667 upvotes · 15 OCT 2026 · Bingxi Hou, Guochao Jiang, Guofeng Quan et al.
This paper develops a method to transfer policy shifts from a reinforcement learning (RL) teacher to a student model, using an intermediate teacher checkpoint to compare early and late policy shifts and gate the endpoint shift based on their agreement. Practitioners in RL and reinforcement learning may care about this paper because it proposes a more effective way to transfer policy shifts, which is crucial for improving student performance in various domains.
452 upvotes · 29 AUG 2026 · Xiyuan Yang, Sheikh Sarwar, Jingru Cheng et al.
This paper proposes a method to scale automatic research agents by replacing environment execution with a world model, which can reduce training costs and improve performance. Practitioners might care about this approach because it can accelerate training times and lead to better results for complex AI tasks.
399 upvotes · 1 SEP 2026 · Ke Yang, Chenglong Wang, Michel Galley et al.
This paper develops a method to create personalized AI tutors that can adapt to individual students' strengths and weaknesses, using a combination of pooled training and per-student fine-tuning. Practitioners in education and AI development may care about this work as it could lead to more effective and personalized learning experiences.
379 upvotes · 26 OCT 2026 · Tao Ren, Hui Yang, Jia Deng et al.
This paper proposes a new optimization technique for long-horizon language model agents, allowing them to make more informed decisions by allocating credit more finely across individual tokens. Practitioners might care because it could lead to better performance in tasks like search, tool-use, and automation.
338 upvotes · 27 JUL 2026 · Kimi Team, Tongtong Bai, Yifan Bai et al.
This paper introduces Kimi K3, a large-scale, open-source AI model that achieves state-of-the-art performance on a range of tasks, including vision and coding, and is designed to be more efficient and scalable than previous models, making it a promising candidate for real-world applications.
258 upvotes · 20 AUG 2026 · Chengsong Huang, Zifeng Wang, Rujun Han et al.
This paper introduces EnvHarness, a tool that lets AI agents learn from static environments by changing how they interact with those environments, without having to rebuild the environments from scratch. This can help improve the performance of AI agents in real-world applications.
256 upvotes · 14 SEP 2026 · Tong Zheng, Xidong Wu, Zheng Zhang et al.
This paper introduces Dream-RSI, a framework for recursive self-improvement in exploration, which helps autonomous AI agents discover high-value solutions more efficiently by using a replay simulator to provide low-cost feedback. Practitioners might care because effective exploration is crucial for AI progress, and Dream-RSI can improve discovery quality and reduce costs.
226 upvotes · 8 OCT 2026 · Zhenglin Wan, Jiayu Chen, Zhengfeiyang Zhang et al.
This paper introduces a new method for reinforcement learning with diffusion models, called DiffusionART, which combines the benefits of forward-process and reverse-process methods to improve efficiency and stability. Practitioners might care about this method because it can train diffusion models more effectively and efficiently than existing methods.
223 upvotes · 5 AUG 2026 · Zhongzhi Li, Yucheng Shi, Zongxia Li et al.
This paper introduces a method to generate long-horizon training tasks for terminal agents at scale, which can be used to improve AI models' performance on tasks like navigation and decision-making. Practitioners might care about this because it could help train more advanced AI models with better performance on complex tasks.
208 upvotes · 8 OCT 2026 · Dianqiao Lei, Kevin Qinghong Lin, Pan Lu et al.
This paper investigates how Large Language Models (LLMs) behave when "told not to think", i.e., when instructed to produce answers without using reasoning. Practitioners might care about understanding how to control this behavior, as it can impact the accuracy of LLMs in certain tasks.
201 upvotes · 13 OCT 2026 · Chuyang Wei, Haoxiang Guan, Minghang Zhu et al.
This paper proposes a framework for self-evolving agents that can adapt to changing environments by modifying their software scaffolding without changing their model weights. This approach has implications for reliability, accountability, and knowledge sharing in various domains, including scientific research and organizational workflows.
192 upvotes · 13 OCT 2026 · Guanting Dong, Xinyu Yang, Jiajie Jin et al.
This paper teaches language agents to reflect on their actions and adapt to changing environments, allowing them to learn from feedback and correct mistakes more effectively. Practitioners can benefit from this approach in real-world applications where agents need to handle noisy feedback and user corrections.
185 upvotes · 26 AUG 2026 · Pengfei Zhou, Hexin Wang, Zhengfeiyang Zhang et al.
This paper proposes a way to improve the efficiency of training world models by using game development as a source of reward signals and trajectory data, allowing for more effective post-training of large language models using reinforcement learning. Practitioners might care about this approach because it could lead to more scalable and effective world models for applications like dialogue systems and visual question answering.
183 upvotes · 9 AUG 2026 · Yijiang Li, Bingyang Wang, Yijun Liang et al.
This paper shows how to make large language models improve themselves without needing external guidance or supervision, by using their own internal consistency to correct mistakes. Practitioners might care about this because it could lead to more robust and self-sufficient AI models.
182 upvotes · 25 OCT 2026 · Ruhao Liu, Qi Li, Shizun Wang et al.
