Tag
This paper introduces MAGE, a framework for analyzing component interactions in prompt optimization, revealing a Prompt Optimization Coupling Effect (POCE) where combining multiple stochastic signals improves performance but increases variance. It provides insights into stability-performance trade-offs and practical guidance for prompt optimization.