【By Observer Net, Yuan Jiaqi】

On July 14 local time, U.S. media outlet The New York Times reported that Meta's newly established "Super Intelligence Lab" (MSL) is planning to make a major adjustment to its artificial intelligence (AI) strategy. According to two sources familiar with the matter, last week several core members of the lab, including 28-year-old new AI head Alexandr Wang, considered abandoning Meta's most powerful open-source AI model Behemoth and instead developing a closed-source model.

This news has left American scholar Ethan Mollick, who has been deeply involved in the open-source field for over two decades, feeling nostalgic. He bluntly said, "In this way, the United States basically steps back from the competition in cutting-edge open-source large language models (LLMs). Europe still has one competitor, while the rest of the market has almost become China's domain."

Mollick is an associate professor at the Wharton School of the University of Pennsylvania, a Ralph J. Roberts Distinguished Scholar, academic director and co-founder of the Wharton Interactive program, and also a co-director of the Penn Generative AI Lab. His main research areas focus on the impact of AI on work, entrepreneurship, and education.

His academic achievements have been published in multiple top-tier journals, and his research in the field of AI is widely applied. He was selected as one of the "Most Influential People in the AI Field" by Time magazine in the United States. He also popularizes AI knowledge among the general public. His book "Co-Intelligence: How to Live and Work with AI" not only topped the New York Times bestseller list, but it was also named "Book of the Year" by The Economist and Financial Times.

According to The Wall Street Journal, because he had early exposure to large language models, Mollick is now the preferred AI expert for U.S. policymakers and corporate leaders.

This January, after the Chinese startup AI company DeepSeek launched the DeepSeek model, Mollick praised it highly after testing it. He once mentioned on a social media post that the Chinese DeepSeek can compete with cutting-edge models and also adopts an open-source model. One of the implications of this phenomenon is that the regulation in the United States (let alone the EU) seems increasingly unlikely to stop the continuous development of AI models.

Recently, Mollick also spoke on the social media platform X about the KIMI K2 open-source model released by Moonshot AI on the 11th. He believed that this model might also spread quickly among developers, but as the impact of "Chinese models" has basically faded, and for most users, the performance of DeepSeek is already sufficient, so KIMI K2 may find it difficult to replicate the popularity of DeepSeek's "viral" spread.

Compared to China's large language models, most U.S. models are not open-source, which is why Meta's news has made Mollick, who closely follows the progress of Sino-U.S. models, feel deeply.

Ethan Mollick

According to The New York Times, for years, Meta has always maintained an open-source approach for its AI models. Meta executives long believed that the open technological development model is more conducive to promoting the rapid development of AI and allows more developers to access related technologies. Shifting to a closed-source model would be both a technical change and a conceptual innovation for Meta.

This strategic adjustment may be related to poor model performance. According to the aforementioned sources, Meta has completed training for the Behemoth model, but due to internal test results being unsatisfactory, the release of the model has been postponed. One of the sources also stated that since Meta announced the establishment of the Super Intelligence Lab last month, the team responsible for Behemoth has stopped conducting new tests on it.

The discussions within the Super Intelligence Lab are still in the initial stage, and the final decision requires approval from Meta CEO Zuckerberg. It has been reported that Meta may retain open-source models while shifting its focus to closed-source models. Regardless of the chosen path, for Meta, which is competing with Google, OpenAI, Anthropic, and other rivals for AI leadership, it will be a significant strategic shift.

A spokesperson for Meta refused to comment on the matter. However, Zuckerberg had previously laid the groundwork in a podcast interview, where he stated that Meta clearly supports open source, "but we have not promised to publicly release every achievement we make."

The U.S. media mentioned that Meta has encountered consecutive setbacks in the AI technology field, hence the Super Intelligence Lab has attracted much attention. To address this, Zuckerberg has initiated a "high-salary talent recruitment" strategy, offering salaries exceeding hundreds of millions of dollars to attract top researchers from companies such as OpenAI, Google, Apple, and Anthropic, causing dissatisfaction among competitors.

In June this year, Meta announced a massive investment of 14.3 billion USD in Alexandr Wang's AI startup Scale AI. According to the agreement, Meta holds nearly half of the shares in Scale AI, and Alexandr Wang and his core team from Scale AI joined Meta and took on leadership positions.

Subsequently, Meta renamed its entire AI department to the "Super Intelligence Lab," with Alexandr Wang serving as chief executive. In this larger AI department, Alexandr Wang leads a dedicated team of approximately 12 people, including several deputies from Scale AI and Nat Friedman, former CEO of the software startup GitHub. This team's office space is separated from other departments and is adjacent to Zuckerberg's office.

A source familiar with the situation revealed that last Tuesday, Alexandr Wang held a meeting with approximately 2,000 AI department employees at Meta. During the meeting, he stated that the work of his small team would remain confidential, but the entire AI department of Meta will now focus on developing super intelligence. However, he did not mention whether the future models would adopt an open-source or closed-source model.

Another source stated that as the next stock vesting period approaches in August, some employees not selected into the Meta Super Intelligence team may face mass departures.

This article is exclusive to Observer Net. Reproduction without permission is prohibited.

Original: https://www.toutiao.com/article/7527150694827131401/

Statement: The article represents the views of the author. Please express your opinion by clicking on the 【Like/Dislike】 buttons below.