Original Shin Chi Yuan Shin Chi Won
Shin Ji Won reports
Editor: Run Lumina
Sam Altman recently revealed an astonishing prediction: AGI, that is, GPT-10, will appear before 2030, when its IQ will exceed the sum of all mankind!
"Humans may develop AGI before 2030."
Sam Altman revealed in a recent podcast interview that GPT-10 is AGI, smarter than everyone in the world combined!
And when the moderator asked, how to define AGI? Altman said:
If we can develop a system that can independently develop scientific knowledge that humans cannot, I will call this system AGI.
The birth of ChatGPT has set off huge shockwaves around the world, far exceeding the response of the AlphaGo man-machine war.
What does OpenAI want, you may ask?
In this cover story, WIRED delves into OpenAI's ambitions, strategy, and attempts to preserve the culture of the lab as the company grows.
The article states that the ultimate goal of OpenAI is to change everything.
What's also interesting is that OpenAI's financial documents even provide for an exit contingency in case AI destroys our entire economic system.
That is, when the real AGI arrives, if it becomes "Skynet", OpenAI has alternatives.
How far is AGI?
How does Sam Altman envision a world where humans and AI coexist?
In his opinion, when will artificial intelligence completely change the way we live?
Recently, Nicolai Tangen, an entrepreneur from Norwegian management funds, opened a conversation with Sam Altman to provide insight into how AI will affect the world.
Sam Altman's vision is for AGI to enable humans to do the work they "don't have to do", so that everyone can do the work they love and can fully devote themselves to.
He used his own example as an example where he said he loves his job and that everything he does now is something he is very willing to invest in. And perhaps in the future where AGI arrives, everyone can do only what they are willing to do, just like him, so that everyone can realize their full potential.
"When GPT-10 was developed, maybe it could be smarter than everyone in the world combined, can you imagine how much human productivity will increase with his help?" Altman said.
And how to reach such an end point, Altman also gave some specific explanations.
"I told the people in the company that our goal is to improve the performance of our prototypes by 10% every 12 months," he said. "If you set this target at 20%, it may be a bit too high."
According to him, in OpenAI, probably 15%-20% of employees actually write code by hand.
How to enable teams to work efficiently is a key factor in whether they can achieve their ultimate goals.
How to lead a team
When asked by the moderator how he manages and promotes employees within the company, Sam Altman said he thinks AI is actually a good fit for the role of leader.
If AGI is developed one day, he will make it the CEO of OpenAI, and he can take a break.
In fact, he says, many leaders tend to fail their companies because they don't do three things well:
The first is not spending enough time and energy on recruitment, so there is no way to build a good team.
The second is not spending enough time and energy communicating with the team about the direction and values of progress.
The third point is that not enough time and energy are spent thinking about strategic issues, which will forever be stuck in the details and unable to move forward.
From a personal point of view, he prefers to promote old employees from within rather than hiring a new person from outside for a leadership position.
But before he puts a new person in a leadership position, he will often talk to him first and then tell him that there are several problems that you may encounter in the future that you need to solve first.
Then tell him, of course, you can completely ignore these questions and do what you want, but I will write them down first, and then in 3 or 6 months we will see if things will develop as I said.
Often after 3 to 6 months, they would come to Altman and say they had noticed these problems. So it's important to tell them about these problems ahead of time and allow them to make mistakes.
In short, getting as good talent as possible and then letting them grow as quickly as possible to build trust as much as possible is a great way to manage genius.
At the same time, enabling users and the public to realize the advantages of OpenAI's continuously developed products, and getting more people to accept and use their products, is also something that must be dealt with before reaching AGI.
When the singularity strikes, humanity needs to be prepared
In his Wired article, Altman describes how he handled ChatGPT and GPT-4 releases at the beginning of the year. Explains why OpenAI released ChatGPT when GPT-4 was nearing completion and security work was underway.
"With ChatGPT, we can introduce chat functionality, but the backend functionality is much weaker and allows people to adapt more gradually," he said. "GPT-4 requires immediate adaptation to a lot of things."
