In April 2023 Greg Brockman took to the TED stage to share the inside story of ChatGPT’s astonishing potential. In a 16-minute talk he showcased upcoming features of the AI tool that many entrepreneurs have been experimenting with and emphasised, “it’s incredibly important that we all become literate in this technology.”
After the talk, which in Brockman’s words covered, “a bit of the future of AI tools, how we teach AIs to follow our intent, and how the tools themselves can help scale our ability to give high-quality feedback” came 14 minutes of questions from TED founder Chris Anderson.
Here, Brockman was probed by Anderson on how everything he shared had been made possible as well as how OpenAI is handling aspects of AI that many people are worried about, which Anderson called “a pandora’s box” of potential problems.
Brockman’s answers hinted at his beliefs and principles, which are presumably woven into the culture of how OpenAI thinks and builds products such as Dall·E and ChatGPT. From Brockman’s responses on the TED 2023 stage, here are some principles that surfaced, for ambitious entrepreneurs to emulate.
1. Get intentional
“We made a lot of deliberate choices from the early days,” said Brockman, who cofounded OpenAI with Sam Altman seven years ago because, “we felt like something interesting was happening in AI and we wanted to help steer it in a positive direction.” The OpenAI approach, explained Brockman, has always been, “to let reality hit you in the face.”
Deal with reality as it is, and work on your mission from there. Let things happen, observe them, draw conclusions and plough on. Become a student of trends, human behaviour and technology and figure out what you can create within that. Brockman said OpenAI is intentional about, “push[ing] to the limits of this technology to really see it in action.”
2. Be prepared to fail
“We tried a lot of things that didn’t work,” said Brockman. Like with many success stories, “you only see the things that did.” For every ChatGPT product that secures 100 million users just two months after launching, there’s hundreds of ideas that didn’t make it out the meeting room, months of cutting losses and moving on, and years of trial and error.
“I don’t think we’re always going to get it right,” said Brockman. But you only need to be right once. Being prepared to fail makes that far more likely, because you take more risks and you know where to draw the line. Plus, failure is nothing to be embarrassed about.
3. Welcome misfits
Brockman believes that a big part of making progress lies in getting, “teams of people who are very different from each other to work together harmoniously.” If everyone is the same, there is no range. The ideas come from the same pool, the worldviews are too similar and the mind-mapping becomes repetitive.
Misfits, rebels and oddballs, each committed to the purpose and prepared to work hard. Strong, supportive management and an underlying feeling that you’re onto something lifechanging. And as Brockman walks through ChatGPT placing a grocery order as well as saving the life of a sick dog, you can see why that might be true.
4. Advance existing knowledge
“We’re all building on the shoulders of giants,” admitted Brockman, pointing to the progress in computers, algorithms and data past and present. OpenAI didn’t invent AI, they just explored it to create a range of products. ChatGPT isn’t the only AI language model, but it was arguably the first to be widely accessible to all.
Engineers work on problems and build on the knowledge of other industries before they make it big, and that’s exactly the message here. See what’s already happening, learn from the work of those that go before you and build right on top. Learn from other people’s mistakes, get a head start, and follow that “ever-increasing up-and-to-the-right trajectory” that Altman swears by.
5. Place big bets
“We always knew we wanted to be a deep learning lab,” said Brockman, but ChatGPT came from an unexpected place. He then told the story of his OpenAI colleague, who created a tool that could predict the next character in Amazon reviews.” When this colleague got a result, the team doubled down on that specific methodology, and that’s when things got interesting. “We knew,” Brockman said, “you gotta scale this thing. You gotta see where it goes.”
Placing big bets is similar to Altman’s method of, “make it easy to take risks,” which he says comes from having, “your basic obligations covered.” OpenAI has been raising funds since 2016, so it was well prepared to tinker and go down rabbit holes. Place bigger bets, win bigger prizes. There’s no proof like ChatGPT.
6. Rebuild old methods
Once OpenAI knew it was onto something with aspects of its tech, Brockman said, “we had to rebuild our entire stack,” adding that, “you have to get every piece of the stack engineered properly.” Paving a successful way forward will require new paths and footholds. Old processes might not be relevant, and the foundations won’t support the growth that’s about to arrive.
Once the new stack is in place, the now-snowballing idea can fully form. In Brockman’s talk, he described the behaviour of ants where, at first, “single ants run around,” and then as you get enough of them together, “you get these ant colonies that show completely emergent, different behaviour.” Rebuilding old methods allows for new results.
7. Scale with caution
“As you scale up, it surprises you,” said Brockman. Anderson then asked about the, “huge risk of something truly terrible emerging.” OpenAI knows the dangers, and Brockman does too. “We think it’s so important to deploy incrementally,” he said. With the ant colonies that act differently at scale in mind, Brockman said they, “take each step as we encounter it,” which gives people ample “time to give input.” They then “figure out how to manage it,” as they add capabilities to their tools and algorithms.
With AI and business in general, you cannot predict everything that will happen with large datasets and multiple moving parts. A small shop might be destroyed by a million customers, but one that already has ten million can handle a 10% increase. Scaling with caution is on the agenda for OpenAI, with Altman confirming that there is currently “no GPT-5 in training.”
Replicate the success of ChatGPT by following the principles that emerged when the cofounder spoke at TED 2023. Get intentional, be prepared to fail, and welcome misfits in your company. Stand on the shoulders of giants so you can place big bets and rebuild your systems as you find potential ways forward. Finally, scale with caution so the whole thing builds in a sustainable way.