Musk verliert einen wichtigen Mann von xAI: Der Schöpfer von Grok 4 kündigt plötzlich, und ein langes Essay enthüllt die spannendsten Geheimnisse des Unternehmensgründungsvorhabens.
Another co-founder of xAI has officially announced his departure! Igor Babuschkin, the father of AlphaStar, wrote a long post to say goodbye to the company. He recalled that he had driven his team hard for 120 days to build the world's most powerful supercomputer. Elon Musk himself personally thanked him: Without you, there would be no xAI today.
The co-founder of xAI, Igor Babuschkin, has officially announced his departure and the founding of his own company!
On his last day at xAI, he recounted in a long post the day in spring 2023 when he met Elon Musk –
We talked for hours about the future and the endless possibilities of artificial intelligence. We agreed that the world needs a new AI company with a different mission.
In two years, they managed to achieve an amazing feat.
They not only built the world's largest AI supercomputer, Colossus, within 120 days but also trained the multimodal Grok 4, which can keep up with the systems of OpenAI and Google DeepMind.
Now, the AI expert Igor Babuschkin officially ends his journey at xAI with his departure.
Next, he will found the venture capital firm Babuschkin Ventures and start a new journey.
Under the post, Musk gratefully said: "Without you, we wouldn't have come this far."
The other co-founder, Toby Pohlen, listed the unforgettable moments:
- On the first day, we flew straight to Austin;
- We all got together to develop Grok - 1;
- We reviewed 15,000 applications in one day;
- And then there was that Sunday at the winery with Kyle and Lilly when we launched the open - source website and code repository of Grok - 1...
"I'm very grateful to you for calling and inviting me to join in early 2023. I'll always be thankful for that."
What kind of person is Igor Babuschkin? What central contributions did he make during his time at xAI?
The father of AlphaStar, who built the largest AI supercomputer in 120 days
Before the founding of xAI, Igor was a leading development engineer at Google DeepMind.
During his time there, he led the AlphaStar project, an AI system for the game "StarCraft". Through training with reinforcement learning, it finally defeated human players and reached a grandmaster level.
He also participated in the development of the WaveNet speech synthesis system, which significantly improved the naturalness and quality of speech synthesis.
After two years at Google, Igor switched to OpenAI, where he was involved in the development of core technologies behind ChatGPT, such as GPT - 4. Then he returned to Google.
Igor's experiences at Google DeepMind and OpenAI have given him rich knowledge in deep learning and architecture design.
He also has a strong background in physics. He worked at the Large Hadron Collider and then switched to this field because of his passion for large models and AI.
According to his personal statement, he obtained a master's degree in physics from the Technical University of Dortmund.
During his time at xAI, Igor led the construction of the supercomputer cluster Memphis – Colossus – which supported the large - scale training of Grok.
Moreover, he also led the development of Grok from the first version to the various iterations. Today, Grok 4 is a top - tier model that can compete with Gemini 2.5 and GPT - 5.
When Grok 4 was released, Musk announced that there are further plans in the coming months.
From now on, a key member will be missing from xAI's future historical releases.
Farewell Post: The Victory at 4:20 a.m.
Developing an AI that drives human progress is my life goal.
After the collapse of the Soviet Union, my parents left Russia to seek a better life for their children.
As migrants, life was not always easy. But despite all the difficulties, my parents always believed in the invaluable importance of human values: Courage, empathy, and curiosity about the world.
As a child, I admired scientists like Richard Feynman and Max Planck, who expanded the boundaries of physics to understand the universe.
As a doctoral student in particle physics at the European Organization for Nuclear Research (CERN), I was excited to be involved in this task.
But the path to discovering new physics became increasingly difficult. The accelerators got bigger, but the new discoveries became fewer.
So I began to think that the key to understanding the universe might not be a bigger accelerator but superintelligence.
Can AI develop a unified theory of quantum gravity? Can AI prove the Riemann hypothesis?
In spring 2023, I was convinced that we are close to the secret of creating superintelligence.
I felt that it was only a matter of time before the logical abilities of AI would surpass humans. But how do we ensure that this technology is used for good?
Elon has been warning about the risks of strong AI for years. I and Elon recognized that we have a common goal: To use AI for the benefit of humanity.
So we hired other like - minded people and founded xAI.
The founding phase of xAI was full of challenges.
Some told us it was too late and that it was impossible to build a first - class AI company from the ground up. But we were convinced that we could make the impossible possible.
Building from scratch requires a lot of initiative:
- Initially, I developed many of the basic tools for the company's internal training tasks.
- Later, I also took over most of the company's projects, including infrastructure, products, and AI applications.
Everyone at xAI was full of passion. Our team worked hard and built the supercomputer cluster Memphis and developed a top - tier model in an incredibly short time.
I learned two valuable lessons from Elon:
- First, one must be willing to roll up one's sleeves and solve technical problems oneself;
- Second, one must have an almost crazy sense of urgency.
xAI's implementation speed was incredibly fast.
Industry experts said that it was impossible to build the supercomputer cluster Memphis in 120 days. But we were convinced that we could make the impossible possible.
Our goal was to fully operate the training facilities in the Memphis cluster as soon as possible.
As the 120 - day deadline approached, we had problems with the RDMA communication between the machines. So Elon decided to fly to the data center himself, and we followed him.
Our infrastructure team arrived in Memphis at midnight and immediately started working.
After we checked thousands of lines of the lspci output line by line, we finally found a wrong BIOS setting – that was the cause of the problem.
Elon accompanied us late into the night. When the training task was finally successful, Elon wrote a post at "4:20 a.m." to announce our victory, which made us laugh.
I'll never forget how adrenaline - filled I was that night and how strong the bond was between all of us. When we went to bed, we felt like we were in the most exciting moment of our lives.
I love everyone in the xAI family.
Our team is so special – you are the most dedicated and professional colleagues I've ever worked with. It wasn't easy to reach the top of the industry so quickly.
All this is due to the tireless will and team spirit of each member.
Thank you to everyone who joined me on this journey.