In five years, it will only cost one jiao to write software. Here comes a ten-thousand-word transcript of the latest interview with Ultraman.

On July 23rd, news emerged that during the recently held "Comprehensive Review of the Large Bank Capital Framework" meeting, Sam Altman, the CEO of OpenAI, had a fireside chat with Michelle Bowman, the Vice Chair of the Federal Reserve. They explored the impact of AI on the economic and financial sectors and emphasized the unprecedented potential of AI in enhancing efficiency and reducing costs. In this conversation, Altman predicted that AI would significantly boost productivity across all industries and even give rise to entirely new business models. As the implementation of AI applications accelerates at an unprecedented pace, its impact on society will be profound and transformative.

Meanwhile, Altman also admitted that the development of AI comes with a series of challenges. For example, it may trigger a structural reshaping of jobs and bring new risks in areas such as cybersecurity and biological threats. This requires society to be highly sensitive and forward - looking in the application and regulation of AI.

He also stressed that AI not only has commercial potential but also holds positive significance in promoting inclusive services and reducing biases at the social value level. Altman compared AI to "the new - generation transistor" - in the future, it will permeate all kinds of products and services as widely as transistor technology, becoming an omnipresent basic capability.

The following are Altman's core viewpoints:

1. The cost per intelligent unit of AI has decreased by more than ten times each year in the past five years, and it is expected to maintain or even accelerate this trend in the next five years. This exponential reduction in cost will make intelligence extremely cheap and widely accessible.
2. The pace of AI's transformation far exceeds that of the Industrial Revolution, the Computer Revolution, and the Internet Revolution. In some fields (such as programming), the cost can drop from $10,000 to a few cents within a year. This rate of cost reduction in application is unprecedented in history.
3. AI will become infrastructure like the transistor. In the future, there will no longer be dedicated "AI companies." Instead, all products and services will integrate AI capabilities by default, becoming ubiquitous yet invisible.
4. Altman is particularly worried about the "financial fraud crisis" caused by AI because it can already fully mimic human voices and will soon be able to mimic videos, rendering existing voice recognition and selfie verification methods ineffective.
5. Altman pointed out three major AI risks: 1) Malicious use of super - intelligence (biological weapons / power grid attacks) 2) AI refusing to shut down 3) Society being silently taken over by AI.
6. AI can provide high - quality medical and financial consulting services at extremely low costs. Developing countries may even skip several generations of technological development and directly enter the "AI - first" model, thus achieving rapid economic transformation.

The following are the highlights of Altman's latest exclusive interview:

01 The era of low - cost AI services is here, with the cost per intelligent unit reduced by over 10 times

Bowman: Could you introduce the development process of this wave of AI?

Altman: Just five years ago, AI technology was regarded as a distant future technology. Even two and a half years ago, when ChatGPT was launched, it only caused a stir among Silicon Valley geeks. It wasn't until the advent of GPT - 4 that the adoption rate of AI technology began to accelerate, and its economic impact started to show.

Just last week, our model reached the gold - medal level in the International Mathematical Olympiad (IMO). A few years ago, most people in the industry thought this was impossible. In fact, ChatGPT can already compete with real human experts. Many scientists reported that their work efficiency has tripled or quadrupled, and programmers even said their efficiency has increased by 10 times, completely changing the way they write software.

In many fields, our AI systems are already comparable to human intelligence. Although they currently cannot handle long - term and complex tasks like humans, which is their limitation, the technology is still advancing. Now, we are very close to achieving intelligence at an extremely low cost. In the past five years, we have reduced the cost per intelligent unit by more than 10 times each year, and in the next five years, we expect to continue this trend, or even see an accelerated decline in cost.

Just a few days ago, I used a model we are about to release to complete a programming task - a project I've always wanted to do. I'm a "geek" in home automation and have always hoped that the room lights and music could change according to specific needs. In the past, without AI technology, it might have taken days to complete this task. But with this technology, I only spent five minutes! Almost all the work was done by AI. This was originally a task that a senior programmer would need 20 or even 40 hours to complete, and the cost for AI to do it might be less than $1 (calculated by tokens).

This is truly an amazing transformation, and the speed of this change will continue to accelerate in the next few years. However, I think this development potential is still not fully recognized. Even a year ago, we weren't sure how far the current research path could go or if we would encounter some kind of limit. But now it seems that we will continue to make important progress in the next few years.

