Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Xuanming, Wang, Sitong, Ma, Jenny, Hwang, Alyssa, Yu, Zhou, Chilton, Lydia B.
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2410.03882
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909620529790976
author Zhang, Xuanming
Wang, Sitong
Ma, Jenny
Hwang, Alyssa
Yu, Zhou
Chilton, Lydia B.
author_facet Zhang, Xuanming
Wang, Sitong
Ma, Jenny
Hwang, Alyssa
Yu, Zhou
Chilton, Lydia B.
contents Human-AI planning for complex goals remains challenging with current large language models (LLMs), which rely on linear chat histories and simplistic memory mechanisms. Despite advances in long-context prompting, users still manually manage information, leading to a high cognitive burden. Hence, we propose JumpStarter, a system that enables LLMs to collaborate with humans on complex goals by dynamically decomposing tasks to help users manage context. We specifically introduce task-structured context curation, a novel framework that breaks down a user's goal into a hierarchy of actionable subtasks, and scopes context to localized decision points, enabling finer-grained personalization and reuse. The framework is realized through three core mechanisms: context elicitation, selection, and reuse. We demonstrate that task-structured context curation significantly improves plan quality by 16% over ablations. Our user study shows that JumpStarter helped users generate plans with 79% higher quality compared to ChatGPT.
format Preprint
id arxiv_https___arxiv_org_abs_2410_03882
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle JumpStarter: Human-AI Planning with Task-Structured Context Curation
Zhang, Xuanming
Wang, Sitong
Ma, Jenny
Hwang, Alyssa
Yu, Zhou
Chilton, Lydia B.
Human-Computer Interaction
Human-AI planning for complex goals remains challenging with current large language models (LLMs), which rely on linear chat histories and simplistic memory mechanisms. Despite advances in long-context prompting, users still manually manage information, leading to a high cognitive burden. Hence, we propose JumpStarter, a system that enables LLMs to collaborate with humans on complex goals by dynamically decomposing tasks to help users manage context. We specifically introduce task-structured context curation, a novel framework that breaks down a user's goal into a hierarchy of actionable subtasks, and scopes context to localized decision points, enabling finer-grained personalization and reuse. The framework is realized through three core mechanisms: context elicitation, selection, and reuse. We demonstrate that task-structured context curation significantly improves plan quality by 16% over ablations. Our user study shows that JumpStarter helped users generate plans with 79% higher quality compared to ChatGPT.
title JumpStarter: Human-AI Planning with Task-Structured Context Curation
topic Human-Computer Interaction
url https://arxiv.org/abs/2410.03882