This repository provides a curated reading list and survey of research on Large Language Model (LLM) based agents, focusing on tool use, planning, and feedback learning. It is targeted at researchers and practitioners in the AI and NLP fields looking for a structured overview of key paradigms and influential papers, with a primary contribution being a survey accepted at CoLing 2025.

How It Works

The project categorizes LLM agent research into a unified framework based on LLM-Profiled Roles (LMPRs). It prioritizes high-quality papers from top-tier conferences (ICML, ICLR, NeurIPS, ACL, COLING) and includes insightful unpublished works marked with a '💡' emoji. The survey also offers links to OpenReview for paper reviews and highlights specific sub-areas like search workflows, planning, tool use, and feedback learning.

Quick Start & Requirements

This is a curated list of papers and does not involve code execution or installation.

Highlighted Details

• Features a survey accepted at CoLing 2025: "A Review of Prominent Paradigms for LLM-Based Agents: Tool Use (Including RAG), Planning, and Feedback Learning."
• Includes a separate survey on LLM Test-Time Compute via Search, targeting TMLR 2025.
• Organizes papers by key agent capabilities: Tool Use, Planning (Base Workflows, Search Workflows, Decomposition), Feedback Learning, Composition, and World Modeling.
• Provides links to papers, code repositories, and related surveys for deeper exploration.

Maintenance & Community

The repository was last updated on March 14, 2025. The author notes a shift in research focus to LLM Inference via Search (LIS) and may not actively update this repo, suggesting other actively updated repositories for the latest papers in the general LLM agent field.

Licensing & Compatibility

The repository itself is a collection of links and does not appear to have a specific license attached. The licensing of the linked papers would be governed by their respective publishers.

Limitations & Caveats

The author indicates a reduced focus on actively updating this repository due to a shift in research interests, suggesting that users seeking the absolute latest advancements in general LLM agents might need to consult other resources. Some papers are marked as unpublished or under review, meaning their content may evolve.