Elon Musk starts XAI to understand universe truth, but faces many challenges

At four o’clock in the morning on July 15th, Musk invited the founding team of XAI to Twitter for a live broadcast. In two hours, he told what this artificial intelligence company, which was only established for two days, was going to do.

Two days ago, the artificial intelligence war in Silicon Valley changed from the rivalry between the two heroes to the Romance of the Three Kingdoms. XAI, founded by Musk, became the third force after Google and OpenAI.

In short, Musk believes that humans have already solved some of the basic questions about the truth of the universe. If artificial intelligence cannot solve some basic questions, such as where are the aliens, then it cannot surpass humans.

He proposes a new view that the ability to solve fundamental questions about the truth of the universe should be regarded as a key threshold of artificial intelligence. He wants to subvert the paradigm of large models in the past. He said that “truth is not a popularity contest. If a model is trained on Internet data, it will give you a popular but wrong answer.”

All 11 members of the team also made an appearance and introduced their backgrounds. What is striking is that five of the team members have Asian faces, and at least four of them are definitely Chinese.

Similar to Musk, they are all non-standard academic masters with good grades and “rebellious”. For example, a young member named Younger dropped out of Harvard twice, the first time to become a DJ, and the second time to create an artificial intelligence smarter than himself and eliminate himself.

Musk is like-minded with them. He likes these newcomers who are not bound by existing experience, and likes to re-walk the road that others have already gone through in a unique way.

Among the researchers who developed ChatGPT, Chinese accounted for one-fifth, but in AXI, they accounted for nearly 40%. A senior industry expert introduced the current talent phenomenon, “If the Chinese account for one-fifth, it is the average level”, but it has exceeded this number, indicating that the Chinese are more suitable for AI.

Perhaps, as Musk recently said in a live broadcast: “I’m a bit pro-China.”

The outcome of the artificial intelligence war will undoubtedly determine the future pattern of Silicon Valley. As the former richest man in the world, although Musk has always been vigilant about artificial intelligence, he is absolutely unwilling to give up this field. Following Tesla, SpaceX, Twitter, Neuralink and The Boring Company, XAI has also become Musk’s sixth company.

Unlike Google and Meta, which gather the originators of the industry, most of the XAI team are young and new faces. Musk hopes that they will create more possibilities. This time, can the legend of Iron Man continue to write a new chapter?

0 1

From space to the universe, Musk continues to star the sea

Two days ago, the slogan of XAI’s debut was very Musk, and the entire page only had three concise words “understand the universe” (find the truth of the universe).

In April of this year, Musk had announced that a TruthGPT would be released. For half a year, he has been criticizing OpenAI, arguing that OpenAI is training ChatGPT to lie for “political correctness”.

Musk has always been wary of AI, and the time is shorter than everyone predicted. Musk believes that in five or six years, AI smarter than humans will be born. “It is difficult to predict when silicon-based organisms that are much smarter than humans will be born.” What happens when it shows up?”

Musk is worried that artificial intelligence will destroy mankind. He called on the government to regulate artificial intelligence. “Artificial intelligence may be more dangerous than nuclear weapons.” The solution he gave was to create an artificial intelligence that understands the universe better. He believes that if XAI grasps the true meaning of the universe, maybe it will find that humans are also an interesting part of the universe.

For a sky-high entrepreneurial project, telling a story of a sea of ​​stars has always been Musk’s standard configuration before starting a business.

A doctor of physics from the Chinese Academy of Sciences who is starting an artificial intelligence business expressed his confusion to the author: “Unless Musk believes that the essence of the universe is mathematics, it is possible to use artificial intelligence to discover mathematical laws.” However, this artificial intelligence technology Obviously, it is not possible for the most popular large models at the moment. “Large models operate according to mathematical logic, so there is no way for them to jump out of mathematical logic.”

“Looking for the truth of the universe”, this is a slogan that confuses the industry leaders.

A top domestic artificial intelligence expert believes that this is probably just a slogan. He told the author: “If AI can assist physics, especially quantum mechanics, to uncover the truth of the universe, it would be great. However, this is only half of the truth. The other half is that the universe will produce more and more complex intelligence, and this truth cannot be revealed. , will only expand endlessly.”

So, what exactly is XAI?

It is still the familiar “X”. The payment company that earned Musk the first pot of gold is called X. The nickname of his favorite child is also X. After acquiring Twitter, he changed his name to X Corp. Of course, there is the most famous rocket company – X space.

This time XAI is still a continuation of Musk’s space dream. According to his explanation, there are still many basic problems that have not been broken in the world, such as the Fermi paradox, such as the nature of gravity. Before the live broadcast on July 14, Musk had announced: “What unsolved mysteries do you think the universe has?”

Musk said: “From the perspective of the universe, there are many such basic problems. I don’t think you can call it general artificial intelligence unless at least one basic problem is solved. Because humans have solved many basic problems, or basic problems. Solved above. If AI can’t solve one of these problems, then I think it is inferior to human. Solve an important problem, and that will be a critical threshold for general artificial intelligence. You can reformulate XAI’s mission statement as: Universe What really happened?”

