Altman believes that AGI was originally only slightly smarter than humans, but in the end it may be 1 million or even 1 billion times smarter than humans, and it will benefit humanity around the world.
Editor’s note: This article is from “Tencent Technology” “Silicon Valley Cover” section, author: Deer, authorized reprint.
[Editor’s note]
Sam Altman is an American entrepreneur, angel investor, programmer, and well-known blogger. As a student at Stanford University, he led the Silicon Valley start-up incubator Y Combinator to help create more than 2,000 emerging technology companies, including unicorn companies such as Dropbox, Airbnb, and Stripe. The total value of incubators is more than $100 billion. However, just as Y Combinator became hot and even regarded as a success criterion in the world, Altman suddenly resigned as president in March and turned to OpenAI, a non-profit artificial intelligence (AI) research organization. And under the leadership of Altman, OpenAI has also begun to turn to a new direction. In fact, everything the “YC Godfather” did was to achieve an ambitious goal of bringing universal artificial intelligence (AGI) into reality.
Focus on focus
1, Microsoft’s $1 billion investment can help OpenAI achieve its exciting goal of creating Universal Artificial Intelligence (AGI), a machine that can simulate any task performed by the human brain.
2, Altman believes that AGI can be built in a similar way to the development of AI technology in “Knife 2”, which strengthens human intelligence.
3, Altman said that most of Microsoft’s $1 billion investment will be spent on the computing power that OpenAI needs to achieve its ambitions. OpenAI is starting with smaller projects, such as the recent development of systems that attempt to understand natural language.
4, Altman believes that AGI was originally only slightly smarter than humans, but in the end it may be 1 million or even 1 billion times smarter than humans, and it will benefit humanity around the world.
(This article is about 5200 words, it takes about 6 minutes to read the full text)
The following is the body of the article:
After 30 days of hard negotiations, Sam Altman finally convinced Microsoft to lead him.The AI Lab OpenAI invested $1 billion, which marks OpenAI’s move to a new direction. In March of this year, Altman resigned as president of Silicon Valley start-up incubator Y Combinator, where he helped Altman quickly become a Silicon Valley elite. Now, at the age of 34, he is the CEO of OpenAI, the AI lab he co-founded in 2015 with Elon Musk, CEO of electric car manufacturer Tesla.
I hope to simulate the human brain mechanism
Muske left the lab last year, focusing on his AI ambitions at Tesla. Since then, Altman has been working to re-engineer the non-profit organization OpenAI into a profitable company so that it can seek financing more aggressively. Now, Altman has found big-name investors to help it achieve exciting and lofty goals. Altman and his research team hope to build Universal Artificial Intelligence (AGI), a machine that can simulate the human brain to perform any task.
AGI still exists in science fiction. But in the agreement between Microsoft and OpenAI, the two sides discussed the possibility of turning it into reality in the same factual language and applied it to any other technology they wish to build, whether it is a cloud computing service or a new type of robot. Arm. Altman recently said in an interview: “The goal of running OpenAI is to successfully create a broad and beneficial AI, and building a partnership with Microsoft is the most important milestone to date.”
In recent years, small but enthusiastic communities of AI researchers are turning their attention to AGI, and they are supported by many of the world’s richest companies. DeepMind, the top AI lab at Google’s parent company, said it is pursuing the same goal. Most experts believe that AGI will not come for decades or even centuries. Even Altman admits that OpenAI may never achieve its goals, but competition continues.
Accepted with AltmanIn a joint telephone interview, Microsoft CEO Satya Nadella later compared AGI’s efforts to build a quantum computer with his company, which would be much faster than current machines. Nadella said: “Whether it is our pursuit of quantum computing or the pursuit of AGI, I think you need these ambitious ‘North Stars’ projects to guide the direction.”
But since its inception, OpenAI has achieved many impressive achievements. It developed open source software that allows other researchers to train and test their AI systems. It invented a new kind of software that can be learned from trial and error rather than through existing data. As one of its most famous achievements, it created five independent AIs that can team up to play the complex video game “Dota 2”. In fact, these AIs performed very well and defeated many of the world’s top human teams.
OpenAI’s technology shines in “Knife 2” thanks to a mathematical technique called reinforcement learning that allows machines to understand tasks through extreme experimentation and error. By playing games over and over again, as well as automated pieces of software (called agents), it can understand which strategies can help achieve success. These “agents” have learned a lot of skills in a few months and accumulated the equivalent of 45,000 years of gaming experience.
OpenAI also teaches a robot to manipulate objects with unprecedented dexterity. Recently, it created an algorithm that uses one or two sentences of human-written text, then automatically generates a novel, and maintains the same style.
