Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ ZENODOarrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
ZENODO
Report
Data sources: ZENODO
addClaim

Prompt Clarity, Correction Churn, and Token Efficiency

Authors: Sonntag, Barrett;

Prompt Clarity, Correction Churn, and Token Efficiency

Abstract

An evaluation of whether stating requirements and output constraints in the initial prompt reduces correction exchanges and tokens per solved task. The study runs 14 machine-checkable tasks across lazy and front-loaded prompt arms, thinking off and default adaptive thinking, with five repetitions per cell and one verifier-guided correction turn after failures. Across 280 first turns and 120 corrections, front-loaded prompts used 45,463 recorded tokens versus 160,799 for lazy prompts, while solving 134 of 140 repetitions versus 96 of 140. The paper reports the method, verifier design, task-category results, counterexamples, limitations, and reproducibility artifacts.

Powered by OpenAIRE graph
Found an issue? Give us feedback