The first researcher Musk recruited for XAI, Igor Babushkin, was originally a physicist. He studied physics at university and worked at the Large Hadron Collider before turning into an artificial intelligence researcher.

Musk wants to subvert the paradigm of large models in the past, and does not want the training of large models to be trapped in Internet data. “Current models just mimic the data they’ve been trained on,” he said. “What we’re really trying to do is shift that paradigm to models discovering truth, not just repeating what they’ve learned from the training data, but generating real new insights.” , new discoveries, from which we cannot benefit”.

In the live broadcast, when someone asked Musk whether he would become a competitor of OpenAI and Google, he gave an affirmative answer.

Although the future of XAI is a sea of ​​stars, Musk still hopes that it will be a tool that everyone can use. “We’re just getting started and it’s going to take some time to actually make something useful, but I hope to make it useful and be a useful tool for people, consumers, businesses or anyone,” he said.

0 2

Chinese chosen by Musk

Just as Musk first found a group of rocket enthusiasts from the private sector when he founded Xspace, this time, the 12-member team of XAI is also all new faces, and there are few familiar technical experts.

This kind of employment strategy has been tried and tested in the entrepreneurship of Tesla and Xspace, and newcomers who are not bound by existing experience can create miracles.

This time, the 11 employees Musk invited, although they used to work for Deepmind and OpenAI, and former Google artificial intelligence head Jeffrey Hinton, Facebook artificial intelligence chief scientist Yang Likun, OpenAI chief scientist Ilya Su Tskevi, and even Deepmind’s chief scientist, Oriol Vinyals, are still relatively unknown little characters.

It is worth noting that 5 of the 11 employees are Chinese faces, they are Yuhuai Wu, Jimmy Ba, Greg Yang, Guodong Zhang and Zihang Dai.

This makes people wonder, what kind of academic background do these Chinese have, and why can they be selected by Musk? These people, without exception, are academic masters.

Ranked first is Yuhuai Wu, whose Chinese name is Wu Yuhuai. He graduated with a Ph.D. from the University of Toronto, and later studied as a postdoctoral fellow at Stanford University. According to the information on Facebook, Wu Yuhuai went to study at Fredericton High School in Canada since high school, and studied at the University of Toronto in 2015.

When he was at the NIPS conference in 2016, three papers co-signed as the first author were accepted. Founded in 1987, the NIPS conference is the world’s top conference in the field of machine learning.

According to reports in 2017, only seven Chinese students won the Google Doctoral Fellowship that year, and Wu Yuhuai, who was studying at the University of Toronto, was among them.

It is worth mentioning that during his study at the University of Toronto in 2017, Jimmy Ba, one of the members of the founding team, co-signed Wu Huaiyu’s paper. Jimmy Ba has an Asian face. In the TED speech video, Jimmy Ba also has a little Chinglish accent, but from the current public information, it is not sure whether he has ever lived in China.

However, in relevant domestic reports, many media reports listed Jimmy Ba as a Chinese scholar. In February of this year, the 2023 Sloan Awards were announced. At that time, Xinzhiyuan and other media reported that 30 Chinese scholars were selected, including Jimmy Ba.

According to the data, Jimmy Ba is an assistant professor of artificial intelligence pioneer Geoffrey Hinton (Geoffrey Hinton) in the Department of Computer Science at the University of Toronto.

The two Chinese, Guodong Zhang (Zhang Guodong) and Zihang Dai (Dai Zihang), both went to the United States for further study after graduating from domestic undergraduate programs.

Zhang Guodong graduated from Zhejiang University with a major in information engineering, and then entered the University of Toronto to study for a doctorate in machine learning, graduating in 2022. Zhang Guodong can be described as a “student master” who received scholarships all the way until graduation. According to the information on the official website of the School of Information and Electronic Engineering of Zhejiang University, Zhang Guodong won the highest scholarship honor of Zhejiang University – Zhu Kezhen Scholarship as early as his undergraduate period.

According to the official website of Zhejiang University, Zhang Guodong was admitted to Zhejiang University in 2013, and his academic performance ranked first in his major for three consecutive years. Prizes. In November 2016, Zhang Guodong went to Microsoft Asia Research Institute (Beijing) for an internship.

In 2022, Zhang Guodong also won the Apple Scholarship and is one of the four Chinese scholars who won the award.

Dai Zihang graduated from Tsinghua University with a bachelor’s degree. In 2020, he graduated with a Ph.D. from the Computer Department of Carnegie Mellon University and joined Google in the same year.

0 3

A “non-standard academic master” similar to Musk

As the second largest country in the field of artificial intelligence, China ranks first in the world in the number of employees in the field of artificial intelligence. It is not difficult to explain why the Chinese can always occupy a place in the top artificial intelligence companies.

Taking OpenAI as an example, there are 9 Chinese in the ChatGPT R&D team, accounting for 10% of the total number of the team, and 5 of them are undergraduates studying in Chinese universities. In the chatGPT-4 R&D team, there are more than 30 Chinese engineers.