Research and development of new technologies is costly
However, these breakthroughs are hard to come by. Some of them use a lot of computing power, which means it costs a lot of money. OpenAI has said that the company suspects that the way to get AGI is to increase the amount of data used in the experiment, and more servers are needed to train the algorithm. To this end, OpenAI spends millions of dollars on tens of thousands of computer chips, which are located in cloud computing services operated by companies such as Google and Amazon. In addition, the “human brain” behind AI is also very expensive, and the salaries of many top machine learning experts are staggering.
Although OpenAI received a $1 billion donation, it feels it needs more money. This non-profit organization is at a distinct disadvantage compared to the corporate AI lab. Many enterprise AI labs are part of a company with a large-scale cloud computing business, so there is no need to pay a high cost for server time. In addition, as the largest initial donor, Musk quit the OpenAI board of directors in early 2018, reducing the investment in OpenAI. OpenAI said that Tesla is increasingly interested in commercializing AI applications, which has created a conflict of interest for Musk.
In order to compete with big companies and startups with huge venture capital reserves, OpenAI’s board made an important decision: if it can’t beat these companies, join them. In March of this year, Altman announced that OpenAI will create an independent profitable company called OpenAI LP to seek external capital. Then this week, it received a huge investment from Microsoft. The specific terms of the transaction have not been disclosed, and this cooperation has left many observers with no idea. Reichter said: “Microsoft already has one of the world’s most respected private research institutions – Microsoft Research.”
In response to email questions, Greg Brockman, the former chief technology officer of the start-up company Stripe and current OpenAI chairman, said that Microsoft “has a large but not a majority” in this profitable entity. Shares. He added that Microsoft also has the right to get a board seat in this non-profit organization, but it has not yet exercised this right.
Microsoft said its Azure cloud computing infrastructure will be the platform of choice for OpenAI research, so some of its $1 billion in cash investment may be in return for Microsoft in the form of cloud computing. But the two companies also said they will cooperate to develop a new AI-specific supercomputer.
Most importantly, Microsoft said it will be the preferred partner of OpenAI, commercializing any technology developed by OpenAI other than AI. Brockman said that this arrangement is not inconsistent with OpenAI’s anti-company founding philosophy.
He said: “OpenAI has the right to choose, but is not obligated to authorize any particular technology. The purpose of authorizing some of our technology is to fund our efforts to build AGI so that we can accomplish our mission. Authorizing certain AGI technologies violates our mission, and we will not do so.”
Oltman and others believe that the development of AGI is a huge project, the cost will never be cheap, and may require billions of dollars in investment. He said: “In the next few years, we may need to invest in large-scale cloud computing to attract and retain talent and build AI supercomputers.”
Building AGI has a long way to go
In the end, Altman and his colleagues believe that they can build AGI in a similar way to Turret 2. If they can collect enough data to describe the various tasks that humans handle every day, and have enough computing power to analyze all of this data, they believe they can enhance human intelligence..
Oltmann describes the deal with Microsoft as a big step in this direction. While Microsoft is investing in OpenAI, the technology giant is also working on building new computing systems to help the lab analyze more and more information. Nadella said: “This is about the establishment of a truly close feedback cycle between the ambitious AGI pursuit and our core business. Our core business is to make the world of computers.”
This work may include computer chips designed specifically for training AI systems. Like Google, Amazon and dozens of startups around the world, Microsoft is already exploring this new chip.
Since AGI doesn’t exist yet, OpenAI is starting with smaller projects that have recently developed systems that try to understand natural language. This technology addresses everything needed from digital assistants such as Alexa and Google Home to automated analysis of file software within law firms, hospitals and other businesses.
This deal is also a way for the two companies to self-promote. OpenAI needs computing power to achieve its ambitions, but it must also attract the world’s leading researchers, which is difficult to do in today’s talent market. And Microsoft is competing with Google and Amazon in the field of cloud computing. In the field of cloud computing, human AI capabilities are becoming more and more important.
The question is how we should take AGI’s ideas seriously. Like everyone else in the technology industry, Altman often said that AGI’s future seems unstoppable. He said: “I think AGI will be the most important technological development in human history.” Altman hinted at the concern of Musk and others that AGI may be out of our control. In this regard, he said: “Finding control methods will be one of the most important social challenges we face.”
But games like “Knife 2” are far from the complexity of the real world. In recent years, AI has made significant progress thanks to many technologies developed by companies such as DeepMind and OpenAI. Some systems recognize images, spoken words, and translate them accurately between languages. But this does not mean that AGI is about to become a reality, or even that it is possible. But Oren Etzioni, chief executive of the Seattle-based Allen AI Institute, said: “We are closer to AGI than ever before.”
