#retrieval-augmented generation
Retrieval-Augmented Generation: 28 AI articles covering retrieval-augmented generation news, analysis, and research
Articles
LEANN: The Open-Source Vector Index That Shrinks a 60M-Chunk RAGβ9
LEANN is UC Berkeley's open-source vector index that shrinks a 60M-chunk RAG index from 201 GB to 6 GB by recomputing embeddings at query time.
A Small Model Asks Better: Fine-Tuning Query Clarification forβ9
ClariEdge, a new dataset of ~3,900 examples, shows a lightweight fine-tuned LoRA detects ambiguous RAG queries with 96.6% accuracy and generates better clarific...
Perplexity Releases pplx-embed-v2-context-9b-preview: Aβ10
Perplexity releases pplx-embed-v2-context-9b-preview, a contextual embedding model that retrieves answers along with their supporting evidence.
How to Build a GraphRAG System with Python, Neo4j, and ServiceNowβ8
Build a GraphRAG system with Python, Neo4j, and ServiceNow: load 11,891 CIs and 68,900 tickets, then benchmark eight retrieval methods for accurate AI answers.
BioPhys-Bridge: A Benchmark for Interdisciplinary Scientificβ8
BioPhys-Bridge is a new benchmark of 500 biophysics cases and 1,517 tasks testing evidence-grounded, interdisciplinary scientific reasoning in language models.
Should You Read the Code? Is RAG Dead? Did Skills Kill MCP?β9
As AI coding tools evolve, we tackle three key questions: should you read generated code, is RAG dead, and did Agent Skills kill MCP? Get a practical 2026 persp...
Google Research Unveils Retrieve-for-Train (R4T): An RL-Compiledβ8
Google Research's Retrieve-for-Train (R4T) uses RL and a diffusion retriever to generate diverse query fan-outs 12β20Γ faster, avoiding paraphrastic collapse in...
R2VC: Modular Fact-Checking with Retrieval, Verification, andβ8
R2VC is a modular fact-checking pipeline combining retrieval, verification, and confidence calibration, achieving 13.74% higher accuracy on FEVER with reliable ...
PRO-Step: Step-level Process Reward Optimization for Retrievalβ9
PRO-STEP: A novel framework optimizing step-level retrieval and reasoning in RAG via process reward models and preference optimization, improving accuracy acros...
A RAG That Says 'Not in This Document' Has to Show Four Kinds ofβ9
A four-part RAG system must prove a "not found" answer with evidence from retrieval, generation, grounding, and verification to avoid misleading users.
