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

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Li, Margaret, Shi, Weijia, Pagnoni, Artidoro, West, Peter, Holtzman, Ari
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866910510185709568
author Li, Margaret
Shi, Weijia
Pagnoni, Artidoro
West, Peter
Holtzman, Ari
author_facet Li, Margaret
Shi, Weijia
Pagnoni, Artidoro
West, Peter
Holtzman, Ari
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2407_02446
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Predicting vs. Acting: A Trade-off Between World Modeling & Agent Modeling
Li, Margaret
Shi, Weijia
Pagnoni, Artidoro
West, Peter
Holtzman, Ari
Computation and Language
Artificial Intelligence
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.
title Predicting vs. Acting: A Trade-off Between World Modeling & Agent Modeling
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2407.02446