Editor/Zhang Nan
Head of Design/Department
The feud between elon musk and Sam Altman, the current CEO of OpenAI, began five years ago.
Recently, because of the popularity of ChatGPT, Musk brought up the old things again.
He said on Twitter, "I still don't understand how a non-profit organization that I donated $654.38 billion has become a profit-making organization with a market value of $30 billion. If this is legal, why don't others do it? "
Altman was interviewed by Lex Friedman, an artificial intelligence researcher at MIT, on March 25th. Lex is also an online celebrity podcast.
The interview lasted nearly two and a half hours, and Altman responded positively to Musk's bombardment of OpenAI's transformation from open source to closed source and from non-profit organization to profit organization.
He said: "I believe that he is deeply nervous about the safety of AGI (Artificial General Intelligence), which is understandable, but I believe that he also has some other motives ... I have always regarded Musk as an idol since childhood. He acted like an asshole on Twitter, but I'm still glad there are people like him in the world. I hope he can see more of our efforts to do this well ... Maybe I should fight back, maybe later, but this is not my normal practice. "
Musk bial Terman 14 years old. 20 1 February 5 1 day When OpenAI was founded, he donated1billion dollars. At that time, he promised to invest $ 1 billion, together with peter thiel, co-founder of PayPal, and Altman, then president of Y Combinator.
At present, Altman has stepped down as President of Y Combinator, and he became OpenAI around 20 18.
It is worth mentioning that Huang Renxun, the founder of NVIDIA, donated the first supercomputing DGX- 1 made in NVIDIA to OpenAI on August 20 16, and the witnesses at that time included Musk.
This supercomputer DGX- 1, in Huang Renxun's words, is the first lightweight small supercomputer that took 3,000 people three years to develop, and its computing and throughput capacity is equivalent to 250 traditional servers. At that time, he had orders from more than 100 companies, but donated the first one to OpenAI.
From 20 15 to 20 18, OpenAI was positioned as a non-profit organization, and its goal was to take the lead in creating a machine AGI with human thinking, learning and reasoning capabilities. Their purpose is not to rule the world, but to hope that technology can develop safely and its benefits can be distributed equally to the whole world.
"We need a leading research institution to give priority to the interests of all ordinary people, not our own interests. We encourage researchers to publish their works, whether they are papers, blog posts or codes, and our patents (if any) will be shared with the world. " This was the initial intention of OpenAI at that time.
OpenAI before the end of 20 18 encourages open source, transparency and sharing. Its main rival at that time was DeepMind. This artificial intelligence laboratory was established five years earlier than OpenAI, and was acquired by Google in 20 14. Its financial strength was far above that of OpenAI at that time.
divergence
For most Americans, the first time I heard about OpenAI was on February 9, 2009, when they released ChatGPT-2. For us in China, ChatGPT seems to have been made public only recently.
On that day, the lab announced an impressive discovery: a model that can generate convincing papers and articles by pressing a button. The Lord of the Rings, give it a sentence, and it will be sent out paragraph by paragraph in the same way.
The change of the organizational nature of OpenAI began in April, 20 18, and the laboratory issued the articles of association. The articles of association restated the core values of the laboratory and subtly changed the original statement. In addition to promising to "avoid using artificial intelligence or AGI to harm human beings or improperly concentrate power", the articles of association also emphasized the necessity of resources this time.
"We need to mobilize a lot of resources to accomplish our mission," the company's articles of association read, "but we will always strive to take actions to minimize the conflict of interests between employees and stakeholders and the broad interests."
But this year, Volkswagen and most employees did not realize the quiet change of OpenAI. It was at this time that Musk chose to quit OpenAI. Perhaps at that time, he already knew the news of Microsoft investment. He thinks the values of both sides will be different.
After quitting OpenAI, he didn't continue to donate money, so he started a new stove and set up his own Agi R&D team.
The real structural change of OpenAI occurred in 2065438+March 2009. It is no longer a purely non-profit organization, but the laboratory has become a profit organization, with a "ceiling profit", which limits the return to investors to 100 times, but from the organizational structure, it is not a completely profit organization, because OpenAI is supervised by the board of directors of a non-profit entity.
