‘Knowledge Compilation is the New RAG’, as RAG is deemed to be obsolete by the firm that built the vector database category.

← Back to Blog

Knowledge Compilation is the New RAG’, as RAG is deemed to be obsolete by the firm that built the vector database category. Pinecone has moved the debate on from RAG, which burns tokens to figure out understand data at query time. The move is now to focus on precompiling task-specific artefacts and querying those. Lifting

Pinecone’s argument is that handing raw chunks to a frontier model and hoping the model figures it out is fragile, slow, and expensive.

Or to put it another way, the AI may get the reasoning wrong due to both the context window, which is more expensive to chunk and reason, and reasoning should be grounded on knowledge. But lifting the level of reasoning up to task-specific, based on information rather than ‘raw data,’ has its issues. Least of all, the AI precompiles task-specific artefacts, and precompiling requires AI to base its compilation on typed, cited, task-specific artefacts. AKA known grounded sources of data.

However, this approach lends itself to domain-specific, ontologically grounded AI models. Those required for AI to address complex reasoning, including Audit Assurance and cybersecurity risk management.

@Parava Security Solutions

Liz Sandwith CFIIA Ryan Buckner Faisal Khan Michael G. Jacobides Georgios Alevras

https://lnkd.in/dkA2QM2T

View Original Post
https://www.linkedin.com/feed/update/urn%3Ali%3Ashare%3A7458423612048441344