"The vision is for humans and AI to form a symbiotic relationship. After Sam Altman finished speaking, he looked at Elon Musk. The latter picked up, "I agree with what Sam said, we're already working like a cyborg. However, the development of AI needs to be approached with caution to ensure that it is ultimately beneficial to humanity. ”
The conversation took place nine years ago, when the two sat next to each other on a talk show in the United States, calmly laying out ambitious blueprints on a wide range of cutting-edge topics such as future innovation, extraterrestrial life, and entrepreneurship. At the time, Musk was interested in artificial intelligence (AI), but it was difficult to find like-minded people because of security concerns. In his eyes, the young man next to him is not only the president of the tech incubator Y Combinator, but also a survivalist who shares a common nervousness about AI. It was also in 2015 that they co-founded a non-profit artificial intelligence research lab "openai" to compete with tech giants such as Google.
However, less than three years later, Musk stepped down from the OpenAI board of directors on the grounds of "conflict of interest". Now, Musk and Ultraman have completely broken up, but the old vision has become the focus of contradictions.
On February 29 this year, Musk suddenly filed a lawsuit against OpenAI and the company's CEO Altman and President Brockman in a San Francisco court, in which he reprimanded Altman for violating his "original intention" and demanded that OpenAI restore open source and give compensation. In response, the five co-founders of OpenAI, including Ultraman, wrote a response, refuting the accusations one by one, and ** a large number of emails exchanged between Musk and them, trying to confirm that Musk is mercenary.
This dispute is not only the extreme pull between Musk and Ultraman, but also a side of the convergence of technology giants under the AI frenzy, involving Microsoft, the "financial master" behind OpenAI, as well as Musk's Tesla and artificial intelligence startup XAI. In addition to watching the business fight, the disagreements and debates on the future development of AI may be more worthwhile.
Non-profit vs. for-profit.
Musk's complaint to the San Francisco court is 35 pages long, and the biggest "crime" points to OpenAI's abandonment of its original mission - "developing artificial intelligence technology for the benefit of mankind" for profit. Musk claims that OpenAI's partnership with Microsoft has transformed the company into a "de facto closed-source subsidiary" of Microsoft focused on profit maximization.
Subsequently, OpenAI posted a counter: "It's a pity that a person we admire very much has come to this point - he inspires us to set higher goals, and then tells us that we will fail, and he founded a competitor himself." He sued us again as we started making meaningful progress toward OpenAI's mission without him. Just a few sentences condense the grievances and hatred between Musk and OpenAI over the years.
OpenAI released the content of the email between Musk and his co-founders on March 5, ** all founders agreed to create a for-profit entity to support OpenAI, but Musk wanted to get a majority stake in the entity, initial board control and serve as CEO, while the other founders disagreed. Musk then suggested merging OpenAI into Tesla, and the OpenAI team also rejected the idea, leading to a break between the two sides.
Part of the "Elon Musk Biography" written by the famous biographer Isaacson confirms this statement. The book mentions that Musk wants to develop artificial intelligence capabilities in his various companies, trying to persuade Altman to merge OpenAI into Tesla, on the grounds that OpenAI was still lagging behind Google for 3 years at that time.
Now Musk and OpenAI rebuke each other for profit, and businessmen are understandable for pursuing profits, but Musk accuses OpenAI of violating the founding agreement, that is, maintaining the company's nature as a "non-profit". According to the content of the email, Musk said that the financing scale of OpenAI should start from $1 billion, and promised that "I will be responsible for filling the gap that occurs due to the failure of others to provide." ”
OpenAI said in a March 5 post that it received less than $45 million from Musk and raised more than $90 million from other donors when it was a nonprofit company. Obviously, this amount is nowhere near enough to support the company in building artificial general intelligence (an AI capable of mimicking and performing any human intelligence activity, and even performing better than humans in a variety of cognitive tasks. Goals. Microsoft became the savior.
Just after Musk stepped out of the OpenAI board of directors in 2018, while remaining non-profit, Altman set up a for-profit division to raise equity** (but there is a cap on the revenue of this for-profit division. After this cap is reached, everything earned by the for-profit entity is returned to the non-profit research laboratory. ）。A year later, Microsoft invested $1 billion in the company, and another $12 billion over the next few years to pay for OpenAI's generative AI technology. In return, OpenAI provides a number of AI services to customers of Microsoft's Azure cloud computing products. Microsoft has held a 49% stake in OpenAI since January 2023 and remains one of the largest shareholders so far, while OpenAI's nonprofit parent company holds a 2% stake. With the advent of chatbot ChatGPT in the same year, Microsoft's stock price soared, and it was logical to have access to GPT's large language model and related software.
In Musk's view, OpenAI has embarked on the path of relying on Microsoft to make profits, which is contrary to the original intention when it was founded. This dates back to 2 years before OpenAI was founded, in 2013, Google took the lead in the field of AI, Musk at his 42nd birthday party, and Google founder Larry Page, who has known him for more than 10 years, fiercely debated the risks of artificial intelligence, the former advocated the establishment of a "firewall" to prevent AI from replacing humans, worried about the demise of human consciousness. The latter disagreed, believing that the superiority of machine intelligence and consciousness over humans meant the next stage of evolution.
Musk's hoarseness has failed to stop Larry Page's ambitions. In January 2014, Google announced the acquisition of Deepmind, a company dedicated to the realization of artificial general intelligence (AGI), and created a safety committee for it.
As Google continues to make efforts in the field of AI, Musk has been running around warning about the threat of AI to human survival, and even made an appointment for a one-on-one conversation with then-** Obama in 2015. Years later, he recalled the content of the meeting and said that instead of taking the opportunity to promote Tesla or Space X, he urged the United States to promote AI regulation, but failed to attract enough attention.
