Zihang Dai, a co-founder of xAI and the researcher behind the influential Transformer-XL paper, has left Elon Musk's AI lab, marking the fourth co-founder departure. Dai is best known for his work at Google Brain on Transformer-XL, a groundbreaking architecture for modeling long-range dependencies in sequences that has accumulated over 3,000 academic citations.
The Transformer-XL paper introduced the concept of segment-level recurrence with a novel positional encoding scheme, enabling transformers to learn dependencies far beyond the fixed-length context windows that limited earlier models. Dai's deep expertise in transformer architectures was a key asset for xAI as the company developed Grok and pushed the boundaries of model capability.
His departure, combined with those of other technically distinguished co-founders, raises fundamental questions about xAI's ability to retain the kind of deep research talent needed to compete at the frontier. The pattern of departures suggests systemic issues rather than isolated decisions, with industry speculation centering on disagreements over research direction and the pace of commercialization.
