In-Context Teacher–Student Guidance for Open-Weight Browser Agents
Wéwerthon Silva Cardoso , Ricardo Marcacini
- 🏛 Institutions
- University of São Paulo
- 📅 Date
- September 2026
- 📑 Publisher
- KDMiLe 2026
- 💻 Env
- Web
- 🔑 Keywords
TLDR
In-Context Learning Guidance (ICLG) has a stronger teacher model solve representative browser tasks offline and converts its reasoning into reusable prompt-level guidance for open-weight student agents, without fine-tuning. On MiniWoB++ with GPT-5.1 as the teacher and ten open-weight students, the number of solved student-task pairs rises from 208 to 271.
Related papers (24)
- Action Conditioned Bisimulation For GUI Agent MemorySeptember 30, 2026 · arXiv
- SPIDES: Information-Flow Control for Browser AgentsOctober 16, 2026 · arXiv
- MHVF: Training-Free GUI Grounding via Verification-Guided Multi-Hypothesis SearchOctober 13, 2026 · arXiv
- BrowseSafe++: Benchmarking Prompt Injection Detection Across Web Contexts, Languages, and Browser ViewsOctober 12, 2026 · arXiv
- The Agent Clicks Approve: Injected Page Text Makes Browser Agents Authorize Access the Task Never NamedOctober 10, 2026 · arXiv
- M^2: Dual-Memory Augmentation for Long-Horizon Web Agents via Trajectory Summarization and Insight RetrievalFebruary 28, 2026 · arXiv
- WebATLAS: An LLM Agent with Experience-Driven Memory and Action SimulationOctober 26, 2025 · NeurIPS 2025 Workshop on Language Agents and World Models
- Improved GUI Grounding via Iterative NarrowingNovember 18, 2024 · arXiv
- Auto-Intent: Automated Intent Discovery and Self-Exploration for Large Language Model Web AgentsOctober 29, 2024 · Findings of EMNLP 2024
- Navigating WebAI: Training Agents to Complete Web Tasks with Large Language Models and Reinforcement LearningMay 1, 2024 · SAC 2024
- Reinforcement Learning on Web Interfaces Using Workflow-Guided ExplorationFebruary 24, 2018 · ICLR 2018 (Poster)
- World of Bits: An Open-Domain Platform for Web-Based AgentsAugust 31, 2017 · ICML 2017
- Absorbed in Inertia: Activation Analysis for Computer-Use AgentsSeptember 29, 2026 · arXiv
- PhoneCLI: From App Interfaces to Callable Commands for Mobile AgentsSeptember 28, 2026 · arXiv
- Relevance Does Not Imply Applicability: Experience Activation for Personal GUI AgentsSeptember 27, 2026 · arXiv
- STaR-KV: Spatio-Temporal Adaptive Re-weighting for KV Cache Compression in GUI Vision-Language ModelsJune 1, 2026 · arXiv
- UI-Zoomer: Uncertainty-Driven Adaptive Zoom-In for GUI GroundingApril 15, 2026 · arXiv
- GPA: Learning GUI Process Automation from DemonstrationsApril 2, 2026 · arXiv
- GUIDE: Resolving Domain Bias in GUI Agents through Real-Time Web Video Retrieval and Plug-and-Play AnnotationMarch 27, 2026 · arXiv
- Zoom to Essence: Trainless GUI Grounding by Inferring upon Interface ElementsMarch 15, 2026 · arXiv
- Trifuse: Enhancing Attention-Based GUI Grounding via Multimodal FusionFebruary 6, 2026 · arXiv
- Darwinian Memory: A Training-Free Self-Regulating Memory System for GUI Agent EvolutionJanuary 30, 2026 · arXiv
- MVP: Multiple View Prediction Improves GUI GroundingDecember 9, 2025 · arXiv
- Zoom in, Click out: Unlocking and Evaluating the Potential of Zooming for GUI GroundingDecember 5, 2025 · arXiv