PRIME: Planning and Retrieval-Integrated Memory for Enhanced Reasoning

2025-09-26

AI Summary

This work addresses the inefficiency and inaccuracy of large language models (LLMs) on multi-hop and knowledge-intensive reasoning tasks. We propose PRIME, the first multi-agent framework that formally integrates the dual-process theory of human cognition—System 1 (intuitive, rapid reasoning) and System 2 (deliberative, sequential reasoning)—into LLM-based inference architectures. PRIME orchestrates dynamically coordinated agents—including fast-response, planning, hypothesis-generation, retrieval, integration, and decision modules—to enable adaptive inter-system scheduling and closed-loop information flow. Built end-to-end atop LLaMA-3, PRIME achieves state-of-the-art performance on multiple multi-hop reasoning benchmarks: it is the first open-weight model to match GPT-4/GPT-4o in accuracy, substantially narrowing the gap with top proprietary models. Our core contribution is the first scalable, interpretable dual-process coordination mechanism, establishing a novel paradigm for efficient and robust complex reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Multiagent Learning
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Inspired by the dual-process theory of human cognition from extit{Thinking, Fast and Slow}, we introduce extbf{PRIME} (Planning and Retrieval-Integrated Memory for Enhanced Reasoning), a multi-agent reasoning framework that dynamically integrates extbf{System 1} (fast, intuitive thinking) and extbf{System 2} (slow, deliberate thinking). PRIME first employs a Quick Thinking Agent (System 1) to generate a rapid answer; if uncertainty is detected, it then triggers a structured System 2 reasoning pipeline composed of specialized agents for extit{planning}, extit{hypothesis generation}, extit{retrieval}, extit{information integration}, and extit{decision-making}. This multi-agent design faithfully mimics human cognitive processes and enhances both efficiency and accuracy. Experimental results with LLaMA 3 models demonstrate that PRIME enables open-source LLMs to perform competitively with state-of-the-art closed-source models like GPT-4 and GPT-4o on benchmarks requiring multi-hop and knowledge-grounded reasoning. This research establishes PRIME as a scalable solution for improving LLMs in domains requiring complex, knowledge-intensive reasoning.

Problem

• Integrates fast intuitive and slow deliberate thinking processes
• Enhances reasoning efficiency and accuracy in language models
• Enables competitive performance on complex knowledge-intensive tasks

Innovation

• Integrates fast intuitive and slow deliberate thinking
• Uses multi-agent pipeline for planning and retrieval
• Enables open-source models to compete with closed-source ones