OpenAI COO's latest interview: Future children will find it hard to imagine a world where they cannot converse with computers, we are still in a very early stage

On July 6th, OpenAI COO Brad Lightcap was interviewed by CNBC senior reporter Julia Boorstin at the Global Conference 2024 summit, where he analyzed the future development of AI and industry trends. He emphasized that OpenAI is establishing partnerships with companies across various industries to help them leverage AI technology to enhance efficiency and innovation. For example, Clarnet and Moderna have successfully applied GPT-4 in customer support, significantly improving work efficiency, and accelerating processes in clinical research and drug development using AI technology. Addressing concerns about the potential mass unemployment caused by AI, Lightcap stated that while some positions may be replaced by AI, this will create opportunities for new job roles. He cited the historical example of the mechanization revolution, highlighting that the adaptability and diversity of economies will help society cope with this technological change. Furthermore, Lightcap also discussed the issue of energy requirements for AI systems, emphasizing the need for a reorganization of the global supply chain to meet the growing demands of data centers and GPUs. He called on AI companies to actively engage in energy discussions to ensure sustainable development. Lightcap mentioned that future AI systems will become more powerful and intelligent, with interactions with users becoming more natural and efficient. He predicted that future AI will become the "perfect team member" for users, helping them solve complex problems and achieve more efficient work patterns. The full content of this conversation is provided below, enjoy~ Julia Boorstin OpenAI is a groundbreaking company. What is most interesting to me is its leadership in innovation and the true AI revolution, not to disrupt existing companies and industries, but to collaborate with them. This collaborative approach is unprecedented in the past waves of technological innovation and startups. Today, I hope we can delve into how OpenAI, as a partner and ally, helps companies in various industries harness this new AI technology and customize it to their datasets. As the COO of OpenAI, Brad, you are responsible for business operations. You have just announced a new partnership. Since we at CNBC typically start with the news, please tell us what the new API partnership between OpenAI and Stack Overflow means in practical terms for your customers? Brad Lightcap OpenAI is a unique company, and we have a very mission-driven approach to AI development. This essentially means that we will be a company fundamentally based on collaboration We are a small company, building a very large technology with a very broad impact. Therefore, how we bring this technology to the world must be achieved through cooperation, establishing a replicable and value-added model, but also core to make our work valuable in the world. Stack Overflow is a good example. We announced this collaboration this morning. ChatGPT as a programming assistant is one of the most popular use cases. If you are a software engineer, at least self-report, many software engineers have ChatGPT open in one window and their programming terminal open in another window. This is a collaboration that allows Stack Overflow to bring their resources into ChatGPT in a very native way. It helps us improve ChatGPT as a programming assistant and programming reasoner. Hopefully, it will make Stack Overflow more efficient in assembling interesting datasets to improve code productivity. Julia Boorstin It looks like you are doing a lot of these collaborations. These collaborations involve different fields, but obviously these collaborations help accelerate your promotion and truly demonstrate the capabilities of this technology. Just announced a partnership with the Financial Times last week, and you have also established partnerships with some publishers, but at the same time, you are facing many lawsuits. Lawsuits from Sarah Silverman, author Jody Pico, and The New York Times. How do you view these partnerships or build tools to address these issues? Brad Lightcap Yes, publishing is an interesting field. The interesting thing about publishing is that we actually think it is the most opportunistic field. If you fundamentally consider what AI is and the opportunities AI has, it is the most advanced tool in the world for understanding, generating information. And publishing fundamentally is about interacting with information, creating and generating information. How to align these two goals and what this technology can bring to this industry, how to enhance journalism, how to improve the quality of reporting, how to enhance user experience and the ability of readers to interact with content. We believe the opportunities are endless. We are fortunate to have many publishers who want to work with us to pursue this goal. We are just starting to explore what this looks like. We are just starting this process. I guess we will have more news to share soon. But we think this is absolutely crucial and a top priority for us. Julia Boorstin It seems like the concept of this partnership with the Financial Times, do you think we