Zuckerberg's billion-dollar talent poaching spree across Silicon Valley has hit a snag this time. The man who gave him a cold shoulder happens to be one of his own: Meta FAIR researcher Zeyuan Allen-Zhu. (Honestly, even LeCun would slowly type out a question mark hearing about Zuckerberg's move)

To be specific, he rejected Zuckerberg's invitation to join MSL (Meta's Super Intelligence Lab) - the new Meta department where Zuckerberg has repeatedly offered hundred-million-dollar annual packages to recruit talent from everywhere.

On Allen-Zhu's homepage, his latest tweet welcomes new FAIR colleagues. Not only is he unmoved by MSL, Allen-Zhu seems reluctant to get involved with Meta's GenAI department either. A particularly striking statement appears on his LinkedIn profile:

Looking back briefly, Allen-Zhu has long been a shining star for the future. He won two IOI gold medals, an ACM-ICPC World Finals gold medal, and ranked among the top globally in Google's programming challenges. His academic credentials are equally enviable: undergraduate degree from Tsinghua University's Physics Department, PhD from MIT's Computer Science and Electrical Engineering Department, and postdoctoral research positions at Princeton University and the Institute for Advanced Study. His first industry job involved research on large language models.

Allen-Zhu joined FAIR over three years ago; now, he chooses to continue working at FAIR. Netizens are moved, noting that some things in this world are more important than piles of money.

The news about Zuckerberg trying to poach Allen-Zhu from Meta's FAIR to Meta's MSL actually came from Allen-Zhu himself. He posted a tweet stating:

The Meta MSL's TBD lab he mentioned is a newly unveiled core component in Meta's recent AI strategy adjustment. TBD Lab stands for "To Be Determined Lab," an elite sub-team under MSL specifically responsible for creating the next generation Llama series models, especially the new versions codenamed Llama 4.X/4.5. The lab is led by Meta's Chief AI Officer Alexandr Wang (former CEO of Scale AI), with the goal of achieving breakthroughs in model reasoning, autonomy, and multimodal understanding capabilities.

Although Allen-Zhu quickly deleted that tweet, Zuckerberg's plan to have Allen-Zhu "transfer departments" within Meta fell through. Allen-Zhu chose to stick with FAIR.

In July 2022, Allen-Zhu left Microsoft after five years and joined Meta FAIR. Three years later, as a FAIR researcher, his research focuses on exploring the physics principles of large models and AI in a broader sense, including:

His proposed reverse training method has also achieved practical results on open-source models like Llama-2.

From Meta's perspective, it's hard to say Allen-Zhu's choice is "wise." In Meta's current AI organizational structure, FAIR and MSL are parallel but completely different departments with very distinct goals.

GPT-5 provides this comparison table:

Although fast-paced sprints and pushing AI theoretical breakthroughs aren't mutually exclusive, and despite Meta denying that it values FAIR less than before, everyone somewhat senses that FAIR seems to have fallen out of favor with Zuckerberg.

This is evidenced not just by Zuckerberg's attempt to persuade Allen-Zhu to transfer internally to MSL. First, MSL is his personally driven AGI dream factory, set up specifically apart from FAIR. High-salary talent poaching, massive model training, rapid product deployment - everything aims to achieve "personal super intelligence."

For this purpose, Meta acquired substantial equity in Scale AI and recruited its founder Alexandr Wang as MSL's new leader. Second, to build the team and strengthen MSL, Zuckerberg has been intensively using compensation packages worth up to $100 million to aggressively recruit talent over a long period.

Anyone can see that while FAIR and MSL are parallel in organizational structure, from actual resource investment and company strategic focus, MSL has become Meta's new core for AI advancement. As many media outlets have stated, FAIR has been noticeably marginalized.

These operations aren't inherently right or wrong. Although Meta isn't short of money, it's fundamentally still a company. FAIR's basic research has a long way to go, and if Zuckerberg wants to seek breakthroughs in product implementation, you can hardly say he's wrong.

But friends wanting to work in AI at Meta now face a choice: either go to the new lab to focus on implementation, or stick to basic research. Allen-Zhu chose the latter.

Last month, he clearly expressed the differences between FAIR and GenAI/MSL:

Allen-Zhu emphasized that FAIR is completely independent from GenAI and MSL in terms of data, code, infrastructure, and computing clusters. The success, failure, or internet memes of GenAI/MSL don't represent FAIR's work situation.

Additionally, he has publicly emphasized the importance of research freedom and long-term basic research. After becoming part of FAIR, Allen-Zhu said he had the opportunity to initiate projects completely autonomously, which is crucial for his "Language Model Physics" project.

For people like Allen-Zhu, academic freedom, research atmosphere, and long-term value are what they value more. (MSL's packages are indeed high, but FAIR's compensation probably isn't very low either... right?)

Whether it's research ideals or financial calculations, this AI talent war has just begun. At least with Ilya, Mira, and Allen-Zhu, we see:

Honestly, Zuckerberg's recent recruitment frenzy has left us spectators numb. But major AI companies definitely feel the pressure - OpenAI no longer allows mentioning specific researcher names in their latest livestreams.

However, this doesn't stop Zuckerberg's golden recruitment shovel. The latest battle report is that an OpenAI Chinese engineer was poached by Zuckerberg as soon as they emerged (covering face for OpenAI with a big cry).

This prolonged recruitment drama has been so entertaining that many friends have overlooked one point - since ChatGPT's launch, many talented individuals have actually been leaving Meta gradually.

Over the past few years, Meta's talent loss has been devastating. Three former Meta AI employees told Forbes that before this new wave of talent joined, Meta was already talent-rich, "already had the best people, but lost them to OpenAI..."

Insiders indicate that during this period, even as Zuckerberg offered stunning packages to top AI researchers, the giant's remaining talent continued to leave.

Feel the severity of Meta's talent loss through one insider close to Meta:

In other words, Zuckerberg wasted the AI talent he originally had, and now he's spending billions of dollars to fill the gaps.

Meanwhile, people don't seem to hide their disdain for Zuckerberg's sky-high recruitment spending spree - Anthropic CEO Dario Amodei stated that some employees who received Meta offers but haven't decided whether to leave have communicated with him, and he clearly stated he wouldn't re-negotiate salaries because of Zuckerberg's sky-high offers.

He said in a podcast last month:

Another unnamed AI company founder said: "Meta hires many decent-level AI scientists with high salaries, then ordinary people mistakenly think these are the world's best AI scientists because of the sky-high compensation."

Now, Silicon Valley commonly circulates the belief that Meta doesn't fit in with real AI companies because Zuckerberg attracts profit-driven "mercenaries," while real AI companies attract true AI believers and "missionaries."

Altman supports this view. In a letter he wrote to OpenAI employees in July, he mentioned:

Altman also added that "I believe OpenAI options have more upside potential than Meta stock."

Of course, huge upside potential should be built on huge success (insert dog head emoji).

Then, while Altman criticized Meta's actions as deeply affecting cultural issues between companies and talent, he started giving his research and engineering teams bonuses worth millions of dollars.

Ha, these two men.