This article is from the WeChat official account: Zimu Bang, author: Xiao Jinya, editor: Wang Jing, the title image is generated by AI

It's really not a good time for Ultraman.

On the afternoon of July 11th local time, two bad news came in a row.

The first bad news is that Zuckerberg has once again taken action. Meta has recruited two researchers from OpenAI again.

And among them is a Chinese person named Lu Liu, who is a key member of the 4o image generation team.

The second bad news is that OpenAI almost acquired the AI coding company Windsurf. Not only did the acquisition fall through, but it also officially announced a collaboration with Google.

It's called a collaboration, but it's almost the same as an acquisition — not only did it get the authorization of some of Windsurf's technology, but also absorbed the company's CEO, co-founder, and some employees into Google DeepMind.

For this, Google spent 2.4 billion dollars.

Indeed, whoever has money can win the game. The two old tech giants, Meta and Google, have poured tens of billions of dollars, leaving OpenAI in the wind.

One

Let's first see who Meta has recruited.

These two people are both researchers at OpenAI.

One is Allan Jabri, who previously interned at Google and DeepMind, earned a Ph.D. in Computer Science from the University of California, Berkeley in 2023, and then joined OpenAI. Jabri has long studied "learning visual common sense without manual labels." He advocates allowing models to discover "different perspectives of the same object" and "cross-frame correspondence" on their own through contrastive learning or random walks in temporally continuous video segments.

The other is Lu Liu, yes, another Chinese person, we will focus on her resume.

Lu Liu is a post-90s person. She graduated with a bachelor's degree in Electronic Engineering from South China University of Technology in 2017, and then went to Australia, where she was an exchange student in the computer science program at the University of New South Wales in Sydney for one year. After that, she continued to study in Australia, entering the University of Technology Sydney, and earned a Ph.D. in Computer Science in 2021, specializing in deep learning.

Her research interests include meta-learning, low-shot learning, graph neural networks and their applications in edge computing, personalized artificial intelligence, and privacy protection under conditions of limited supervision, weak supervision, and unsupervised learning.

After completing her doctorate, Lu Liu moved to the United States, working successively as a researcher at Microsoft, Google, and Character.AI.

At the beginning of last year, Lu Liu joined OpenAI.

Before leaving OpenAI, she was a member of the 4o image generation team, often sharing content related to 4o image generation on X and LinkedIn, showing great enthusiasm for her work.

The image generation of 4o was a big success, and in March it once sparked a wave of "Ghibli" trend, with netizens around the world using its function to convert images into the style of Studio Ghibli.

Now even the X account avatar of OpenAI CEO Altman is in the Ghibli style.

According to The Information, the two people who were recruited, Allan Jabri and Lu Liu, will also join Meta's newly established Superintelligence Labs.

With Lu Liu being recruited, the number of Chinese people Meta has openly recruited from various companies has reached 11, including 9 from OpenAI, and one each from Google DeepMind and Apple. Among them are top researchers like Yu Jiahui, and it is widely rumored that Meta has offered them a staggering salary of $300 million over four years, with some people able to unlock $100 million in the first year. Just recently, Ruoming Pang, the former head of Apple's basic research team, was recruited, and Bloomberg reported that Meta offered him a salary of $200 million.

Although these salaries are not cash, but total compensation including stock, signing bonuses, and performance-linked incentives, the total face value of several hundred million dollars is still astonishingly high.

Two

Zuckerberg is pouring resources into Meta to stand out in the AI race, and OpenAI is being stripped bare.

Bad luck doesn't stop, when the news of Meta recruiting two more researchers from OpenAI spread, another disaster came:

OpenAI was about to acquire Windsurf, but it was snatched away by Google!

Windsurf grew from a browser plug-in to an AI coding platform valued in the billions of dollars in just four years.

Last August, Windsurf completed a $150 million Series C funding round, reaching a valuation of $1.25 billion. By April of this year, Windsurf launched a new funding round, aiming to raise $3 billion.

Then, multiple media reports stated that OpenAI was negotiating to fully acquire the company for $3 billion, hoping to quickly make up for the three missing pieces: "IDE + code RLHF data + enterprise distribution."

However, on the local time of July 11th, Google and Windsurf suddenly announced a shocking message: the transaction of OpenAI acquiring Windsurf had been canceled, and Google would hire Windsurf's CEO Varun Mohan, co-founder Douglas Chen, and some of Windsurf's R&D employees, and incorporate them into Google DeepMind's team.

Among them, Mohan and Windsurf's employees will focus on Google DeepMind's agent coding work and mainly contribute to the development of the Gemini project. Google will not have any control or equity in Windsurf, but will receive non-exclusive licensing rights to some of Windsurf's technologies.

It should be noted that this is not an acquisition, Windsurf is still in existence, with a new temporary CEO and new president.

