PeopleofSuperintelligence

Builder

Tri Dao

Tri Dao works on the systems that help AI models use memory and compute efficiently. His papers include the 2022 structured-matrix work Monarch and FlashAttention, followed by Mamba and a framework that connects state-space models with attention. He completed a Stanford computer science PhD in 2023, became Together AI Chief Scientist in July, and earned COLM and MLSys paper honors and a 2025 AI2050 Early Career Fellowship. As of October 2026, Dao is Together AI’s co-founder and Chief Scientist and a visiting research scholar at Princeton; the research program he leads includes FlashAttention-4.

Now at
Together AI and Princeton University
Works
70

Publication numbers from OpenAlex, as of 14 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Photo of Tri Dao

Timeline

  1. 2022

    Structured matrices for training

    Tri Dao and his coauthors received the ICML 2022 Outstanding Paper Runner Up for Monarch, which represents common transforms with products of two block-diagonal matrices.

  2. 2022

    Tiling attention for speed

    With four coauthors, he introduced FlashAttention, an exact attention algorithm that uses tiling to reduce data movement between GPU high-bandwidth memory and on-chip SRAM.

  3. 2023

    A Stanford doctorate

    He completed a Stanford computer science PhD in 2023, co-advised by Christopher Ré and Stefano Ermon.

  4. July 2023

    Joining Together AI

    Together AI announced him as Chief Scientist in July 2023; he joined to make open-source AI more accessible and cost-competitive.

  5. 2023–2024

    A model for long sequences

    With Albert Gu, he co-authored Mamba, a selective state-space model designed for linear-time sequence modeling.

  6. 2024–2025

    Connecting state spaces and attention

    The 2024 paper Transformers are SSMs, co-authored with Albert Gu, connects state-space models with variants of attention. His work received COLM’s 2024 Outstanding Paper award and an MLSys 2025 Outstanding Paper Honorable Mention.

  7. 2025

    An AI2050 fellowship

    Schmidt Sciences named him a 2025 AI2050 Early Career Fellow for a project that uses compilers and simulators to help AI learn through experimentation.

  8. As of October 2026

    Research at Together AI

    He is a visiting research scholar at Princeton and a co-founder and Chief Scientist of Together AI. The research program he leads introduced FlashAttention-4, a new algorithm and kernel co-design tuned for NVIDIA Blackwell GPUs.

Sources and credits

Photo: tridao.me; Tri Dao; creator not stated. License: Not stated by source. Source: tridao.me. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. ICML 2022 Awards icml.cc. Checked 14 October 2026.
  2. [2205.14135] FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness arxiv.org. Checked 14 October 2026.
  3. Introducing Together AI Chief Scientist Tri Dao, as he releases FlashAttention-2 to speed up model training and inference together.ai. Checked 14 October 2026.
  4. Tri Dao cs.princeton.edu. Checked 14 October 2026.
  5. [2312.00752] Mamba: Linear-Time Sequence Modeling with Selective State Spaces arxiv.org. Checked 14 October 2026.
  6. [2405.21060] Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality arxiv.org. Checked 14 October 2026.
  7. Tri Dao - AI2050 ai2050.schmidtsciences.org. Checked 14 October 2026.
  8. Inside the Together AI kernels team together.ai. Checked 14 October 2026.
  9. Key research and product announcements at the AI Native Conf together.ai. Checked 14 October 2026.