Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Proposer-Agent-Evaluator (PAE): Autonomous Skill Discovery For Foundation Model Internet Agents
Yifei Zhou, Qianlan Yang, Kaixiang Lin, Min Bai, Xiong Zhou, Yu-Xiong Wang, Sergey Levine, Li Erran Li
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
placeholder
Problem
placeholder
Approach
placeholder
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply PAE instead of SFT fine-tuning | LLaVa-7B · WebVoyager | success rate (average): 14.9% to 22.3% | Improved |
| apply PAE instead of SFT fine-tuning | LLaVa-7B · WebArena Easy | success rate (average): 18.0% to 24.6% | Improved |
| apply PAE instead of SFT fine-tuning | LLaVa-34B · WebVoyager | success rate (average): 22.2 to 33.0 | Improved |
| apply PAE instead of SFT fine-tuning | LLaVa-7B · WebVoyager unseen websites | success rate: 9.1 to 16.3 | Improved |
| use Qwen2VL-7B instead of Claude as proposer and evaluator (ablation) | LLaVa-7B PAE · WebArena Easy | success rate (average): 18.0% to 23.1% | Improved |
| use Qwen2VL-7B instead of Claude as proposer and evaluator (ablation) | LLaVa-7B PAE · WebVoyager unseen websites | success rate: 9.1% to 13.7% | Improved |
| add 10 user-demo screenshots as task-proposer context (ablation) | LLaVa-7B PAE · WebArena Easy OpenStreetMap | success rate: 19.5% to 21.7% | Improved |
| add 10 user-demo screenshots as task-proposer context (ablation) | LLaVa-7B PAE · WebArena Easy PostMill and OneStopMarket | success rate: PostMill: 21.1 to 21.5; OneStopMarket: 42.3 to 42.1 | No clear effect |
| include a reasoning step before web operations (ablation) | PAE agent policy · unseen human-written evaluation set | performance | Improved |
| replace outcome-based evaluation with step-based evaluation (ablation) | PAE · evaluation task set | performance | Hurt |
| replace outcome-based evaluation with function-based evaluation (ablation) | PAE · evaluation task set | performance | Hurt |
| apply PAE instead of SFT fine-tuning | LLaVa-34B · WebVoyager unseen websites | success rate: 16.1 to 21.4 | Improved |
Closest papers by meaning
- A Subgoal-driven RL Framework for Improving Long-Horizon Web Agents (NeurIPS 2026 · similarity 0.88)
- REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites (NeurIPS 2025 · similarity 0.88)
- WebGym: Scaling Training Environments for Long-Horizon Visual Web Agents with Realistic Tasks (CVPR 2026 · similarity 0.87)
- WebArena: A Realistic Web Environment for Building Autonomous Agents (ICLR 2024 · similarity 0.87)
- Learning to Adapt: Self-Improving Web Agent via Cognitive-Aware Exploration (CVPR 2026 · similarity 0.87)
- RealWebAssist: A Benchmark for Long-Horizon Web Assistance with Real-World Users (AAAI 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
