PeopleofSuperintelligence

Researcher

John Thickstun

John Thickstun is a computer scientist whose research on machine learning and generative models centers on controlling what the models produce, with applications in music. After studying applied mathematics at Brown University, he earned a PhD at the University of Washington's Allen School, where he introduced the MusicNet dataset in 2016 with his advisers Sham Kakade and Zaid Harchaoui. The MAUVE measure for comparing machine-written and human-written text, which he coauthored, received a NeurIPS Outstanding Paper Award in 2021, and Backpack Language Models received an ACL Outstanding Paper award in 2023. As a Stanford postdoctoral scholar advised by Percy Liang, he developed the Anticipatory Music Transformer, which lets a composer write part of a piece and ask the model to fill in the rest. He joined Cornell in 2024 and is an assistant professor of computer science there.

Timeline

  1. 2016

    MusicNet and a Washington PhD

    He studies applied mathematics at Brown University, advised by Eugene Charniak and Björn Sandstede, then earns a PhD in the Allen School of Computer Science & Engineering at the University of Washington, co-advised by Sham Kakade and Zaid Harchaoui. In 2016, he and those two advisers introduce MusicNet, a dataset of freely licensed classical recordings with over 1 million temporal labels on 34 hours of chamber music.

  2. 2021

    MAUVE and a NeurIPS award

    In 2021, MAUVE, which he coauthors with Krishna Pillutla, Swabha Swayamdipta, Rowan Zellers, Sean Welleck, Yejin Choi and Zaid Harchaoui, receives a NeurIPS Outstanding Paper Award. The measure compares the distribution of a text generation model with the distribution of human-written text using divergence frontiers.

  3. 2022

    The Diffusion-LM language model

    In 2022, he coauthors Diffusion-LM with Xiang Lisa Li, Ishaan Gulrajani, Percy Liang and Tatsunori Hashimoto. The non-autoregressive language model denoises Gaussian vectors into word vectors step by step, which allows gradient-based control over fine-grained properties of the text, and the paper is a NeurIPS 2022 oral presentation.

  4. 2023

    Backpacks and text watermarks

    In 2023, he coauthors Backpack Language Models with John Hewitt, Christopher Manning and Percy Liang, an architecture that gives each word several sense vectors that can be inspected and edited to steer the model; the ACL paper receives an Outstanding Paper award. With Rohith Kuditipudi, Tatsunori Hashimoto and Percy Liang, he also proposes robust watermarks for text from autoregressive language models that leave the distribution over text undistorted.

  5. 2023

    The Anticipatory Music Transformer

    Having learned the cello as a child, he develops the Anticipatory Music Transformer in 2023 with David Hall, Chris Donahue and Percy Liang at Stanford's Center for Research on Foundation Models, while a postdoctoral scholar at Stanford advised by Liang. The model uses a method called anticipation to interleave events and controls, so a composer can write part of a piece and ask the model to fill in the rest; it is trained on the Lakh MIDI dataset. Human evaluators report accompaniments with similar musicality to human-composed music over a 20-second clip.

  6. 2024

    Assistant professor at Cornell

    He joins Cornell's Ann S. Bowers College of Computing and Information Science in fall 2024 as an assistant professor of computer science, based on the Ithaca campus. His work covers machine learning and generative models.

  7. 2025

    AI policy and a screenplay

    In March 2025, he joins six Cornell colleagues in a proposal for the White House's AI Action Plan that emphasizes the need for sustained investment in AI education and research. In April 2025, Stanford HAI publishes his conversation with filmmaker Sophie Barthes about a screenplay on text watermarking that the two began at Stanford's Stories for the Future workshop. In October 2025, he speaks at TEDAI on research and investment in technical control mechanisms for generative AI.

  8. As of October 2026

    Controllable generation at Cornell

    He is an assistant professor of computer science at Cornell, working on methods that control the behavior of generative models, including applications in music technologies. In 2026, he coauthors Esoteric Language Models, an ICML paper that fuses autoregressive and masked diffusion language models.

Career

  1. Since 2024Cornell UniversityAssistant professor of computer science
  2. Stanford UniversityPostdoctoral scholar

Education

  1. 2016University of WashingtonPhD
  2. Brown University

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes John Thickstun. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. 11% of AI doctorate holders from 2010 to 2016 founded or led a company or lab by 10 years after graduationIncludes Cornell University. We estimate how often AI doctorate holders founded or led a company or lab by 10 years after graduation, how waiting times compare between founders and top jobs, and which universities trained the most.
  3. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorateIncludes Cornell University. 84% of people shaping AI with recorded undergraduate and doctoral schools changed institutions for their doctorate. Across 2,116 people with both an undergraduate degree and a doctorate recorded, we trace which undergraduate schools feed which doctoral programs, how often students stay at the same school, and how paths differ by birth-country region.
  4. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Cornell University. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.

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 John Thickstun yet. Send us a free-licensed photo.

  1. John Thickstun | Department of Computer Science | Cornell Bowers cs.cornell.edu. Checked 21 October 2026.
  2. Learning Features of Music from Scratch (arXiv:1611.09827) arxiv.org. Checked 21 October 2026.
  3. NeurIPS 2021 Awards neurips.cc. Checked 21 October 2026.
  4. Announcing the NeurIPS 2021 Award Recipients blog.neurips.cc. Checked 21 October 2026.
  5. John Thickstun johnthickstun.com. Checked 21 October 2026.
  6. Backpack Language Models - ACL Anthology aclanthology.org. Checked 21 October 2026.
  7. Robust Distortion-free Watermarks for Language Models | Stanford CRFM crfm.stanford.edu. Checked 21 October 2026.
  8. A Composer's Helper: Using AI to Create New Harmonies | Stanford HAI hai.stanford.edu. Checked 21 October 2026.
  9. Anticipatory Music Transformer: A Controllable Infilling Model for Music | Stanford CRFM crfm.stanford.edu. Checked 21 October 2026.
  10. Anticipatory Music Transformer (arXiv:2306.08620) arxiv.org. Checked 21 October 2026.
  11. Seven faculty hired during recruiting season | Cornell Bowers bowers.cornell.edu. Checked 21 October 2026.
  12. Securing America's AI Future: Cornell faculty offer recommendations for White House's AI Action Plan ai.cornell.edu. Checked 21 October 2026.
  13. Exploring the Ethics of AI through Narrative | Stanford HAI hai.stanford.edu. Checked 21 October 2026.