What is 'agentic RAG' or self-RAG?
hardAnswer
- The LLM decides at each step whether it needs to retrieve, reformulates queries dynamically, and iteratively fetches more context.
- Contrast with 'static RAG' (one retrieval → one generation).
- Techniques: Self-RAG (Asai 2023), FLARE (Jiang 2023 — retrieve when the next token has low confidence), Corrective RAG (Yan 2024 — grade retrieved chunks, discard bad ones, do web search on fallback).
- Trade-off: variable / higher latency, higher cost, but much better on multi-hop and out-of-domain queries.
Check yourself — multiple choice
- Static RAG only
- LLM autonomously decides when + what to retrieve, reformulates, retries — Self-RAG, FLARE, Corrective RAG; better on multi-hop
- No decision-making
- Only for classification
Agentic RAG: LLM-driven retrieval loop → better multi-hop, higher latency.
#rag#agents
Practise LLMs & GenAI
214 interview questions in this topic.