6thWave: AI News Hub

More companies are incorporating retrieval augmented generation (RAG) systems in their technology stack, and new enhancements are emerging to optimize these systems. Vector database company Qdrant has developed a new search algorithm, BM42, which promises to make RAG more efficient and cost-effective. Founded in 2021, Qdrant aims to offer hybrid search capabilities by combining semantic and keyword search. Co-founder and CTO Andrey Vasnetsov explained that BM42 is an update to the BM25 algorithm, traditionally used to rank the relevance of documents in search queries. Since RAG works with chunks of information, BM25 is no longer suitable. BM42 uses a language model to extract information from documents, turning it into tokens that are then scored for relevance. This allows Qdrant to accurately pinpoint information needed to answer queries. While BM42 is not the only method aimed at improving hybrid search and RAG, it is considered more cost-efficient compared to alternatives like Splade. As RAG becomes a crucial topic in enterprise AI, companies like Microsoft and Amazon are offering infrastructure for building RAG applications, and OpenAI has acquired Rockset to enhance its RAG capabilities. Despite the advancements, RAG systems still face challenges such as susceptibility to hallucinations.

Source.