Controllable method improves adaptive retrieval and reasoning in large language models
CWM: Controllable White-Box Meta-Prompting for Adaptive Retrieval-Augmented Generation and Reasoning Ability
Artificial Intelligence
Summary
Large language models are very good at understanding and generating text, but most improvements focus on specific tasks only. The authors introduce a new method called Controllable White-Box Meta-Prompting (CWM) that works well across many different tasks involving retrieving information and reasoning. Unlike previous methods, CWM is transparent and adjustable without needing extra decision tools. It works well on several tests and can even regulate how the model retrieves information internally.
What this means in practice
- •For natural language processing engineers: Enhance adaptive retrieval capabilities in language generation systems without external decision modules by employing CWM for better task generality.
- •For ai product managers: Integrate controllable retrieval mechanisms into AI products to improve information accuracy and adaptivity across diverse user requests.$Commercial implications: CWM enables customized retrieval decisions for AI services, creating more reliable and versatile conversational agents for end users.
Authors
Keuntae Kim, Eunhye Jeong, Yong Suk Choi
Abstract
Recently, Large Language Models (LLMs) have gained significant attention due to their strong language understanding and generation capabilities, demonstrating impressive reasoning abilities as well as effective utilization of external knowledge. Many studies have proposed methods that specialize in improving performance for individual tasks. However, ironically, only a limited number of attempts have explored general-purpose, task-agnostic methods. In this work, we present a unified framework integrating reasoning and Retrieval-Augmented Generation (RAG) tasks. We further propose Controllable White-Box Meta-Prompting (CWM), a low-cost white-box method for adaptive RAG tasks previously dominated by black-box approaches, without requiring external decision modules or multi-sampling. CWM achieves state-of-the-art performance on three adaptive RAG benchmarks across recent LLMs, including GPT-oss-20b, Qwen3-14b, and Llama3.1-8b, while also demonstrating strong generality by extending to reasoning tasks. In addition, CWM provides controllability by enabling retrieval decisions to be regulated through the manipulation of internal model signals. Our code is available at https://github.com/JeongEunhye00/CWM.