Different from other teams, the Chinese selected by Musk, although they are also academic masters, are very similar to himself, and they are all non-standard academic masters.

For example, Younger is the only one of the five XAI Chinese founders who started working without a doctorate, and like Musk, he also likes to leave school.

Young’s first drop-out from school was to become a DJ. At the end of his sophomore year, Younger decided to choose to pursue his own path in music.

This time the suspension of school lasted for a year and a half, perhaps during this year and a half, Young discovered that music may be much more difficult than mathematics.

His enthusiasm for learning and researching mathematics had been numbed by Harvard’s predictable series of courses, assignments, and competitions. But this break from school rekindled his passion for mathematics.

After returning from school, Young published his first paper. It didn’t take long for his passion for mathematics to fade again. This time he became fascinated by neural Turing machines and discovered an interesting connection between algebraic topology and machine learning theory.

At this time, Young set a goal for himself—to use general artificial intelligence (AGl) to create something smarter than himself and eliminate himself.

After two rounds of suspension from school, Yang Ge finally found “true love” this time. Two years later, Young returned to Harvard and was lucky to meet his “Bole” Qiu Chengtong. In 2018, he was nominated for the Morgan Award, the highest honor in the field of undergraduate mathematics.

Yang Ge said in an interview: “It is no exaggeration to say that it is even a one-in-a-million probability—if Qiu Chengtong hadn’t randomly become his mentor at the beginning, it would not be possible to have what he is today.”

Mathematics has always been the weak point of ChatGPT. There have been a large number of cases circulating on the Internet. ChatGPT’s mathematics level is even wrong for elementary school math problems.

Improving the mathematical ability of artificial intelligence will obviously become the future direction of XAI.

Wu Huaiyu in the team, his main research interest is to build a machine capable of reasoning. He uses mathematics as the starting point for researching reasoning, with the goal of creating an automated AI “mathematician”.

In October last year, Wu Yuhuai introduced in an online sharing organized by the heart of the machine that he was determined to create an artificial intelligence good at reasoning to solve all mathematical problems.

With the popularity of large language models, people began to realize that they may not be trustworthy, they sometimes make illogical statements, or confidently tell lies as truth. “We don’t want to create a language model that can only speak like a human being.” Wu Yuhuai thinks about this, “We want it to understand what it is saying.”

Wu Huaiyu’s goal obviously coincides with Musk’s.

0 4

Call on others to stop, and do it quietly

Back in March, Musk also called for a six-month moratorium on large-scale model studies. However, business documents show that just half a month before Musk called for a moratorium on research, Musk and Jared Birchall, a former Morgan Stanley banker who managed his wealth, established a company called X.AI in Nevada. .

At the same time, Musk also purchased thousands of chips from Nvidia that can be used for large model training. Today, the official announcement of XAI is also expected.

Of course, there are still many questions waiting to be answered, for example, where does the money come from? Musk just spent more than 40 billion US dollars to acquire Twitter. Since then, Twitter advertising has plummeted, and its market value has fallen by nearly half. At present, this loss-making business has not seen any signs of improvement.

According to the Financial Times, Musk may invite investors from Tesla and Twitter to continue investing in his XAI.

Another problem is that it took OpenAI six years to iterate ChatGPT, and XAI has to start over if it wants to catch up with the opponent or even surpass it.

Musk is definitely the most successful investor in the field of artificial intelligence. When OpenAI was founded in 2015, he was one of the main investors. The current chief scientist Ilya Sutskevi was also invited by him after doing a lot of work. “OpenAI would not have succeeded without Ilya Sutskevi”.

Back then, he and Google founder Larry Page were close friends, often staying at his home, discussing artificial intelligence security issues late into the night. The most fundamental reason for founding OpenAI is “I don’t think Larry has paid enough attention to the safety of AI”, but Google acquired Deepmind and formed a monopoly.

However, in 2018, he announced his departure from OpenAI. The official explanation given was that there was a potential conflict with Tesla’s business. Afterwards, Musk repeatedly expressed criticism of OpenAI, believing that they did not take artificial intelligence security issues seriously enough.

In addition, he believes that GPT-4 also has an ideological bias, and there will be human-supervised reinforcement learning in the process of training it. “A lot of people look at the output of GPT-4, and then tell the model whether the output is correct. In other words they are training the AI ​​to lie.”

In its concise official introduction, XAI also introduced its own security consultant. XAI will be advised by Dan Hendrycks, director of the Center for Artificial Intelligence Safety, which aims to “reduce the societal-scale risks associated with artificial intelligence”.

The group released a statement in May signed by hundreds of AI scientists and experts, as well as the CEOs of some of the top AI companies, in which it said that “mitigating the risk of AI extinction should be compared with other efforts of similar magnitude.” Things like pandemics, nuclear war, become global priorities.”

In addition to the biggest favorable factor of Musk himself, XAI’s pace from scratch will obviously be faster than others.

In researching autonomous driving, Tesla already has a strong artificial intelligence team and a large number of digital processing computers, which are crucial for training complex “big language models”. Plus, Musk has Twitter in his hands, a trove of data that could help train any large language model.