Query-Efficient Planning with Language Models

Gonzalo Gonzalez-Pumariega, Wayne Chen, Kushal Kedia, Sanjiban Choudhury · 2024

Read Query-Efficient Planning with Language Models by Gonzalo Gonzalez-Pumariega, Wayne Chen, Kushal Kedia, Sanjiban Choudhury (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-06162.txt.

Description

Gonzalo Gonzalez-Pumariega and co-authors compare two ways of using large language models to plan in complex settings. Planning means finding a series of steps from a start point to a goal while asking a model of the world as few questions as possible. The team tests the language model in two competing roles.

In the first role, the model acts as a guide inside a search planner, picking which states look promising. In the second, the model writes out a whole plan, checks it against the world model, and revises it from feedback. Tests on a robot task benchmark and on standard planning problems showed both roles beat plain baselines. The planning role, though, needed far fewer checks of the world model. The authors say the planner role adapts faster because it can reshape its whole strategy at once.

Catalog record

CREATORGonzalo Gonzalez-Pumariega, Wayne Chen, Kushal Kedia, Sanjiban Choudhury
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-06162

Browse the full library · About the Tower of Babel