AI People Wiki

Builders & open source

Matt Deitke

AI researcher, leading a data pod, Meta

  • AI researcher, leading a data pod, Meta1

Matt Deitke is an AI researcher at Meta, where Deitke leads a data pod. Deitke was previously a research scientist at the Allen Institute for AI (Ai2) while studying at the University of Washington's Allen School, and is co-first author of the Molmo and PixMo paper on open vision-language models. Deitke is also first author of ProcTHOR, on procedurally generated environments for embodied AI, and of the Objaverse dataset of annotated 3D objects.1234

At a glance

Current role
AI researcher, leading a data pod
Company
Meta
Sector
Company
Education
UW
Work listed
3
Honors
2
Sources cited
4
Profile reviewed

Career

Deitke worked as a research scientist at Ai2 while completing undergraduate and doctoral studies at the University of Washington's Allen School, where Deitke's PhD work was with Ali Farhadi. Deitke later co-founded Vercept, a company building computer-use agents, and is now an AI researcher at Meta, leading a data pod.1

Research

Deitke's work at Ai2 included ProcTHOR (2022), which used procedural generation to scale the diversity of simulated environments for embodied AI, and Objaverse (2022), a dataset of annotated 3D objects. Deitke's site notes that ProcTHOR received an Outstanding Paper Award at NeurIPS 2022.341

With Christopher Clark, Deitke is co-first author of the Molmo and PixMo paper (2024), which released open vision-language models together with new caption, question-answer and pointing datasets collected without the use of external vision-language models; the paper received a CVPR 2025 Best Paper Honorable Mention.21

Work and contributions

  • Molmo Open weights

    First-listed co-first author, with Christopher Clark, of the Molmo and PixMo paper introducing the Molmo open vision-language models and the PixMo datasets.21

  • ProcTHOR Published research

    First author of the paper on scaling embodied AI with procedurally generated houses.3

  • Objaverse Open data

    First author of the paper introducing Objaverse, a dataset of annotated 3D objects.4

Timeline

  1. Posted ProcTHOR, on large-scale embodied AI using procedural generation, as first author.3
  2. ProcTHOR received an Outstanding Paper Award at NeurIPS 2022.1
  3. Posted Objaverse, a dataset of annotated 3D objects, as first author.4
  4. Posted the Molmo and PixMo paper as co-first author with Christopher Clark.2
  5. Announced the co-founding of Vercept, focused on computer-use agents.1
  6. The Molmo and PixMo paper received a Best Paper Honorable Mention at CVPR 2025.1

Other documented roles

Roles the sources document that are not confirmed as current.

Honors and recognition

  • NeurIPS Outstanding Paper Award (ProcTHOR) · Conference on Neural Information Processing Systems (NeurIPS), 20221
  • CVPR Best Paper Honorable Mention (Molmo and PixMo) · IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 20251

Where they work

  • Meta 106 people in the wiki

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

4 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026

  1. Matt Deitke — homepageMatt Deitke · Official page · accessed Oct 22, 2026
  2. Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models (arXiv 2409.17146)arXiv · Paper · published Sep 25, 2024 · accessed Oct 22, 2026
  3. ProcTHOR: Large-Scale Embodied AI Using Procedural Generation (arXiv 2206.06994)arXiv · Paper · published Jun 14, 2022 · accessed Oct 22, 2026
  4. Objaverse: A Universe of Annotated 3D Objects (arXiv 2212.08051)arXiv · Paper · published Dec 15, 2022 · accessed Oct 22, 2026