PeopleofSuperintelligence

Researcher

Huazuo Gao

Also known as 高华佐, Gao Huazuo

Huazuo Gao is a DeepSeek researcher who works on the attention and mixture-of-experts designs inside the company's open language models. He earned a physics degree from Peking University in 2017 and worked at Megvii Technology for five years, then joined DeepSeek. DeepSeek credited him and Wangding Zeng with key innovations in the Multi-head Latent Attention architecture of DeepSeek-V2 in 2024, and he co-authored a 2024 paper on auxiliary-loss-free load balancing, a strategy DeepSeek-V3 also uses. Native Sparse Attention, a paper he co-wrote as second author, won a Best Paper award at ACL 2025. At DeepSeek, he co-authored DeepSeek-V4 and DeepSeek-V4.1-Flash in 2026.

Timeline

  1. 2017

    Peking University physics and Megvii

    He graduates from Peking University in 2017 with a physics degree. Before DeepSeek, he works for five years at Megvii Technology, an AI company that the U.S. Department of Commerce sanctions in 2019.

  2. January 2024

    DeepSeek LLM and DeepSeekMoE

    In January 2024, he co-authors DeepSeek LLM, a project on open-source language models at 7B and 67B scale, and DeepSeekMoE, a mixture-of-experts architecture designed for expert specialization.

  3. May 2024

    Multi-head Latent Attention in DeepSeek-V2

    DeepSeek-V2, released in May 2024, is a 236-billion-parameter mixture-of-experts model whose Multi-head Latent Attention (MLA) compresses the key-value cache into a latent vector and reduces it by 93.3%. DeepSeek credits him and Wangding Zeng with key innovations in the research of the MLA architecture.

  4. August–December 2024

    Loss-free balancing and DeepSeek-V3

    In August 2024, he co-authors a paper with Lean Wang, Chenggang Zhao, Xu Sun and Damai Dai on Loss-Free Balancing, which evens out expert load in mixture-of-experts models by adding an expert-wise bias to routing scores, with no auxiliary loss. He is also among the authors of DeepSeek-V3, a 671-billion-parameter model released in December 2024 that uses an auxiliary-loss-free strategy for load balancing.

  5. 2025

    An ACL Best Paper for NSA

    He is second author of Native Sparse Attention (NSA), a February 2025 paper with Jingyang Yuan, Wenfeng Liang, Wangding Zeng and others on a natively trainable sparse attention mechanism for efficient long-context modeling. The paper wins a Best Paper award at ACL 2025 in Vienna.

  6. September 2025

    DeepSeek-R1 in Nature

    In September 2025, Nature publishes "DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning", a paper he co-authors showing that pure reinforcement learning can develop reasoning in large language models with no human-labelled reasoning trajectories.

  7. December 2025–April 2026

    mHC and DeepSeek-V4

    In December 2025, he co-authors mHC (Manifold-Constrained Hyper-Connections), which projects the residual connection space of hyper-connections onto a manifold to restore the identity mapping property. He also co-authors DeepSeek-V4, previewed in April 2026, whose 1.6-trillion-parameter Pro and 284-billion-parameter Flash models support one-million-token contexts and use mHC.

  8. As of October 2026

    Researcher at DeepSeek

    He works at DeepSeek. In September 2026, he is among the DeepSeek-AI authors of DeepSeek-V4.1-Flash, a multimodal mixture-of-experts model that shrinks its global KV cache to 890 bytes per token, about a quarter of the footprint of DeepSeek-V4-Flash.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Huazuo Gao yet. Send us a free-licensed photo.

  1. China's AI disrupter DeepSeek bets on low-key team of 'young geniuses' to beat US giants tech.yahoo.com. Checked 19 October 2026.
  2. DeepSeek: The Secretive AI Company That's Turning Heads in Silicon Valley chinaaifrontier.substack.com. Checked 19 October 2026.
  3. [2401.02954] DeepSeek LLM: Scaling Open-Source Language Models with Longtermism arxiv.org. Checked 19 October 2026.
  4. [2401.06066] DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models arxiv.org. Checked 19 October 2026.
  5. [2405.04434] DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 19 October 2026.
  6. [2408.15664] Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts arxiv.org. Checked 19 October 2026.
  7. [2412.19437] DeepSeek-V3 Technical Report arxiv.org. Checked 19 October 2026.
  8. [2502.11089] Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention arxiv.org. Checked 19 October 2026.
  9. Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention - ACL Anthology aclanthology.org. Checked 19 October 2026.
  10. Awards - ACL 2025 2025.aclweb.org. Checked 19 October 2026.
  11. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 19 October 2026.
  12. [2512.24880] mHC: Manifold-Constrained Hyper-Connections arxiv.org. Checked 19 October 2026.
  13. [2606.19348] DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence arxiv.org. Checked 19 October 2026.
  14. [2609.19969] DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression arxiv.org. Checked 19 October 2026.