Sol Messing
Research Scientist, Google DeepMind
Research Associate Professor, New York University
Affiliate, ML Commons
Working on evaluation science and the socioeconomic impacts of AI systems.
Current work
- Measurement — hidden error in LLM evals & benchmarks and how to reduce it. Measuring the spread of ideas across languages with fine-tuned LLMs; estimating ideology via forced-choice LLM-as-judge.
- AI & Democracy — what LLMs refuse and the political consequences. How state media in training data shapes model output. Alignment choices and democratic outcomes.
- Economic impacts — simulating real-world economic tasks to understand model capabilities and labor-market impacts.
Featured Publications
All publications →State Media Control Influences Large Language Models
Nature, 2026.
PDF DOI
Quantifying Narrative Similarity Across Languages
Sociological Methods & Research, 2025.
PDF DOI
A 2 million-person, campaign-wide field experiment shows how digital advertising affects voter turnout
Nature Human Behaviour, 2023.
PDF
Projecting confidence: How the probabilistic horse race confuses and demobilizes the public
Journal of Politics, 2019.
PDF
Writing
Browse & filter →
No matching items