Databricks says its Instruction Retrieval offers better AI answers than RAG in the enterprise

"Databricks is joining the AI software vendors quietly admitting that old-fashioned deterministic methods can perform much better than generative AI's probabilistic approach in many applications. Its new "Instructed Retriever" architecture combines old-fashioned database queries with the similarity search of RAG ( retrieval-augmented generation) to offer more relevant responses to users' prompts. Everything about retrieval-augmented generation (RAG)'s architecture was supposed to be simple."

"But as enterprises push AI systems closer to production, that architecture is starting to break down. Real-world prompts come with instructions, constraints, and business rules that similarity search alone cannot enforce, forcing CIOs and development teams into trade-offs between latency, accuracy, and control. Databricks has an answer to that problem, Instructed Retriever, which breaks down requests into specific search terms and filter instructions when retrieving documents to augment the generative prompt."

Databricks introduced Instructed Retriever, combining deterministic database queries with similarity search to deliver more relevant responses. RAG retrieves similar documents and relies on a language model to reconcile prompts with retrieved data. Real-world prompts include instructions and business rules that similarity search alone cannot enforce, creating trade-offs among latency, accuracy, and control. Instructed Retriever breaks requests into explicit search terms and filters to enforce constraints during retrieval, for example restricting reviews to those under a year based on metadata. Databricks reports improved performance over RAG and faster paths to production, while analysts warn the approach can reveal data, governance, and budget gaps CIOs must address.

Read at InfoWorld