SteP: Stacked LLM Policies for Web Actions

TL;DR

SteP introduces a dynamic, stack-based policy composition method for large language models to effectively perform diverse web tasks, improving performance over state-of-the-art approaches.

Contribution

This paper presents SteP, a novel approach that enables dynamic control of multiple policies in web tasks, surpassing static hierarchy methods.

Findings

SteP improves WebArena performance by 14.9% to 33.5% over SOTA.

SteP is competitive on MiniWoB++ with less data.

Code and data are publicly available.

Abstract

Performing tasks on the web presents fundamental challenges to large language models (LLMs), including combinatorially large open-world tasks and variations across web interfaces. Simply specifying a large prompt to handle all possible behaviors and states is extremely complex, and results in behavior leaks between unrelated behaviors. Decomposition to distinct policies can address this challenge, but requires carefully handing off control between policies. We propose Stacked LLM Policies for Web Actions (SteP), an approach to dynamically compose policies to solve a diverse set of web tasks. SteP defines a Markov Decision Process where the state is a stack of policies representing the control state, i.e., the chain of policy calls. Unlike traditional methods that are restricted to static hierarchies, SteP enables dynamic control that adapts to the complexity of the task. We evaluate SteP against multiple baselines and web environments including WebArena, MiniWoB++, and a CRM. On WebArena, SteP improves (14.9\% to 33.5\%) over SOTA that use GPT-4 policies, while on MiniWob++, SteP is competitive with prior works while using significantly less data. Our code and data are available at https://asappresearch.github.io/webagents-step.