Musk sets a goal for xAI: to achieve general artificial intelligence in 2029, this is the deadline

Twitter, Tesla and xAI work together, can AGI be a big deal?

"The goal of xAI is to build a decent AGI (artificial general intelligence) whose first purpose is to try to understand the universe," said Elon Musk.

A few hours ago, the co-founders of xAI were all unveiled, and Musk's mysterious artificial intelligence company announced its plans.

Two days ago, Musk announced the establishment of xAI on Twitter. Today, xAI’s self-introduction used the Twitter Spaces audio conference, and more than 34,000 listeners participated in the live broadcast. Musk seems to want to add a bit of office platform attributes to Twitter, but there are so many listeners, the platform was once crowded, and the live broadcast event was delayed for a while.

During an hour-and-a-half long introduction and QA session, the world's richest man and the leader of Tesla, SpaceX, and Twitter outlined the goals of the new company, xAI.

In general, including but not limited to the following points:

• xAI will release new tools to the public as soon as possible, first product details will be announced in a few weeks *Companies will work on building artificial intelligence that will combine text, images and the real world
• The company will work closely with companies such as Tesla and Twitter
• Any ordinary person can vote for the future of xAI

It seems that Musk is not only building a company that benchmarks OpenAI, but also has a new concept of running a company.

Goal: Explore the unknown universe

Musk said that the fundamental goal of xAI is to understand "what the hell is really going on".

Based on this goal, the original vision of xAI seems to be more focused on answering deep scientific questions, rather than competing head-on with OpenAI, Google, and Microsoft to create consumer-grade AI products. Musk made it clear that his new company will replace the current large artificial intelligence companies, but xAI is still in the "nascent stage" and it will take time to catch up with OpenAI and Google.

In recent times, the capabilities of a new wave of AI tools such as ChatGPT have amazed experts, and some in the industry have advanced the timetable for achieving AGI. But the current technology is still a conventional form of natural language processing: models output content by predicting the next word, and there is no evidence that this approach can solve unknown scientific problems. Many artificial intelligence researchers say that human-level artificial intelligence, if at all, may be decades away.

Musk said that unresolved questions about "the nature of dark matter" and "how gravity works" may be the goals xAI is trying to explore. He talked in detail about the "Fermi Paradox" - a theoretical problem that explains the contradiction between the overestimation of the existence of extraterrestrial civilizations and the lack of relevant evidence.

Jimmy Ba, a professor at the University of Toronto who joined Musk's company and the inventor of the Adam optimizer, said that xAI is committed to creating an artificial intelligence that can help solve a series of problems faced by humans, "(What needs to be discussed) is how to build a general problem Solving machines to help all, overcoming the most challenging and ambitious problems, and how we can use these tools to enhance ourselves."

Musk has criticized existing chatbots for bias, though he said xAI's artificial intelligence could offend people because it tells the "truth." Musk has been outspoken about the risks of artificial intelligence for years, saying in 2014 that inventing a superintelligence would be like "summoning a demon" that could pose an existential threat to humanity.

Others are a challenge, but for me this is an advantage

The main challenges in building large models lie in data and computing power. Musk said that xAI will work closely with Twitter and Tesla. But on the other hand, the latter two are listed companies, and this tendency of left-handed and right-handed has attracted criticism from investors.

In the past few months, Musk has complained that OpenAI and other artificial intelligence companies grab Twitter data to help train large models, and then limit the number of daily visits by Twitter users, causing an uproar.

Musk claimed that every AI organization on the planet "has been trained on Twitter data, which is illegal in all cases." In contrast, xAI will use public data on Twitter (not limited to text, but also images and videos) to train the "most curious" artificial intelligence systems and products he hopes to build.

In addition to Twitter, Musk also owns another company closely related to AI, Tesla, which has a strong artificial intelligence team and has been working on exploring autonomous driving for years.

"We're going to work with Tesla on chips and maybe on artificial intelligence software," Musk said of Tesla's custom supercomputing chip. "Obviously, any connection to Tesla has to be an arm's length deal, because Tesla is a public company with a diverse investor base."

Training the “big language models” that form the backbone of modern AI tools requires massively parallelized AI computing power to process data—an extremely complex and expensive process. Tesla announced its self-developed supercomputer Dojo last year and claimed that it will be used to build self-driving technology.

Musk said that the "4th generation hardware" that Tesla has shipped is three to five times more powerful than the 3rd generation, and promised that the "5th generation hardware" will be launched in a few years and will be four times more powerful than the current version. to five times.

At the xAI event, Musk said that the new company would of course need computing power, but hinted that it would not use as much as other companies. Igor Babuschkin, a former Google and OpenAI researcher who joined xAI, said the team will remain relatively small.

