WILBUR: Adaptive In-Context Learning for Robust and Accurate Web Agents

arXiv:2404.05902v1 Announce Type: new
Abstract: In the realm of web agent research, achieving both generalization and accuracy remains a challenging problem. Due to high variance in website structure, existing approaches often fail. Moreover, existing fine-tuning and in-context learning techniques fail to generalize across multiple websites. We introduce Wilbur, an approach that uses a differentiable ranking model and a novel instruction synthesis technique to optimally populate a black-box large language model's prompt with task demonstrations from previous runs. To maximize end-to-end …

abstract accuracy agent agents arxiv context cs.ai cs.cl fine-tuning in-context learning multiple realm research robust type variance web website websites