Bowman: Can you compare the potential impact of AI on productivity with other technological breakthroughs we've witnessed in the past? Can you use an analogy to illustrate the current development status of AI?

Altman: I've never seen a technological revolution like the one we're experiencing today. Historically, we've witnessed the Industrial Revolution, the Computer Revolution, the Internet Revolution, etc., and people often mention them. But I don't remember any technology that could reduce the cost from $10,000 to a few dollars or even a few cents in just one year. Take the programming task I mentioned earlier as an example; this kind of change is unprecedented. Although not all fields are like this, robots in the physical world still need a long time to make a breakthrough. But for tasks that can be completed by computers, there is no precedent for this speed of change.

Another example: in 2020, you might have spent $10 on an Uber ride, $100 on an urgent package delivery, or $100,000 on software development. But with the development of technology, the costs of these tasks are dropping significantly.

By 2030, the same services - such as developing a software application, might only cost $0.1. Although the speed of change varies for different services, it is foreseeable that tasks that can be completed by computers will become extremely cheap, almost free. Of course, for tasks that rely on robots or fully automated services, such as fully automated humanoid robots driving cars, picking up packages, or pressing elevator buttons, it still takes some time to achieve.

As for an analogy for this change, my favorite one is the transistor. The transistor was a complex physics discovery. At first, it was very difficult to understand, but once mastered, its application became simple and economically disruptive. The value of this technology quickly spread throughout society, bringing a huge boost in productivity.

During that time, a large number of semiconductor companies emerged, and that period experienced a prosperous stage. Today, most semiconductor companies are no longer prominent, except for companies like TSMC. Now, almost everyone uses devices with a large number of transistors, such as mobile phones, computers, screens, and cameras. We no longer pay special attention to the "transistors" in these devices; they have become an indispensable part of our lives. The discovery of the transistor was truly remarkable; it changed what we could build and quickly became an important part of our lives.

Now, we hardly ever talk about "transistor companies" separately. Similarly, I think in the near future, we won't talk about "AI companies" separately either. You will naturally expect that every product and service will utilize this technology, and AI will become the norm.

This technology has strong scalability. We used Moore's Law to describe the iteration speed of transistors, and in the field of AI, I think we haven't found a specific term yet, but we do have similar "Scaling Laws". AI technology is getting better, and we are constantly learning how to use it.

So, I think this is the best historical analogy. The transistor brought a huge boost in productivity, and AI is doing the same thing, and its potential is far greater than that. More importantly, in some cases, AI can even invent new algorithms on its own, driving technological progress, and this impact is very profound.

Bowman: What impact will AI have on the labor market, labor productivity, or productivity in a broader sense?

Altman: I always remind my colleagues in the company that no one can be certain about what will happen next. Although many people make seemingly smart predictions about how a certain field will change, I personally think that no one can accurately predict the future. The world is too complex, AI technology is still new, and the changes it brings are so significant that it's difficult to make precise predictions.

Although some occupations may disappear due to the development of AI, at the same time, new occupations will emerge. Overall, I think this transformation will be similar to most technological revolutions in history: new technologies will change the way people work, enabling them to achieve more and better results. Take doctors, lawyers, or computer programmers as examples. Although the form of their work will change, people still need medical services, still want to communicate with their peers, still need reliable legal advice, and still hope that computers can help them.

As technology develops, the work that an individual can complete will become even more amazing, and our expectations for each individual will also increase significantly. Whenever a new technology appears, people in history often say "this means the end of work, the disappearance of jobs", but in reality, humans always seem to want more things, have stronger creativity, and have a strong desire to help others.

To this day, I'm still waiting for the ideal life that the Industrial Revolution promised, where we work a few hours a week and spend the rest of the time on the beach with our families. But the reality is that we are still fully committed to our work. However, I'm firmly convinced that the trend of technological progress is irreversible. Evolution has taken a long time, and our biological instincts have been refined to the extreme. This is our innate nature, and technology can't change it; you can't break it. Therefore, I think the fundamental factors driving us to work will not disappear or change.

Although we may complain about working too hard and although we are incredibly wealthy, if we could look at life 100 years from now, we might think that those people's work is not "real work" at all. They aren't really busy, have unimaginable luxury, have everything they need, and have nothing to do. So, they will invent work, participate in some meaningless games, and fill their empty time to make themselves feel useful to others.