By the time the excitement of ChatGPT wears off, users may be ready for GPT-4, which can pass the bar exam, plan a course outline, and write a book in seconds.
Conspiracy theorists might say that the pace of new product launches is tied to the company's commitment to investors and shareholding employees so that it is guaranteed to make money.
But OpenAI insists its real strategy is to provide an environment where the singularity can land softly.
"It doesn't make sense to secretly build AGI and put it out into the world," Altman said. "Looking back at the Industrial Revolution, everyone thought it was great for the world, but the first 50 years after the pre-Industrial Revolution were really painful. Massive unemployment, massive poverty, and then the world adapted. We are trying to think about how we can make it as easy as possible for AGI to adapt to the period before the world adapts."
How regulation is handled will determine whether the goal is achieved
OpenAI employs a lot of people who can help itself communicate well with regulators, hoping that they can help OpenAI avoid some of the difficulties and challenges that tech companies have encountered before.
A Russian-born Washington, D.C., insider, Makanju has held foreign policy positions at the U.S. Mission to the United Nations, the U.S. National Security Council and the Department of Defense, and served in office positions when Joe Biden was vice president.
"I have a lot of relationships with both the U.S. government and European governments," she said. She joined OpenAI in September 2021. At the time, few people in government cared about generative AI.
Knowing that OpenAI's products would soon change that, she began introducing Altman to government officials and lawmakers to make sure they heard OpenAI's good news and bad news first.
Senate Judiciary Committee Chairman Richard Blumenthal said, "Altman was very helpful and very shrewd in dealing with members of Congress." He contrasted Altman's actions with that of the younger Bill Gates, who unwisely obstructed lawmakers when Microsoft was under antitrust investigation in the '90s.
"In contrast, Altman is happy to spend an hour or more sitting next to me and trying to teach me," Blumenthal said. "He didn't have an army of lobbyists or caretakers. He demonstrated ChatGPT. It's really exciting."
Tom Rubin, an elite intellectual property lawyer who officially joined OpenAI in March, is optimistic that the company will eventually find a balance between its own needs and the needs of creators, including those who sue OpenAI like comedian Sarah Silverman.
Use their content to train their models. One hint of the OpenAI path: working with news and photo agencies like the Associated Press and Shutterstock to feed their models without disputes over ownership.
The relationship between risk and return
In the interview, Altman also talked a lot about his thoughts on entrepreneurship, management, and possible future changes in the AI and technology industries.
Altman said, "Probably most people overestimate the risk and underestimate the benefits."
He further explains, "A lot of people are less likely to make risky choices, such as starting a company, fighting for a job that they are willing to put into it, or realizing a product idea. They always think it's too risky."
"But when you tell him that we're going to look specifically at what the risks are, people say because the company may go out of business, but a stable job keeps you stable."
"But in my opinion, just like many people worry about other risks, I personally feel that if a person spends 20 years doing a job that does not improve, does not produce new things, does not learn new knowledge, the risk of such a job is catastrophic."
And for the global visit he just completed, he also talked about his feelings.
He feels that in most developing countries, people are most concerned about the economic impact of AI technology.
In more developed countries, the conversation is more about how AI is and what shortcomings it might cause, and what damage it might cause to existing things.
Relationship with Microsoft
When asked about the relationship between Open AI and Microsoft, Sam Altman replied: From a macro perspective, the two companies are more like cooperative partnerships. Microsoft provides hardware and computing power for OpenAI, OpenAI trains the model, and then the resulting model is shared by the two companies.
Referring to the recent news that OpenAI and Microsoft will also have some business conflicts, it seems that the relationship is not as close as it seems, Sam Altman said:
"Like any relationship in life and business, nothing is perfect, but the most important thing is that we are consistent at the top, so all I can say is that our relationship is very good."
"Because there's no relationship where we agree on everything, but we maintain a very strong ability to understand and compromise with each other for a long time."
Being able to think long-term is a huge competitive advantage
When the host asked him, what do you think is your greatest strength?
Sam Altman replied that he believes that his greatest ability is the ability to think long-term. Because almost no one thinks this way.
