CHOIR: Collaborative Harmonization fOr Inference Robustness

TL;DR

CHOIR is a test-time framework that harmonizes multiple persona-conditioned reasoning signals in LLMs, improving inference robustness across demographics without additional training.

Contribution

It introduces CHOIR, a novel method that leverages persona variations as constructive signals to enhance reasoning robustness in LLMs at test time.

Findings

Up to 26.4% performance improvement for individual demographics.

Average 19.2% performance boost across five demographics.

Effective across various models, scales, and tasks without extra training.

Abstract

Persona-assigned Large Language Models (LLMs) can adopt diverse roles, enabling personalized and context-aware reasoning. However, even minor demographic perturbations in personas, such as simple pronoun changes, can alter reasoning trajectories, leading to divergent sets of correct answers. Instead of treating these variations as biases to be mitigated, we explore their potential as a constructive resource to improve reasoning robustness. We propose CHOIR (Collaborative Harmonization fOr Inference Robustness), a test-time framework that harmonizes multiple persona-conditioned reasoning signals into a unified prediction. CHOIR orchestrates a collaborative decoding process among counterfactual personas, dynamically balancing agreement and divergence in their reasoning paths. Experiments on various reasoning benchmarks demonstrate that CHOIR consistently enhances performance across demographics, model architectures, scales, and tasks - without additional training. Improvements reach up to 26.4% for individual demographic groups and 19.2% on average across five demographics. It remains effective even when base personas are suboptimal. By reframing persona variation as a constructive signal, CHOIR provides a scalable and generalizable approach to more reliable LLM reasoning.