Extending Your Agent with RAG to Prevent Hallucinations
In earlier chapters, we saw what an LLM is, and in the previous chapter, we saw how it can control different tools to succeed at completing a task. However, some of the limitations of LLMs prevent their deployment in sensitive fields such as medicine. For example, LLMs crystallize their knowledge at the time of training, and rapidly developing fields such as medical sciences cause this knowledge to be outdated in a short time. Another problem that has emerged with the use of LLMs is that they can often hallucinate (produce answers that contain factual or conceptual errors). To overcome these limitations, a new paradigm has emerged: retrieval-augmented generation (RAG). RAG, as we will see in this chapter, allows for the LLM to refer to memory that is external to the model; thus, it allows knowledge to be found and kept updated. Similarly, providing contextual guidance to the model’s response allows for the reduction of...
