الملخص
Long-horizon language agents often receive supervision only from terminal task outcomes, leaving little signal for distinguishing productive intermediate behavior from stagnation or even regression. Rather than learning a separate value function or process reward model for every task, we ask whether pretrained models can recognize task progress from their existing world knowledge. We formalize this capability with a World Potential Model (WPM), a goal-conditioned evaluator of task-relative realized progress in agent contexts. In ALFWorld and ScienceWorld, off-the-shelf pretrained models substantially outperform chance at recovering realized-progress structure without task-specific evaluator fine-tuning. We further anchor these progress judgments to task-specific milestones to obtain scalar world potentials, whose temporal differences provide process-sensitive step-level credit for policy optimization. Under matched comparisons, WPM-guided optimization improves success over outcome-only GRPO across all evaluated configurations. Together, these results provide initial evidence that pretrained world knowledge can support reusable realized-progress evaluation and provide useful supervision for long-horizon agents.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Zhao, J., Tang, J., Shu, Y., Wu, J., Lu, Y., Tong, J., Xu, H., Ge, W., & Zhang, Q. (2026). World Potential Model: Pretrained World Knowledge as Progress Potentials. https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials
MLA 9
Zhao, Jun, et al. "World Potential Model: Pretrained World Knowledge as Progress Potentials." https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials.
شيكاغو (المؤلف–التاريخ)
Zhao, Jun, Jixin Tang, Yang Shu, Jinyang Wu, Yuyang Lu, Jingqi Tong, Hao Xu, Weifeng Ge, and Qi Zhang. 2026. "World Potential Model: Pretrained World Knowledge as Progress Potentials." https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials.
هارفارد
Zhao, J., Tang, J., Shu, Y., Wu, J., Lu, Y., Tong, J., Xu, H., Ge, W. and Zhang, Q. (2026) 'World Potential Model: Pretrained World Knowledge as Progress Potentials', Available at: https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials.
فانكوفر
Zhao J, Tang J, Shu Y, Wu J, Lu Y, Tong J, et al. World Potential Model: Pretrained World Knowledge as Progress Potentials. https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials
IEEE
J. Zhao, J. Tang, Y. Shu, J. Wu, Y. Lu, J. Tong, H. Xu, W. Ge, and Q. Zhang, "World Potential Model: Pretrained World Knowledge as Progress Potentials," https://omanscience.com/ar/articles/world-potential-model-pretrained-world-knowledge-as-progress-potentials.