PeopleofSuperintelligence

Researcher

Ian Magnusson

Ian Magnusson is a University of Washington PhD student whose research on the science of language modeling centers on evaluation, scaling behavior and robustness across text domains. He holds a BA in cultural anthropology from Bard College and an MS in computer science from Northeastern University, and he worked at Ai2 as a predoctoral young investigator (PYI). As first author, he introduced Paloma, a 2023 benchmark of language-model fit across 546 English and code domains, and DataDecide, a 2025 suite of more than 30,000 checkpoints for testing how small experiments predict the best pretraining data. Along the way, he coauthored OLMo, whose paper won a Best Theme Paper Award at ACL 2024, and the fully open Olmo 3 model family in 2025. Working with Pang Wei Koh and Noah Smith, he released COUNT in 2026, a method that builds n-gram embeddings from counts stored in host memory.

Timeline

  1. 2021

    Knowledge graphs of scientific claims

    With a BA in cultural anthropology from Bard College, he earns an MS in computer science at Northeastern University and interns at AWS AI Labs and SIFT. In 2021, he and Scott Friedman publish an EMNLP 2021 paper that introduces SciClaim, a dataset of scientific claims drawn from Social and Behavior Science, PubMed and CORD-19 papers.

  2. 2023

    Auditing the reproducibility checklist

    He works at Ai2 as a predoctoral young investigator (PYI). In 2023, he publishes "Reproducibility in NLP: What Have We Learned from the Checklist?" with Noah A. Smith and Jesse Dodge in Findings of ACL 2023, the first analysis of the NLP Reproducibility Checklist that *CL conferences created in 2020.

  3. December 2023

    The Paloma benchmark

    In December 2023, he and 15 coauthors release Paloma, a benchmark that measures language-model fit to 546 English and code domains, along with six baseline 1B models built for fair comparisons of pretraining corpora. The paper appears at NeurIPS 2024.

  4. 2024

    OLMo and Dolma awards

    In February 2024, he coauthors OLMo, an open language model released with its training data and its training and evaluation code. At ACL 2024, the OLMo paper wins a Best Theme Paper Award, and Dolma, a three-trillion-token pretraining corpus he also coauthors, wins a Best Resource Paper Award.

  5. April 2025

    Releasing DataDecide

    In April 2025, Ai2 releases DataDecide, with him as first author: a suite of models pretrained on 25 corpora at 14 sizes from 4M to 1B parameters, more than 30,000 checkpoints in all. Ranking models of a single small size, such as 150M parameters, picks the best data for the 1B scale with about 80% accuracy, and the paper appears at ICML 2025. In May 2025, he tells The New Stack that "you can make predictions about what will be the best choice using surprisingly little compute."

  6. December 2025

    Coauthoring Olmo 3

    In December 2025, he is among the coauthors of Olmo 3, a family of fully open language models at 7B and 32B parameters. The release includes every stage, checkpoint, data point and dependency used to build the models.

  7. As of October 2026

    PhD student at Washington

    He is a PhD student at the University of Washington, working with Pang Wei Koh and Noah Smith on the science of language modeling, with a focus on evaluation, scaling behavior and robustness across text domains. In October 2026, the three release COUNT, which builds n-gram embeddings from precomputed n-gram counts stored in host memory, without adding learned parameters.

  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 Ian Magnusson yet. Send us a free-licensed photo.

  1. Ian Magnusson ianmagnusson.github.io. Checked 19 October 2026.
  2. Ian Magnusson - ACL Anthology aclanthology.org. Checked 19 October 2026.
  3. [2312.10523] Paloma: A Benchmark for Evaluating Language Model Fit arxiv.org. Checked 19 October 2026.
  4. [2402.00838] OLMo: Accelerating the Science of Language Models arxiv.org. Checked 19 October 2026.
  5. Best Paper Awards - ACL 2024 2024.aclweb.org. Checked 19 October 2026.
  6. DataDecide: How to predict best pretraining data with small experiments | Ai2 allenai.org. Checked 19 October 2026.
  7. [2504.11393] DataDecide: How to Predict Best Pretraining Data with Small Experiments arxiv.org. Checked 19 October 2026.
  8. New Tools Help LLM Developers Choose Better Pre-Training Data - The New Stack thenewstack.io. Checked 19 October 2026.
  9. [2512.13961] Olmo 3 arxiv.org. Checked 19 October 2026.
  10. Pang Wei Koh koh.pw. Checked 19 October 2026.
  11. Researchers in Noah’s ARK - Noah’s ARK noahs-ark.github.io. Checked 19 October 2026.
  12. [2610.20456] The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation arxiv.org. Checked 19 October 2026.