PeopleofSuperintelligence

Researcher

Joshua V Dillon

Also known as Josh Dillon, Joshua Dillon

Joshua V Dillon is a machine-learning researcher whose work spans probabilistic modeling, model uncertainty, and generative AI. He earned a Ph.D. at Georgia Tech in 2011 and received the DHS Fellowship in Data Analysis and Visual Analytics for 2010–2012; his early work studied stochastic composite likelihood. From 2013 to 2019, he designed and wrote ContentAds’s auction mechanism; in 2017, he created TensorFlow Probability and coauthored Deep Variational Information Bottleneck. His 2019 paper studied predictive uncertainty under dataset shift; he contributed to Gemini and co-created the Veo prototype in 2024, and coauthored VideoPoet in 2023, whose paper won ICML’s 2024 Best Paper award. His January 2026 solo preprint Speed is Confidence reports higher Sudoku-Extreme accuracy with fewer reasoning steps; he led foundational-model pre-training at Luma AI and, as of October 2026, is a co-founder of rekursiv.ai, which builds AI systems for machine-learning research.

Now at
rekursiv.ai
Badges

Timeline

  1. 2009–2010

    Stochastic composite likelihood

    As a machine-learning researcher, he and Guy Lebanon publish Statistical and Computational Tradeoffs in Stochastic Composite Likelihood at AISTATS in 2009. A 2010 arXiv report proposes estimators that trade computation cost against statistical accuracy.

  2. 2010–2012

    A Ph.D. and DHS fellowship

    At Georgia Tech, he completes a Ph.D. under Guy Lebanon in 2011. He receives the Department of Homeland Security Fellowship in Data Analysis and Visual Analytics for 2010–2012.

  3. 2013–2019

    ContentAds and TensorFlow Probability

    From 2013 to 2019, he designs and writes the auction mechanism used by ContentAds. In 2017, he creates TensorFlow Probability and coauthors TensorFlow Distributions; with Alex Alemi, he also publishes Deep Variational Information Bottleneck.

  4. 2019

    Predictive uncertainty under shift

    In 2019, he coauthors Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift. The study benchmarks uncertainty methods on classification problems and finds that some approaches averaging across models perform strongly under dataset shifts.

  5. 2023–2024

    Gemini, Veo and VideoPoet

    In 2023, he contributes to VideoPoet, a multimodal language model for zero-shot video generation; its paper wins ICML’s 2024 Best Paper award. In 2024, he contributes to Gemini and co-creates the prototype that becomes Veo.

  6. January 2026

    Speed is Confidence

    In January 2026, he posts the solo preprint Speed is Confidence. On Sudoku-Extreme, its halt-first selection reaches 97% accuracy against 91% for probability averaging, with ten times fewer reasoning steps.

  7. By 2026

    Pre-training at Luma AI

    His career moves from Staff Research Scientist roles at Google Research and Google DeepMind to foundational-model pre-training at Luma AI.

  8. As of October 2026

    Co-founding rekursiv.ai

    As of October 2026, he is a co-founder of rekursiv.ai, where AI scientists form hypotheses, run experiments, review each other's work, and track results on leaderboards.

Career

  1. Nowrekursiv.aiCo-founder
  2. 2013–2019ContentAds
  3. Google ResearchStaff Research Scientist
  4. Google DeepMindStaff Research Scientist

Education

  1. 2011Georgia Institute of TechnologyPh.D.

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 Joshua V Dillon. 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. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Sources and credits

Photo: We have no published photo of Joshua V Dillon yet. Send us a free-licensed photo.

  1. Joshua V. Dillon, rekursiv.ai rekursiv.ai. Checked 21 October 2026.
  2. Statistical and Computational Tradeoffs in Stochastic Composite Likelihood arxiv.org. Checked 21 October 2026.
  3. TensorFlow Distributions arxiv.org. Checked 21 October 2026.
  4. Deep Variational Information Bottleneck arxiv.org. Checked 21 October 2026.
  5. Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift arxiv.org. Checked 21 October 2026.
  6. VideoPoet: A Large Language Model for Zero-Shot Video Generation arxiv.org. Checked 21 October 2026.
  7. Award-winning paper in ICML 2024 on “VideoPoet: A large language model for zero-shot video generation.” | Irfan Essa irfanessa.gatech.edu. Checked 21 October 2026.
  8. Speed is Confidence arxiv.org. Checked 21 October 2026.
  9. rekursiv.ai: Scale AI scientists whose own breakthroughs accelerate the next. ycombinator.com. Checked 21 October 2026.
  10. rekursiv.ai, Autonomously Create Knowledge rekursiv.ai. Checked 21 October 2026.