David Wadden

Career

David Wadden was a PhD student at the University of Washington advised by Hanna Hajishirzi, then worked as a research scientist on the AllenNLP and Semantic Scholar teams at the Allen Institute for AI. Wadden is now a research scientist at Google DeepMind working on post-training for Gemini, with a stated interest in reinforcement learning approaches for long-horizon agentic tasks.¹

Research

Wadden's earlier research centred on information extraction and scientific text. DyGIE++ (2019) is a multi-task framework for named entity recognition, relation extraction and event extraction that scores text spans in context, and Fact or Fiction (2020) introduced the task of scientific claim verification together with the SciFact dataset of expert-written claims paired with evidence.⁵⁶

MultiVerS (2021) extended claim verification by predicting labels and rationales from full-document context with weak supervision. At Ai2, Wadden then worked on open language models, co-authoring the Tülu 2 adaptation paper and contributing to base-model development for OLMo 2 and to evaluations for the OLMoE mixture-of-experts model.⁷²³⁴

Work and contributions

• OLMo 2 and OLMoE (OLMo family) Open weights

  Credited in the OLMo 2 paper with base-model development, including training and data curation, and in the OLMoE paper with running DCLM evaluations and Weights & Biases reports.³⁴

• Tülu 2 Published research

  Co-author of "Camels in a Changing Climate: Enhancing LM Adaptation with Tulu 2" (2023).²

• SciFact and MultiVerS Published research

  First author of the SciFact-Open and MultiVerS papers on scientific claim verification, as listed on Wadden's homepage.¹

Timeline

1. Sep 2019Posted DyGIE++, a span-based framework for entity, relation and event extraction, as first author.⁵
2. Apr 2020Introduced the scientific claim verification task and the SciFact dataset as first author.⁶
3. Dec 2021Posted MultiVerS, a scientific claim verification model using full-document context.⁷
4. Nov 2023Co-authored the Tülu 2 paper on adapting open language models.²
5. Sep 2024Contributed DCLM evaluations to the OLMoE paper on an open mixture-of-experts language model.⁴
6. Dec 2024Credited with base-model development in the OLMo 2 paper.³

Other documented roles

Roles the sources document that are not confirmed as current.

• Research scientist on the AllenNLP and Semantic Scholar teams, Ai2 (Allen Institute for AI)¹³

Honors and recognition

• Outstanding Study Design Paper, ICWSM 2021 · International AAAI Conference on Web and Social Media, 2021¹
• Best Paper Honorable Mention, AKBC 2022 · Conference on Automated Knowledge Base Construction, 2022¹

Where they work

• Google DeepMind 64 people in the wiki

Related people

• Shixiang Shane GuAlso at Google DeepMind
• Sahil DuaAlso at Google DeepMind
• Azalia MirhoseiniAlso at Google DeepMind
• Henrique Schechter VeraAlso at Google DeepMind
• Pushmeet KohliAlso at Google DeepMind
• Brian HieAI for science · Language models
• Ali MadaniAI for science · Language models
• Sergey EdunovLanguage models · AI for science
• Garyk BrixiAI for science · Language models
• Jakob UszkoreitLanguage models · AI for science