Recently, Google researcher Geoffrey Hinton won the Turing Award for his contributions to AI research over the past few years. Recently, someone asked him about the AGI competition. He said: “This is a very general problem. I am more willing to focus on things that you can come up with.” He added that another question about AGI is: Why do you need it?
Can you redeem your promises?
OpenAI was founded less than four years ago and took a brave stance against the idea that “a powerful new technology must be controlled by a large enterprise.” The goal of the non-profit organization at the time was: “Developing artificial intelligence in ways that are most likely to benefit all humanity, without being bound by the need to generate economic returns.”
However, earlier this week, OpenAI, which was already quite different from the start-up, accepted Microsoft’s $1 billion investment, one of the world’s most valuable technology companies. The deal gives OpenAI a huge source of funding and can compete with well-funded competitors such as Google, Facebook, Tencent and Baidu.
Under Altman, OpenAI has established a profit-oriented department. As for the idea of opposing the establishment of advanced technology for large companies, the situation seems a bit subtle. So how do you benefit humanity? This is still the mission of OpenAI, but Microsoft now needs to make some money in the process.
As a non-profit organization, OpenAI has received $1 billion in donations from many wealthy supporters, most notably Tesla CEO Elon Musk. According to a biography written by Bloomberg writer Ashlee Vance for Musk, Musk worried that AI might get out of hand, including “blowing an AI-enhanced robotic army capable of destroying humans.”
Muske is also concerned about the intellectual property rights of a few companies with General Artificial Intelligence (AGI). AGI means a technology that excels in a variety of different tasks, like humans, rather than the narrow AI that can only do some tedious work, such as playing chess or recognizing faces.
OpenAI echoed Musk’s worldview in his first blog, saying: “As a non-profit organization, our goal is to create value for everyone, not shareholders. We believe that AI should be an extension of personal will, and The spirit of freedom is deployed as widely and evenly as possible.”
Although OpenAIIt makes people feel good, but some people are skeptical about whether the organization is a suitable vehicle for democratizing AI benefits. Ben Recht, a professor of computer learning algorithms at the University of California at Berkeley, is one of his critics, and he is concerned about the key roles played by Musk and Altman. He said: “Anyone who believes that Musk and Altman are benefiting humanity in non-profit organizations is deceiving themselves. This is a weird Silicon Valley vanity project.”
Joining OpenAI is not just for money
So, why did Altman resign as the president of Y Combinator and join OpenAI as CEO?
AGI is an interesting bet, because it doesn’t exist yet, and the technology industry is full of controversy about when AGI will become an implementation. Some people think it will appear soon, others think it may never happen.
In Altman’s view: “We are working hard to build a secure AGI. So I believe that in the next few decades, someone or someone will build smarter than humans in every way. And a more capable software system. Initially, it may only be a little stronger than humans, but over time, its capabilities may exceed 1 million or even 1 billion times humans. So we are trying to figure out how to achieve it technically. At the same time, AGI wants to ensure its security and fairness, and share its benefits.”
When asked how the OpenAI program made money, Altman replied, “The honest answer is that we don’t know. We have never made any income, and we don’t have any plans to generate income. We don’t know if we will How to generate income.” However, Altman has made a soft commitment to investors: “Once we have established an AGI system, basically we will let it find ways to bring you a return on investment.”
Oltmann wants OpenAI to set an example for other companies to replicate the so-called “capped profit” business model. OpenAI said that its first profit supporters have a profit limit of 100 times the initial investment, and any excess profits will be returned to the non-profit organization. The company said that the profit ceiling of subsequent investors will be different.
It’s worth noting that 100 times return is a very high threshold, and most investors who invest in ordinary for-profit companies rarely get such high returns. It is said that institutional investor Sequoia Capital has given WhatsApp invested $60 million. When the latter was bought by Facebook for $22 billion, Sequoia Capital received a 50-fold return, which is already an amazing number.
OpenAI said that Microsoft did not participate in the first round of financing, the first round of financing including Khosla Ventures (Khosla Ventures) and LinkedIn co-founder Reid Hoffman’s charitable foundation. Hoffman is also one of the co-founders of OpenAI, but the company declined to disclose Microsoft’s ceiling. According to OpenAI, most of the value will ultimately be owned by humanity, as required by the company’s charter. But at the same time, OpenAI may make Microsoft, which is among the richest companies in the world, richer.
Therefore, you first need to build AGI, and secondly use it to make money. Altman insisted: “I really believe that the work we do at OpenAI will not only far exceed the work I did at YC, but will far exceed the work done by anyone in the technology industry.” /p>