Jesse Dodge

Timeline

1. 2011–2012

   The Midge image describer

   In 2011, he takes part in the JSALT workshop, on its vision and language team. In 2012, he is among the coauthors of Midge: Generating Image Descriptions From Computer Vision Detections, presented at EACL 2012 in Avignon with Margaret Mitchell, Tamara Berg and Hal Daumé III, among others.

2. 2015

   Best student paper at NAACL

   In 2015, he coauthors Retrofitting Word Vectors to Semantic Lexicons with Manaal Faruqui, Sujay Kumar Jauhar, Chris Dyer, Eduard Hovy and Noah A. Smith of Carnegie Mellon University, and it wins a best student paper award at NAACL HLT 2015. He also spends six months of 2015 as a research intern at Facebook AI Research in New York with Jason Weston and Antoine Bordes, where he builds the Movie Dialog dataset and the MovieQA dataset.

3. 2020

   A PhD at Carnegie Mellon

   He earns his PhD in 2020 from the Language Technologies Institute in the School of Computer Science at Carnegie Mellon University, with Noah A. Smith as his advisor. He spends much of the PhD as a visiting student at UW CSE in Seattle.

4. 2020–2021

   The NLP Reproducibility Checklist

   Working on the AllenNLP team at the Allen Institute for AI since 2019, he creates the NLP Reproducibility Checklist. All submissions to EMNLP 2020, NAACL 2021, ACL 2021 and EMNLP 2021 fill it out, more than 10,000 in total.

5. 2022

   A Test-of-Time Paper Award

   ACL gives its 10-year Test-of-Time Paper Award for 2022 to Midge: Generating Image Descriptions From Computer Vision Detections, the 2012 EACL paper he coauthored with Margaret Mitchell, Tamara Berg, Hal Daumé III and others.

6. 2022

   Measuring AI's carbon intensity

   He is first author of Measuring the Carbon Intensity of AI in Cloud Instances, posted to arXiv in June 2022 with Emma Strubell, Alexandra Sasha Luccioni, Noah A. Smith and others. The paper proposes a framework for measuring software carbon intensity, reports measurements that include the pretraining of a 6.1 billion parameter language model, and tests ways to cut emissions on Microsoft Azure, such as using different regions or times of day.

7. 2024–2025

   Awards for Dolma, OLMo, OLMoTrace

   In 2024, two papers he coauthors win at ACL: Dolma, an open corpus of three trillion tokens for language model pretraining research, takes Best Resource Paper, and OLMo takes Best Theme Paper. In 2025, OLMoTrace, a Playground feature that traces model outputs back to their training data, wins Best Demo Paper at ACL; he helps build it and is research lead of Ai2's Playground. He also leads the evaluation team for OLMo.

8. Since 2025

   Joining Meta Superintelligence Labs

   In 2025, he moves from the Allen Institute for AI to Meta Superintelligence Labs as a research scientist. His work there covers building large language models, evaluating them throughout training, and studying pretraining data, the environmental impact of AI, transparency and reproducibility.