OpenAI founder: We are a very truth-seeking organization

Source: China Entrepreneur Magazine

Organization｜Wu Ying

Image source: Generated by Unbounded AI tool

On July 24, OpenAI CEO Sam Altman (Sam Altman, hereinafter referred to as "Altman") launched WorldCoin, a cryptocurrency project developed for many years, which once again attracted widespread attention. Altman is, as he claims, working hard for a better life for humans in an AI world.

Three months ago, Sam Altman and OpenAI co-founder Greg Brockman (Greg Brockman, hereinafter referred to as "Brockman") accepted an interview with the American podcast. During the interview, the two core figures of OpenAI explained their views on AI. In addition, they also discussed the impact of AI on various industries, security issues, and their investment layout outside of AI.

When the two explained the vision of OpenAI, they agreed that they want to develop AGI (General Artificial Intelligence) that is beneficial and safe for all mankind. Regarding the hotly debated AI security issues, they also acknowledged the risks and said they are working hard to overcome them. In addition, the two also talked about investments in other fields. For example, Altman is particularly interested in the clean energy industry, while Brockman is interested in the medical and education industries.

Here are the highlights from the interview:

1. We are working hard to develop beneficial and safe general artificial intelligence for all humanity.

2. Fundamentally, what we want is to provide this technology to everyone, remove the barriers for them to create, complete new tasks, and push humans to solve problems they are really interested in.

3. We no longer just stop at the stage of having primitive knowledge, but move forward to the second stage-guiding the machine to perform the tasks we give it, and operate according to social values.

4. We've built a culture of really paying attention to the details and trying to get them right that attracts very careful engineering and scientific research from those who want to work that way.

5. I think that abundant and clean energy can bring benefits in other ways than just solving the climate crisis, and that investing in energy is a more important thing than investing in any other single input I can think of. Funding a fundamental change in the cost of energy acquisition is a great thing.

6. AI is a productivity tool for everyone that will make our best scientists better and advance scientific research.

Which industries will AI affect in the next 3 to 10 years

**Moderator:**Let’s start with the most basic questions. Can you briefly introduce what is the mission of your OpenAI? What is the purpose of all your decisions? **Altman: We are working hard to develop beneficial and safe general artificial intelligence for all human beings. **This is an unprecedented project, I know how difficult it is to do it, we will make many mistakes along the way, but this is the direction we stick to. I firmly believe that AI will be the most transformative technology ever developed by human beings, in a way, it will help us develop all future technologies, and the future is bright.

Brockman: About 8 years ago, when we started OpenAI, I really believed that artificial intelligence was going to happen, and I knew it would take a long time to achieve it, but thinking that we have the opportunity to influence this technology I feel very excited about my performance. Our vision is that AGI is beneficial to all mankind, and we are trying to achieve this goal in various ways.

At the beginning we had such ideals, but we didn't know how to combine them with practical actions. After years of hard work, we figured out how to build a system that really works, and how to integrate our values into this system. Over the past few months, we've been delighted that our technology has slowly become mainstream and that people are starting to reap the benefits of our technology. **This technology is still in its early stages and there is a lot more to be done, but we strongly believe that everyone should be involved and given a voice. ** Figuring out how to actually achieve effective global governance is as important as technology development.

Moderator: About 1 billion people in the world belong to the middle class with superior economic conditions. They are also freely using AI to help them work and study. So how can AI benefit the remaining 7 billion people?

**Brockman: An important feature of the AI we developed is that there are no access restrictions, and everyone can access and use this technology, input their preferences and feedback into the system, and use it to improve them life, which is the original intention of our development of this technology. **

Now there are a lot of questions about how this technology will work. Our company has a special structure, it is a non-profit organization, so if the capital is highly concentrated in OpenAI, it will not be held by shareholders, but owned by the company, and eventually distributed to the world. When profits are distributed to the whole world, UBI (Universal Basic Income) can be considered and resources allocated in this way. **Fundamentally, what we want is to provide this technology to everyone, remove the barriers for them to create, complete new tasks, and push humans to solve problems they are really interested in. **

**Altman: All truly sustainable economic growth comes from technological progress, and sociopolitical management and culture can help us achieve technological progress. ** Therefore, we are advancing how to help the remaining 7 billion people. In the future, our technology will go further. I believe that technology is the key to improving the quality of life for everyone.

