, not even a bachelor's.
"Recursive knowledge-filling"
In a podcast interview on November 28, 2025, Petersson explained his learning approach. He calls it "recursive knowledge-filling," which sounds fancy but is actually pretty simple.
He starts with something he doesn't understand. Say, diffusion models. Instead of opening a textbook and working from fundamentals up, he asks ChatGPT to help him build a small video generation demo. He looks at the code. Doesn't understand most of it. Then he starts asking questions.
What does this block do? That's a ResNet. Why does it help the model learn? Okay, but what's the math behind that? Can you explain it like I'm 12?
He copies the code, runs it. Gets errors. Screenshots them, sends them back to ChatGPT. Repeat until it works.
The key insight, he says: you need to be able to notice when you don't actually understand something. Most people skim past that feeling. He leans into it.
The visa problem
Here's a detail I hadn't considered. How does a 23-year-old Swedish high school dropout get a US work visa?
The H-1B requires a college degree. That was out. So Petersson went for the O-1, which is for people with "extraordinary ability." His evidence: Stack Overflow contributions that had been viewed by millions of people, plus recommendation letters from industry contacts.
He worked with a company called Plymouth that specializes in visa cases like his. The Stack Overflow answers, of all things, qualified as "academic contributions." Convenient interpretation, but apparently it worked.
The broader pattern
Petersson's story fits a moment. Sam Altman, who dropped out of Stanford, said at DevDay in October 2025 that he's "envious of the current generation of 20-year-old dropouts" because of how much they can build with current tools.
And then there's Palantir. In April 2025, they launched something called the Meritocracy Fellowship, a four-month paid internship for high school graduates who choose not to go to college. You need SAT scores above 1460 to apply. 22 students made the cut out of 500+ applicants. According to Fortune, fellows who "excelled" will get interviews for permanent positions.
CEO Alex Karp's pitch: "Skip the debt. Skip the indoctrination. Get the Palantir degree."
So we're seeing major tech companies actively recruit teenagers while simultaneously criticizing universities for failing to prepare students for real work. Make of that what you will.
What he actually does
Petersson works on Sora 2, OpenAI's video and audio generation model. According to his own description, his daily workflow looks something like this: observe generated videos, identify problems, use GPT-4 to analyze potential architectural issues, read and rewrite core module code, debug with ChatGPT, retrain, evaluate.
He was credited on the Sora 2 release page. Some of the demo videos featured his likeness.
But I want to flag something. Sora handles billions of parameters and requires coordination across multiple modules. This isn't hobby-level stuff. The gap between "taught myself with ChatGPT" and "doing research at one of the top AI labs" is enormous. Petersson himself doesn't claim the path was easy. In the podcast, he talks about sleeping on sofa cushions in his cousin's student apartment for a year, staying up all night, having what he calls a "distorted perception of reality."
The survivorship bias question
The uncomfortable part of stories like this: they get amplified precisely because they're unusual. Petersson made it. Thousands of other dropouts who tried the same approach didn't, and we don't hear about them.
When I searched for follow-up data on what happens to tech dropouts who don't land at OpenAI, I couldn't find much. One Harvard researcher cited in coverage of the Palantir fellowship estimated that fewer than 1 in 700 workers actually benefits when degree requirements are dropped.
Petersson's advice, skip HR and write directly to company executives with an MVP and an offer to work free for a week, works great if you're unusually talented and unusually lucky. For everyone else?
The article I pulled the Russian-language source from positions this as a story about how "universities have lost their monopoly on knowledge." That's true in some sense. But universities provide more than knowledge access. They provide credentials, structure, networking, and, let's be honest, a safety net for failure.
What happens next
Petersson is credited on the Sora team and seems to be doing well at OpenAI. No indication he's going anywhere.
The larger trend to watch: whether more companies follow Palantir's lead in explicitly recruiting teenagers over college graduates. The first cohort of Meritocracy Fellows wrapped up in November 2025. If any of them get hired into permanent engineering roles and perform well, expect more programs like this.
For now, Petersson's story remains an outlier. An interesting one.
