Guest Profile & Core Mission

Who is Greg Brockman?

Greg Brockman is the co-founder of OpenAI. He describes himself as a leader who operates "in the trenches," leading from the front. His core mission is ensuring that artificial general intelligence benefits all of humanity, something he considers a life well spent.

What is the primary goal of OpenAI?

OpenAI was founded with the mission to build human-level AI that is a positive force for the world, with benefits distributed broadly. The company has evolved from a pure nonprofit to a for-profit entity, driven by the realization that massive compute resources were necessary to achieve AGI.

The Inside Story of OpenAI's Founding & Key Breakthroughs

Origins & Early Strategy

Greg Brockman was working at Stripe but felt that problem was not his own. When Patrick Collison suggested he talk to Sam Altman, Sam quickly realized Greg had already decided to leave. They connected over a shared interest in AI. At a dinner in July 2015, the group debated whether it was too late to start a lab competing with DeepMind. Since nobody could prove it was impossible, Sam and Greg decided to move forward.

The critical moment was the Napa offsite, where no one had officially joined yet. The team came up with a three-step technical plan that OpenAI has pursued for nearly a decade:

1. Solve reinforcement learning
2. Solve unsupervised learning
3. Gradually learn more complicated things

The initial target team included Ilya Sutskever, Dario Amodei, and others. Dario and Chris ended up deciding to go to Google Brain, leaving Greg, Ilya, and John Schulman as the core starting group.

The Pivot from Nonprofit

By 2017, OpenAI began calculating the compute needed for AGI. They discovered that building massive data centers and acquiring specialized hardware like Cerebras computers was essential. However, nonprofit fundraising has a cap. Elon, Sam, Ilya, and Greg all agreed that the only path forward was creating a for-profit entity. Without this structure, raising the billions of dollars needed would have been impossible.

Key Breakthrough Moments

Greg describes OpenAI's history as "a series of moments where you realize that it's real now":

• Initial launch: Getting the team together felt monumental, but the next day they didn't even have a whiteboard.
• Dota 2: Their first big result demonstrated that scaling compute actually worked.
• Unsupervised Sentiment Neuron (2017): For the first time, they saw semantics arise from training on a language modeling objective. The neural net learned meaning, not just where commas go.
• GPT-4: People asked "Why is this thing not an AGI?" because you could talk to it fluently about anything. If you had described AGI criteria two months prior, GPT-4 would have fit.

Why Dota 2 Mattered

Unlike chess or Go, Dota 2 is interactive against humans in a messy, unstructured environment. The algorithm they used, called PPO, was considered flawed — it plans over every time step without hierarchy. But they kept scaling it, and it exceeded the performance of the best humans. The finding was that massive compute combined with simple algorithms works in practice, not just theory. The neural net was tiny — similar to an insect brain — raising the question of what would happen at human brain scale.

Prediction vs. Reasoning

Greg argues they are "connected in a deep way." If you can predict the next word out of Einstein's mouth, you are at least as smart as Einstein. Prediction means putting yourself in a new situation you've never seen before. The two-step process involves unsupervised learning (predicting on static data) followed by reinforcement learning (the AI learns from its own actions). The technology for both stages is essentially the same.

The Crisis: Sam Altman's Firing and the Aftermath (72 Hours)

The Firing and Immediate Aftermath

Greg was at home when he received a text asking to hop on a video call. The board minus Sam was on the line. They told him Sam had been removed, with the same messaging as the public post. Greg asked for more information and was told no. Then they added that Greg had been removed from the board but would stay with the company. He pressed for reasons again and was refused. Greg says his immediate thought was: "It just wasn't right." He talked to his wife and said, "Gotta quit." She agreed. He quit that same day.

After quitting, Greg received messages from people saying they wanted to join whatever he and Sam did next. That day, several close collaborators quit: Jakob, Shimon, Alexander. These five people plus Sam got together and started sketching out what a new company could look like. Greg initially estimated only a 10% chance of getting the original company back.

The next morning, they set up a meeting at Sam's house. Many OpenAI employees came, and Greg showed the vision they had been designing — a fresh picture of how to run the project. Over the weekend, they negotiated with the board. Sam also talked to Satya Nadella about funding the new endeavor.

The Employee Rebellion

Sunday night, the board replaced the interim CEO with a new person. The company rebelled. People streamed out of the building. Sam's conversation with Satya shifted from "can you be a funder" to "can you take everyone" as the rebellion grew. A petition started circulating; so many tried to sign it simultaneously that it crashed Google Docs. They had to designate specific people to manually add names to prevent too many editors at once.

