Goodbye, xAI: Co-founder Babuschkin Leaves. A Long Letter on Twitter Tells the Story of Ideals and Farewell with Musk

Co-founder of xAI Igor Babuschkin announced that his last day at xAI has ended.

The news was announced on his personal Twitter. Babuschkin said that this artificial intelligence company, which he co-founded with Elon Musk in 2023, once carried the shared vision of the two for the future.

He recalled that when he first met Musk, the two talked for hours, discussing artificial intelligence and the future of humanity. Both of them believed that the world needs an AI company with a different mission.

Babuschkin admitted that AI that drives human progress has always been his lifelong dream.

His parents left Russia after the collapse of the Soviet Union and immigrated overseas in pursuit of a better life for their children. Amid the hardships of immigrant life, his parents always adhered to human values: courage, compassion, and curiosity about the world.

As a child, he admired scientists like Richard Feynman and Max Planck, who constantly explored the boundaries of physics just to understand the universe. As a doctoral student in particle physics at CERN (European Organization for Nuclear Research), he initially believed that large colliders could reveal new physics.

However, there were fewer and fewer new discoveries, while the experimental equipment required larger and larger investments. So he began to think: Perhaps the key to unlocking the mysteries of the universe is not a larger collider, but superintelligence.

Can artificial intelligence propose a unified theory of quantum gravity? Can it prove the Riemann Hypothesis? By early 2023, he was convinced that humanity was close to the formula for creating superintelligence and realized that this technology must be used for good.

Musk had been warning about the potential dangers of powerful AI for years. The two hit it off and began to recruit like - minded engineers to establish xAI.

01 The Sprint from Zero to the Peak

xAI didn't get off to a smooth start. Some people said bluntly that they were too late to enter the game, and it was almost impossible to build a top - tier AI company from scratch.

But Babuschkin and the team chose to face the challenges head - on. In the early days of the company, he personally built many core tools for starting and managing training tasks.

Subsequently, he was in charge of the entire engineering system, including infrastructure, products, and applied AI projects.

xAI's team is well - known in the industry for its execution speed. They completed the construction of the Memphis super - cluster in 120 days, which was an impossible task in the eyes of many industry veterans.

In the final stage, a mysterious RDMA communication problem occurred in the cluster. Musk made a prompt decision and flew to the data center himself.

The team arrived in Memphis in the middle of the night and immediately started troubleshooting. They went through thousands of lines of lspci output and finally identified the problem: an incorrect BIOS setting.

When the training task ran successfully, Musk announced their victory at 4:20 am. This moment made everyone burst into laughter.

Babuschkin said that the adrenaline rush and the team spirit that night are still unforgettable to him.

He said he learned two things from Musk: First, have the courage to personally solve technical problems; second, have an almost crazy sense of urgency.

Under his leadership, xAI not only caught up with the forefront but also set a record for the fastest launch of a cutting - edge model in history.

Babuschkin is very grateful to the team. He said these people are the most dedicated partners he has ever met. They worked day and night, burning the midnight oil, and turned the "impossible" into reality.

02 The Mission of the Next Chapter

When leaving xAI, Babuschkin described himself as a parent sending a child to college. There are both tears and pride in his heart, and he looks forward to the company's continued growth.

He said that he will carry on the spirit of his parents in pursuing a better world.

A dinner with Max Tegmark, the founder of the Future of Life Institute, made his direction clearer. Tegmark showed him photos of his two young sons and asked, "How can we build AI safely so that our children can thrive?"

This question deeply touched Babuschkin. He was once the technical leader of the AlphaStar project at DeepMind and witnessed the power of large - scale reinforcement learning.

He foresees that future cutting - edge models will have stronger capabilities over longer time spans and a wider range of tasks.

Therefore, AI safety research will become crucial.

Now, he announces the establishment of Babuschkin Ventures, which is committed to supporting AI safety research and investing in AI and autonomous system startups that drive human progress and explore the mysteries of the universe.

03 OMT

The following is the original tweet, translated by DeepSeek:

Today is my last day at xAI.

