PeopleofSuperintelligence

Researcher

Che Zheng

Che Zheng is a researcher whose work spans neural text generation, transformer design, syntax, and multimodal AI. He completed an MS in computational data science at Carnegie Mellon in 2015, with Eric Nyberg as his advisor, and went to Google after graduation. His papers explored how to identify generated text, rethink transformer attention, and infer grammatical structure. In 2024, he coauthored Reka's report on the Core, Flash, and Edge models and Vibe-Eval, a benchmark for multimodal chat.

Now at
Google
Works
13
Citations
351
h-index
7

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2015

    A Carnegie Mellon master's

    In 2015, he completes an MS in Computational Data Science at Carnegie Mellon under advisor Eric Nyberg, then moves to Google after graduation.

  2. 2020

    Testing generated-text fingerprints

    In 2020, he and his coauthors introduce a task for identifying which variant of a neural text-generation model produced a passage.

  3. 2021

    Rethinking transformer attention

    In 2021, he coauthors Synthesizer, which tests how much dot-product self-attention contributes to Transformer performance.

  4. 2021

    Inducing syntactic structure

    In 2021, he coauthors StructFormer, a model that induces dependency and constituency structure at the same time.

  5. By 2024

    Coauthoring Reka reports

    In 2024, he coauthors Reka's technical report introducing the Core, Flash, and Edge multimodal models. He also coauthors Vibe-Eval, a multimodal chat benchmark with 269 visual-understanding prompts.

Career

  1. 2015–2023Google

Education

  1. 2015Carnegie Mellon UniversityMS in Computational Data Science

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Synthesizer: Rethinking Self-Attention in Transformer Models2020 · 247 citations
  2. StructFormer: Joint Unsupervised Induction of Dependency and Constituency Structure from Masked Language Modeling2021 · 24 citationsWith Aaron C. Courville
  3. Choppy: Cut Transformer for Ranked List Truncation2020 · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · 21 citations
  4. Reverse Engineering Configurations of Neural Text Generation Models2020 · Annual Meeting of the Association for Computational Linguistics (ACL) · 19 citations
  5. Synthesizer: Rethinking Self-Attention for Transformer Models2021 · International Conference on Machine Learning · 17 citations

From OpenAlex, as of 19 October 2026.

  1. Google has the highest combined h-index among AI organizationsIncludes Google. Our estimates of combined and median h-index at AI organizations, using current affiliations and OpenAlex records.
  2. Google alumni founded or led 81 other organizations, the most among the labs we compareIncludes Google. We trace 194 organizations founded or led by AI lab alumni, and 202 career moves between the labs.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  4. Of 435 people shaping AI who worked at Google, 178 work there nowIncludes Google. Our estimates of where former Google and Google DeepMind people work now, including moves within Google.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

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

  1. Che Zheng - Language Technologies Institute - School of Computer Science - Carnegie Mellon University lti.cmu.edu. Checked 18 October 2026.
  2. Reverse Engineering Configurations of Neural Text Generation Models - ACL Anthology aclanthology.org. Checked 18 October 2026.
  3. Synthesizer: Rethinking Self-Attention for Transformer Models - Proceedings of Machine Learning Research proceedings.mlr.press. Checked 18 October 2026.
  4. StructFormer: Joint Unsupervised Induction of Dependency and Constituency Structure from Masked Language Modeling - ACL Anthology aclanthology.org. Checked 18 October 2026.
  5. [2404.12387] Reka Core, Flash, and Edge: A Series of Powerful Multimodal Language Models arxiv.org. Checked 18 October 2026.
  6. [2405.02287] Vibe-Eval: A hard evaluation suite for measuring progress of multimodal language models arxiv.org. Checked 18 October 2026.