GenTool: Enhancing Tool Generalization in Language Models through Zero-to-One and Weak-to-Strong Simulation

Jie He, Jennifer Neville, Mengting Wan, Longqi Yang, Hui Liu, Xiaofeng Xu, Xia Song, Jeff Z. Pan, Pei Zhou · 2025

Read GenTool: Enhancing Tool Generalization in Language Models through Zero-to-One and Weak-to-Strong Simulation by Jie He, Jennifer Neville, Mengting Wan, Longqi Yang, Hui Liu, Xiaofeng Xu, Xia Song, Jeff Z. Pan, Pei Zhou (2025) 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-2502-18990.txt.

Description

Jie He leads a nine-author team presenting GenTool, a training framework that teaches language models to use unfamiliar tools, in a paper published in two thousand twenty-five. Tool use lets AI assistants reach information beyond their training. Models are usually fine-tuned on tool examples, yet their grip on new tools stays shaky.

GenTool trains for two kinds of generalization. Zero-to-one generalization covers queries where no suitable tool exists at first, then one appears. Weak-to-strong generalization covers upgraded tools replacing weaker versions. Simulated training scenarios exercise both situations. The goal is an assistant that adopts new tools gracefully instead of failing or ignoring them. That robustness matters as tool libraries keep growing and changing.

Catalog record

CREATORJie He, Jennifer Neville, Mengting Wan, Longqi Yang, Hui Liu, Xiaofeng Xu, Xia Song, Jeff Z. Pan, Pei Zhou
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-18990

Browse the full library · About the Tower of Babel