Rémy Portelas

Stick to your Role! Stability of Personal Values Expressed in Large Language Models

2024

Standard Large Language Models (LLMs) evaluation contains many different queries from similar minimal contexts (e.g. multiple choice questions). Conclusions from such evaluations are little informative about models' behavior in different new contexts (e.g. in deployment). …

Offline Reinforcement Learning of High-Quality Behaviors Under Robust Style Alignment

2026 · HAL (Le Centre pour la Communication Scientifique Directe)

We study offline reinforcement learning of style-conditioned policies using explicit style supervision via subtrajectory labeling functions. In this setting, aligning style with high task performance is particularly challenging due to distribution shift and inherent conflicts …