Why Prompt Design Matters and Works: A Complexity Analysis of Prompt Search Space in LLMs

Xiang Zhang, Juntai Cao, Jiaqi Wei, Chenyu You, Dujian Ding · 2025

Read Why Prompt Design Matters and Works: A Complexity Analysis of Prompt Search Space in LLMs by Xiang Zhang, Juntai Cao, Jiaqi Wei, Chenyu You, Dujian Ding (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-2503-10084.txt.

Description

Xiang Zhang and co-authors give a theory of why prompt wording steers language models so strongly, in this 2025 paper. Step-by-step prompting helps models reason, but a single stock phrase is often used for all tasks. That forces the model to search a vast space of paths on its own.

The authors argue that a prompt acts as a selector, pulling task-relevant facts from the model's hidden state. Each prompt sets a unique path through the answer space. They measure how large that search space is and show why stock phrases can badly hurt scores. In tests, a well-searched prompt raised reasoning scores by more than half.

Catalog record

CREATORXiang Zhang, Juntai Cao, Jiaqi Wei, Chenyu You, Dujian Ding
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-10084

Browse the full library · About the Tower of Babel