After repeated setbacks, Musk decided to find a partner to enter the AI field together, and he found Ultraman, who was 14 years younger than him, and the two hit it off, believing that one of the ways to improve the security of AI systems was to have a large number of competing systems to check and balance each other. So they decided to set up a non-profit AI company to develop an AI that is not controlled by any individual or company.
Moreover, Musk named this company OpenAI in the hope of achieving true openness and allowing others to build their own AI systems according to the source ** to achieve the purpose of checks and balances. Nowadays, whether it is open source has also become the focus of controversy.
OpenAI or ClosedAI
Change the name of OpenAI to ClosedAI, and I will cancel the lawsuit. On March 6, local time, Musk posted on social platform X and attached a retouched **, in which the "closedai" icon was displayed on the badge held by Ultraman in the picture.
Musk used this to ridicule OpenAI for not being open enough, but the precise definition of openness is not the same as that of both sides.
Screenshot. Ilya Sutskeve, chief scientist at OpenAI, wrote in an email in 2016: "Openness in OpenAI means that everyone should benefit from the results of AI builds, but also without sharing the scientific results." In response, Musk replied: "Yes. OpenAI believes that this email is enough to confirm that the meaning of openness is broad, not absolutely "open source".
In March 2023, OpenAI released the AI language model GPT-4 and shared a large number of GPT-4 test results as well as some interesting demos, but basically did not provide information about the data used to train the system, the specific hardware or methods used to create it. Many in the AI community have criticized the practice, pointing out that it undermines the company's founding spirit as a research organization. What's more, in the face of the threat posed by AI systems such as GPT-4, it becomes more difficult to develop targeted risk protection measures.
In an interview with US media, Ilya Sutskovi explained the reason why OpenAI does not share more information about GPT-4 - concerns about competition and security are "self-evident". In terms of security concerns, he noted that these models are very powerful and will get more and more powerful. At some point, it can be very easy for some people to do great damage if they take advantage of these models. As the model's capabilities increase, it makes sense not to want to expose them"。
At that time, Musk "interviewed" Altman on the issue of open source, saying that OpenAI was founded as an open-source non-profit company to compete with Google, but now it has become a "closed-source" company, "and now it has fallen into the hands of ruthless monopolies". In response, Altman said that he did not profit significantly from the startup, and he understood that Musk's criticism was a real concern, but believed that he had not fully recognized the complexity of AI security issues.
Musk is not only a staunch open sourcer, but also Logan Kilpatrick, the head of developer relations at OpenAI. He announced his resignation on social platforms on March 1, and lamented that "an open-source AI will be a pure victory for developers, businesses, and all mankind." Some developers speculated that the disagreement around "open source" was the reason for its departure. The previous OpenAI "palace fight" controversy is essentially a "battle of ideas" about AI development - whether to accelerate the commercialization process of AI, or to ensure the safety of AI and then promote it to the world.
However, at this time, for Musk, even if he can't win the lawsuit against OpenAI, he still has other "** to defend the AI security he insists on."
Find out the answer to the ultimate question.
Through lawsuits and bold statements on social platforms, Musk punched tech giants such as OpenAI, Google, and Microsoft one by one, like a gladiator rushing into the battle of artificial intelligence, hoping that XAI, the artificial intelligence company he founded last year, would kill all sides.
In Musk's vision, XAI not only wants to create an AI robot that can write computer programs and a competitor that can compete with OpenAI's GPT series of chatbots, but more importantly, to achieve "an artificial intelligence that can pursue the truth to the maximum, and he cares about the understanding of the universe, and the result is that it wants to protect humanity." ”
There is no doubt that XAI got a late start and is at a disadvantage in this ever-changing field, and the only product that stands out so far is the chatbot Grok. The XAI team highlighted that Grok has a greater sense of humor than other chatbots, and that it aims to mimic the Hitchhiker's Guide to the Galaxy, with "a little bit of wisdom" to answer questions and a "little bit of rebellion." The book, which rescued Musk from a depressed state of adolescence, defines the supercomputer's mission as "finding answers to the ultimate questions about life, the universe, and all things."
The current GRAK is the same as the GPT-3 released by OpenAI5 is in a similar format, but the test results show that it is better than GPT-3 in math, **, and multidisciplinary knowledge assessments5. Perform better. And the inherent advantage of GROK is that it can obtain real-time information from the X platform.
Musk is sitting on a huge data asset - the information flow of platform X (formerly Twitter), which can be used to "feed" artificial intelligence, which is trained on real-world human speech and dialogue information. He has said that it is a fringe perk of the acquisition of Twitter. In addition, he has navigation and ** data from Tesla, which may allow his team to develop a chatbot that is not limited to text generation.
According to US media reports, Musk plans to raise $1 billion to $3 billion through a round of private financing in late March, and he may also invest personally, and XAI may be valued at as much as $20 billion by then. But earlier this year, in response to inquiries from **, Musk repeatedly denied that XAI was raising money. Even if it reaches a valuation of $20 billion, it is still only a small part of OpenAI's current situation. In February, OpenAI's valuation has increased to more than $80 billion as it reached its latest *** agreement with venture capital firm Thrive Capital.
Musk, who is active in today's world, sometimes looks like an unreasonable weirdo, speaks amazingly, makes enemies on all sides, and sometimes looks like a naïve child full of scientific curiosity and boundless ideas. His obsession with security around artificial intelligence is difficult to judge, but the future development of this field cannot be ignored.
For AI to help us understand the true nature of the universe, it must be able to discard what is popular but wrong and choose what is not popular but is right. Musk said on platform X on March 5.
The Paper reporter Chen Wenjing also contributed to this article).
The Paper, for more original information, please **"The Paper" app).