#Elon Musk lays off 500 employees over the weekend

What kind of situation does the person who helped Musk train the big model say they are unemployed?

Last Thursday evening, xAI conducted an internal surprise test and required employees to complete and submit it by the next morning.

This is not a simple in class test——

As of now, the elimination rate of xAI’s internal testing is as high as 33%, and over 500 employees have been notified to leave.

##Musk’s layoff style assessment

The data annotation team was once the largest team in xAI and played a key role in the development process of Grok. Their job is to label, classify, and place raw data in specific contexts, teaching AI how to better understand the world.

Since the establishment of xAI, the size of the data annotation team has been continuously growing.

**Unlike most artificial intelligence companies, many data annotators at xAI are hired directly, rather than outsourced. Through this approach, companies can have more control over model training and better privacy.

But correspondingly, the cost is also higher.

In February of this year, xAI disclosed plans to hire thousands of people to help train Grok and added about 700 data annotators within six months.

However, in early September, the LinkedIn page showed that at least 9 out of the dozen managers responsible for managing the data annotation team had been fired.

This unusual personnel change has planted the seeds for the upcoming intense turbulence.

Afterwards, xAI began conducting one-on-one conversations with some employees of the data annotation team.

During the communication process, these employees were asked to showcase their work results and how they can create value for xAI.

The rain is about to come and the wind is blowing all over the building. According to some employees, this series of interviews has created a sense of panic within xAI.

As expected, last Thursday evening, xAI released an announcement on Slack * (a team collaboration tool similar to Feishu):

Please ask some employees to put down their current work and participate in at least one test before tomorrow morning to determine their future roles in the company.

It is reported that the test covers traditional fields such as STEM, programming, finance, and medicine, as well as some more alternative directions, such as Grok’s “Personality and Model Behavior” and “Joker and Addicted Screen Walker”.

In addition, there are tests aimed at improving chatbot security.

After the announcement was released, over 100 employees expressed confusion on Slack, with one employee openly expressing dissatisfaction with the testing schedule.

It’s really shameful to do such things after people come home from work.

Rebellion did not have any effect, let alone prevent the bad news from spreading——

Not long after the test ended, on Friday night, many data annotators received dismissal emails announcing that their journey at xAI had come to an end.

The dismissal email reads as follows:

After a comprehensive review of our human data work, we have decided to accelerate the expansion and prioritization of data annotators, while reducing our focus on ordinary AI data annotator positions.
This strategic shift will take immediate effect. As part of this key shift, we no longer need most of the regular data annotator positions, and your employment with xAI will also be terminated.

The data annotation team of xAI mainly consists of two parts: one is professional data annotators responsible for vertical fields such as STEM, programming, finance, law, and media; Another large part is the general data annotators responsible for tasks such as video annotation, audio annotation, and writing. ）

According to the statistics on Slack chat rooms, the number of members in the data annotation team has dropped from 1500 to over 1000, which means nearly one-third of the employees have been laid off.

In sharp contrast, xAI’s new recruitment notice:

The company plans to expand the scale of professional data annotators by 10 times based on the existing foundation.

##Elon Musk turns direction: betting on vertical AI fields

For a long time, people have believed that more data is better.

However, the large tilt of xAI from a general data annotator to a professional data annotator this time may reflect Musk’s bet on the opposite viewpoint - * * quality is more important than quantity * *.

The benefit of doing so is that it can enhance Grok’s ability and credibility in specific fields.

But the drawbacks are also obvious. Relying on a smaller expert database may limit the diversity of data that the model can see. Artificial intelligence systems thrive on diversity, and overly shrinking training data may bring blind spots.

For model manufacturers, the focus is on how to find a new balance between quality and quantity when the constraints change.

Since the high-quality corpus on the Internet has been exhausted, Scaling Law has ushered in the inflection point of diminishing marginal effects, the cost of model training has risen rapidly, and the development of large models has also entered the platform period.