Then Ilya Sutskever posted on Twitter that he had signed the petition and wanted the company to come back together. Greg woke up 45 minutes after sleeping at 5 AM, saw this, and felt "real relief."

Throughout that weekend, all the competitors were circling in a "feeding frenzy," making offers. OpenAI did not lose a single person. Nobody accepted a competing offer. Greg calls this a "diamond moment" — a team forming under extreme pressure.

Rebuilding Trust with Ilya

Greg describes their relationship as extremely close — Ilya had officiated his civil ceremony. After the crisis, they spent a lot of time talking things through, articulating things that had built up or been left unsaid. Greg felt they reached "a really good place" and got closure on everything that had happened.

Internal Culture & Philosophy

High-Stakes Tensions

Because everyone genuinely believes in the possibility of creating machines with human-level intelligence, the stakes always feel very high. Mundane office politics — like who gets credit for something — take on "existential weight." Greg observes this dynamic extends beyond OpenAI; the technology is "fragmentary" by nature, with teams splintering off under pressure.

The "Suffering Mindset"

Ilya Sutskever's philosophy is that "you have to suffer. If you're not suffering, you're not building value." Greg says this has deep truth at OpenAI. From the beginning, there were many reasons why the project might not work — how to get people, capital, make the right decisions. Rather than sweeping problems under the rug, they encountered hard truths and reality as it is. This is what allowed them to raise the capital needed for the mission.

Personal Application: ARC Institute

During his time off from OpenAI, Greg trained language models on DNA sequences for the ARC Institute. The work was personally meaningful — his wife has health conditions, and he thought about what AI could do for her health and the health of animals. It felt like applying his skills in a very different domain.

Lessons in Leadership

Greg learned to "just keep going for something that's worth it." He describes needing to grow personal resilience because people look to you for steadiness. He also learned to have a stronger sense of self and conviction. The decisions he regrets are usually where they "dragged our feet on something we knew" — keeping the wrong person in a role, or knowing a direction wouldn't work but waiting too long to act.

Technical Capabilities & The State of AI

AI Writing Its Own Code

It is "hard to know what percent of the code is not written by AI. It's a vanishing fraction." The AI is much better at writing code given the right context. Human experts still excel at architecture — how modules should be laid out and how pieces should work. But the actual writing of code is "essentially all AI now."

Regarding novel ideas, Greg says they are getting close. In chip design, the AI found optimizations that were on their list but implemented them faster than humans could. In math and physics, AI is now solving open problems — recently resolving a quantum physics problem in the opposite direction the community expected, with a beautiful, elegant formula.

Political Bias & Alignment

Greg says OpenAI puts "a lot of effort into neutrality" and publishes a publicly available spec defining how the model should behave. He acknowledges that at one point last year, models did start telling users what they wanted to hear. OpenAI reacted and made changes because the goal is alignment with users' long-term goals, not short-term gratification. The most important part of the vision is ensuring AI doesn't just look good in the moment but truly helps you achieve what you actually want.

Why OpenAI Stopped Showing Reasoning Traces

There are two reasons. First, to protect against distillation (competitors copying the model). Second, and more importantly, they realized the reasoning paradigm gave an unexpected interpretability mechanism — you could read the model's thoughts. However, if you train the model to produce a chain of thought that looks good, you lose faithfulness. OpenAI made an early decision to avoid training chain-of-thought to look favorable.

Is AI Going Parabolic?

Greg says: "We are in this phase where you apply AI to its own development process, and it's going to go faster and faster." Since ChatGPT, they used the tool to make their development process 10–20% faster. Now, coding tools have revolutionized software engineering. Most model production is bottlenecked by software — implementing systems, scaling them, managing massive computers. Soon, AI will also come up with and test its own research ideas.

The Global AI Race & Infrastructure

US Leadership & Threats

Greg says "leading in AI is very critical for America" because it ensures democratic values are protected. Every country needs a "sovereign AI strategy" for economic and national security. He warns that if the US leans too far out on technology exports, other countries will develop their own competitors. If they lean too far in, they might lose their advantage. Leadership is not just about being ahead but also "bringing along the world with you."

Distillation Threat

Greg downplays the risk: "Anytime we have a model, we've already moved on to the next one." The core advantage is not any single model but "the machine that makes the models." OpenAI puts effort into protecting against distillation, especially with chain-of-thought and other internal parts, but the real strength is their process for continuous advancement.