Shortly thereafter, OpenAI announced that it had received an investment of $6,543.8+billion from Microsoft, but the two sides did not disclose whether the investment was cash or whether the Microsoft Azure platform was also discounted.
Altman became CEO of OpenAI during this period. This time, Musk once again criticized the organizational nature of OpenAI on Twitter, but it was a continuation of dissatisfaction five years ago.
Why from non-profit to profit
At the intersection of San Francisco 18 Street and folsom Street, OpenAI's office looks like a mysterious warehouse. This historic building has monotonous gray paneling and colored windows, and most of the curtains are pulled down. The letter "Pioneer Building" shows the remains of its former owner, Pioneer Truck Factory.
Inside the office, the space is bright and transparent. There are several public areas and two meeting rooms on the first floor. One suitable for large-scale meetings is called "space roaming"; The other room is more like a beautified telephone booth, called "Infinite Jokes".
Three years ago, Karen Hao, a writer of MIT Science and Technology Review, had an in-depth interview with this company. At that time, she was only allowed to move on the first floor, and the second and third floors were restricted areas and kept secret. These two floors are the office area for employees and the place where robots are placed.
Karen met met Greg Brockman, co-founder of OpenAI, on the first floor. He looks nervous and alert, she wrote in the article. He wears casual clothes and, like many people in OpenAI, has an irregular hairstyle, which seems to reflect an efficient and simple mentality.
Greg Brockman, a member of OpenAI core team) ▼ Second row ▼
Brockman said that when the team studied the trend of AGI, they realized that it was not feasible to maintain the nature of non-profit organizations in terms of the sustainability of funds. The computing resources used by other competitors in this field to achieve breakthrough results double every 3.4 months, which requires sufficient capital to match or exceed this exponential growth. This requires a new organizational model, which can not only accumulate funds quickly, but also be loyal to the mission to some extent.
This change from 2065438 to April 2009 has caused many people to accuse OpenAI of violating its mission.
Shortly after the announcement, in a post on AGI related forums, a user asked how the 65,438+000-fold limit was limited: "Early investors in Google achieved a return on capital of about 20 times," they wrote. "Can you bet that you will have a corporate structure with a rate of return several orders of magnitude higher than that of Google? At the same time, you said you didn't want to' over-concentrate power'? How will this work? If it is not the concentration of resources, what is power? "
This move also made many employees feel uncomfortable, and they expressed similar concerns.
Brockman said: "We spent a long time internally iterating with employees to make the whole company accept a set of principles and a set of values, which means that even if we change the structure, we must remain the same."
In the summer of 20 19, within a few weeks after switching to the profit ceiling mode and Microsoft's capital injection of 10 billion dollars, the management assured employees that these updates would not change the research methods of OpenAI functionally. The values of Microsoft and Lab are very consistent, and any commercialization efforts will be in the foreseeable future, and seeking basic problems will remain the core of the work.
For a while, these assurances seemed correct and the project went on as usual. Many employees don't even know what commitment OpenAI made to Microsoft.
But at the beginning of 2020, the pressure of commercialization of OpenAI is increasing, and making money is no longer out of reach. When privately sharing his vision for the laboratory in 2020 with employees, Altman conveyed a clear message: OpenAI needs to make money to do research.
The management said that this is a difficult but necessary trade-off. Due to the lack of wealthy charitable donors, it has to make such a trade-off.
But the real reason behind it is the trade-off that OpenAI faces, not only the financial pressure, but also the strategic choice. OpenAI tried to get to AGI before others.
This pressure forces it to make decisions that seem farther and farther away from its original intention. In the process of attracting capital and talents, it tends to hype, protect its own research in order to maintain its advantages, and pursue the strategy of large computation-not because it is regarded as the only way to AGI, but because it seems to be the fastest.
In 20021year, Microsoft invested again, this time as much as 10 billion dollars. Microsoft has thus become the exclusive cloud service provider of OpenAI. In 2023, Microsoft officially announced that the cooperation with OpenAI has entered the third stage.
This article was originally published by auto business review.
Reprint or content cooperation, please contact the instructions.
Illegal reprinting will be investigated.
This article is from auto business review, written by Yi Chehao, and the copyright belongs to the author. Please contact the author for any form of reprint. The content only represents the author's point of view and has nothing to do with the car reform.