We've productionized query-aware compression for faster, cleaner, more-accurate search.
Perplexity has developed a new method for Retrieval-Augmented Generation (RAG) that prioritizes query-aware context compression. This approach significantly reduces the amount of text processed by cutting context tokens by up to 70%, while simultaneously improving answer quality and reducing noise. The company claims this leads to a 63% increase in vital content per snippet and maintains frontier-level performance with a 50x compression ratio on SimpleQA. AI
IMPACT Perplexity's new RAG technique could lead to more efficient and accurate AI-powered search experiences.