This paper introduces DynaFlow, a framework that enables multi-agent systems to adapt to diverse task requirements by learning to orchestrate sub-agents in a free-form way. Practitioners might care about DynaFlow because it can improve the flexibility and performance of multi-agent systems in real-world applications.
168 upvotes · 28 SEP 2026 · Julianna Piskorz, Antonin Berthon, Mihaela van der Schaar
This paper compares the effectiveness of on-policy and off-policy learning in a distillation setting, finding that rollout policy is not as crucial as previously thought, and that learning rate and token-level KL direction play more significant roles in task performance and forgetting.
158 upvotes · 14 AUG 2026 · Yijiang Li, Yijun Liang, Yunjie Tian et al.
This paper proposes a new method called Self-Supervised Visual On-Policy Distillation (S^2VOPD) that generates learning signals from asymmetric augmented views of images, allowing for effective on-policy learning without privileged information. Practitioners might care about this paper because it presents a simple yet effective way to improve performance on various perception benchmarks.
154 upvotes · 14 OCT 2026 · Panrong Tong, Quyu Kong, Pengxiang Li et al.
This paper introduces MAI-Gym, a platform for training and evaluating GUI agents in various environments, allowing for scalable reinforcement learning and enabling the creation of diverse and realistic tasks. Practitioners can use MAI-Gym to train agents that can perform well in both simulated and real-world environments.
150 upvotes · 25 OCT 2026 · Xuan He, Sining Zhoubian, Ruozhen Yang et al.
This paper investigates the net value of agent memory in long-horizon agents, finding that dedicated memory systems often underperform without memory in many cases. Practitioners may care because understanding the net value of memory can help design more reliable and effective memory systems.
142 upvotes · 23 JUL 2026 · Shuqi Lu, Chaofan Li, Kun Luo et al.
This paper introduces AREX, a self-improving agent that recursively refines its answers by verifying intermediate results and using the partially verified state to guide subsequent refinement, aiming to improve the efficiency of deep research. Practitioners might care about AREX's approach to reducing the cost of discovery and verification in complex research tasks.
142 upvotes · 17 AUG 2026 · Yixuan Wang, Yifei Chen, Haichao Zhang et al.
This paper improves reinforcement learning for post-training language model reasoners by addressing two problems in existing methods: identical advantages for distinct reward profiles and fixed relative weights for all objectives. A new method, SA-MRPO, dynamically reallocates optimization effort toward under-optimized objectives while maintaining performance on well-satisfied objectives.
142 upvotes · 17 AUG 2026 · Xin Ding, Liang Mi, Mingzhe Huang et al.
This paper introduces Zetta, a system that enables embodied agents to learn and adapt in real-time while executing physical tasks, allowing for more efficient and effective learning. Practitioners in robotics and AI may care about Zetta's approach as it could lead to more reliable and scalable physical intelligence.
141 upvotes · 15 OCT 2026 · Yijiang Li, Bingyang Wang, Nuno Vasconcelos
This paper helps vision-language models (VLMs) think and explore in 3D space by interacting with the environment through actions and observations, rather than receiving pre-rendered images. Practitioners might care because it can improve the accuracy of VLMs on 3D benchmarks.
138 upvotes · 28 SEP 2026 · Xin Li, Hao Jiang, Xin Gao et al.
This paper proposes a new approach to multi-teacher on-policy distillation that allows language models to learn multiple skills by balancing feedback from different specialists. A practitioner might care about this paper because it could lead to more versatile and effective language models that can handle a wide range of tasks.
127 upvotes · 8 OCT 2026 · Taoyang Jia, Weikai Huang, Linxin Song et al.
This paper develops a new AI system called Reflex that helps robots catch thrown objects, like boxes, using only visual input from the camera. A robot that can catch boxes is useful for many real-world applications, such as robots working in warehouses or factories.
124 upvotes · 21 OCT 2026 · Mingze Sun, Ziang Cao, Yuan Fei et al.
This paper develops a system that can track both hands of a robot in complex, real-life situations, using demonstrations from humans. Practitioners might care because it could enable robots to perform tasks like assembly, cooking, or surgery with greater precision and flexibility.
120 upvotes · 3 AUG 2026 · Ziyu Ma, Hailang Huang, Shun Zou et al.
This paper proposes a framework, LongHorizon-Harness, to help large language model agents tackle long-horizon tasks by explicitly tracking task states and verifying facts from the environment. Practitioners might care because it can improve the performance of these agents on real-world tasks.
118 upvotes · 22 SEP 2026 · Wenbo Pan, Zhichao Liu, Shujie Liu et al.
This paper develops a benchmark to measure an agent's ability to make good decisions along the way, or "taste", in long-horizon tasks. Practitioners might care because better taste can lead to improved overall performance and success in complex tasks.
110 upvotes · 2 SEP 2026 · Yuling Shi, Zhensu Sun, Junsen Dong et al.
This paper introduces EarlyEval, a method to reduce the cost of evaluating large language model (LLM) agents by predicting their outcomes early, allowing for earlier termination of agent runs. Practitioners may care because it can significantly reduce the computational cost of agent development.
109 upvotes · 5 SEP 2026 · Hao Liang, Mingrui Chen, Hengyi Feng et al.
This paper introduces DataFlex-RL, an evaluation platform for comparing different data policies in reinforcement learning with verifiable rewards (RLVR), and finds that uniform data policy leads to better performance and is more reproducible than other methods. Practitioners might care about this because it provides a way to compare different data policies and ensure that their RLVR models are performing well.
109 upvotes · 13 OCT 2026 · Yu Wang, Rui Dai, Wei Chen et al.
This paper proposes a new approach to 3D scene grounding in city-scale images, where an agent actively searches for objects in a scene using a vision-language framework, allowing for more efficient and flexible grounding. Practitioners might care about this research because it could improve the accuracy and speed of object detection in large outdoor scenes.
107 upvotes · 12 AUG 2026 · Cheng Qian, Wenting Zhao, Liangwei Yang et al.
This paper explores a new approach to transferring capabilities from strong AI models to weaker ones at test time, rather than just during training. Practitioners might care about this because it could lead to more efficient and effective use of powerful models in real-world applications.
101 upvotes · 19 SEP 2026 · Meng-Hao Guo, Zhe-Han Mo, Jia-Jun Wang et al.
This paper investigates whether the intelligence of vision-language models (VLMs) can be transferred to robotic control, enabling robots to learn and adapt in both digital and physical worlds. Practitioners might care because it could lead to more human-like robots that can seamlessly adapt to new environments and tasks.
98 upvotes · 20 AUG 2026 · Yunheng Li, Guohong Mu, Hao Li et al.
This paper introduces a method called OraRL to improve the efficiency and scalability of reinforcement learning for multimodal large language models (MLLMs) trained on video data. By leveraging annotations as a source of high-quality rollouts, OraRL can significantly reduce the number of required rollouts, leading to faster training times and better performance on video understanding tasks.
93 upvotes · 29 JUL 2026 · Hao Fei, Yiran Zhao
This paper develops a new framework for world modeling that takes into account the mental state of agents, which is essential for predicting human decisions. Practitioners caring about human decision-making and planning might find this research useful.
92 upvotes · 19 AUG 2026 · Yunhao Yang, Yuexin Bian, Yunjie Tian et al.
This paper introduces Co-RL, a framework for unsupervised multi-agent reinforcement learning that enables diverse and accurate reasoning in language and vision-language models. Practitioners can use Co-RL to improve their models' ability to reason and respond without relying on expensive ground-truth supervision.
90 upvotes · 18 AUG 2026 · Zhi Zheng, Rongsheng Chen, Yunpeng Ba et al.
This paper proposes a new method for fine-tuning large language models (LLMs) in reinforcement learning (RL) tasks with long horizons, using evolution strategies (ES) instead of traditional backpropagation-based training. Practitioners might care because it allows for more efficient and flexible fine-tuning of LLMs, enabling them to tackle complex tasks with larger models and longer interactions.
89 upvotes · 31 AUG 2026 · Yi Ding, Ruqi Zhang
This paper investigates whether on-policy distillation (OPD) truly improves student policies by analyzing the effects of noisy teacher supervision. It finds that OPD works by suppressing low-probability tokens, which can be achieved without a teacher, and introduces a new method called On-Policy Self-Adaptation (OPSA) that outperforms OPD and traditional reinforcement learning methods.
89 upvotes · 9 SEP 2026 · Yanzhe Chen, Zechen Bai, Zhijun Cao et al.
This paper shows how a vision-language model (VLM) can control robots without needing extensive pretraining or specialized hardware, by providing a compact interface that links the model's intentions to specific actions. Practitioners might care about this because it could make robots more accessible and user-friendly.
89 upvotes · 27 OCT 2026 · Muxi Diao, Wenbin Xu, Lele Yang et al.
This paper evaluates how well multimodal agents can generate videos that include synchronized code, sounds, and on-screen events, and why a practitioner might care because it could lead to more realistic video generation in applications like educational content, music videos, and video games.
87 upvotes · 13 AUG 2026 · DreamX Team, Rui Chen, Xiangxiang Chu et al.
This paper introduces a new AI model called DreamX-Phi 1.0 that can predict what will happen in a robotic manipulation scenario, given an initial state and instructions. This model is useful for robotics developers because it can help them design more reliable and efficient robotic systems.
84 upvotes · 8 JUL 2026 · Xinyu Geng, Xuanhua He, Sixiang Chen et al.
This paper introduces a framework called DeepSearch-Evolve, which helps train self-improving web agents by iteratively refining their performance using their own experience. Practitioners might care because this approach can lead to more efficient and effective agents that can learn from their own mistakes.
82 upvotes · 18 SEP 2026 · Bowen Ye, Lei Li, Shicheng Li et al.
This paper creates a system called CodeMidas that turns existing open-source code into environments for training coding agents using reinforcement learning. Practitioners might care because it could lead to more diverse and effective coding agents that can learn to fix issues, construct code, and verify their own work.