Tom Bewley

Interpreting Language Reward Models via Contrastive Explanations

Representation Consistency for Accurate and Coherent LLM Answer Aggregation

To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models

Language Models Can Help to Learn High-Performing Cost Functions for Recourse

Sequential Harmful Shift Detection Without Labels

Zero-Shot Reinforcement Learning from Low Quality Data

Conservative World Models