Former Lead World Modeler of xAI Leaks Musk's R&D Insider Information: The Wanker Cluster Burns Through Computing Power Every Hour, High-Pressure Environment Forces R&D to Initiate a "Marathon"

According to Voyant Beating monitoring, former xAI World Model Lead Ethan He decrypted the reason for his departure on the Latent Space podcast. He stated that the startup's shifting business priorities and objectives too frequently made it challenging for long-term cutting-edge research to align with short-term commercial deliverables. To break free from the cycle of relying solely on computational power in the video generation direction, he decided to focus on fundamental exploration in language models (LM) currently unexplored in xAI.

Ethan admitted that the operating costs of the Colossus ten-thousand GPU cluster were extraordinarily high, and any idle computational power per hour would create tremendous financial pressure. This ultra-large-scale infrastructure cost brought significant computational anxiety and psychological burden to the developers. Under a high-intensity work rhythm, the team had to maintain long-term operation with a "preserve the body for a marathon mindset."

The video team was able to go live with the Grok Imagine 0.9 audiovisual model in just 3 months, mainly achieving a qualitative leap through flattened management and data flow correction. During the development process, Musk was highly hands-on in technical decision-making. Ethan mentioned how fortunate he was to participate in a technical closed-door workshop convened by Musk, engaging in deep discussions on product and R&D details.