AI
Masked Diffusion Language Models:エージェント型RLのための強力で制御可能なテキストベース世界モデル
Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL
arxiv2026年7月21日
日本語要約
本論文は、Masked Diffusion Language Models(MDLM)を、エージェント型強化学習(RL)のための強力で制御可能なテキストベース世界モデルとして提示する。この進歩は、AIエージェントの計画および意思決定能力を大幅に向上させる可能性がある。
English Summary
This paper presents Masked Diffusion Language Models (MDLM) as powerful and steerable text-based world models for agentic Reinforcement Learning (RL). This advancement could significantly improve the planning and decision-making capabilities of AI agents.