Reproduce “From Local Skills to Long-Horizon Web Tasks: Progressive Skill Exploration for LLM Web Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xueyang Feng, Xiaohe Bo, Xu Chen, Quanyu Dai, Feng Wen
SCOPE reuses locally successful web actions as skills, replays them to a useful frontier, and uses failures from that state to guide exploration. On WebArena, its OPSI variant reaches 54.3% task success with Llama3.1-8B, versus 45.8% for WEBRL, a reported 18.6% relative gain from a WEBRL-checkpoint starting point. Evaluation spans two backbones and three web benchmarks, mostly with one run.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.