On a quiet block in San Francisco, where a delivery driver might still struggle to find the right door, a small group of teenage and college-age students are sitting in a loose circle. Some are scribbling equations into worn notebooks; a few are hunched over laptops running dozens of experiments at once; others are staring at the ceiling, which, if you know anything about serious mathematical thinking, is not a sign of boredom but of deep thought. There are no progress bars, no congratulatory pop-ups, no “You are 12% complete!” pings. It looks less like an online course than like a monastery that happens to have excellent Wi-Fi. This is the strange and human new world of Eren Bali, one of the cofounders of Udemy, the global online education marketplace that taught millions of people through the internet. After spending more than a decade trying to make learning available to everyone, Bali has pivoted back to the oldest kind of education there is: in-person, communal, apprenticeship-style learning, gathered inside a new and deliberately small academy for what he calls AI phenoms. It may sound like a giant U-turn, but in a way it is the natural conclusion of everything he learned about why most online courses fail and why the most brilliant young minds are still starving at the exact moment they seem most successful.
Bali’s own story explains why this matters so much. He grew up in a small town in Turkey, far away from the great technology centers, and in that way, he has said, the internet changed his life. He taught himself to code from free resources and eventually cofounded Udemy with the idealistic belief that the world’s best teachers could be put on a platform and shared with anyone with a laptop. Udemy did grow enormously, and it did help a lot of people. But the longer Bali worked in online education, the more he began to notice a serious flaw: access is not the same as connection. People would sign up by the millions, but far too many would drift away. They would watch a lecture, then another, then quietly disappear. They were not failing because they lacked talent. They were failing because learning, especially deep learning, is not just the transfer of information. It is a social event. It requires stumbling, asking a question at the right moment, being seen by a teacher who knows your face, and sitting in a room with other people who are equally confused and equally excited. Those things are very hard to replicate in a video stream. At the same time, Bali began seeing a new kind of learner emerging all over the world, especially in artificial intelligence. These were young people who had taught themselves deep learning from papers, who had built projects that would impress working researchers, who were incredibly bright, stubborn and creative. And many of them were desperately lonely. They lived in Discord servers and notebook environments. They had never had a mentor who could look them in the eye. They were being raised by algorithms, and an algorithm can teach you a lot, but it cannot teach you how to belong.
That gap is exactly what the new academy is designed to fill. It is intentionally small. There are no giant lecture halls, no standardized semesters, no multiple-choice exams. The model is more like an old-fashioned guild crossed with a modern research lab. Students do not just study artificial intelligence; they live it. They work on real problems that matter to them, side by side with researchers and practitioners who have spent years in the field. The curriculum, if you can call it that, is not a list of modules but an ongoing sequence of impossibly difficult questions. A student might start the morning by reading a new paper no one fully understands yet, spend the afternoon breaking it apart and trying to rebuild a small version of it, then spend the evening arguing with classmates about what it means. There are no grades to chase, because the only real test is whether you can build something that works and explain why it works to people who care just as much as you do. Admission is not about credentials. It is not about how many prizes you have won or how well you scored on a standardized test. It is about curiosity, stamina, and the strange willingness to sit with a hard problem until it starts to reveal itself. The academy also tries to remove financial barriers. It wants to take in students who may not have had access to elite universities or expensive tutors, the kids with extraordinary minds who got lucky enough to find an electricity outlet and a library card. The hope is to gather a group of people who are weird in all the right ways, and then to give them something that cannot be downloaded: a shared life.
To outsiders, the idea of a school for AI prodigies can sound like an elite fantasy, a club for child geniuses. But the deeper truth is that these phenoms are often not fine at all. A lot of them have spent their childhood being told they are too intense, too weird, too difficult. They have learned to hide the parts of themselves that are most alive because those are the parts that make other people uncomfortable. Some have been diagnosed as disruptive when they were actually just bored. Some have been praised as geniuses, which is often just another way of saying they are isolated. Being exceptionally good at something can feel a lot like being exceptionally alone. It is not uncommon for an AI phenom to be able to explain the internals of a transformer model but completely unable to feel at ease in a group of ordinary teenagers. The academy is not just a place to make these young people smarter; it is a place to let them exhale. It turns out that when you put a roomful of them together, something almost magical happens. They look at each other and realize they are not aliens after all. They begin to share not only code but jokes, snacks, late-night anxieties, and the tiny micro-wars over who gets the good whiteboard markers. They learn to fail in front of each other. They learn to ask for help. They learn that there is a difference between being brilliant and being invulnerable. Their minds are incredibly strong, but their hearts still need the same things every human being needs: acceptance, warmth, and a reason to leave their desk at three in the morning.
The “in-person” part of this story is not just a nice theme. It is the entire point. Artificial intelligence is often imagined as a solitary sport, one person alone in a dark room talking to a glowing machine, but the actual history of AI is full of messy human rooms, long whiteboards, furious arguments, and sudden breakthroughs at odd hours. The most important research happens not when someone is alone but when two people are staring at the same mistake and one of them says, “Wait, what if we try…” There are certain kinds of knowledge that simply do not transmit through a video call. There is knowledge in the way a mentor pauses before answering, in the way a peer’s face lights up when they finally understand something. There is courage in being in the same physical space as someone who is not afraid to be wrong. There is also something deeply important about learning to work with people who are not like you. An online group chat can filter out body language, tone, and context. In person, you have to learn to read a room, to recognize when someone is tired, to make peace and move forward. Those skills are not secondary; they are the heart of real work and real life. For AI phenoms, who may have spent too many of their formative years safely behind screens, the academy is partly about reminding them that they have bodies, that they exist in the world, and that the world itself is full of problems they cannot solve by taking gradients alone. The most powerful algorithm in the building is still the ordinary chemistry of people who trust each other enough to say “I’m stuck” out loud.
In the end, Bali’s pivot from online education to a tiny in-person academy is a kind of correction, not a betrayal of his old ideas. The internet did not fail. It opened doors, and for many people around the world it remains the only door they have. But an open door is not the same as a warm home. The contents of a library are not the same as a friend who sits beside you and wonders with you. Bali built Udemy to tear down the walls around education, and he succeeded. But the new academy suggests that tearing down walls was always only half the job. The other half is building rooms where people want to stay, communities where being smart is not a reason to be ashamed, and cultures where talent is treated as the start of a relationship rather than a medal. The academy may be small, maybe only dozens of students at a time, but it could become a model for a new kind of elite education that is not built on status or exclusion. It could remind us that the most human thing about education is not the transfer of facts; it is the fact that another human being looked at your work, listened to your confusion, and believed in you before the world did. In the age of artificial intelligence, when almost any answer can be generated in seconds, the scarcest resource in the world will not be information. It will be attention, compassion, and the courage to be present. That is what this academy is really trying to grow. And if it works, the AI phenoms who pass through it will do more than build great models. They will become the kind of people who remember that behind every advance is a room full of humans, asking questions, making mistakes, and learning, at last, not to do it alone.