should expect more of these types of collaborations? Brad Lightcap Yes. We are indeed looking for ways to diversify the perspectives we bring to ChatGPT, and how we view the ability of AI to interact with the industry. For example, the form of financial reporting is very different from current affairs and political reporting, and very different from lifestyle reporting Having every type of content, every perspective is very important. These tools are essentially reasoning engines. Helping people explore the full space of written content, public opinion, we see this as a great opportunity. Julia Boorstin I hope I won't be replaced too soon by an OpenAI tool. But if I and my colleagues are really replaced, I believe it will be a very good tool. So, let's talk about the current situation in enterprise adoption. The Wall Street Journal just published a wonderful case study about your collaboration with Moderna. I also saw fantastic analyses about Oscar Health and Klarna. If you had to describe where we are in terms of enterprise adoption 18 months after the launch of ChatGPT, how is the situation? Brad Lightcap Yes, this is where I spend a lot of time. This may be the most interesting thing this year. When I was on this stage or a similar stage last year, I swear, most people in the audience had not heard of OpenAI. They had heard of ChatGPT. They thought the company I work for is called ChatGPT, they knew it was being used in their companies, but they didn't know exactly how, but they wanted to understand better. In fact, almost from that moment, we started this journey, understanding how people are using this type of technology in the enterprise and what tools we need to build to accelerate enterprise adoption. A year later, we have built an enterprise-grade ChatGPT. Now we see that about 92% of the Fortune 500 companies are using ChatGPT as a tool. We now have over 600,000 enterprise users. The adoption is very significant. It is a very versatile technology, and we do not believe any industry has a higher adaptability than others. It is a tool for everywhere. We expect this growth curve to continue. But we are still in a very early stage. The founder of a top U.S. hedge fund, an early investor in Tencent & ByteDance, said that at the beginning of these new waves, the new winners in 10 or 20 years, some are existing companies, some are new companies. But the new companies that grow larger, there may be only one each year, not more... Julia Boorstin So, how many companies prioritize interacting with AI or adopting AI tools? It's also interesting, but fewer than the number of individual users. So, who is driving the adoption of ChatGPT? Is it top-down or driven by employees? We discussed this before, for example at Box, this drive is more from employees. What is driving this wave of adoption? What do you think is most effective? Brad Lightcap Yes. Our perspective has changed in this regard. When we first started OpenAI, our first business was to build a developer platform, basically exposing our models to developers through APIs, allowing them to build these highly customized tools We still have this business, and it's doing well. Most companies believe they have to build AI through this. Most companies contact us and say, "I have a goal, I want to increase revenue by x%, I want to cut customer support costs by y%, etc." That's no problem, we can still do that work. But what we gradually realize, and what most companies are realizing, is that getting your employees to engage with these tools is actually one of the biggest accelerators. It won't neatly show up on financial statements, it won't be an item in the second-quarter earnings report, but getting people familiar with how these tools work, enabling them to shape workflows around models, experiencing the capabilities of next-generation models, is almost like putting training wheels on general AI. This is very important, and companies are gradually realizing this. We are very encouraged. Julia Boorstin So enabling broad experiments, this is interesting because OpenAI's iterative development philosophy and the idea of introducing things along the way are different from traditional ways of working. How do you explain this concept and its significance for businesses? Brad Lightcap Yes. The reason we do this globally is that we believe the pace of progress in this field and the models we train will accelerate from here. If you can believe it, people actually think the pace of change has already been dazzling. It will further accelerate in the coming years. So, how do you think about this from a deployment perspective, bringing models to users, whether consumers or businesses. Last year we changed this approach a bit, we no longer do these big releases. It used to be, "Well, the world has GPT-3, no one knows what GPT-4 looks like." Then suddenly we release GPT-4, and everyone has to adapt. As the models become more powerful, we believe it's important for people to gradually understand the improvements in capabilities and trajectory. Therefore, iterative deployment is basically a practice to make people familiar with incremental improvements in the model, making the trend line look smoother rather than discontinuous. Julia Boorstin So, what value does a company gain from taking an AI-first perspective or