Awesome-Self-Improving-Agents: A Curated List for Agentic AI Self-Improvement

Introduction

The Awesome-Self-Improving-Agents repository, maintained by selfimproving-agent, is a meticulously curated and continuously evolving resource hub dedicated to the field of self-improvement in foundation model-based agentic systems. This repository provides a central collection of essential materials, including research papers, benchmarks, insightful blogs, podcasts, interviews, videos, workshops, and courses. It serves as a companion to the survey paper "Self-Improvements in Modern Agentic Systems: A Survey", offering a deep dive into the definition and scope of foundation model-based agents, self-improvement loops, and the distinct categories of foundation model improvement and scaffolding improvement.

Why Use It & Key Benefits

This repository offers unparalleled value for anyone interested in the rapidly advancing domain of self-improving AI agents. Its key benefits include:

• Comprehensive Overview: Provides a structured and exhaustive collection of resources, making it easy to grasp the current state and future directions of agentic AI.
• Structured Taxonomy: Features a clear taxonomy that categorizes self-improving agents along two orthogonal axes: Foundation Model Improvement and Scaffolding Improvement, aiding in a deeper understanding of the mechanisms involved.
• Up-to-Date Literature: Continuously updated with the latest research, including upcoming publications, ensuring users have access to the most recent advancements.
• Evaluation and Benchmarking Insights: Offers dedicated sections on how agent progress is measured and groups benchmarks by improvement mechanism and application domain, crucial for assessing agent performance.
• Diverse Resource Formats: Beyond academic papers, it includes blogs, podcasts, and videos, catering to different learning preferences and providing broader context.
• Community Engagement: Facilitates connection with other researchers and enthusiasts through its Discord community.

How to Explore

As an awesome list, Awesome-Self-Improving-Agents does not require any traditional installation. To benefit from this valuable resource:

1. Visit the GitHub Repository: Navigate directly to the repository's GitHub page to browse its contents.
2. Read the README.md: The README.md file provides a comprehensive, structured overview of the entire list, including definitions, taxonomy, and direct links to resources.
3. Explore the Survey Paper: For a foundational understanding, delve into the accompanying survey paper, linked within the repository, which details the theoretical underpinnings.
4. Engage with the Community: Join the Discord server to discuss topics, ask questions, and connect with other members of the self-improving agents community.

Examples

The repository meticulously organizes key literature and resources. Here are a few examples of the types of content you'll find:

• Foundation Model Improvement: Discover papers like Self-Instruct: Aligning Language Models with Self-Generated Instructions (2023), Constitutional AI: Harmlessness from AI Feedback (2022), and RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation (2023), illustrating how foundation models can enhance themselves through various feedback mechanisms and interactions.
• Scaffolding Improvement: Explore advancements in Prompt Optimization with works such as Self-Refine: Iterative Refinement with Self-Feedback (2023), Memory enhancements seen in Generative Agents: Interactive Simulacra of Human Behavior (2023), and Tool integration exemplified by Voyager: An Open-Ended Embodied Agent with Large Language Models (2023).
• Evaluation & Benchmarking: Understand how agents are assessed through benchmarks like SWE-Bench for software engineering tasks and Mind2Web for web navigation, providing critical insights into agent capabilities and progress.
• Related Resources: Access a rich collection of supplementary materials, including blogs from Anthropic and Google DeepMind, podcasts featuring AI pioneers like Jürgen Schmidhuber, and academic courses from institutions like Stanford University.

Links

• GitHub Repository: https://github.com/selfimproving-agent/Awesome-Self-Improving-Agents
• Survey Paper (arXiv): https://arxiv.org/abs/2607.13104
• Project Page: https://selfimproving-agent.github.io/
• Discord Community: https://discord.gg/fn5rYJhgaz