Ethan He Reveals Core Reasons for Leaving xAI
Ethan He, former head of world models at xAI, stated on the Latent Space podcast that the frequent adjustments in business priorities and goals at startup large model companies make it difficult to align long-term cutting-edge research with short-term commercial delivery.
He decided to focus on foundational exploration of language models to break free from the cycle of video generation that relies solely on computational power. The operational costs of the Colossus cluster are extremely high, creating immense pressure from idle computing power, and the team needs to maintain operations with a "marathon mindset for health."
Market Mechanism: Top AI talents are accelerating their movement, with capital gathering towards companies with more stable research environments and clear long-term goals. xAI is currently facing critical talent loss and team stability pressures in the short term, while competing AI labs benefit from attracting high-level researchers.
Supplementary Data: The video team launched Grok Imagine 0.9 in just 3 months, with Musk being highly involved in technical decisions.
Source: Public Information
ABAB AI Insight
Ethan He previously led the world models direction at xAI, and his departure continues his pursuit of long-term foundational technology exploration. He has accumulated extensive experience in large model research at institutions like Google DeepMind and has a clear understanding of the conflict between commercial pace and research.
On the capital path, xAI has heavily invested in video generation through the Colossus cluster, but Ethan pointed out that frequent priority adjustments and enormous computational costs have led to excessive R&D pressure. His motivation is to find an environment that allows for deep investment in foundational innovations in language models, while also exposing the coordination challenges of ultra-large-scale infrastructure in startups.
Similar to many top researchers leaving high-intensity commercial labs for more focused research, xAI is currently in a phase of rapid expansion, facing a key transition stage in balancing Musk's hands-on decision-making style with the stability of long-term research.
Structural Judgment: This essentially represents a technological shift. Ethan He's departure reflects a potential transition in the video generation field from mere computational power stacking to more efficient architectural exploration, driving talent from high-cost cluster projects towards foundational research in language models, and shifting pricing power from large-scale training to innovative algorithms and architectures. The mechanism is that the enormous infrastructure costs accelerate the re-optimization of R&D resources.
ABAB News · Cognitive Law
The higher the computational costs, the harder it is to retain top talents doing foundational research.
The faster the commercial pace, the easier long-term innovation is sacrificed.
The best research environment is one where geniuses don't have to worry about GPU idling every day.