Textual Gradients are a Flawed Metaphor for Automatic Prompt Optimization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Melcer, Daniel, Chen, Qi, Chiang, Wen-Hao, Garg, Shweta, Garg, Pranav, Bock, Christian
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909963227496448
author Melcer, Daniel
Chen, Qi
Chiang, Wen-Hao
Garg, Shweta
Garg, Pranav
Bock, Christian
author_facet Melcer, Daniel
Chen, Qi
Chiang, Wen-Hao
Garg, Shweta
Garg, Pranav
Bock, Christian
contents A well-engineered prompt can increase the performance of large language models; automatic prompt optimization techniques aim to increase performance without requiring human effort to tune the prompts. One leading class of prompt optimization techniques introduces the analogy of textual gradients. We investigate the behavior of these textual gradient methods through a series of experiments and case studies. While such methods often result in a performance improvement, our experiments suggest that the gradient analogy does not accurately explain their behavior. Our insights may inform the selection of prompt optimization strategies, and development of new approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2512_13598
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Textual Gradients are a Flawed Metaphor for Automatic Prompt Optimization
Melcer, Daniel
Chen, Qi
Chiang, Wen-Hao
Garg, Shweta
Garg, Pranav
Bock, Christian
Computation and Language
Machine Learning
A well-engineered prompt can increase the performance of large language models; automatic prompt optimization techniques aim to increase performance without requiring human effort to tune the prompts. One leading class of prompt optimization techniques introduces the analogy of textual gradients. We investigate the behavior of these textual gradient methods through a series of experiments and case studies. While such methods often result in a performance improvement, our experiments suggest that the gradient analogy does not accurately explain their behavior. Our insights may inform the selection of prompt optimization strategies, and development of new approaches.
title Textual Gradients are a Flawed Metaphor for Automatic Prompt Optimization
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2512.13598