From the course: Hands-On AI: RAG using LlamaIndex

Unlock the full course today

Join today to access over 25,200 courses taught by industry experts.

FLARE

FLARE

- [Instructor] Let's continue on our journey with advanced RAG by talking about FLARE, Forward-Looking Active REtrieval Augmented Generation. Most RAG methods only retrieve information once based on the initial user query. This is limiting for generating longer outputs where we need to continually gather relevant information throughout the generation process. As you know, large language models have a tendency to hallucinate and generate factually incorrect information, especially when it comes to longer outputs. And so there's a need for the language model to kind of actively decide when and what to retrieve at each step of the generation process, rather than just retrieving once upfront. This is where active retrieval augmented generation comes into play. This was introduced in a paper from, I believe, 2023, as it states here, on archive called Active Retrieval Augmented Generation. And this approach is promising because it's looking to enhance the factual accuracy of a LLM by…

Contents