Tri Dao

A key researcher for understanding why state-space models became a serious alternative to standard transformer stacks rather than a recurring side path.

One of the more useful people to follow for the systems side of modern model building, especially where better kernels and sequence methods translate directly into frontier-model training and inference speed.

Worth following because he brings a real theory background into the model-systems layer, especially where structured linear algebra and sequence methods end up mattering for practical modern architectures.

Important because he sits at a productive seam between machine learning, data systems, and model infrastructure, with work that ranges from weak supervision to some of the most important efficiency breakthroughs in modern training stacks.

A high-signal figure for understanding how DeepMind turned ambitious research systems into durable products, especially across reinforcement learning, speech, and code generation.

Foundational less for any single public paper than for shaping the infrastructure, engineering culture, and systems thinking that make frontier-model research possible.