But in his view, that's why they were finally able to develop a necessary capability for AGI.
Because AGI may be something that will happen 50 years from now, no one will aim for this thing to direct and schedule their daily work, but if you don't, it will be difficult to reach that goal in the end.
Outside of AI, he's most interested in controlled nuclear fusion
He felt that if humanity is to create a prosperous and carefree future, the two most important things are:
The first is to make the cost of intelligence drop significantly, and the second is to make the cost of energy drop significantly.
OpenAI is doing the first thing, and the second thing may only be achieved by controlled nuclear fusion.
Giving humans access to unlimited clean energy will change everything.
OpenAI Developer Conference
Let's go back in the future time Altman described and see what OpenAI's recent progress is in reality:
Just today, OpenAI announced that it will host OpenAI's first global developer conference, OpenAI DevDay, on November 6 in San Francisco, California.
The OpenAI team will exchange ideas and showcase the latest OpenAI research with hundreds of developers from around the world.
In addition, on-site developers can participate in breakout sessions led by OpenAI technicians.
While OpenAI DevDay is primarily a live event, some sessions, including keynotes, will be streamed live.
Offline attendees are capped at a few hundred developers to ensure an intimate experience for attendees.
Registration Address: https://devday.openai.com/
OpenAI did not disclose the research that will be presented at the conference.
But GPT-5 is less likely. Because after a March 28 letter signed by tens of thousands of technical experts suspending AI stronger than GPT-4, Altman publicly stated in April that OpenAI stopped training GPT-5.
The conference is likely to present an update on OpenAI's acquisition of Global Illumination in August, as well as the availability of GPT-4 image understanding capabilities.
Previously, due to privacy concerns, the image understanding function had not been live.
OpenAI's developer conference also foreshadows OpenAI's vision of building a developer community.
Since opening up the API in 2020, OpenA has been constantly updating and upgrading it to incorporate state-of-the-art models.
As a result, developers around the world can integrate cutting-edge AI into their projects with simpler API calls than ever before.
Today, more than 2 million developers are using GPT-4, GPT-3.5, DALL-E, and Whisper in a wide range of applications:
From integrating smart assistants into existing applications to building entirely new applications and services that weren't possible before.
CEO Sam Altman said, "We look forward to showcasing OpenAI's latest research at the developer conference to help developers realize new ideas and ideas."
But for OpenAI, the developer conference is also a great marketing opportunity.
According to The Information, while ChatGPT is known around the world, OpenAI spent $540 million last year alone to develop ChatGPT.
But because of ChatGPT's outstanding performance and the launch of more OpenAI products, such as enterprise-level ChatGPT, OpenAI's revenue is expected to reach $1 billion next year.
But with the intensification of AI competition, soaring AI hardware costs, GPU shortages and a series of other challenges, OpenAI must find more and better ways out.
The convening of DevDay is a milestone in the development of OpenAI.
The conference not only provides an opportunity for OpenAI to interact with the developer community, promote the latest achievements of OpenAI, but also solidify its position as an AI industry leader.
For this developer conference, netizens are full of expectations and have said that they can't wait!
"Can't wait!"
"This is so hot, can't wait!"
Some people hope that OpenAI can broadcast all the content of the conference live because of the limited number of seats in the offline conference or cannot attend the conference.
"Is it possible for you to broadcast the whole conference live, because there are a lot of interested developers who are not in San Francisco and they definitely want to know everything about the developer capabilities and experiences?"
Looking forward to OpenAI DevDay is also out of netizens' love for OpenAI products and services.
They speak highly of OpenAI:
"I've been using their service since May, and over time, OpenAI's service has gotten better and better: it's faster and there are no inaccuracies.
For us, this is the best proof that they place great importance on the construction of the platform.
While I'm not a big customer, and because of the need to sign a Data Protection Agreement (DPA), I actually need to take a different approach for some of our customers when using OpenAI's products.
But choosing OpenAI is well worth it."
Resources:
https://openai.com/blog/announcing-openai-devday
https://www.youtube.com/watch?v=O77UyYK51s4&t=1670s