LatentPrompt: Optimizing Promts in Latent Space

Mateusz Bystroński, Grzegorz Piotrowski, Nitesh V. Chawla, Tomasz Kajdanowicz · 2025

Read LatentPrompt: Optimizing Promts in Latent Space by Mateusz Bystroński, Grzegorz Piotrowski, Nitesh V. Chawla, Tomasz Kajdanowicz (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-2508-02452.txt.

Description

Mateusz Bystroński, Grzegorz Piotrowski, Nitesh Chawla, and Tomasz Kajdanowicz tune prompts in a hidden space in a 2025 paper. The prompt, the text given to a language model, strongly shapes its answers. Most tuning methods rely on hand rules or trial and error. Their framework, LatentPrompt, automates the search without fixed rules.

Seed prompts are mapped into a smooth semantic space, where nearby points hold similar meaning. The system explores that space, tests new prompts, and keeps the best scorers. In a trial on a finance sentiment task, accuracy rose by about three points after one round. The method needs only black-box access to a model plus an automatic score. That makes it portable across tasks and fields.

Catalog record

CREATORMateusz Bystroński, Grzegorz Piotrowski, Nitesh V. Chawla, Tomasz Kajdanowicz
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-02452

Browse the full library · About the Tower of Babel