Good morning. Recently, AI has been evolving at an incredible pace. I think many people still believe that humans are smarter than AI and that AI has various limits. For example, some say that creativity is one of the wonderful functions remaining to humans. Of course, views on calculation ability and knowledge have changed a lot recently, but AI has rapidly evolved into the world of video, storytelling, and art. I believe that in the near future, AI will not only understand people's various emotions but will also begin to possess things equivalent to its own emotions and consciousness. Today, I will give a presentation later, but before that, for the first 30 minutes, I will be having a dialogue with Sam from OpenAI. I hope you will listen with us. Thank you.
Sam, how are you? What do you think of agents? What do you think about the latest developments in agents?
First, thank you for inviting me. I want to express my gratitude to you, who has been one of our greatest visionaries and strongest partners. We're going to talk about agents, but if it weren't for your belief in computing and AI, we wouldn't be here talking. We've come this far thanks to you. Thank you. So, the first era of AI, the early days of ChatGPT, was about an AI that could answer any question. And now, with the advent of these agents, AI can write actions, compile, and do various things. It can write reports for you, do knowledge work, and use the internet. It's really amazing, isn't it? It's incredible progress. We don't fully understand what's happening yet. But when you give complex instructions to a computer in natural language, it understands and executes them. This expands productivity and possibilities, and I think it's a tremendous step forward for the world. Let's talk about Stargate too. We announced it together in January this year. What is your vision for it?
It's a shared vision. We want to think about how to create the most efficient and effective computer in the world through mutual cooperation. We are making amazing progress. We are working on a big project together, and the first goal is 10 gigawatts of compute. If that goes well, we'll continue further. You're putting a lot of effort into it. Now, thinking about scaling beyond 10 gigawatts in the future, new technologies and new approaches will be needed. We need new approaches that leverage new technologies to make building compute at this scale possible. I feel that providing the abundance of compute needed around the world can bring the greatest positive impact to the AI revolution. In short, some people say that the effects will diminish, that such a large amount of compute isn't necessary, but you're still bullish, aren't you?
There is a scale for growing compute, and scaling laws exist. The first reason is that scaling laws will continue for a long time. We will see wonderful new research in the near future, which will take us very far. If we continue to make better models, as the cost of AI decreases, more people will want to use it. So if we lower the cost of AI, if the cost of AI becomes one-fourth, people will want to use it 30 times more. Then the demand for intelligence will become huge worldwide, and it will keep increasing as we progress. It's a wonderful thing.
Yes, that's right. After the keyboard, smartphones became widespread, right?
Yes, that's right. So far, there have been basically two major revolutions in computer interfaces. Before that, there were older methods like command lines and punch cards, but the two big ones in the mass market were the keyboard and mouse, and the monitor. And then the smartphone, the touchscreen smartphone. And before AI came along, that was sufficient. Now we have systems that can truly understand us and perform complex tasks. It's like an agent that acts on our behalf, communicating with people, filtering information, and so on. Imagine it. Something completely new. And I don't think current hardware can handle such interactions. So I've been talking with Jony Ive for a long time about what comes next, and I think it's something dreamlike. It's a very exciting problem. I hope it works out well. It would be great for everyone.
Now, then... I think AGI will definitely be important in 5 years. That's not the goal, though. In 10 years, 30 years... I might not be around in 30 years, but what about America? What do you think? Could you tell me your vision in three stages: 5 years, 10 years, and 30 years?
That's difficult. But in 5 years, I think we will definitely surpass AGI, and systems will definitely be able to do things they can't do now. In 10 years, I have very high expectations. I think the speed of scientific progress and economic growth will be completely different in the future. Technology and science will advance very quickly. Social change might not be that fast. Self-improvement loops have already been achieved in various fields and are progressing in a very important way. In 30 years, it's completely unknown. It's really difficult. It's hard to imagine, but it's hard to think of what superintelligence couldn't do. It will probably become a superintelligent robot connected to a robot. It will be able to move things and produce things. So I understand that too. We both believe in the future of robotics, but what is your view?