Safe artificial intelligence in the eyes of Musk

For a long time, realizing AGI has been the ultimate goal of many researchers. Humans, however, are both excited and terrified of this goal, as AGI could threaten the dominance of humans or other species on this planet.

Musk spent a lot of time explaining his point of view on how to build what he calls a secure "superintelligence." This approach, as he calls it, promotes human survival, not extinction.

Musk said that the safest way is to build an AGI that is extremely curious and truth-seeking, and minimizes the difference between the real and the actual truth.

For a truth-seeking superintelligence, having human traits is much more interesting than not having them, so this is the safest way to create a safe intelligence. Musk used the example of space and Mars to illustrate, "For example, when you look at the various planets, moons and asteroids in the solar system, although they are very interesting, they are dwarfed by the interesting level of humans." Musk said.

Musk also emphasized that he has spent years thinking about and worrying about the safety of artificial intelligence, and claimed that he has been one of the most vocal voices calling for the regulation of artificial intelligence. In his view, AI and the humans who oversee it can be kept safe by maximizing curiosity and truth-seeking processes.

After the rise of ChatGPT, Musk has repeatedly warned people to be vigilant against AI and called on the society to strengthen the supervision of AI. In March of this year, Musk also signed an open letter suspending the training of artificial intelligence systems stronger than GPT-4 for at least 6 months. However, almost back and forth, also in March, Musk registered a company called X.AI Corp in Nevada, which listed Musk as the sole director. Now, Musk officially announced the establishment of xAI. Looks like a big reversal.

The reason Musk gave was that the pause no longer seemed realistic, and he hoped xAI would provide an alternative path.

"If I could pause AI or advanced AI superintelligence development, I would. But that doesn't seem realistic, so xAI is essentially building an artificial intelligence...in a good way, sort of hopefully way," he said.

Reiterate the story with OpenAI

Under the influence of ChatGPT, OpenAI quickly became the most popular AI research institution in the world.

And Musk has a deep relationship with OpenAI, and he is one of the co-founders of the company. OpenAI was founded in 2015 as a non-profit research institution. Its vision is to build a safe and beneficial general artificial intelligence (AGI) for humans. At that time, Musk, Sam Altman and others invested $1 billion to establish it.

But three years later, in 2018, Musk left. The public reason was that he wanted to focus on Tesla's operations.

In 2016, Nvidia CEO Huang Renxun personally delivered the first DGX-1 supercomputer to OpenAI and took a photo with Musk.

In 2019, OpenAI transformed into a for-profit company called OpenAI LP, controlled by a parent company called OpenAI Inc.

Months after the change of format, Microsoft announced a $1 billion capital injection. OpenAI's partnership with Microsoft means allowing the latter to commercialize some of its technology, and the next thing is a series of big events in the recent technology field that we are familiar with.

OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella.

When talking about the original intention of founding OpenAI, Musk recalled that he was a close friend with Google co-founder Larry Page. After Google acquired DeepMind in 2014, Musk said he had a series of conversations with Page about AI safety. According to Musk, those conversations have not been smooth, and his and Page's views differ greatly.

As a result, Musk realized that there needed to be what he called a counterweight to Google and its influence on artificial intelligence. This counterbalancing force is OpenAI. Musk's original goal of setting up OpenAI was open source and non-profit.

“Now, because fate likes irony, OpenAI is closed source. It is frankly greedy for profit,” Musk said.

Musk hopes that OpenAI will not stray from its founding vision, which is to help humans exist.

Musk: AGI will come out in 2029, that's the deadline

Musk emphasized that, in his view, it is clear that AGI is on the horizon, and it will happen soon. Therefore, he realized that he had two choices, either to be a bystander or to be a participant. As a participant, he can influence the outcome, become a competitor.

Musk said: "I think we can create a competitive alternative, hopefully better than Google DeepMind, OpenAI or Microsoft." In this regard, xAI needs to develop technology that "understands the physical world, not just the Internet" , he believes Tesla's driving data will help in this regard.

While Musk didn't specify how effectively xAI will compete with rivals, he did outline a specific timeline in which he expects AGI to be practical: about 2029.

However, we can't fully believe Musk's target time setting, because he has set goals such as sending humans to Mars with SpaceX, launching Tesla's self-driving taxi service, allowing car owners to rent out their own self-driving cars Etc. None of these are implemented.

In general, xAI, a company that wants to do big things, is still in the early stages of construction, and there are many details to be discussed. Despite the lack of clarity on the direction, the company appears to be offering some transparency, with Musk saying he will continue to accept outside feedback as the work progresses, a lesson he learned on Twitter.

"As with everything, I think we're very open to critical feedback and welcome it," Musk said of his work at xAI. "Actually, one of the things I love about Twitter is that there's a lot of negative feedback on the platform, which helps motivate myself."

Reference content: