RNG-Bench evaluates MLLMs on hidden-observation reconstruction in non-Markov games, finds forgetting as the dominant error source, and shows fine-tuning on optimal rollouts improves performance with transfer to other benchmarks.
hub
Seagent: Self-evolving computer use agent with autonomous learning from experience
21 Pith papers cite this work. Polarity classification is still indexing.
21
Pith papers citing it
hub tools
citation-role summary
background 2
baseline 1
citation-polarity summary
years
2026 21representative citing papers
CUA-Gym generates 32,112 verified RLVR tuples across 110 mock environments, enabling trained models to reach 62.1% and 72.6% on OSWorld-Verified while transferring to WebArena.
OLIVIA treats LLM agent action selection as a contextual linear bandit over frozen hidden states and applies UCB exploration to adapt online, yielding consistent gains over static ReAct and prompt-based baselines on four benchmarks.
MAGE uses a four-subgraph co-evolutionary knowledge graph plus dual bandits to externalize and retrieve experience for stable self-evolution of frozen language-model agents, showing gains on nine diverse benchmarks.
PACT achieves perfect security and utility under oracle provenance by enforcing argument-level trust contracts based on semantic roles and cross-step provenance tracking, outperforming invocation-level monitors in AgentDojo evaluations.
PlanViz is a new benchmark with three sub-tasks and PlanScore metric to evaluate planning-oriented image generation and editing by unified multimodal models for computer-use tasks.
Failure-driven self-improvement raises OpenCUA-72B success rate on OSWorld from 42.3% to 48.9% via LLM diagnosis and inference-time code patches, without retraining.
GUICrafter uses curriculum learning on unannotated GUI screenshots for visual grounding followed by RL calibration on limited labels to match or exceed prior GUI agents with far less annotation.
Presents DocFormBench benchmark and DocFormFlow workflow for content-aware LLM document formatting, claiming higher accuracy and lower token use via decoupled localization and modification.
VESTA introduces dynamic tool creation for VLMs that outperforms static-tool and no-tool baselines on distribution fitting, time series, and astronomy tasks in the new DAWN benchmark.
LearnWeak specializes small CUAs via weakness detection by a reference agent, targeted task synthesis, and error-aware training, delivering 11+ point gains on OSWorld.
SEAL co-evolves LLM agents and environments via shared turn-level failure diagnoses, yielding +8.25 to +26.25 point gains on tool-use tasks with only 400 samples.
MementoGUI introduces a modular memory-control framework with working and episodic memory operators that improves long-horizon GUI agent performance over history-replay and text-only baselines.
Skill-R1 applies bi-level group-relative policy optimization to evolve skills recurrently from verified outcomes, yielding gains over baselines on multi-step tasks.
HiLSVA shows that a human-in-the-loop LLM agent system can help novices and experts complete scientific visualization tasks, while human oversight adds measurable execution time.
AliyunConsoleAgent-32B reaches 63.52% success on a 278-task cloud console benchmark, closing to 1.82pp of frontier models at 92% lower cost via SFT distillation and GRPO RL.
Q-Evolve unifies automatic process-reward labeling via advantage estimation and behavior-proximal policy optimization inside an in-distribution RL loop to enable self-evolving LLM agents on interactive tasks.
Survey of auto-research systems identifies objective, validation, and acceptance collapses, concluding that workflow closure does not equal scientific closure and advocating non-autonomous epistemic control.
Empirical comparison of domain-specific, computer-use, and general-purpose LLM agents plus CLI/GUI modalities on SciVis tasks reveals general-purpose agents highest in success rate but costliest, domain-specific agents more efficient, and persistent memory beneficial depending on mode.
Agent skills are the emerging standard for giving LLM agents specialized procedures, but the reported 26.1% vulnerability rate in community skills argues for provenance-based, tiered trust governance.
citing papers explorer
-
Beyond the Current Observation: Evaluating Multimodal Large Language Models in Controllable Non-Markov Games
RNG-Bench evaluates MLLMs on hidden-observation reconstruction in non-Markov games, finds forgetting as the dominant error source, and shows fine-tuning on optimal rollouts improves performance with transfer to other benchmarks.
-
CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents
CUA-Gym generates 32,112 verified RLVR tuples across 110 mock environments, enabling trained models to reach 62.1% and 72.6% on OSWorld-Verified while transferring to WebArena.
-
OLIVIA: Online Learning via Inference-time Action Adaptation for Decision Making in LLM ReAct Agents
OLIVIA treats LLM agent action selection as a contextual linear bandit over frozen hidden states and applies UCB exploration to adapt online, yielding consistent gains over static ReAct and prompt-based baselines on four benchmarks.
-
MAGE: Multi-Agent Self-Evolution with Co-Evolutionary Knowledge Graphs
MAGE uses a four-subgraph co-evolutionary knowledge graph plus dual bandits to externalize and retrieve experience for stable self-evolution of frozen language-model agents, showing gains on nine diverse benchmarks.
-
The Granularity Mismatch in Agent Security: Argument-Level Provenance Solves Enforcement and Isolates the LLM Reasoning Bottleneck
PACT achieves perfect security and utility under oracle provenance by enforcing argument-level trust contracts based on semantic roles and cross-step provenance tracking, outperforming invocation-level monitors in AgentDojo evaluations.
-
PlanViz: Evaluating Planning-Oriented Image Generation and Editing for Computer-Use Tasks
PlanViz is a new benchmark with three sub-tasks and PlanScore metric to evaluate planning-oriented image generation and editing by unified multimodal models for computer-use tasks.
-
Learning from Failure: Inference-Time Self-Improvement for Computer-Use Agents
Failure-driven self-improvement raises OpenCUA-72B success rate on OSWorld from 42.3% to 48.9% via LLM diagnosis and inference-time code patches, without retraining.
-
GUICrafter: Weakly-Supervised GUI Agent Leveraging Massive Unannotated Screenshots
GUICrafter uses curriculum learning on unannotated GUI screenshots for visual grounding followed by RL calibration on limited labels to match or exceed prior GUI agents with far less annotation.
-
What to Format and How: A Benchmark and Workflow Approach for Document Formatting
Presents DocFormBench benchmark and DocFormFlow workflow for content-aware LLM document formatting, claiming higher accuracy and lower token use via decoupled localization and modification.
-
VESTA: Visual Exploration with Statistical Tool Agents
VESTA introduces dynamic tool creation for VLMs that outperforms static-tool and no-tool baselines on distribution fitting, time series, and astronomy tasks in the new DAWN benchmark.
-
Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents
LearnWeak specializes small CUAs via weakness detection by a reference agent, targeted task synthesis, and error-aware training, delivering 11+ point gains on OSWorld.
-
SEAL: Synergistic Co-Evolution of Agents and Learning Environments
SEAL co-evolves LLM agents and environments via shared turn-level failure diagnoses, yielding +8.25 to +26.25 point gains on tool-use tasks with only 400 samples.
-
MementoGUI: Learning Agentic Multimodal Memory Control for Long-Horizon GUI Agents
MementoGUI introduces a modular memory-control framework with working and episodic memory operators that improves long-horizon GUI agent performance over history-replay and text-only baselines.
-
Skill-R1: Agent Skill Evolution via Reinforcement Learning
Skill-R1 applies bi-level group-relative policy optimization to evolve skills recurrently from verified outcomes, yielding gains over baselines on multi-step tasks.
-
HiLSVA: Design and Evaluation of a Human-in-the-Loop Agentic System for Scientific Visualization
HiLSVA shows that a human-in-the-loop LLM agent system can help novices and experts complete scientific visualization tasks, while human oversight adds measurable execution time.
-
AliyunConsoleAgent: Training Web Agents in Real-World Cloud Environments via Distillation and Reinforcement Learning
AliyunConsoleAgent-32B reaches 63.52% success on a 278-task cloud console benchmark, closing to 1.82pp of frontier models at 92% lower cost via SFT distillation and GRPO RL.
-
Self-evolving LLM agents with in-distribution Optimization
Q-Evolve unifies automatic process-reward labeling via advantage estimation and behavior-proximal policy optimization inside an in-distribution RL loop to enable self-evolving LLM agents on interactive tasks.
-
Workflow Closure Is Not Scientific Closure in Auto-Research Systems
Survey of auto-research systems identifies objective, validation, and acceptance collapses, concluding that workflow closure does not equal scientific closure and advocating non-autonomous epistemic control.
-
Exploring LLM Agent Designs and Interaction Modalities for Scientific Visualization
Empirical comparison of domain-specific, computer-use, and general-purpose LLM agents plus CLI/GUI modalities on SciVis tasks reveals general-purpose agents highest in success rate but costliest, domain-specific agents more efficient, and persistent memory beneficial depending on mode.
-
Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward
Agent skills are the emerging standard for giving LLM agents specialized procedures, but the reported 26.1% vulnerability rate in community skills argues for provenance-based, tiered trust governance.
- MobileForge: Annotation-Free Adaptation for Mobile GUI Agents with Hierarchical Feedback-Guided Policy Optimization