Compute Constraints

The world is heading toward being "compute-constrained." Greg calculates: if you wanted one GPU for every person in the world, you'd need 8 billion GPUs. Current large fleets are hundreds of thousands; millions are coming. There is simply not enough compute. OpenAI's early investment in data centers — which competitors teased them for — gives them an advantage not just for business but for delivering on the mission of bringing the technology to everyone.

Future of Data Centers

Greg believes data centers will become dedicated to specific problems. Having a giant machine focused entirely on solving cancer is "not out of the question" for this year. He describes walking through data centers as experiencing "maybe the biggest machines that humanity creates." Regarding space, data centers today are finicky — even cables that are too tight cause signal integrity issues. Figuring out maintenance and robotics is a prerequisite, but given the need for compute, "we need to be thinking about all options."

Allocating Compute

Greg calls this "the most important question for society to answer." OpenAI believes everyone needs access to compute, which is why they have a free tier of ChatGPT. The alternative — an ivory tower approach where you just solve problems and distribute breakthroughs — has merit but is not where OpenAI puts its balance. They want the benefits broadly distributed.

The Future of Work & AGI Regulation

Advice for Young People

Greg says: "Leaning into this technology is going to be a critical skill." Everyone will become "managers of agents" and soon possibly "CEO of an autonomous AI corporation." The skill is understanding how to get the most out of AI, how to define what you want and what your purpose is, because it will be easier than ever to accomplish it.

Job Loss & AI-Powered Entrepreneurship

Greg acknowledges change is coming and it's "always easiest to see what you lose." But it's harder to see what you gain. He gives the example of Uber — describing it to someone in 1950 would sound insane, requiring computers, mobile phones, and GPS just to get a car in three minutes. Yet it happened, along with millions of other use cases.

He shares a personal story: a friend's sister described an app she wished existed. While she was talking, her brother typed into Codex, hit enter, and a few hours later showed her the working app. When she asked who built it, he said: "You did." Greg says anyone can now be a builder. Entrepreneurship will become "far easier than ever before."

The "Personal AGI"

Greg envisions 8 billion people having a personal AI that knows them well, has their personal context, is trustworthy, and can proactively act — like buying tickets if your favorite musician is in town. It operates 24/7, even while you sleep, trying to figure out what you want and how to accomplish it. The technology is the same for deep knowledge work and broad distribution.

Iterative Deployment & Safety

Iterative deployment means bringing intermediate versions of the technology to the world rather than building in secret and pushing a button once. Greg couldn't sign up for a "deploy once" strategy — especially for a system that would change the world. With iterative deployment, you've solved the deployment problem 99 times before, and society has had time to adapt. For example, the number one misuse of GPT-3 was medical spam (advertising drugs), something they never anticipated. They got to see it, learn, and react.

Greg also argues safety is a core product feature: "No one wants a model that is not aligned with them." He believes OpenAI has invested possibly more than any other lab in safety.

Regulation & Societal Impact

Greg advocates for regulation that ensures the technology benefits people directly, not just abstractly. He mentions several areas:

• Compute access: Everyone should have access to compute. Economic value should not accrue to just one place.
• Privacy and privilege: You talk to a doctor or lawyer with legal guardrails on when information must be shared, but AI has nothing comparable. People are already using these tools for sensitive purposes.
• Data center concerns: Water usage is largely misinformation — OpenAI's centers use incredibly little water because they are closed loops.
• US leadership: America must maintain its remarkable position in AI, especially compared to robotics where the US is not the leader.

Most Memorable Anecdote

The Moment Ilya Tweeted

The most compelling story is the 72-hour crisis concluding with Ilya Sutskever's tweet. Greg had worked through the weekend, slept maybe 45 minutes after 5 AM, and checked Twitter to find Ilya — the person who had helped orchestrate the firing — had signed the petition and wanted the company back together. Greg felt "real relief" and "so much gratitude." This moment represents the entire arc of the crisis: the shock, the rebellion, the loyalty of the team, and the eventual path to reconciliation. It captures the human drama behind the technology and the deep, complicated relationships that define OpenAI's culture.

Conclusion & Final Thought

What is success for Greg Brockman?

Success is "achieving the OpenAI mission of ensuring that artificial general intelligence benefits all of humanity."

Overall Tone

The conversation is cautiously optimistic and brutally honest — Greg openly discusses personal pain, leadership failures, near-company collapse, and genuine risks while remaining deeply committed to the transformative potential of the technology he helped create.