Autor original: Sam Altman
Compilação original: Deep Chao TechFlow
O segundo aniversário do ChatGPT acabou de passar há pouco mais de um mês e agora entramos em uma era de novos modelos capazes de raciocínios complexos. O Ano Novo sempre traz reflexão e gostaria de aproveitar esta oportunidade para compartilhar algumas reflexões pessoais sobre esta jornada e alguns dos aprendizados que aprendi ao longo do caminho.
À medida que nos aproximamos da AGI, agora é um momento importante para refletir sobre o progresso que a nossa empresa fez. Ainda há muitas incógnitas a explorar, muitos problemas que ainda não foram resolvidos e tudo ainda está nos seus primórdios. Mas aprendemos muito mais do que quando começamos.
We founded OpenAI about nine years ago because we believed AGI was possible and had the potential to become the most influential technology in human history. We hoped to find a way to build AGI and ensure it can broadly benefit humanity; at the same time, we also yearned to write our own page in history. Our goals are extremely ambitious, and we firmly believe that this work can give back to society in equally grand ways.
At that time, hardly anyone cared about AGI. If anyone did, most thought we would fail.
In 2022, OpenAI was still a low-profile research lab developing a project temporarily named 'Chat With GPT-3.5.' (We were indeed far ahead in research than in naming.) We noticed that developers really enjoyed the experience of conversing with the model through the API's playground feature. Thus, we thought that building a demonstration around this conversational experience would not only showcase the future possibilities but also help us improve the model's performance and safety.
Ultimately, we decided to name it ChatGPT and officially launched it on November 30, 2022.
We have always vaguely known that the AI revolution would come one day, but we did not know what that moment would look like. What surprised us was that this release became that turning point.
The release of ChatGPT triggered an unprecedented wave of growth—both within our company, across the industry, and on a global scale. We finally saw the immense potential that AI brings and can foresee that many more breakthroughs are on the way.
The road is not smooth, and making the right choices has never been easy.
Over the past two years, we have built an entire company around this new technology from almost zero. In facing such challenges, the only way to learn is to explore through practice. When the technology itself is brand new, no one can tell you exactly how to proceed.
Building a company at such a rapid pace, with almost no existing experience to draw on, is bound to be chaotic. We often take two steps forward and one step back (sometimes even one step forward and two steps back). While mistakes are continuously corrected in the process, there is nearly no existing manual or clear direction to guide us when venturing into uncharted territory. Moving rapidly through unknown fields is an exhilarating experience, but it also brings immense pressure to all participants. Conflicts and misunderstandings are inevitable.
These years have been the most fulfilling, interesting, and enriching years of my life, while also being the most exhausting and stressful. Especially the past two years, one could even say it has been the toughest time, but my deepest feeling is still gratitude. I know that one day, when I retire and sit on a farm watching plants grow, feeling a bit bored, I will recall the days when I had the opportunity to realize my childhood dreams—what a fortunate thing that is. I try to remind myself of this every Friday, even when, by 1 p.m., seven things have already gone wrong.
About a year ago on a Friday, what caught me off guard was that I was suddenly fired during a video call, with no warning whatsoever. Immediately after we hung up, the board published a blog post announcing the news. I was in a hotel room in Las Vegas at the time, and the feeling was almost indescribable, like a beautiful dream suddenly turned into a nightmare.
Being publicly dismissed without any signs triggered hours of chaos, and the following days were filled with uncertainty. The strangest part was that state of 'information chaos.' None of us could get a clear answer about what had happened or why.
In my view, this whole incident is a governance failure, despite the good intentions of everyone involved, including myself. Looking back, I certainly wish I had handled things differently at that time. I also want to believe that I am more mature and thoughtful today than I was a year ago.
This event made me realize that a board needs to have diverse perspectives and rich experiences to tackle complex challenges. Good governance needs to be built on a foundation of deep trust and credibility. I am grateful that many people worked together to establish a stronger governance system for OpenAI, allowing us to continue pursuing the mission of AGI to benefit all of humanity.
Over the past year, my biggest realization is that I have so much to be grateful for and so many people to thank: thanks to all the colleagues at OpenAI who chose to invest their time and energy into realizing this dream; thanks to the friends who supported us in times of crisis; thanks to the customers and partners who trusted us, collaborated with us, and relied on our technology for success; and thanks to those in my life who expressed care for me.
After that, we returned to work in a more united and positive manner. I am incredibly proud of the focus we have shown since then. We completed some of the most outstanding research work in OpenAI's history. Our weekly active users grew from about 100 million to over 300 million. Most importantly, we continued to launch technologies that people truly love and that can solve real-world problems.
