Sam Altman's Retrospective and Reflections on the Coming of Super Artificial Intelligence

 ChatGPT's second birthday was just over a month ago, and now we have entered the next paradigm - the era of models capable of complex reasoning. The New Year always puts people in a reflective mood, and I'd like to share some personal reflections on the journey so far, and some of the insights I've gained along the way.

As we get closer to generalized artificial intelligence (AGI), it seems like an important time to look back at the company's progress. There's still a lot to understand, a lot of uncharted territory, and we're still in the early stages. But we have accumulated more knowledge than when we started.

We founded OpenAI nearly nine years ago because we believe that general artificial intelligence is possible and that it could be the most impactful technology in human history. We want to figure out how to build it and make it broadly accessible to humanity; we want to make our mark on history. Our ambitions were extremely ambitious, and we were convinced that this work could benefit society in equally extraordinary ways.

At the time, few people were paying attention to the field, and when they were, most of them didn't think we'd make it.

In 2022, OpenAI was an obscure research lab developing a project tentatively named “Chat With GPT-3.5”. (We're much better at research than we are at names.) We've been watching people use the playground features of our API and have found that developers really enjoy talking to the models. We thought that building a demo system around this experience would show people important features for the future and help us make models better and safer.

We ended up being lucky enough to name it ChatGPT and release it on November 30, 2022.

We've always known in the abstract that at some point we would reach a tipping point from which the AI revolution would unfold. But we didn't know what that moment would be. To our surprise, it is that moment.

The launch of ChatGPT has set off a wave of unprecedented growth - in our company, our industry, and the world at large. We're finally seeing some of the great advantages we've been waiting for from AI, and we can foresee more breakthroughs in the near future.

The journey has not been easy. The road has not been smooth and the right choice has not been obvious.

Over the past two years, we've had to build an entire company around this new technology almost from scratch. There is no way to train people other than hands-on, and when the technology category is completely new, there is simply no one who can tell you exactly what to do.

The process of building a company at such a high rate of growth and with little to no training to follow is destined to be chaotic. It's usually two steps forward and one step back (sometimes one step forward and two steps back). Mistakes are corrected along the way, but there really isn't any manual or guidebook to refer to when you're doing groundbreaking work. Moving at high speed through uncharted waters is an incredible experience, but it's also extremely stressful for all involved. Conflict and misunderstandings abound.

These have been some of my most rewarding, interesting, exciting, exhausting, and stressful years to date, and the last two years in particular have been full of unpleasantness. But the most overwhelming feeling is gratitude; I know that someday when I retire to our ranch to watch the plants grow, I might get a little bored and think back on how cool it was to be able to do the job I've dreamed of since I was a kid. I try to remember this every time seven bad things have gone wrong before 1:00 p.m. on a given Friday.

One Friday about a year ago, the worst thing that happened that day was that I was abruptly fired from my job during a video call, and right after we hung up, the board immediately published a blog post about it. I was in a hotel room in Las Vegas at the time. The feeling, which is hard to describe in any words, was like a dream that suddenly turned into a nightmare.

Being publicly fired without any warning, followed by an extremely crazy few hours, and a pretty crazy few days. The strangest part was the “fog of war”. None of us can get a satisfactory answer as to what happened and why.

In my view, the whole thing was a major failure of governance, and that includes all people of goodwill, including me. Looking back, I certainly wish I had done things differently then, and I would like to believe that I am a better and more thoughtful leader now than I was a year ago.

I also recognize the importance of a board with diverse perspectives and a wealth of experience in managing complex challenges. Good governance requires a great deal of trust and credibility. I am grateful that so many people have worked together to build a stronger governance system for OpenAI so that we can continue to pursue our mission of ensuring that general-purpose AI benefits all of humanity.

My biggest realization is that I have so much to be thankful for, and so many people to be thankful to: everyone at OpenAI who has chosen to invest their time and energy in pursuing this dream; our friends who have helped us in times of crisis; our partners and customers who have supported us and trusted us to help them succeed; and the people in my life who have shown me that they care! We are all working together in a more united and positive way.

We all returned to work in a much more cohesive and positive way, and I'm very proud of how focused we've been since then. We completed some of arguably our best research ever. We grew our weekly active user base from about 100 million to over 300 million. Most importantly, we continue to introduce the world to technology that people really love and that solves real problems.

Nine years ago, we really had no idea what we'd end up becoming; even now, we only have a vague idea. The evolution of AI has taken many turns, and we expect more changes in the future.

Some of the twists and turns have been exhilarating; others have been difficult. It's been fun to watch the research miracles continue to emerge, and many skeptics have become true believers. We've also seen some colleagues go their separate ways and become competitors. Teams tend to change as they get bigger, and OpenAI is expanding really fast.

I think some change is inevitable - startups typically experience a lot of turnover at each major new stage of scale, and at OpenAI the numbers are growing by an order of magnitude every few months. The last two years have been like a decade for the average company. When any company grows and evolves this quickly, interests naturally diverge. And when any company is leading in an important industry, many people will attack it for a variety of reasons, especially when trying to compete with it.

Our vision will not change; our strategy will continue to evolve. For example, in the beginning we didn't think at all that we would need to build a product company; we thought we would just need to do great research. We also didn't realize that we would need such a huge amount of capital. Now we have to build new things that we didn't even understand a few years ago, and there will be new things in the future that we can barely imagine now.

We are proud of our research and deployment accomplishments to date and are committed to continuing to advance our thinking on security and benefit sharing. We continue to believe that the best way to make AI systems secure is to release them into the world in an iterative and incremental way, giving society time to adapt and co-evolve with the technology, learn from experience, and continue to make the technology more secure. We believe in the importance of being a world leader in security and alignment research and have guided this research with feedback from real-world applications.

We are now confident that we know how to build general-purpose AI as traditionally understood. We believe that in 2025, we may see the first AI agents “join the workforce” and materially change a company's output. We continue to believe that putting great tools in people's hands iteratively will lead to widely distributed great results.

We're starting to move beyond that toward true superintelligence. We love our current products, but we're here for that glorious future.

With superintelligence, we can do anything else. Superintelligent tools can dramatically accelerate scientific discovery and innovation far beyond our own capabilities, and in turn dramatically increase abundance and prosperity.

Now that this sounds like science fiction, it seems a little crazy to talk about it. That's okay - we've been through this before and we wouldn't mind going through it again. We're pretty confident that everyone will see what we're seeing over the next few years, and the need to maintain great caution while maximizing broad benefits and empowerment is so important. Considering the possibilities of our work, OpenAI cannot be an ordinary company.

How fortunate and awesome to be a part of this work.