news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Kong Haili, Xiao Xiao, intern Wang Tian, ​​Liu Xin

Edited by Zhu Yimin and Wang Jun

Image source: Tuchong

There is an "earthquake" among OpenAI's top management!

Core team members left one after another! One of the founders resigned and the president stopped working and took a vacation

August 5, local time,John Schulman, one of the co-founders of OpenAI, announced his resignation on social media and will move to Anthropic, the latter of which was founded by former OpenAI researchers and is considered a strong competitor to OpenAI. Anthropic has always claimed to be more security-conscious than OpenAI.

Meanwhile, another OpenAI co-founder,President Greg Brockman is reportedly taking an extended vacation until the end of the year to "relax and recharge"Peter Deng, who joined OpenAI as head of product last year, also left after saying that OpenAI’s models were released with their most powerful features suppressed to ensure safety.

In response to John Schulman's departure, CEO Sam Altman expressed his gratitude in a reply post on social media, saying that he "developed a large part of OpenAI's initial strategy."

At this point, only three people remain of OpenAI’s 11 co-founders: CEO Sam Altman, Wojciech Zaremba, head of OpenAI’s language and code generation team, and president Greg Brockman, who is on a long vacation.

In fact,Since OpenAI's CEO Sam Altman was ousted in November last year and rehired within a week, the company's leadership has undergone frequent personnel changes and has been slow to stabilize.

In February this year, another co-founder, Andrej Karpathy, resigned and announced last month the establishment of an AI+education company, Eureka Labs, which will serve as a new type of "AI native school" to tutor students.

In May, Ilya Sutskever, former chief scientist of OpenAI, officially announced his resignation. He announced that the new company he founded will focus on pursuing AI capabilities through revolutionary engineering and scientific breakthroughs while prioritizing safety.

In July, OpenAI moved another security chief, Aleksander Madry, to another position. Recently, OpenAI also hired two new executives, Facebook product veteran Kevin Weil will serve as the newly created chief product officer position, and former Nextdoor CEO Sarah Friar will also serve as chief financial officer.

Industry insiders believe that as a super-leading AI company, OpenAI’s series of personnel changes will not only affect the company’s operations and strategic development, but will also, to a certain extent, affect the industrial layout of the entire artificial intelligence industry.

At the same time, OpenAI is facing fierce competition from competitors in the commercial battlefield. Microsoft has reached a strategic cooperation with Anthropic, a competitor of OpenAI, and Meta will open source its powerful model for free.OpenAI may also be experiencing significant operating losses.During this stage, frequent changes in leadership may dampen employee morale within the company.

The industry believes that the instability of OpenAI's senior management may be related to decision-making differences, strategic direction or personal career planning within the company. When more than one senior executive experienced resignation, the remarks made on social media more or less conveyed concerns that "OpenAI focuses too much on commercialization and ignores safety."

Schulman, who has already left, wrote in a social media post: "Company leadership has been committed to investing (in alignment research). My decision is a personal one, based on what I want to focus on in the next stage of my career." He also said, "This choice is because I want to focus deeply on AI alignment research and start a new chapter in my career at Anthropic, returning to actual technical work."

It was previously revealed that the Super Alignment Team was disbanded

In May this year, OpenAI was also exposed to have disbanded the Super Alignment Team, which was committed to "guiding and controlling artificial intelligence systems smarter than us through scientific and technological breakthroughs." Jan Leike, the leader of the Super Alignment Team, also joined Anthropic at that time.However, OpenAI later denied rumors that it had disbanded the super alignment team and ignored safety.

FutureSearch, an artificial intelligence (AI) research organization, released a report in July this year, saying that it collected all available financial information of OpenAI and concluded through professional calculations that OpenAI has about 9.88 million paying users per month and annual recurring revenue of $3.4 billion. This is almost exactly the same as the annualized revenue of OpenAI previously revealed by Altman.

However, due to the huge costs of training, developing and running models, it will take some time for OpenAI to become profitable. According to FutureSearch's research and other relevant information, OpenAI's total operating costs this year may reach $8.5 billion, of which inference and training costs are $4 billion and $3 billion respectively. If OpenAI's annual revenue is $3.4 billion, its revenue and expenditure gap will be as high as $5 billion. As the company's CEO, Sam Altman has a heavy burden on his shoulders. On the one hand, he has to try to obtain revenue as much as possible, and on the other hand, he has to seek new financing. The turmoil in the company's management has added some uncertainties.

Musk sues OpenAI

On the same day that John Schulman resigned (August 5th local time), Elon Musk reopened the lawsuit against OpenAI, again accusing OpenAI and its two founders, Sam Altman and Greg Brockman, of violating the company's founding mission of developing artificial intelligence technology for the benefit of humanity.

Transcribe YouTube videos to train AI? More than 100 creators decide to sue collectively

On August 2, local time, a representative of a YouTube anchor formally submitted a class action lawsuit in the U.S. District Court for the Northern District of California.More than 100 people are currently involved in the class action lawsuit, accusing OpenAI of arbitrarily transcribing millions of YouTube videos to train large models.

As of press time, OpenAI has not responded to the class action lawsuit.

The lawsuit states that creators own the YouTube videos, OpenAI's actions violated YouTube's platform terms, and that it has gained unfair benefits from the creators' losses, and demands OpenAI pay more than $5 million in compensation.

According to the 21st Century Business Herald,Over the past six months, OpenAI has been embroiled in controversy over the “stealing” of YouTube videos.: In April of this year, the New York Times published an article stating that OpenAI exhausted all text training data in 2021 and turned to developing Whisper, a large text-to-speech model, to transcribe YouTube videos, podcasts and other video data into text, and then continue to optimize the large model.

Half a month ago, the media discovered that many AI companies used an open source dataset called the Pile in their training, one of which was called "YouTube Subtitles". The dataset consists of plain text, mainly text descriptions in videos and text subtitles automatically transcribed by YouTube, covering 173,500 YouTube videos and more than 48,000 channels.

As the world's largest video website, YouTube has long banned the crawling of videos, and even has strict restrictions on commercial and batch downloads of videos.Google spokesman Matt Bryant has publicly commented that Google's terms of service and robots.txt files (a web page file that tells crawlers what content cannot be crawled) clearly prohibit unauthorized crawling of YouTube content. Google will take "technical and legal measures" to prevent such use when there is a clear legal or technical basis.

However, OpenAI executives have always refused to answer directly whether they use YouTube videos to train their AI products, especially Sora. Earlier this year, OpenAI Chief Technology Officer Mira Murati spoke out for the first time in an interview, saying that she was "not sure" whether YouTube videos were used.

Li Yunkai, partner of Tianyuan Law Firm, previously analyzed with a reporter from 21st Century Business Herald:The copyright dispute over training materials cannot be resolved, and the fundamental problem is the technical black box"How can we prove whether the big model has taken your materials? What kind of training has it done?" Li Yunkai said that in this case, the only way to make the platform less responsible and more secure is to "keep silent". As long as the platform does not say anything, the creator cannot prove it. For copyright holders, the biggest difficulty is that they cannot provide evidence.

SFC

Editor of this issue: Jiang Peipei

21 Recommended Reading