Does a Turing Award laureate report to a guy born in 1997? Mark Zuckerberg hires a rising star with an annual salary of 100 million, and a veteran laments staying up late to scavenge for GPUs.

What kind of magical plot is it that a Turing Award laureate reports to a guy born in 1997? The "teenager in data annotation" hired by Zuckerberg with $14.3 billion has been promoted to Meta's Chief AI Officer. On one hand, Zuckerberg offers new employees annual salaries in the hundreds of millions of dollars. On the other hand, veteran Meta employees are in urgent need of GPUs and have to stay up late borrowing cards, almost going bald. Netizens exclaimed in pain: They really feel sorry for the employees of Meta FAIR...

I woke up suddenly and saw LeCun reporting to Alexandr Wang!

One is one of the three Turing giants and a long - standing academic authority, and the other is a guy born in 1997 who has successfully become a billionaire through data annotation. This scene is really too magical.

For Wang, Zuckerberg spent a full $14.3 billion to acquire 49% of the equity of Scale AI. It can be said that it is truly "buying a company for a person".

Now, Alexandr Wang, the former CEO of Scale AI, has risen to prominence and become Meta's "Chief AI Officer", and will lead Meta's new dream team, the "Super Intelligence Laboratory".

All the top talents from OpenAI that Zuckerberg hired with hundreds of millions of dollars will report to him.

Even... including LeCun?

Just thinking about this possibility makes everyone gasp.

Just yesterday, Alexandr Wang officially posted on X, saying that he was very excited to become Meta's Chief AI Officer and also reposted the magnificent list of 11 people.

Clement, the co - founder and CEO of HuggingFace, immediately congratulated him, saying that due to the addition of Wang and Nat Friedman, Meta's influence in the AI field will increase by 100 times.

And LeCun also very "sensibly" reposted this post on X.

Is it a polite act of going through the motions of "having to bow one's head under someone else's roof", or is it a sincere belief? Or, is it a kind of irony that "you should think it over carefully"?

We have no idea what LeCun really thinks deep down.

At least, at the center of the storm yesterday, he posted a thought - provoking post on X: "Chief Artificial Intelligence Scientist, has been since 2018."

Meta's veteran employees imply that Zuckerberg is overly biased

In short, Zuckerberg's recent actions have, on one hand, allowed many researchers to suddenly achieve the myth of getting rich overnight, and on the other hand, have broken the hearts of many hard - working veteran Meta employees.

Zeyuan Zhu, a researcher at the FAIR laboratory, said that due to limited access to GPUs, his research progress was relatively slow, although he had achieved very remarkable results.

Zhu said that this is the Galileo moment in LLM design.

Just as the Leaning Tower of Pisa experiment triggered modern physics, they revealed the real limitations of the LLM architecture in a controllable synthetic pre - training environment.

And this is likely to become a major watershed in LLM research.

Specifically, in real - world pre - training (with a scale of hundreds of billions of tokens), the differences between model architectures are often masked by excessive data noise.

But in the controllable synthetic data environment he built, the depth of inference has doubled, advanced capabilities have emerged early, and even high - quality data can predict the design path of future models.

For this reason, he designed five synthetic pre - training tasks to ensure that LLM can achieve real thinking and reasoning (i.e., System 1), rather than just the Chain of Thought (CoT).

For this reason, the researcher proposed the Canon layer, a lightweight horizontal residual structure. It can be seamlessly integrated into any model, but significantly improves the inference ability (the inference depth increases by 2 - 4 times and the breadth increases by 30%), and the overhead is extremely small, with a disruptive effect.

To complete this research, Zhu stayed up all month and conducted 1.9 million GPU - hours of experiments, and was once exhausted.

Zhu is an undergraduate from the Tsinghua University's basic science class, then studied at MIT and Princeton, and later became a researcher at the Meta FAIR Lab.

This research has also attracted the attention of many researchers on X, but it's a pity that it cannot be advanced quickly due to limited GPU resources.

Zhu even revealed that in order to carry out the experiment to the end, he had to look for idle GPUs in other teams' clusters and constantly change GPUs. Moreover, in order not to delay the progress of other teams, he could only use them at night and on weekends.

After this incident was discovered by Reddit users, they all said that they really felt sorry for the veteran employees of FAIR.

It is said that the FAIR laboratory has 300,000 GPUs. Who on earth are they all allocated to?

Polar opposites! On one hand, hundreds of millions in annual salary, on the other hand, immediate lay - offs

Recently, Zuckerberg has offered extremely outrageous salaries to recruit his desired employees.

It is said that the top - notch researchers have received a compensation package of $300 million over four years, and the total compensation in the first year alone exceeds $100 million.

Altman even said bluntly that they "have changed from a group of nerds in the corner to at least the most eye - catching figures in the tech industry."

Many researchers have suddenly received treatment similar to that of NBA stars.

However, a senior engineer at Meta said that his annual salary at Meta is only $850,000.

According to the