In addition, although general models have stronger breadth, vertical models with relatively small parameter quantities are more likely to ensure accuracy and compliance in industries such as finance, healthcare, and law.

In this context, the image of a unified universal model in the past is becoming increasingly blurred, while vertical scenes are flourishing.

Musk may have taken notice of this.

Therefore, the layoffs at xAI this time may not be a simple team restructuring, but a huge gamble aimed at the AI blue ocean of vertical industries.

If successful, Grok is expected to establish advantages in vertical application areas of the industry.

But if they fail, the Silicon Valley Iron Man team may stall on the Scaling Law track and be left behind by OpenAI and Anthropic.

##Why does Musk like to lay off employees?

As is well known, Musk’s drastic dismantling of the company team is not a new thing.

During the Model 3 production hell, Angry Old Ma laid off over 7000 employees within two years, which is just the number publicly disclosed by Tesla.

During that time, unless absolutely necessary, Tesla employees would not easily pass by Musk’s desk to avoid drawing his attention and endangering their careers with a single mistake.

Tesla employees who were deeply affected by it even coined a term specifically for it - Elon’s Angry Dismissal.

This skill has been continuously released by Musk from Tesla to X, and now to xAI.

In 2022, when Lao Ma took over Twitter, he also activated the “Angry Elon” mode, and within a week, he fired 3700 old Twitter employees, causing multiple departments of Twitter to be paralyzed.

Why does Musk like driving people so much?

This is closely related to his hardcore philosophy——

**I enjoy leading a strong team in combat, with a particular emphasis on performance and efficiency. If I can’t handle high-intensity work or can’t achieve results, I’ll leave.

Specifically, Musk prefers small teams with strong technical capabilities and firmly opposes expanding the team size. He believes that smaller teams with strong abilities and proactive progress usually perform better than much larger teams for three main reasons:

Firstly, small teams can mobilize fewer human resources, which can actually make them more creative, greedy, and busy.

Secondly, small teams almost do not require middle-level managers, making it easier to access first-hand information and achieving higher communication efficiency.

Thirdly, small teams run faster and can greatly shorten the iteration cycle.

And the employees who have been carefully selected by Lao Ma are not always safe.

Musk places great emphasis on performance and efficiency, with a low tolerance for low performers. Andrej Karpathy, former head of artificial intelligence at Tesla, mentioned that he sometimes even has to ‘fight to keep some people’.

Most Silicon Valley companies tend to “pamper” their employees, while Musk tends to “force” them to give their all, and those who perform poorly will soon be removed from the team. I need to work hard to convince him to keep some people in the team.

At the end of 2022, Musk issued a similar ultimatum to former Twitter employees via email at midnight, ‘Focus on hardcore culture, otherwise, please leave with severance pay’.

Looking to the future, to build a breakthrough Twitter 2.0 and succeed in an increasingly competitive world, we need to be very hardcore, which means working long hours of high-intensity work, and only outstanding performance can be considered passing.

But it has been proven that this approach does help stimulate the potential of the team.

So will the layoffs at xAI make Grok Great rise?

Reference link: [1] https://economictimes.indiatimes.com/news/international/us/mass-layoffs-at-elon-musks-xai-500-data-workers-axed-as-grok-chatbot-faces-big-changes/articleshow/123870379.cms
[2] https://www.businessinsider.com/xai-management-shakeup-data-annotation-team-2025-9
[3] https://x.com/xai/status/1966677385056841884
[4] https://www.theguardian.com/technology/2025/may/05/elon-musk-doge-federal-government
[5] https://www.wired.com/story/elon-musk-tesla-life-inside-gigafactory/
[6] https://www.theverge.com/2022/11/16/23462026/elon-musk-twitter-email-hardcore-or-severance
[7] https://medium.com/the-low-end-disruptor/elon-musks-leadership-style-424343fe6cb3