a top-down approach? Moderna seems to be a good example, they decided to prioritize deploying AI. What does this mean in practical terms? What advantages does it bring if a CEO prioritizes it as Moderna's CEO did? Brad Lightcap Yes, it's difficult to do this in a fragmented way. This is not to say that you need to fully AI-ize your company tomorrow. But I think if you consider introducing the internet into business, and today no modern enterprise can operate without the internet, it's a bit like starting by saying, "Okay, I'll only give my finance team access to the internet." We hold the same view here, trying to take a comprehensive approach, allowing everyone to access, diversifying the number of projects we are working on to cover different parts of the company Let's consider our priorities, which include both our internal operations and the types of products we can build and what we can offer to customers. We believe that all these types of work are important, and taking this comprehensive view is crucial. Julia Boorstin So, take us through the use of AI at Moderna and how they compare to Clarnet. Clarnet is the first fintech company to deploy OpenAI. The potential of these tools is so diverse, it's really amazing. Brad Lightcap Of course. Clarnet is a great example. Clarnet has implemented GPT-4 into their customer support team. They have built a fully customized toolset that can now handle the work of about 700 customer support agents, reducing the time to process each ticket from about 11 minutes to around 2 minutes, and reducing repeat queries from any specific customer by about 25%. They have seen amazing results. They are very mature in AI adoption, and Clarnet's CEO Sebastian is very enthusiastic about AI, and this implementation is groundbreaking. But we are now seeing more templates emerge, allowing other companies to adopt this approach in a similar way. Moderna is slightly different, as Moderna is clearly a pharmaceutical company. They don't do much direct customer-facing work, unlike Clarnet, which serves many users. But for them, the real key is how to start leveraging the efficiency brought by AI to drive the clinical research process, drug development process, and all compliance and administrative work. As a company, they are actually relatively small in the pharmaceutical industry, with about 6,000 people. So, Moderna's real focus is on how to maintain this very focused and narrow culture, accelerate drug development, without having to burden the company with a lot of administrative work that most pharmaceutical companies have accumulated over the years. So, they approach this issue in a different way. Julia Boorstin Yes, very interesting. You mentioned Clarnet, where your technology does the work of 700 employees. That's the crux of the issue. I think for many CEOs, this raises the question of how many jobs will be replaced by AI? Will there be large-scale layoffs that impact society as a whole? What is the real impact of AI on unemployment that we are seeing? We have discussed this backstage, but I was a bit surprised to hear that you don't think there will be job losses. Why not? Brad Lightcap I don't think there will be. As labor realigns in this world, there will be a normal turnover in the labor market. In a few years, some areas may wonder why we need people to do these jobs, but other areas will emerge because of the implementation of this technology We have a very dynamic global economy. As I mentioned before, someone told me last night that when the mechanical reaper appeared at the end of the 19th century and the beginning of the 20th century, about 90% of the population in the United States and Europe were engaged in agricultural work. There is a risk of massive unemployment and job loss. You do see this happening. These economies have adapted and developed. You won't see the same situation here. We have a more diversified economy, a more resilient economy. The nature of the introduction of this change in the global economy. The pace of technological development will be very fast. But ultimately, you need to figure out all the networks and channels to allow technology to work locally. This takes time. This is not something that worries me. What actually worries me is that we are not adopting it fast enough. We are an economy where our actual GDP growth rate has not reached the required level. Especially in the United States. We see many areas in the economy where this technology can bring benefits. Our focus is actually on how to introduce it faster and enable it in a way that can have an impact. Julia Boorstin I know we don't have much time left, so I'm going to do some quick Q&A. OpenAI is an open-source company. What does that mean to you? What misconceptions do you think enterprise customers have about this? Brad Lightcap We do open-source some models, but not all models. We take a balanced view on this and believe that open-sourcing is very healthy for the ecosystem. We will continue to open-source models at a certain pace. But for cutting-edge systems, for larger systems, we believe it is important to understand how these systems work, and by accessing these systems through more centralized providers, we can work with companies to implement them securely. A partner at OpenAI's investor Thrive Capital