Sewon Min

What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data

Frustratingly Simple Retrieval Improves Challenging, Reasoning-Intensive Benchmarks

Constantly Improving Image Models Need Constantly Improving Benchmarks

Reliable Fine-Grained Evaluation of Natural Language Math Proofs

RLAC: Reinforcement Learning with Adversarial Critic for Free-Form Generation Tasks

Spurious Rewards: Rethinking Training Signals in RLVR

FlexOLMo: Open Language Models for Flexible Data Use

OLMoE: Open Mixture-of-Experts Language Models

ReasonIR: Training Retrievers for Reasoning Tasks

Organize the Web: Constructing Domains Enhances Pre-Training Data Curation

Leveraging Set Assumption for Membership Inference in Language Models