This is a classic operation of American tech giants in recent years, not in the name of acquisition, but in the actual practice of "swallowing." Last year, Microsoft's move against the startup InflectionAI was exactly the same. This can avoid the risk of monopoly and also absorb technology and talent.

OpenAI went to great lengths, but the company it wanted to acquire ended up partnering with the competitor Google, which is very embarrassing.

Even more embarrassing is that according to reports from the Wall Street Journal and others, the main reason OpenAI failed to acquire Windsurf was due to the interference of its "big investor" Microsoft.

In short, OpenAI didn't want Microsoft to use Windsurf's technology, which violated the contract between OpenAI and Microsoft. So Microsoft raised objections to certain terms of the acquisition, leading to a deadlock in negotiations, and after the 60-day lockout period, the deal fell through.

Three

To say the least, OpenAI is really not easy now.

This startup company was founded in the form of a "non-profit organization," and is currently undergoing a transformation, wanting to turn the commercial entity into a PBC (Public Benefit Corporation). However, the transition process is very difficult, and Microsoft is a major obstacle.

Microsoft and OpenAI have been together since the early days, investing hundreds of billions of dollars in OpenAI. The contract gives Microsoft the right to veto equity dilution, major acquisitions, and company reorganization, so if OpenAI wants to restructure and introduce new investors, it must first get Microsoft's approval.

In addition, there is an "AGI clause": once OpenAI announces the achievement of AGI and expects profits exceeding 100 billion dollars, Microsoft will lose subsequent model exclusivity, and cannot develop AGI on its own.

According to reports from Reuters and Business Insider, Microsoft has yet to approve due to the need to renegotiate profit sharing, causing delays in OpenAI's restructuring. Wired disclosed that Microsoft is using the approval of the restructuring as leverage, requiring OpenAI to withdraw or weaken the AGI clause that allows termination in the future.

OpenAI is naturally very anxious. In the 6.6 billion dollar, 15.7 billion dollar valuation financing in October of last year, OpenAI explicitly wrote "complete the company structure transformation within two years" into the terms. If it cannot fulfill this, then the 660 million dollars will be converted into debt.

On this difficult path of transformation, OpenAI also has to withstand the fierce attacks from competitors.

On one hand, GPT-5 of OpenAI keeps postponing its release, and it has not been launched yet. Competitors such as Google, Meta, and xAI keep announcing model iterations, although OpenAI's current models are still among the top, but the clock is ticking, and there is little time left for it.

On the other hand, the competition for AI talents has always been fierce, and now it has been pushed to a new height by Meta.

OpenAI is not without a counterattack.

First, Altman did not remain silent, but directly expressed his dissatisfaction with Meta's behavior of offering exorbitant salaries to recruit people, describing it as directly robbing someone's house.

Second, OpenAI allowed its employees to take a week off and promised to reconsider the issue of salaries.

The most important thing is that OpenAI also took action to recruit people.

On July 9th, multiple media reported that OpenAI had recruited four key employees.

However, it's awkward that while Meta struck OpenAI hard, the talents OpenAI recruited were only one from Meta.

This person is also a Chinese person named Angela Fan, who majored in statistics at Harvard University, and later obtained a Ph.D. from the Nancy Research Center of the French National Institute for Research in Computer Science and Automation (INRIA), with a research direction in text generation. She has worked at Meta for 9 years.

Other than that, the talents OpenAI has recruited are all related to Musk.

They include Uday Ruddarraju, one of the "Twelve Heroes" of xAI's startup team, and xAI researcher Mike Dalton. Another new employee joining OpenAI is also a Chinese person named David Lau, the former head of software engineering at Tesla, who has visited China multiple times to oversee the Shanghai and Beijing R&D centers and meet with the media face-to-face.

How much money OpenAI spent to recruit these talents is unknown, but it probably won't be as extravagant as Meta.

The Information recently reported that OpenAI's stock incentive in 2024 reached as high as 4.4 billion dollars, accounting for 119% of its annual revenue of about 3.7 billion dollars — meaning "the cost of issuing stock is higher than the money earned." This number is more than five times that of 2023. That is to say, before Meta's aggressive attack, OpenAI had already defensively paid high costs for talent.

What else can OpenAI do next? Let's wait and see.

This article is from the WeChat official account: Zimu Bang, author: Xiao Jinya, editor: Wang Jing

This content represents the independent views of the author and does not represent the position of Huxiu. Reproduction without permission is prohibited. For authorization matters, please contact hezuo@huxiu.com.

This article is from Huxiu, original link:

https://www.huxiu.com/article/4570814.html?f=jinritoutiao

Original: https://www.toutiao.com/article/7526298895257698855/

Declaration: The article represents the views of the author. Please express your attitude by clicking on the [top/down] buttons below.