Reliable information · Science communication · Sustainable AI
Dustin Wright
Tenure-Track Assistant Professor
Dustin studies how AI changes the information people encounter, from the accuracy of individual claims to the diversity of knowledge available through language models.
His work tests whether generated claims are supported by source documents and how scientific information changes through summaries and reports. He also studies the diversity of claims and opinions expressed by language models, and whether evaluation metrics remain reliable when the data changes.
Dustin joined AAU NLP in February 2026, following a Danish Data Science Academy postdoctoral fellowship at the University of Copenhagen. His sustainability research combines technical work on efficient models with analysis of AI’s wider environmental consequences. He co-supervises Luc Raszewski’s DARA fellowship on diversity-preserving alignment with Johannes Bjerva.
Evidence and reliable information
Factuality, faithfulness, attribution, uncertainty, and evaluation that remains valid under controlled change and distribution shift.
Science communication and diversity
Fact-checking, information change, public-facing science, model values, and the diversity of claims made accessible through AI.
Sustainable AI
Combining efficient modelling with systems-level analysis of environmental and social consequences.
Selected work
5 representative papers
What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models
Measures the diversity of claims in language-model outputs across topics and cultural contexts.
- Communities, Domains & Consequences
- Factuality, Reliability & Evaluation
Unstructured Evidence Attribution for Long Context Query Focused Summarization
- Knowledge, Evidence & Reasoning
- Factuality, Reliability & Evaluation
Efficiency Is Not Enough: A Critical Perspective on Environmentally Sustainable AI
- Communities, Domains & Consequences
BMRS: Bayesian Model Reduction for Structured Pruning
- Communities, Domains & Consequences
Generating Scientific Claims for Zero-Shot Scientific Fact Checking
- Knowledge, Evidence & Reasoning
- Factuality, Reliability & Evaluation
Supervision and collaboration
Research with Dustin Wright
Dustin welcomes projects on evidence attribution, information diversity and science communication, alongside efficient modelling and AI sustainability.