**Moderator:**What kind of help can AI bring in the field of education?

**Brockman:**If I hadn’t started an AI company, I would have started a programming education company. Just imagine if the future can provide children with a teacher who really understands them, takes the time to get to know them, knows how to motivate them, and this teacher is available 24/7 for free, doesn't it sound like very good.

You can look at institutions like Khan Academy [a non-profit educational institution], they really started to take GPT-4 and put it into the classroom, and they figured out how to channel this technology so that It becomes a useful mentor. If the children asked it to "help me with my homework", it would say "no, I can't do that". And also trying to figure out what kids are interested in and how to really motivate them to do more, and now we're getting into the stage of how to make the most of it.

**Moderator:**Many people may say that there is Khan Academy now, and there are still many online courses, but what you put forward are just ideas, and there are no actual products. Do you think uneducated people still need your products?

Brockman: The way we are currently educated is through those smart people who teach us knowledge and problem-solving methods. In the future, if there is a machine that can be smarter than humans, will it be possible? more convenient. **We have developed a lot of virtual assistants, like Alexa (the smart assistant launched by Amazon) and Siri (the AI assistant launched by Apple), but for some problems that require brain thinking, it is still not enough, we need more advanced Technology to enhance our ability to think and solve problems. **

Undoubtedly, functions like GPT-4 have been seen by everyone, **We no longer just stay at the stage of having original knowledge, but move forward to the second stage-guiding the machine to perform the tasks we assigned, and according to Social values operate. **All of these technologies are currently being researched and developed by us. The next phase of technology and product testing will take place over the next few years as the use of the technology continues to expand.

**Moderator:**In the next 3 to 10 years, which industries will AI affect?

Brockman: I think this technology can be applied to the legal industry, and it can bring a lot of benefits by providing legal services. My favorite service from GPT-3 is a tool called Augrented, which helps tenants who receive eviction notices understand contract issues. Because there are a lot of people in that demographic who don't necessarily have access to a lawyer, this technology can help them do things they wouldn't otherwise be able to. That's one of the very effective use cases we've seen.

Moderator: Now there is a saying in the market that AI amplifies human intelligence. For example, GPT-4 can be used to write a book. What do you think about this?

Altman: I very much agree with this statement, and this is the direction we are doing our best to promote. I saw a tweet that stuck out to me, **Someone said he never thought he would coexist with a powerful intelligence like GPT-4. You'll find yourself deep in the realm of AGI, and it turns out that it's entirely possible to build an AI that's really good at amplifying the human will and making us stronger. **This will make our work more efficient and better. What's more, it turns out that people really like it. For me, the most gratifying thing about **OpenAI is that people love its products and get all kinds of unexpected benefits from it. **

"We are a very truth-seeking organization"

Moderator: When your OpenAI team is researching and developing AI products such as ChatGPT, did you encounter any surprises?

**Brockman: The first surprise dates back to 2017. At the time we produced this paper called "Unsupervised Emotional Neurons", but it didn't get much attention. But this is a real breakthrough for us, and we really need to pursue this paradigm. **We trained a model to predict the next character in an Amazon review, and it was able to learn where nouns are and where verbs are, but even more amazingly, it learned state-of-the-art sentiment analysis and classification, which can tell Whether your comment is positive or negative. We never told the machine the meaning of these languages, it somehow figured it out by performing this task. **Machine learning is not perfect, we still have a long way to go, it is still early days, but we can start to solve problems. **

It is very meaningful to me to let more people feel the joy of programming. People who have never programmed before can enter the field, and those who are already good programmers can complete more creative activities. Reaching this ability and level can accelerate our coding workflow.

**Moderator:**Can you expand on how to speed up coding?

Brockman: During the coding process, we spend a lot of time remembering exactly which libraries do what you're looking for, how to implement them, and what the parameters are. And you want to build a system or an application, this is an advanced task, the previous Copilot (assistant programming tool) is just boilerplate, they are only rote. Now the model is starting to get better, and if you don't know much about a particular programming language, or if you don't know a particular set of libraries, the model will happily provide that. And the interesting thing about GPT-4 is that it can point out where the encoding is wrong, and these error messages are sometimes cryptic, and you have to dig around for answers, like looking at various websites, trying to piece together from what other people have said , now GPT-4 can help you solve these problems.

