He Was Told Deep Learning Would Mean a Bad Career
A decade later, he helped start OpenAI to pursue artificial general intelligence despite being hammered by leading figures in the field and dismissed for betting on large language models.
David Senra · Sam Altman on Building OpenAI & Betting on the ImpossibleIn the summer after his freshman year of college, Sam Altman, the head of artificial intelligence company OpenAI, was working in the campus artificial intelligence lab. Nothing was working.
Altman had arrived at college hoping to study AI. He had been a nerdy kid in St. Louis who spent Friday nights on his computer and reading science fiction, and always thought AI would be the most amazing, craziest thing. In that lab, sometime around 2005, a professor gave him memorable guidance.
"You could try all of these things. There's all these directions. The one thing we know doesn't work is deep learning," the professor said. Deep learning, a method of training large artificial neural networks on vast amounts of data, had been tried for a long time. Pursuing it, the professor told him, was "the most guaranteed way to have a bad career."
Altman believed him. He later described himself as an impressionable freshman, and he assumed the verdict was true. He set AI aside for years and fell almost by accident into startups, first as an investor and then running a company.
That dismissed idea is what he later bet his career on. In 2015, Altman helped start OpenAI to pursue artificial general intelligence, AI that can do the intellectual work a human can do. In that year, there were only a handful of such efforts in the world, and Altman said OpenAI was hammered by "the intellectual giants of the field for saying" that it was going after AGI.
When the young company then focused on large language models, systems trained on enormous troves of text to predict and generate language, critics in general called that approach completely ridiculous, Altman said.
Altman went ahead anyway. His rule from startup investing, Altman said, was that "risk bets are okay as long as you take the ones where if they work, it's super valuable." As a kid, he said, the more impossible something seemed, the more intrigued he was, and the idea of inventing the one technology that would let people do everything else had always appealed to him. So he bet on AGI, and on language models to get there.
Questions people ask
What is deep learning, exactly?
Deep learning trains a large network of simple mathematical units by showing it vast amounts of data and gradually adjusting the connections until it gets good at a task, such as recognizing images or predicting the next word. It is the core method behind modern chatbots, image generators, and speech recognition.
What is AGI?
Artificial general intelligence, or AGI, means an AI system that can do most intellectual work that a person can do, rather than just one narrow job. OpenAI was founded to build it, but no lab says it has fully arrived yet.
Why did so many experts dismiss deep learning and AGI back then?
Progress was slow, funding was scarce, and many leading researchers thought other approaches were more promising. Neural networks needed far more data and computing power than were widely available around 2005, so results were weak and skepticism was rational at the time.