PeopleofSuperintelligence

Researcher

Xuezhi Wang

Xuezhi Wang is a research scientist at Google DeepMind. Her primary interests are reasoning and systematic generalization in language models. She completed a computer science PhD at Carnegie Mellon University in August 2016, with Jeff Schneider advising the thesis Active Transfer Learning. In January 2021 she and Ankur Parikh presented ToTTo at Google Research, and in 2022 she worked on chain-of-thought prompting and on PaLM, a 540-billion parameter language model. Self-consistency, a method that samples diverse reasoning paths and keeps the most consistent answer, appeared at ICLR in 2023, and in 2024 she and Denny Zhou proposed CoT-decoding.

Works
3

Publication numbers from OpenAlex, as of 17 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. August 2016

    A PhD on transfer learning

    She completes a PhD in computer science at Carnegie Mellon University in August 2016. Jeff Schneider advises the thesis Active Transfer Learning, which focuses on active transfer learning under the model shift assumption.

  2. October 2020

    Gendered correlations in BERT

    In October 2020 she and Kellie Webster study gendered correlations in the pre-trained models BERT and ALBERT.

  3. January 2021

    ToTTo at Google Research

    In January 2021 she and Ankur Parikh, research scientists at Google Research, present ToTTo, an open domain table-to-text dataset for assessing model hallucination.

  4. January 2022

    Chain-of-thought prompting

    In 2022 she, Jason Wei, and Denny Zhou describe chain-of-thought prompting, where a series of intermediate reasoning steps helps large language models perform complex reasoning.

  5. April 2022

    The Pathways language model

    In 2022 she contributes to PaLM, a 540-billion parameter language model. Training scales to 6144 chips with the Pathways system.

  6. 2023

    Self-consistency at ICLR

    In 2023 she and Jason Wei present self-consistency at ICLR. The method samples a diverse set of reasoning paths and selects the most consistent answer, with a 17.9% gain on GSM8K.

  7. 2024

    Reasoning without a prompt

    In 2024 she and Denny Zhou propose CoT-decoding. Studies on reasoning benchmarks show that it elicits reasoning capabilities from language models that standard greedy decoding had obscured.

  8. As of October 2026

    Research scientist at DeepMind

    She is a research scientist at Google DeepMind. Her primary interests are reasoning and systematic generalization in language models.

Sources and credits

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

  1. Xuezhi Wang | Carnegie Mellon University Computer Science Department csd.cmu.edu. Checked 16 October 2026.
  2. Measuring Gendered Correlations in Pre-trained NLP Models research.google. Checked 16 October 2026.
  3. ToTTo: A Controlled Table-to-Text Generation Dataset research.google. Checked 16 October 2026.
  4. Language Models Perform Reasoning via Chain of Thought research.google. Checked 16 October 2026.
  5. [2201.11903] Chain-of-Thought Prompting Elicits Reasoning in Large Language Models arxiv.org. Checked 16 October 2026.
  6. Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance research.google. Checked 16 October 2026.
  7. PaLM: Scaling Language Modeling with Pathways research.google. Checked 16 October 2026.
  8. [2203.11171] Self-Consistency Improves Chain of Thought Reasoning in Language Models arxiv.org. Checked 16 October 2026.
  9. Self-Consistency Improves Chain of Thought Reasoning in Language Models research.google. Checked 16 October 2026.
  10. [2402.10200] Chain-of-Thought Reasoning Without Prompting arxiv.org. Checked 16 October 2026.
  11. NeurIPS Poster Chain-of-Thought Reasoning Without Prompting neurips.cc. Checked 16 October 2026.
  12. Xuezhi Wang research.google. Checked 16 October 2026.