Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agent-Centric Projection of Prompting Techniques and Implications for Synthetic Training Data for Large Language Models

Dhruv Dhamani, Mary Lou Maher

TL;DR

Problem: There is no unified framework to characterize prompting techniques or relate them to multi-agent LLM systems. Approach: The paper introduces linear and non-linear contexts and develops an agent-centric projection to connect prompting methods with multi-agent architectures, accompanied by three conjectures. Contributions: formal classification of contexts, demonstration of projection between single-LLM prompting and multi-agent systems, and implications for synthetic training data generation. Significance: The framework enables cross-domain transfer of results and offers new directions for designing, training, and data-augmentation strategies for future LLM systems.

Abstract

Recent advances in prompting techniques and multi-agent systems for Large Language Models (LLMs) have produced increasingly complex approaches. However, we lack a framework for characterizing and comparing prompting techniques or understanding their relationship to multi-agent LLM systems. This position paper introduces and explains the concepts of linear contexts (a single, continuous sequence of interactions) and non-linear contexts (branching or multi-path) in LLM systems. These concepts enable the development of an agent-centric projection of prompting techniques, a framework that can reveal deep connections between prompting strategies and multi-agent systems. We propose three conjectures based on this framework: (1) results from non-linear prompting techniques can predict outcomes in equivalent multi-agent systems, (2) multi-agent system architectures can be replicated through single-LLM prompting techniques that simulate equivalent interaction patterns, and (3) these equivalences suggest novel approaches for generating synthetic training data. We argue that this perspective enables systematic cross-pollination of research findings between prompting and multi-agent domains, while providing new directions for improving both the design and training of future LLM systems.

Agent-Centric Projection of Prompting Techniques and Implications for Synthetic Training Data for Large Language Models

TL;DR

Problem: There is no unified framework to characterize prompting techniques or relate them to multi-agent LLM systems. Approach: The paper introduces linear and non-linear contexts and develops an agent-centric projection to connect prompting methods with multi-agent architectures, accompanied by three conjectures. Contributions: formal classification of contexts, demonstration of projection between single-LLM prompting and multi-agent systems, and implications for synthetic training data generation. Significance: The framework enables cross-domain transfer of results and offers new directions for designing, training, and data-augmentation strategies for future LLM systems.

Abstract

Recent advances in prompting techniques and multi-agent systems for Large Language Models (LLMs) have produced increasingly complex approaches. However, we lack a framework for characterizing and comparing prompting techniques or understanding their relationship to multi-agent LLM systems. This position paper introduces and explains the concepts of linear contexts (a single, continuous sequence of interactions) and non-linear contexts (branching or multi-path) in LLM systems. These concepts enable the development of an agent-centric projection of prompting techniques, a framework that can reveal deep connections between prompting strategies and multi-agent systems. We propose three conjectures based on this framework: (1) results from non-linear prompting techniques can predict outcomes in equivalent multi-agent systems, (2) multi-agent system architectures can be replicated through single-LLM prompting techniques that simulate equivalent interaction patterns, and (3) these equivalences suggest novel approaches for generating synthetic training data. We argue that this perspective enables systematic cross-pollination of research findings between prompting and multi-agent domains, while providing new directions for improving both the design and training of future LLM systems.
Paper Structure (11 sections, 3 figures)

This paper contains 11 sections, 3 figures.

Table of Contents

  1. INTRODUCTION
  2. DEFINITIONS AND PRIOR WORK
  3. Minimal Task-oriented LLM System
  4. Prompting Techniques
  5. FRAMEWORK AND CONJECTURES
  6. Linear and Non-linear Context in LLM Systems
  7. Agent-Centric Projection of Prompting Techniques
  8. Implications for Synthetic Training Data
  9. CONCLUSIONS
  10. Core Arguments
  11. Implications for Future Research

Figures (3)

  • Figure 1: A bare-bones LLM system.
  • Figure 2: A minimal LLM system that includes a context store.
  • Figure 3: An example of a prompting technique with non-linear context.

Theorems & Definitions (3)

  • Conjecture 1
  • Conjecture 2
  • Conjecture 3