Predicting Vs. Acting: A Trade-off Between World Modeling & Agent Modeling

Li Margaret, Shi Weijia, Pagnoni Artidoro, West Peter, Holtzman Ari. Arxiv 2024

[Paper]    
Agentic Applications Interpretability And Explainability Language Modeling Prompting Reinforcement Learning Training Techniques

RLHF-aligned LMs have shown unprecedented ability on both benchmarks and long-form text generation, yet they struggle with one foundational task: next-token prediction. As RLHF models become agent models aimed at interacting with humans, they seem to lose their world modeling – the ability to predict what comes next in arbitrary documents, which is the foundational training objective of the Base LMs that RLHF adapts. Besides empirically demonstrating this trade-off, we propose a potential explanation: to perform coherent long-form generation, RLHF models restrict randomness via implicit blueprints. In particular, RLHF models concentrate probability on sets of anchor spans that co-occur across multiple generations for the same prompt, serving as textual scaffolding but also limiting a model’s ability to generate documents that do not include these spans. We study this trade-off on the most effective current agent models, those aligned with RLHF, while exploring why this may remain a fundamental trade-off between models that act and those that predict, even as alignment techniques improve.

Similar Work