I am a Research Scientist in the Emergent Artificial Intelligence Lab at Intel Labs.
The substantive focus of my research is characterizing latent social biases in large generative models and techniques for correcting/steering these biases.
In practice, my work is up and down the AI Research/Engineering stack, from optimizing training for large multimodal models on hundreds of accelerators, to writing custom models and pipelines, to designing experiments and evaluation metrics.
Prior to Intel, I was postdoc at Princeton University working with Professor Brandon Stewart, and did my PhD at the University of Oxford with Professors Andy Eggers and Raymond Duch.
News
- Feb 2025: New dataset on HuggingFace: IssueBench
- Dec 2024: Scholar Award for top Intel Labs Academic Author
- Dec 2024: Spotlight Paper at the Creativity and AI Workshop at NeurIPS 2024
- Oct 2024: 3 papers accepted to NeurIPS 2024 Workshops
- Sep 2024: 2 papers accepted to EMNLP 2024
- Aug 2024: ACL Outstanding Paper Award!
- Mar 2024: New model on HuggingFace: intel/llava-gemma-2b
- Feb 2024: Started at Intel Labs as AI Research Scientist
Publications
For a full list of papers, see here.
Highlights
Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models
Naoki Egami, Musashi Hinck, Hanying Wei and Brandon Stewart.
NeurIPS 2023 (Main) 📄 Paper
Political Compass or Spinning Arrow? Towards More Meaningful Evaluations for Values and Opinions in Large Language Models
Paul Röttger, Valentin Hofmann, Valentina Pyatkin, Musashi Hinck, Hannah Rose Kirk, Hinrich Schütze, Dirk Hovy.
ACL 2024 (Main) ⭐Outstanding Paper Award⭐ 📄 Paper
Why do LLaVA Vision-Language Models Reply to Images in English?
Musashi Hinck*, Carolin Holtermann, Matthew Lyle Olson, Florian Schneider, Sungduk Yu, Anahita Bhiwandiwalla, Anne Lauscher, Shaoyen Tseng, Vasudev Lal.
EMNLP 2024 (Findings) 📄 Paper
AutoPersuade: A Framework for Evaluating and Explaining Persuasive Arguments
Till Raphael Saenger, Musashi Hinck, Justin Grimmer, and Brandon M Stewart.
EMNLP 2024 (Main) 📄 Paper
Preprints
IssueBench: Millions of Realistic Prompts for Measuring Issue Bias in LLM Writing Assistance
Paul Röttger, Musashi Hinck, Valentin Hofmann, Kobi Hackenburg, Valentina Pyatkin, Faeze Brahman, Dirk Hovy.
📄 Paper
ClimDetect: A Benchmark Dataset for Climate Change Detection and Attribution
Sungduk Yu, Brian L White, Anahita Bhiwandiwalla, Musashi Hinck, Matthew Lyle Olson, Tung Nguyen, Vasudev Lal.
📄 Paper
Debias your Large Multi-Modal Model at Test-Time with Non-Contrastive Visual Attribute Steering
Neale Ratzlaff, Matthew Lyle Olson, Musashi Hinck, Estelle Aflalo, Shao-Yen Tseng, Vasudev Lal, and Phillip Howard.
📄 Paper
Semantic Specialization in Moe Appears with Scale: A Study of DeepSeek-R1 Expert Specialization
Matthew Lyle Olson*, Neale Ratzlaff*, Musashi Hinck*, Man Luo, Sungduk Yu, Chendi Xue, Vasudev Lal.
📄 Paper