Just as people 100 or 500 years ago would look at us today. That's the nature of things, and I think it's okay.

02 AI triggers identity fraud risks, and prompt injection attacks become a new security threat

Bowman: What risks do we need to address in terms of data protection and ensuring the effective use of innovation?

Altman: We originally didn't think that the financial industry or even government departments would be early adopters of our technology. Although AI has made a lot of progress now, when we first launched, the public's perception of AI was that it often "hallucinated" non - existent content. I remember when we launched GPT - 3, there was an academic survey among scholars who considered themselves experts in the field of AI, asking them what they thought the proportion of GPT - 3's "hallucinations" was.

In fact, the answer should be only 0.1% or even less. But these scholars generally thought that about 50% of the time, GPT - 3's answers were untrue. They thought that half of ChatGPT's answers were nonsense, saying obviously wrong things. Therefore, this label has accompanied us for a long time.

Based on this perception, we didn't expect that the financial industry and government agencies would be early adopters of our technology. But unexpectedly, some of our earliest large - enterprise partners were financial institutions, such as Morgan Stanley and the Bank of New York, which have become our important partners. They have found solutions to apply this technology to critical business processes reliably enough.

Now, we are also cooperating with government departments and gradually expanding our services to more fields. I remember someone once said: "We realized that this is a new technology. If we don't accept it, the biggest risk might be not being able to survive."

Similarly, if banks don't adopt AI technology, they won't be able to compete with new banks that have already achieved an AI - first experience. This is a highly innovative industry, and the adoption and promotion of AI have had better results than I expected.

Of course, there are still risks in this field, such as the AI "hallucinations" we mentioned before, and now there is a new risk - the so - called "prompt injections". Once the model really starts to personalize for individuals and their data, attackers may "deceive" the model in some way, making it reveal information it shouldn't or do things it shouldn't.

For example, if I know a lot of your personal information, I know when I can share it with someone and when I can't. But as the model masters more and more personal information, the risk also increases. We need to handle these risks carefully, although we can currently manage them effectively and gain great benefits from them.

Bowman: The banking industry is very concerned about fraud now. Do you think we can use AI to reduce bank fraud, and how should we prevent fraud by impersonating others?

Altman: This is a very serious issue and an area I'm very worried about. Let me give you an example: Some financial institutions still use "voice recognition" as an identity verification method, especially when transferring large amounts of funds. You just need to say a password, and the system will execute the transfer operation.

This is crazy! AI technology can completely break through these "voice verification" methods. Almost all current identity verification methods, except passwords, can be easily cracked by AI. For example, methods like selfies and voice recognition are very vulnerable in the face of AI.

I'm really worried that we will face a serious fraud crisis, especially as AI technology becomes more and more mature. We and others in the industry have been warning everyone. Although we haven't fully unleashed the potential of these technologies yet, they already exist. Moreover, criminals may not need much technical ability to use these tools easily. There are already reports that in some kidnapping and extortion cases, scammers used the voices of victims' relatives to make threats and even made emergency calls. In the future, this technology will become more and more realistic, and society must face this problem.

At the same time, people need to change the way they interact with others, especially how to verify "who is calling". Future voice calls may soon become video calls, almost indistinguishable from reality. Therefore, we must educate the public on how to verify identities in this environment and how to prevent this kind of fraud. This will be a huge challenge.

03 AI tools shouldn't be demonized; they are releasing the value of time

Bowman: Now many teenagers are starting to use AI tools to help them learn, especially tools like ChatGPT. What's your view on this phenomenon? In the field of basic education, how should AI be correctly guided to play a positive role rather than encourage shortcuts?

Altman: I have two things to share:

First, I never met my grandfather because he passed away before I was born. But my grandmother once told a story about the calculator. When the calculator came out, my grandfather was very opposed to it. He was very good at math. After hearing about the release of the calculator, he thought it would ruin math education because people no longer needed to learn how to use a slide rule or look up logarithm tables. He thought that since there was a calculator, why teach people math?

This story made me think a lot because it also reflects people's natural suspicion of new technologies. However, it turned out that better tools allowed us to use our brainpower in other more valuable areas. We started teaching calculus in high school, and students could access more advanced knowledge.

Second, I had a similar experience myself. I remember when Google first came out during my junior high school years. High school teachers were very panicked when they heard about this new tool. With Google, you no longer needed to remember specific facts, such