**Moderator:**Altman, what was the surprise you felt?

Altman: I quote a quote from Noam Shazeer (Founder of Character.AI): "We don't explain why these architectures seem to work, we attribute these results to the mercy of God." This is for me A recurring theme, **I consider myself an empiricist, I believe that if something works, the more you do it, the better it will be. **The "unsupervised affective neurons" that Greg just mentioned was a very big surprise, and now some of the things that GPT-4 does start to explain why AI should work. I totally believe in AI and I think it will go far. **Before we started OpenAI, I was amazed when I saw the results of AlexNet (Convolutional Neural Network). **

Moderator: With only a few hundred people, you have surpassed the level that can only be achieved by the world's largest technology company. This requires a lot of hard work and some luck. I'm curious to know, what is driving you to achieve this goal? Are there key leaders within OpenAI?

**Altman: We do a lot of things, some on purpose, some by luck. One of the big factors is that we've built a culture of really paying attention to the details and trying to get them right, attracting very careful engineering and scientific research from those who want to work that way. Another reason is that I think we're a very truth-seeking organization, finding what works and doing more. **

When we started OpenAI, we were different from other AI research that existed in the field, they had different incentive systems, and things to consider were scattered, and we took a highly focused risk, doing what was considered unimaginable at the time things. We put most of our resources into this project, and although there are risks involved, we have studied it carefully and are confident that we can deliver results.

**Brockman:**My idea is to start from "first principles". In the very early days, Ilya (Chief Scientist of OpenAI) would debate with me for an hour or so every day. For example, we had a long discussion about who we should hire, should we hire a traditional PhD in machine learning, or should we hire a software engineer who has never done this before, or someone in between? For every decision, we must see their consequences, and then we will continue to learn and update.

We cannot know everything in advance, but we can learn by doing. Just like the OpenAI-Five we developed and continued to expand its scale, in the game competition of Dota2, it does not fall behind at all. (Note: OpenAI-Five, an artificial intelligence that can play Dota2 games, Dota2 is a multiplayer online combat arena game with a large number of fanatical fans. In 2018, they challenged the world champion team OG team with all their strength.)

This is our first system that really scales, and the system gets better every week as we put more computing power into it. Our goal is to exhaust the space of existing algorithms so that we can do new algorithm development, which is what we really want to do. When you try to pursue this direction, another direction pulls you forward and says it works. ** Responding to what we see in front of us, constantly updating, pivoting, changing, is very important from a technical point of view. **

Altman: When we scaled up the OpenAI-Five project, basically no one thought it would be successful. The fact that you can throw any amount of computation at an AI system to achieve the desired level of intelligence, and this covers an incredible range of computation and intelligence, is like an amazing scientific discovery.

Brockman: We also looked at the advancement of algorithms, combining the best systems with the best algorithms, and systems like GPT-4 were born. **When we think about this problem, we realize that if all the information of the GPU is input into the supply chain of the new algorithm, or even uploaded to the large-scale data processing system, then this is a big project, and no company will way to do it alone. In a way, it's actually a whole-of-human project that involves many companies. **

"AI is a productivity tool for everyone"

Moderator: Altman, someone tweeted you a tweet: "The next thing you're going to say is that we should regulate AI." You reply: "Yes, we should regulate artificial intelligence." Smart.” So what do you think of the many people who say, “You guys are moving too fast,” what do you think about regulation, global governance, or talent acquisition?

Altman: I understand everyone's anxiety and fear. People are afraid of the speed of technological change. We also agree to proceed with caution and to strengthen the regulation of artificial intelligence, but there is no point in pausing research.

**What we really need to do is figure out what regulatory approaches, rules and safety standards really work in the context of the messy reality. The rate at which the world changes should be kept relatively constant, rather than building AGI in secret and then at some point immediately making it public. **Institutions don't get great overnight, they should gradually be part of the technological development, provide critical feedback to scientific research, tell us when stupid mistakes are made, identify areas of great benefit and potential harm , make mistakes and learn from them when the risk is lower than expected.

