Matt Deitke
Matt Deitke is an AI researcher whose work spans embodied AI, 3D data, and vision-language models. He grew up in a Chicago suburb, built computer-graphics projects in high school, then studied at the University of Washington and worked at AI2. His work included ProcTHOR’s simulated environments, the annotated 3D-object dataset Objaverse, and the open vision-language model Molmo with its PixMo data. He co-founded Vercept in November 2024 and joined Meta’s Superintelligence Lab by August 2025; Anthropic acquired Vercept in 2026. As of October 2026, he is an AI researcher at Meta and leads the data pod in TBD labs.
Timeline
-
2022
ProcTHOR wins at NeurIPS
He co-authors ProcTHOR, a paper on procedural generation for embodied AI. The paper wins the NeurIPS 2022 Outstanding Paper Award.
-
2022
Objaverse expands 3D data
In 2022, he co-authors Objaverse, a large dataset of annotated 3D objects.
-
By 2023
UW studies and AI2 research
He grows up in a Chicago suburb and completes computer-graphics projects in high school. By 2023, he studies at UW while working as a research scientist at AI2.
-
2024–2025
Molmo and PixMo
In 2024, he co-authors Molmo and PixMo, open vision-language models and data. Molmo receives a CVPR 2025 Best Paper Honorable Mention.
-
November 2024
Co-founding Vercept
In November 2024, he co-founds Vercept to develop autonomous agents for open-ended tasks in real-world environments.
-
2025
Working at Meta’s Superintelligence Lab
As of August 2025, he works at Meta’s Superintelligence Lab.
-
2026
Vercept joins Anthropic
In 2026, Anthropic acquires Vercept. Deitke is one of its initial, full-time founders.
-
As of October 2026
Leading Meta’s data pod
As of October 2026, he is an AI researcher at Meta and leads the data pod in TBD labs.
Sources and credits
Photo: mattdeitke.com; Matt Deitke; creator not stated. License: Not stated by source. Source: mattdeitke.com. Cropped to a square and resized. Ask us to remove or replace this photo.
- NeurIPS 2022 Awards
- Objaverse: A Universe of Annotated 3D Objects
- Matt Deitke
- Inspired by the “pure magic” of AI, Allen School undergraduate Matt Deitke begins his next act
- Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models
- Best Papers at CVPR Reveal New Results with Neural Networks for Real-Time Applications and Novel Ways to Manipulate Light for Scene Recovery
- Matt Deitke
- Vercept Joins Anthropic: A Winning Combination of Human + AI Collaborative Reasoning