I co - founded this company with Elon Musk in 2023. I still clearly remember the day I first met Elon. We talked for hours, discussing artificial intelligence and the possibilities of the future. We both felt that the world needed an AI company with an extraordinary mission.

My lifelong dream is to create artificial intelligence that can drive human progress. My parents left the Russian Federation after the collapse of the Soviet Union in search of a better life for their children. As immigrants, our lives were not always easy, but my parents always believed in courage, compassion, and curiosity and the thirst for knowledge about the world.

I've admired scientists like Richard Feynman and Max Planck since I was a child. They tirelessly explored the boundaries of physics just to understand the essence of the universe. During my doctoral studies, I was engaged in particle physics research at CERN, hoping to contribute to this mission. But finding "new physics" became increasingly difficult, requiring larger and larger colliders, while real discoveries became fewer and fewer. That's when I started to think that perhaps the key to unlocking the mysteries of the universe is not a larger collider, but "superintelligence".

So, can AI derive a consistent theory of quantum gravity? Can it prove the Riemann Conjecture?

In early 2023, I became increasingly convinced that we were approaching the "formula" for achieving superintelligence. I saw the trend: AI would soon have reasoning abilities beyond humans. Then, how to ensure that this technology is used for good? Elon has been warning about the risks of powerful AI for years, and during our communication, I realized that our visions were the same, which is to make AI benefit humanity. So, we gathered more like - minded engineers and started building xAI.

The early days were not easy. Many people said we were too late, and it was impossible to build a top - tier AI company from scratch. But we believed that "impossible" was just a challenge. When the company was just established, I personally built many core tools for starting and managing training tasks. Later, I was in charge of most of the company's engineering work, including infrastructure, products, and applied AI projects. The xAI team is full of dedication. With hard work and perseverance, we built the Memphis super - cluster at an amazing speed and launched a cutting - edge model at an unprecedented speed. I learned two invaluable lessons from Elon: First, don't be afraid to get your hands dirty and solve technical problems in - depth; second, always maintain a crazy sense of urgency.

xAI's execution speed is "crazy". Industry veterans said it was impossible to build the Memphis super - cluster in 120 days. But we chose to challenge the "impossible" again. Our goal was to get the large - scale training system running on the Memphis cluster as soon as possible. As the 120 - day deadline approached, we encountered a mysterious RDMA communication problem. Elon decided to fly to the data center himself, and we went with him. The team arrived in Memphis in the middle of the night and immediately got to work. We sifted through thousands of lines of lspci output and finally found the root of the problem: an incorrect BIOS setting. That night, Elon stayed with us until the training task finally ran through. He posted a celebration at 4:20 am, and we all laughed out loud. The adrenaline rush and the camaraderie that night are unforgettable for me.

I love the xAI family deeply. This is the most dedicated team I've ever worked with. It's not easy to catch up with the industry forefront in such a short time. This is the result of everyone's perseverance and team spirit. I thank every teammate for your time, effort, and sacrifice. I'll always remember those all - nighters. When I drove away today, I felt like a parent sending a child to college, with both reluctance and pride in my heart.

Entering the next chapter, I'm reminded of my parents' courage to immigrate for their children. Recently, I had dinner with Max Tegmark, the founder of the Future of Life Institute. He showed me photos of his two young sons and asked me, "How should we build AI safely to ensure that our children can thrive?" This question really touched me.

Early in my career, I was the technical leader of AlphaStar, the StarCraft AI at DeepMind, and witnessed the powerful force of large - scale reinforcement learning. As cutting - edge models become more "autonomous" over longer time spans and a wider range of tasks, their capabilities will become stronger and stronger. Therefore, researching and promoting AI safety will become extremely important.

I will continue my mission to promote safe and beneficial AI for humanity. Today, I announce the establishment of Babuschkin Ventures, which focuses on supporting AI safety research and investing in AI and autonomous system startups that drive human progress and explore the mysteries of the universe.

This article is from the WeChat official account "Big Data Digest", and is published by 36Kr with authorization.