Naive RAG beats complex agentic memory on conversation benchmarks
Original title: Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We Thought
Raw-RAG ranked first in six of seven settings, ahead of the best agentic system by 0.13 to 5.53 points.
Researchers test Raw-RAG, dense retrieval and reranking over raw conversation history, against agentic memory systems that mutate or reorganize memory. On LoCoMo, LongMemEval, MEME, and MemoryAgentBench, broader coverage of needed evidence offset weaker use of that evidence. Even full access to transformed memory stores left an accuracy gap versus raw history.
- Raw-RAG feeds retrieved raw history to a reader LLM.
- It placed first in six of seven benchmark-model settings.
- The lead over the strongest agentic system was 0.13 to 5.53 points.
- Useful evidence includes answer values, relations, and dependency paths.
From the original article
1 min read at huggingface.co
LLM memory is widely used in long-term conversation understanding. Agentic memory systems often use mutation, structural organization, learned retrieval strategies, or other complex operations. Is this complexity always necessary?
By Zhengye Bao, Shengtao Zhang, Haokai Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.