Nine years ago, we had no idea what we would ultimately become. Even today, we only have a vague understanding of the future. The development of AI is full of ups and downs, and we expect more challenges and opportunities ahead.
Among these changes, some are delightful, while others are full of challenges. The continuous emergence of research breakthroughs is invigorating, and many former skeptics have now become staunch supporters. Of course, we have also witnessed some colleagues choosing to leave and become our competitors. As the team continues to grow, personnel turnover is inevitable, and OpenAI's pace of expansion is particularly astonishing. Startups usually experience turnover during each scale-up, and at OpenAI, this change occurs almost on a monthly basis. The changes over the past two years are almost equivalent to a decade of development for an ordinary company. When an industry-leading company grows at such a rapid pace, team members' goals and priorities are bound to diverge. At the same time, as a leader in the industry, we inevitably become the target of various criticisms and attacks, especially under pressure from competitors.
Our vision has always remained unchanged, but our strategies will continuously adjust and evolve over time. For example, in the beginning, we never thought we needed to build a product company; we believed we only needed to focus on achieving excellent research. Similarly, we did not anticipate the need for such a massive amount of funding to support our work. Now, we need to build many things that were unimaginable just a few years ago. And in the future, there will certainly be more new challenges that we currently find difficult to foresee.
We are proud of the achievements we have made in research and technology applications so far and are committed to continuing in-depth exploration in safety and benefit-sharing. We have always believed that the best way to make AI systems safe is through gradually and rhythmically introducing them to the world, allowing society ample time to adapt and evolve with the technology. Continuously learning from practical applications and improving the safety of the technology. We are well aware that being a leader in global safety and alignment research is crucial, and this research needs to be guided by real-world feedback.
Today, we are confident that we have grasped a method for building AGI as traditionally defined. We believe that by 2025, we may see the first AI agents 'joining the workforce,' significantly enhancing productivity in businesses. This will be a historic moment. We still firmly believe that gradually putting excellent tools in people's hands is the best way to achieve widespread and profound impact.
We have begun to turn our sights toward greater goals, namely the pursuit of true superintelligence. While we are passionate about our existing products, our mission is to work hard for that beautiful future. With superintelligence, we will be able to achieve all possible goals. Superintelligent tools will greatly accelerate scientific discovery and technological innovation, helping us tackle complex problems that are beyond human capabilities alone, thereby significantly enhancing societal wealth and prosperity.
Now, all of this may sound like science fiction, and even discussing it seems a bit crazy. But that's okay—we have faced similar situations before, and we are excited to embrace such challenges again. We are very confident that in the coming years, everyone will see the prospects we see and recognize how important it is to remain highly cautious while maximizing universal benefits and empowerment. Given the potential and possibilities of our work, OpenAI is destined not to be an ordinary company.
Being able to participate in this endeavor makes me feel both fortunate and honored.
(Thanks to Josh Tyrangiel for inspiring me. I really wish we had more time to talk.)
During those days, many people made tremendous efforts for OpenAI and for me personally, but two individuals stood out in particular.
The help from Ron Conway and Brian Chesky has far exceeded the call of duty, to the point where I don't even know how to express my gratitude. I had long heard of Ron's capabilities and tenacity, and over the past few years, I have received countless help and advice from Brian.
But only in times of true crisis, fighting side by side with others, can we truly see their extraordinary abilities. I can say without exaggeration that without their help, OpenAI might have collapsed. They worked tirelessly for several days until the issue was completely resolved.
Despite their intense workload, they remained calm, demonstrated clear strategic thinking, and provided extremely valuable advice. They prevented me from making several mistakes while not making any themselves. They leveraged their extensive network to meet all demands and successfully navigated many complex situations. I am confident they also did many things I was unaware of.
However, what touched me the most was their care, compassion, and support.
I once thought I knew how to support a founder and a company, to some extent, I did have some understanding. But I had never seen, nor even heard of, support like theirs. Now I finally understand why they have such a legendary reputation. They are indeed extraordinary and fully deserve that evaluation. They all possess remarkable abilities to provide help against all odds at critical moments, demonstrating unwavering commitment. Our tech industry is better because of them.
Of course, there are others like them. This is particularly striking in the tech industry, more important than people realize, and it allows the entire industry to operate smoothly. I look forward to passing on this spirit.
From a more personal perspective, I especially want to thank Ollie for his support that weekend and all along. He is unparalleled in every way, and no one could find a better partner than him.