The second anniversary of ChatGPT just passed a little over a month ago, yet it feels like we’ve already leaped into a new era of AI models capable of sophisticated reasoning. As the new year prompts reflection, I wanted to share some personal insights on this incredible journey so far and some of the key lessons I’ve learned along the way.
Approaching Artificial General Intelligence (AGI), it’s crucial to assess our company’s progress. There is so much to learn and see, vast unknowns still lie ahead, and we are truly just at the beginning. However, our understanding has grown exponentially since our inception.
Nearly nine years ago, OpenAI was founded on the conviction that AGI was achievable and held the potential to be the most transformative technology in human history. We aimed to explore its creation and ensure its widespread benefits, driven by an ambition to leave a significant mark and a strong belief in its societal advantages.
Initially, skepticism was rife. Few believed in our vision, many doubting our chances of success.
Back in 2022, OpenAI was a quiet research entity, developing what was then tentatively named “Chat With GPT-3.5”. (Research is definitely our forte, naming things, not so much.) We observed user engagement with our API’s playground feature, noting developers’ enthusiasm for model interaction. We envisioned a demo to highlight AI’s future potential and refine our models for enhanced performance and safety.
We thankfully settled on “ChatGPT” and launched it on November 30, 2022.
We always anticipated a tipping point that would ignite the AI revolution, though the exact catalyst remained unclear. To our surprise, ChatGPT became that pivotal moment.
Its launch triggered an unprecedented growth trajectory for our company, the AI industry, and globally. We are now witnessing the substantial benefits we’ve long anticipated from AI, with even greater advancements on the horizon.
This journey has been far from simple. The path has been uneven, and the optimal choices have not always been clear-cut.
Over the past two years, we essentially built an entire company from the ground up around this groundbreaking technology. Experience is the only real teacher in this domain, and in such a nascent field, there are no definitive guides.
Building a company at such breakneck speed with minimal precedent is inherently messy. Progress is often iterative—two steps forward, one step back (and occasionally, one step forward and two steps back). Mistakes are course-corrected on the fly, as established protocols are nonexistent when pioneering. Navigating uncharted territory at high speed is exhilarating yet intensely stressful for everyone involved. Conflicts and misunderstandings are inevitable.
These years have been the most fulfilling, enjoyable, rewarding, fascinating, exhausting, stressful, and—especially the last two—challenging years of my life. Overwhelmingly, gratitude prevails. I imagine future retirement days at a ranch, perhaps a bit bored, fondly recalling the incredible experience of pursuing a childhood dream. This perspective helps on Fridays when multiple crises hit before lunchtime.
About a year ago, one particular Friday brought the unexpected jolt of being fired via video call, immediately followed by a public blog post from the board. Stranded in a Las Vegas hotel room, it felt surreal, like a nightmare unfolding.
This abrupt and public dismissal initiated a chaotic few hours and days. The “fog of war” was disorienting. Satisfactory explanations for the events and their causes were elusive.
In hindsight, the entire episode was a significant governance failure, involving well-intentioned individuals, myself included. Reflecting, I wish I had acted differently and believe I’ve evolved into a more thoughtful and effective leader.
I also deeply appreciate the value of a board with diverse perspectives and extensive experience in navigating complex challenges. Robust governance hinges on trust and credibility. I am grateful for the collective effort in establishing a stronger governance framework for OpenAI, enabling us to advance our mission of ensuring AGI benefits all of humanity.
My paramount takeaway is profound gratitude and indebtedness to countless individuals: to every OpenAI team member dedicating their time and energy to this vision, to friends who supported us through crises, to partners and customers who believed in us and entrusted us with their success, and to my personal circle who showed unwavering care. [1]
Following this period, we refocused with renewed cohesion and positivity, achieving some of our most impactful research to date. Our weekly active users surged from approximately 100 million to over 300 million. Most importantly, we continued deploying technology that users genuinely appreciate and that addresses real-world problems.
Nine years ago, our trajectory was unclear; even now, our path is only partially defined. AI development is characterized by unexpected turns, and more are anticipated.
Some pivots have been joyous, others challenging. Witnessing consistent research breakthroughs and the conversion of skeptics into believers has been exhilarating. We have also seen colleagues branch out to become competitors. Team evolution is natural with rapid scaling, and OpenAI’s growth is exceptionally rapid. Startup turnover at each major scaling phase is typical, but at OpenAI, these escalations occur in orders of magnitude every few months. The past two years have felt like a decade in a conventional company. With such rapid growth and evolution, diverging interests are natural. Furthermore, industry leadership attracts scrutiny and competition, leading to various forms of attacks.
Our vision remains constant; our strategies will continue to adapt. Initially, we didn’t foresee becoming a product-focused company, believing our role was purely in research. Nor did we anticipate the immense capital needs. New necessities are emerging that were previously unforeseen, and future requirements are likely beyond our current imagination.
We take pride in our research and deployment achievements and are dedicated to advancing our understanding of safety and equitable benefit sharing. We remain convinced that gradual, iterative release of AI systems into the world is the optimal approach to ensure safety, allowing societal adaptation, experiential learning, and ongoing safety enhancements. We are committed to leading in safety and alignment research, guided by real-world application feedback.
We are now confident in our ability to build AGI as traditionally conceived. We anticipate that by 2025, the first AI agents may “join the workforce,” significantly impacting company outputs. We continue to believe that empowering individuals with powerful tools leads to substantial, widely distributed benefits.
Our focus is now extending beyond, towards true superintelligence. While we value our current products, we are driven by the potential of a transformative future. Superintelligent tools could revolutionize scientific discovery and innovation, far exceeding human capabilities, leading to unprecedented abundance and prosperity.
This vision might seem like science fiction, even audacious to discuss. But we’ve been in this position before and are comfortable with it. We are confident that in the coming years, the broader world will recognize the importance of acting with caution while maximizing widespread benefit and empowerment. Given the profound potential of our work, OpenAI cannot operate as a conventional company.
It is an incredible privilege and humbling experience to contribute to this endeavor.
[1]
Countless individuals contributed immensely to OpenAI and to me personally during those critical days, but two figures stand out:
Ron Conway and Brian Chesky went so far above and beyond, it’s hard to articulate. I had heard of Ron’s capabilities and tenacity for years and had benefited from Brian’s guidance over the past couple of years.
However, true character reveals itself in crisis. I am reasonably certain OpenAI would have collapsed without their intervention. They worked tirelessly, around the clock, for days until stability was restored.
Despite the immense pressure, they remained calm, strategic, and provided invaluable advice. They prevented me from making several missteps and made none themselves. They leveraged their extensive networks to address every need and navigated numerous complex situations. And I’m sure much of their support remains unseen.
What resonates most is their extraordinary care, compassion, and unwavering support.
I believed I understood founder and company support, and in a limited sense, I did. But I have never witnessed, or even heard of, anything comparable to their actions. I now fully grasp their legendary status. They are unique and richly deserve their reputations, sharing a remarkable capacity to move mountains and provide assistance, coupled with steadfast commitment in times of crisis. The tech industry is profoundly better for their presence.
Others like them exist; this exceptional quality within our industry is more vital to its functioning than often recognized. I look forward to reciprocating this support in the future.
On a personal note, special thanks to Ollie for his constant support, particularly that weekend; his incredible nature and unwavering partnership are invaluable.
