Learning to Select In-Context Demonstration Preferred by Large Language Model

TL;DR

This paper introduces GenICL, a generative framework that uses LLM feedback to directly optimize demonstration selection for in-context learning, significantly improving performance across diverse tasks.

Contribution

GenICL is the first method to leverage LLM feedback for directly optimizing demonstration selection in ICL, outperforming retrieval-based approaches.

Findings

GenICL outperforms existing methods on 19 datasets across 11 task categories.

Direct optimization via LLM feedback improves demonstration quality for ICL.

GenICL enhances in-context learning performance significantly.

Abstract

In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks during inference using only a few demonstrations. However, ICL performance is highly dependent on the selection of these demonstrations. Recent work explores retrieval-based methods for selecting query-specific demonstrations, but these approaches often rely on surrogate objectives such as metric learning, failing to directly optimize ICL performance. Consequently, they struggle to identify truly beneficial demonstrations. Moreover, their discriminative retrieval paradigm is ineffective when the candidate pool lacks sufficient high-quality demonstrations. To address these challenges, we propose GenICL, a novel generative preference learning framework that leverages LLM feedback to directly optimize demonstration selection for ICL. Experiments on 19 datasets across 11 task categories demonstrate that GenICL achieves superior performance than existing methods in selecting the most effective demonstrations, leading to better ICL performance.