使大型语言模型能够有效执行需要长期规划和多次交互的长时程任务,对于开放世界自主性至关重要。传统方法采用基于动作的规划(planning with actions, PwA),即提供可执行的动作列表作为参考。然而,当环境动作空间组合爆炸(例如,开放式的现实世界)时,这种动作表示选择将变得不切实际。这自然引出一个问题:随着环境动作空间的扩展,长时程智能体的最优动作表示是什么?在本文中,我们系统地研究了两种不同动作表示的有效性。第一种是传统的基于动作的规划(PwA),因其在现有基准上的有效性而被广泛采用。另一种是基于模式的规划(planning with schemas, PwS),它将动作模式实例化为动作列表(例如,"移动[物体]到[物体]" -> "移动苹果到桌子"),以确保动作空间简洁且可可靠扩展。这种替代方案的动机在于其与人类认知的一致性及其对环境施加的动作格式限制的遵从。我们提出认知带宽视角作为概念框架,以定性理解这两种动作表示之间的差异,并经验性地观察到ALFWorld(约35个动作)和SciWorld(约500个动作)之间的表示选择拐点,这为可扩展表示的需求提供了证据。我们进一步进行受控实验,研究该拐点位置如何与不同模型能力相互作用:更强的规划能力使拐点右移,而更好的模式实例化使其左移。最后,鉴于PwS智能体的次优性能,我们为构建更强大的PwS智能体以更好地实现可扩展自主性提供了可操作的指南。
英文摘要:
Enabling LLMs to effectively operate long-horizon task which requires long-term planning and multiple interactions is essential for open-world autonomy. Conventional methods adopt planning with actions where a executable action list would be provided as reference. However, this action representation choice would be impractical when the environment action space is combinatorial exploded (e.g., open-ended real world). This naturally leads to a question: As environmental action space scales, what is the optimal action representation for long-horizon agents? In this paper, we systematically study the effectiveness of two different action representations. The first one is conventional planning with actions (PwA) which is predominantly adopted for its effectiveness on existing benchmarks. The other one is planning with schemas (PwS) which instantiate an action schema into action lists (e.g., "move [OBJ] to [OBJ]" -> "move apple to desk") to ensure concise action space and reliable scalability. This alternative is motivated by its alignment with human cognition and its compliance with environment-imposed action format restriction. We propose cognitive bandwidth perspective as a conceptual framework to qualitatively understand the differences between these two action representations and empirically observe a representation-choice inflection point between ALFWorld (~35 actions) and SciWorld (~500 actions), which serve as evidence of the need for scalable representations. We further conduct controlled experiments to study how the location of this inflection point interacts with different model capacities: stronger planning proficiency shifts the inflection rightward, whereas better schema instantiation shifts it leftward. Finally, noting the suboptimal performance of PwS agents, we provide an actionable guide for building more capable PwS agents for better scalable autonomy.
Visual Large Language Models Exhibit Human-Level Cognitive Flexibility in the Wisconsin Card Sorting Test
Laboratory of Brain Atlas and Brain-inspired Intelligence, Institute of Automation, Chinese Academy of Sciences(脑图谱与脑启发智能实验室,自动化研究所,中国科学院);School of Artificial Intelligence, University of Chinese Academy of Sciences(人工智能学院,中国科学院大学);Inria centre at the university of Bordeaux, Neurodegenerative Diseases Institute(波尔多大学Inria中心,神经退行性疾病研究所)
University of Toronto(多伦多大学);Nanyang Technological University(南洋理工大学);Acceleration Consortium(加速联盟);Vector Institute for Artificial Intelligence(向量人工智能研究所);Canadian Institute for Advanced Research (CIFAR)(加拿大高等研究院(CIFAR));NVIDIA(英伟达)
Beyond ID Embeddings: Process-Grounded Language Modeling for Cognitive Diagnosis
State Key Laboratory of AI Safety, Institute of Computing Technology, CAS(中国科学院计算技术研究所人工智能安全国家重点实验室);University of Chinese Academy of Sciences(中国科学院大学)
Exploring Spatial Schema Intuitions in Large Language and Vision Models
Ludwig-Maximilian-University, Munich(慕尼黑大学);Munich Center for Machine Learning (MCML)(慕尼黑机器学习中心);King’s College London(伦敦国王学院);Imperial College London(帝国理工学院)
CogniDir: Combating Cognitive Malicious Comments via Adaptive Distributional Learning for Robust Fake News Detection
Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所);University of Minnesota(明尼苏达大学);Queen Mary University of London(伦敦玛丽女王大学);New Laboratory of Pattern Recognition (NLPR), Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所模式识别新实验室)
GCAgent: Long-Video Understanding via Schematic and Narrative Episodic Memory
Integrated Vision and Language Lab., School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST)(整合视觉与语言实验室,电气工程学院,韩国科学技术院(KAIST));Visual Intelligence Research Section, Superintelligence Creative Research Laboratory, Electronics and Telecommunications Research Institute (ETRI)(视觉智能研究部,超智能创意研究实验室,电子电信研究院)
Towards Understanding and Measuring COGNITIVE ATROPHY in LLM Behaviour
York University(约克大学);Vector Institute(向量研究所);Rotman Research Institute(罗特曼研究学院);Dalhousie University(达尔豪斯大学);Centre for Addiction & Mental Health(成瘾与心理健康中心);KITE Research Institute(KITE研究机构)
ChatGPT and Gemini participated in the Korean College Scholastic Ability Test -- Earth Science I
Institute for Research Excellence in Learning Sciences(学习科学卓越研究所);Seoul National University(首尔大学);National Taiwan Normal University(台湾师范大学);Universitas Negeri Malang(玛琅国立大学)
Rethinking Supply Chain Planning: A Generative Paradigm
Tsinghua University, Beijing, China(清华大学);JD.com, Beijing, China(京东);Faculty of Engineering and Faculty of Business and Economics, The University of Hong Kong, China(香港大学工程学院和商学院);College of Engineering, University of California, Berkeley, CA, USA(加州大学伯克利分校工程学院)