RCScore: Quantifying Response Consistency in Large Language Models
This addresses the need for more robust LLM evaluations for real-world deployments, though it is incremental as it builds on existing benchmarking methods.
The paper tackles the problem of LLM evaluations overlooking sensitivity to instruction style by introducing RCScore, a framework that quantifies how instruction formulation affects responses, showing that instruction style can shift accuracy by up to 16.7% points across ten LLMs on four reasoning benchmarks.
Current LLM evaluations often rely on a single instruction template, overlooking models' sensitivity to instruction style-a critical aspect for real-world deployments. We present RCScore, a multi-dimensional framework quantifying how instruction formulation affects model responses. By systematically transforming benchmark problems into multiple instruction styles, RCScore reveals performance variations undetected by conventional metrics. Our experiments across ten LLMs on four reasoning benchmarks demonstrate that instruction style can shift accuracy by up to 16.7% points. We introduce Cross-Response Similarity (CRS), a method applying RCScore metrics to measure stylistic self-consistency, and establish its strong correlation with task accuracy, suggesting consistency as a valuable proxy for model reliability. Additional findings show that deterministic decoding produces more stylistically stable outputs, and model scale correlates positively with cross-style consistency. RCScore offers a principled approach to assess instruction robustness.