Augmenting LLM Reasoning with Dynamic Notes Writing for Complex QA

Hong-Wei Wu published on 2025-06-28 included in category Paper Introduction

Discover NotesWriting, a simple yet effective technique to boost LLM performance in Multi-Hop RAG. Learn how it refines retrieved documents to create a clean, focused context, solving key challenges in complex question answering.

AutoMind: Adaptive Knowledgeable Agent for Automated Data Science

Hong-Wei Wu published on 2025-06-28 included in category Paper Introduction

Explore AutoMind, a state-of-the-art AI agent designed to master complex data science challenges. This deep dive breaks down its powerful LLM-based framework, which uses an expert knowledge base, agentic tree search, and adaptive coding to achieve top results in Kaggle competitions with remarkable efficiency. Learn how the future of automated data science is being shaped.

OctoTools: An Agentic Framework with Extensible Tools for Complex Reasoning

Hong-Wei Wu published on 2025-06-02 included in category Paper Introduction

Introducing Agentic AI: OctoTools! Understand how OctoTools enhances the performance of LLM Agents on complex tasks through its well-defined and extensible Tool Cards, and the ingenious interplay between its Planner and Executor.

Steering Large Language Models Between Code Execution and Textual Reasoning

Hong-Wei Wu published on 2025-05-25 included in category Paper Introduction

Explore an ICLR 2025 paper on guiding Large Language Models (LLMs) between code execution and textual reasoning. Learn why models like GPT-4o may prefer text-based approaches, sometimes leading to errors, and how combining both reasoning methods yields the best performance.

Pre-Act: Multi-Step Planning and Reasoning Improves Acting in LLM Agents

Hong-Wei Wu published on 2025-05-20 included in category Paper Introduction

Discover Pre-Act, an approach enhancing Large Language Model (LLM) agent performance through multi-step planning and reasoning. Learn how Pre-Act overcomes ReAct's limitations in long-term planning by generating and modifying plans at each thinking step, thereby improving acting capabilities.

MemGPT: Towards LLMs as Operating Systems

Hong-Wei Wu published on 2025-05-18 included in category Paper Introduction

Explore MemGPT, an innovative approach treating Large Language Models (LLMs) as operating systems. This article explains how MemGPT overcomes LLM long-term memory and context window limitations through its core 'Prompt Compilation' technique and unique memory management mechanisms (Core Memory, Recall Memory, Archival Memory) to enable more persistent conversational interactions.

Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory

Hong-Wei Wu published on 2025-05-13 included in category Paper Introduction

Explore the paper 'Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory.' Learn how Mem0 addresses long-term memory challenges in LLMs, detailing its core architectures (Mem0 & Mem0^g), memory management techniques, and experimental results on the LOCOMO dataset.