PeopleofSuperintelligence

Researcher

Jacob Morrison

Jacob Morrison is a language-model researcher and Ph.D. student at the University of Washington whose work spans open models, post-training and the environmental cost of AI. After software engineering at Tableau and Google, they became a predoctoral researcher on Ai2's Open Ecosystem team and served on Seattle's Generative AI Policy Advisory Team, whose policy the city released in November 2023. They coauthored OLMo, whose paper won an ACL 2024 Theme Paper Award, were second author of the Tülu 3 post-training recipe in November 2024, and coauthored Olmo 3 in December 2025. Their first-author studies estimated the carbon and water cost of building language models in 2025 and found in 2026 that reasoning models take 17 times more datacenter energy to post-train than instruction-tuned ones. Advised by Noah Smith, they are also a visiting researcher at Meta FAIR and an AI Fellow at Madrona Venture Group.

Timeline

  1. By 2022

    Tableau, Google and NAACL 2022

    Their early career includes work as a software engineer at Tableau and Google. In 2022, they coauthor two NAACL papers with Jungo Kasai, Yejin Choi, Noah A. Smith and others, one on bidimensional leaderboards for generating and evaluating language and one on transparent human evaluation for image captioning.

  2. 2023

    Public policy lead at Ai2

    As a predoctoral researcher on Ai2's Open Ecosystem team, formerly AllenNLP, they work mainly on language modeling and also serve as a public policy lead at the institute. They sit on the City of Seattle's Generative AI Policy Advisory Team, whose policy defining responsible generative AI use by city employees the city releases in November 2023.

  3. 2024

    OLMo and two ACL awards

    In February 2024, they coauthor OLMo, a fully open language model built to enable the scientific study of language models and released with its training data and its training and evaluation code. They also coauthor Dolma, an open corpus of three trillion tokens for language-model pretraining research. At ACL 2024, the OLMo paper wins a Theme Paper Award and the Dolma paper a Best Resource Paper Award.

  4. November 2024

    Second author of Tülu 3

    In November 2024, they are second author, after Nathan Lambert, of Tülu 3, a family of fully open post-trained models built on Llama 3.1 base models and released with its data, code and training recipes. Its training methods include supervised finetuning, Direct Preference Optimization and a new method called Reinforcement Learning with Verifiable Rewards (RLVR).

  5. March 2025

    Carbon and water accounting

    In March 2025, they are first author, with Clara Na, Jared Fernandez, Tim Dettmers, Emma Strubell and Jesse Dodge, of a study on the environmental impact of creating language models. Counting hardware manufacturing, model development and final training runs, they estimate that a series of models emits 493 metric tons of carbon and consumes 2.769 million liters of water. The paper is an ICLR 2025 spotlight.

  6. December 2025

    Coauthoring Olmo 3

    In December 2025, they coauthor Olmo 3, a family of fully open language models at the 7B and 32B parameter scales. The release includes every stage, checkpoint, data point and dependency used to build the models, and its flagship is the thinking model Olmo 3 Think 32B.

  7. 2026

    BAR and the cost of thinking

    In April 2026, they are first author of BAR (Branch-Adapt-Route), an Ai2 recipe for modular post-training that trains separate domain experts and composes them into one mixture-of-experts model. In May 2026, with Noah A. Smith and Emma Strubell, they estimate that Olmo 3 development consumes about 12.3 GWh of datacenter energy and that reasoning models are 17 times more expensive to post-train than instruction-tuned models. They receive the UW Hacherl Endowed Fellowship in CSE in 2026.

  8. As of October 2026

    Ph.D. student at UW

    They are a Ph.D. student in computer science and engineering at the University of Washington, advised by Noah Smith. Their research covers the full language-model pipeline, from model architectures to pretraining and post-training, and they are also a visiting researcher at Meta FAIR and an AI Fellow at Madrona Venture Group.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

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 Jacob Morrison yet. Send us a free-licensed photo.

  1. Jacob Morrison jacobmorrison.com. Checked 19 October 2026.
  2. Publications | Jacob Morrison jacobmorrison.com. Checked 19 October 2026.
  3. City of Seattle Releases Generative Artificial Intelligence Policy Defining Responsible Use for City Employees - Office of the Mayor harrell.seattle.gov. Checked 19 October 2026.
  4. [2402.00838] OLMo: Accelerating the Science of Language Models arxiv.org. Checked 19 October 2026.
  5. [2411.15124] Tulu 3: Pushing Frontiers in Open Language Model Post-Training arxiv.org. Checked 19 October 2026.
  6. [2503.05804] Holistically Evaluating the Environmental Impact of Creating Language Models arxiv.org. Checked 19 October 2026.
  7. [2512.13961] Olmo 3 arxiv.org. Checked 19 October 2026.
  8. [2604.18473] Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts arxiv.org. Checked 19 October 2026.
  9. Train separately, merge together: Modular post-training with mixture-of-experts | Ai2 allenai.org. Checked 19 October 2026.
  10. [2605.01158] The Hidden Cost of Thinking: Energy Use and Environmental Impact of LMs Beyond Pretraining arxiv.org. Checked 19 October 2026.