Our ability to continuously acquire, organize, and leverage knowledge is a key feature of human intelligence that AI systems must approximate to unlock their full potential. Given the challenges in continual learning with large language models (LLMs), retrieval-augmented generation (RAG) has become the dominant way to introduce new information. However, its reliance on vector retrieval hinders its ability to mimic the dynamic and interconnected nature of human long-term memory. Recent RAG approaches augment vectorembeddings with various structures like knowledge graphs to address some of these gaps, namely sense-making and associativity. However, their performance on more basic factual memory tasks drops considerably below standard RAG. We address this unintended deterioration and propose HippoRAG 2, a framework that outperforms standard RAG comprehensively on factual, sense-making, and associative memory tasks. HippoRAG 2 builds upon the Personalized PageRank algorithm used in HippoRAG and enhances it with deeper passage integration and more effective online use of an LLM. This combination pushes this RAG system closer to the effectiveness of human long-term memory, achieving a 7% improvement in associative memory tasks over the state-of-the-art embedding model while also exhibiting superior factual knowledge and sense-making memory capabilities. This work paves the way for non-parametric continual learning for LLMs. Code and data are available at https://github.com/OSU-NLP-Group/HippoRAG.

Humans are excellent at learning continuously—integrating new experiences into memory and drawing on them later. In contrast, today's large language models (LLMs) are static: they can’t incorporate new information unless retrained on massive datasets. To work around this, LLMs often rely on external search tools to access up-to-date knowledge. While effective in simple cases, this approach falters as the volume of information grows and tasks become more complex.Our new system, HippoRAG 2, tackles this limitation by giving LLMs a more human-like memory. It combines existing methods for structuring new information into a web of entities and relations between them a knowledge graph with a powerful algorithm called Personalized PageRank, which ranks different parts of the graph based on their relevance to a user's query. Unlike prior methods which also allow LLMs to structure new information, which sometimes hurt performance in simpler tasks, HippoRAG 2 improves LLM performance across both simple and complex tasks. Though challenges remain, this approach marks a step toward LLMs that can learn continuously and efficiently, without needing full retraining.