Matt Deitke

Matt Deitke
Born	2001 Chicago, Illinois, U.S.
Nationality	American
Alma mater	University of Washington
Occupation(s)	AI Researcher, Computer Scientist
Employer	Meta Superintelligence Lab
Known for	Computer vision, Embodied AI, Multimodal AI systems
Notable work	ProcTHOR, Objaverse, Molmo
Awards	NeurIPS 2022 Outstanding Paper Award

Matt Deitke (born c. 2001) is an American artificial intelligence (AI) researcher and computer scientist recognized for his contributions to computer vision, embodied AI, and multimodal AI systems. He has developed large-scale datasets and models for 3D vision and interactive AI environments, advancing research in robotics and generative modeling. In 2025, Deitke gained attention when Meta offered him a $250 million compensation package to join its Superintelligence Lab, reflecting the high demand for top AI talent.

Early life and education

Matt Deitke was raised in a Chicago suburb, where he developed an interest in computer graphics, interface design, and visualization. During high school, he collaborated on AI-driven image processing projects with Ohio State University and the University of Cincinnati. His fascination with AI image colorization, which he described as "pure magic," inspired him to pursue computer science at the University of Washington (UW).

At UW's Allen School of Computer Science & Engineering, Deitke excelled as an undergraduate. As a freshman, he enrolled in a graduate-level computer vision course taught by professors Richard Szeliski, Steve Seitz, and Harpreet Sawhney, contributing insights on transformers and text-to-image generation. He graduated in 2023 and briefly pursued a PhD at UW before leaving to focus on AI research and entrepreneurial projects.

Career

Allen Institute for AI

During his sophomore year at UW, Deitke joined the Allen Institute for Artificial Intelligence (AI2) as a researcher on the Perceptual Reasoning and Interaction Research (PRIOR) team, under Professor Ali Farhadi and later Ani Kembhavi. His work at AI2 included:

• ProcTHOR: A platform for generating interactive 3D environments to train robotic agents, improving scalability and performance in tasks like object navigation.
• Objaverse and Objaverse-XL: Datasets containing over 10 million 3D objects, enabling advancements in 3D vision and models like Zero123 for novel view synthesis.
• Molmo: A series of open vision-language models supported by PixMo datasets, featuring detailed image captions and 2D pointing data, achieving top performance among open-source multimodal models.

Deitke's contributions earned him an Outstanding Paper Award at NeurIPS 2022 for his work in 3D vision and embodied AI.

Vercept

In 2024, Deitke co-founded Vercept, a Seattle-based AI startup focused on embodied and multimodal AI agents. The company raised $16.5 million in a seed round led by Fifty Years, with investors including Eric Schmidt and Jeff Dean. Deitke played a key role in shaping Vercept’s vision before leaving in 2025.

Meta Superintelligence Lab

In July 2025, Meta recruited Deitke for its Superintelligence Lab, led by CEO Mark Zuckerberg to advance artificial general intelligence (AGI). After initially declining a $125 million four-year offer due to his commitment to Vercept, Deitke accepted a revised $250 million package following a personal meeting with Zuckerberg. He joined a team including researchers like Lucas Beyer and Ruoming Pang, describing Meta’s AGI initiative as “the most exciting bet I’ve seen in tech history.”

Research contributions

Deitke’s research centers on computer vision, deep learning, and embodied AI, with over 3,950 citations, 70% as first author. Notable contributions include:

• CoSyn: A framework for generating synthetic multimodal data, enhancing vision-language model performance in visual question-answering and multi-image reasoning.
• AI2-THOR and RoboTHOR: Platforms for training AI in realistic 3D indoor scenes, supporting visual AI and simulation-to-real transfer.
• Crowdsourced studies: Using platforms like Prolific, Deitke collected verbal and pointing data to train AI in image description and grounding, advancing multimodal AI systems.

Recognition

Deitke’s work on Objaverse and Molmo has set benchmarks for open-source AI datasets and models, broadening access to advanced tools. His 2025 recruitment by Meta highlighted the escalating value of AI expertise, with his compensation package compared to those of top athletes. Posts on X in 2025 described him as a “rising star” in AI, particularly for his impact on embodied and multimodal systems.

Personal life

Deitke maintains a private personal life, focusing publicly on his professional work. He is active on GitHub and Reddit, engaging with the AI research community and sharing project updates.