mentioned that the debate between open-source and closed-source is quickly coming to a conclusion; where there is a huge opportunity between small models and endpoints, the next billion-dollar company will be toC... Julia Boorstin Do you think this is misunderstood? Brad Lightcap What is misunderstood is that most people actually just want to access them in the simplest way possible. Open-sourcing is good, but using open-source technology is very difficult. Our goal is to open access as widely as possible. We hope that anyone in the world can pick up OpenAI technology, from experimentation to building to production, in just a few hours. We will continue to build products in this direction, that is our mission. Julia Boorstin Do you think there will be legislation from the EU (which usually leads tech legislation) or the US that could significantly impact or hinder your growth potential? Brad Lightcap Of course, Europe is paying attention to this issue, perhaps more closely than the United States right now. We are pleasantly surprised by the global regulatory dialogue. This dialogue is quite balanced, and I hope it continues The discussion about future systems needs to continue, especially as we understand more about how these systems work. Over-regulating existing systems is a mistake, and we need to be careful not to stifle innovation. But overall, we are very encouraged by the nature of this conversation. Julia Boorstin I know there has been a lot of discussion at this meeting about the global energy ecosystem. How do you view the management of energy demand not only for today's systems but also considering the growth in demand as you continue to expand? Not to mention the demand for GPUs. Brad Lightcap Yes, we see this as a significant risk. If you look at the rate of growth in today's supply chain compared to the demand for this technology that we anticipate in five or ten years, there will be a significant gap. Therefore, how the world reorganizes itself to assemble and power data centers, readjust supply chains, this will be a real issue. We are involved in this discussion. Every AI company should be part of this discussion. But this is a major challenge that society will need to address in the coming years. Julia Boorstin Yes, this will be a big issue. So, as a closing question, what new developments can we expect from OpenAI in the next six to twelve months? We don't expect any proprietary information, but can you give us some insights, provide some advice to the companies here, to understand the prospects of AI in the next twelve months and the next three to four years? For me, twelve months is short-term, but for you, it may be long-term. And three to four years may still seem short-term to me, but in the field of AI, it may feel like eternity. Brad Lightcap Yes, I have stopped making four-year predictions, but at least for the next twelve months. The systems we are using today will be laughably bad. They are indeed bad. We will look back in a year and see how bad they are. Systems like ChatGPT, which are almost turn-based and oracle-like, we don't think that is the long-term interaction pattern for these systems. We believe we are moving towards a more intelligent world, where they will be able to handle more complex tasks and have a more collaborative relationship with users. It's almost like having a great team member working with you. This will be a different way of using software. People today don't really understand what it looks like, and they can't easily internalize it. But a child born today looking back in ten years won't be able to imagine not being able to converse with a computer like with a friend, team member, or project collaborator. This is a profound change that we have not fully understood yet. Julia Boorstin It sounds like one important aspect is transitioning from a text interface to a verbal interface. Brad Lightcap Not necessarily just a verbal interface. The amazing thing about these systems is their multimodal capabilities. They can engage in visual reasoning, handle very complex problems. We've talked about whether an AI could reach a certain level of comedy on the road Julia Boorstin Can AI be a funny AI? Brad Lightcap It can be a stand-up comedian AI. Yes. We are just scratching the surface of the full capabilities of these systems, which will amaze us. Julia Boorstin The pace of AI development is faster than that of enterprises. The speed of innovation is faster than the usual speed of companies. How can companies and investors keep up with this usually uncomfortable rapid pace? Brad Lightcap This is the hardest part. This is indeed something I spend time thinking about, how to help the world adapt to the speed of change in the research lab. For all businesses and partners, how do we implement technology in a scalable, future-oriented way while keeping up with the changes that are happening. We are still learning, we think we may have some understanding. Of course, the more we work with companies, the more we learn. But this is a major challenge for the coming years. Julia Boorstin Looking back at the innovation of the past year, the innovation from text to images, text to videos is amazing. I can't imagine what we will be discussing this time next year. Brad, thank you very much for sharing with us here today.