Brockman: We're really trying to figure this out. **When we founded OpenAI, what we really considered was how to develop a safe AGI, which determines whether it will benefit all mankind. There are two issues here, one is technical security, and the other is social security. It is necessary to ensure that artificial intelligence technology is guided to be used safely. **

I'm relatively technical, so I'll focus on solving the first problem. We've started deployments in the past four or five years, and when GPT-3 has problems, we've patched it the way GPT-3 has been distributed, using filters and non-model-level interventions. Now that our technology is maturing to actually be able to intervene at the model level, GPT-4 can understand the results in the model in an automated way, so it does a very good job of security filtering.

Moderator: One of the criticisms of artificial intelligence is the use of energy. Altman you think there will be not only infinite intelligence in the future but also infinite energy, you have invested heavily in Helion Energy (a clean energy company), can you talk about why you invested in it and what results you expect ?

Altman: I don’t think there will be infinite intelligence or energy in the future of AI. I think it should be falling prices and increasing rich products. It is impossible for us to reach this level of infinity. I find these energy criticisms of AI systems a vulgar one. **I think that abundant and clean energy has other benefits than just solving the climate crisis, and that investing in energy is a more important thing than investing in any other single input I can think of. Funding a fundamental change in the cost of energy acquisition is a great thing. **Helion is working on this and I hope to hear good news next year.

**Moderator: **Brockman We understand that you are investing in some companies in the fields of health care and drug development. What do you think?

Brockman: There is a fund within OpenAI that invests in some technology startups. Startups like Ambience Healthcare (an artificial intelligence company in the medical field) are actually trying to apply AI to actual medical activities. Even Microsoft is beginning to implement its AI strategy in the healthcare industry through Nuance (artificial intelligence software company). If you think about the array of problems doctors face, many of them are administrative. My parents were both doctors, and I heard all the stories about how they were forced to use iPads to talk to patients. **We're going into a world where we can look at Nuance's instant transcripts of physician information and get physician notes after the fact, reducing the amount of explaining effort for the physician to the patient's family and allowing the physician to truly focus on the patient's care. **

We've already seen the benefits of people using ChatGPT to see a doctor on Twitter. Another person saved his dog’s life through ChatGPT. It was reported that he went to the veterinarian, but the veterinarian did not treat him in time. Netizens anxiously uploaded the medical records to ChatGPT. However, ChatGPT is pretty sure about "I'm not a veterinarian, you need to talk to a veterinarian". However, he was also given some advice and explanations, which gave him the confidence to go to a second veterinarian. I think it's a really interesting example because he really points to the problem of how we want AI to step in. Humans are very dependent on doctors in the field of medicine, but humans also make mistakes, and the first doctor made a wrong decision that could have fatal consequences. So figuring out how to get the right oversight by the right people.

Moderator: You are investing in companies using your platform through OpenAI, will this make a difference? Will other companies use your API (application calling interface)? Curious to know what is the best way to use AI to make all industries better?

Altman: We now see that people call OpenAI’s API everywhere, and as the model becomes more and more intelligent, the capabilities of products and services can be continuously improved, so we can answer this question with confidence , and this situation will continue.

**Moderator:**What do you think the use of artificial intelligence or modern artificial intelligence technology will bring to scientific research within 3 to 10 years?

**Brockman:**Chinese mathematics genius Terence Tao recently said that GPT-4 has accelerated his speed. Terence Tao is a well-known mathematician. He said that using GPT-4 to help him write funding claims in project declarations can bring great help and solve some tedious tasks. In fact, people spend a lot of time and energy in some monotonous and repetitive tasks every day. The smartest people in the world are spending so much of their time doing this not-so-creative work, it's such a waste. AI can help them solve these unimportant tedious tasks, freeing people to do things of real value, and they can achieve more.

**Altman: Yes, I think AI is a productivity tool for everyone that will make our best scientists better and advance scientific research. **Ultimately, these systems could help us store more knowledge than the human brain and discover new connections, new ideas, etc.