So, you might be constructing a RAG system or utilizing an LLM to talk with paperwork. However customers usually ask: how can we belief the solutions?
Furthermore, we regularly hear about hallucinations, which undermine customers’ belief.
If we construct an utility however fail to indicate customers the place the solutions come from, the appliance may turn out to be unusable in some circumstances.
On this article, I’ll share an strategy to deal with this concern. By linking each reply generated by the LLM to its supply textual content within the doc, we are able to construct transparency and belief. This methodology not solely offers clear proof for the solutions but in addition permits customers to confirm the outcomes immediately inside the PDF.
Typically, the generated reply will not be completely correct, however having the ability to find the right supply textual content is already useful for the person.
Let’s take an instance of this paper from arxiv.org. We will think about this use case:
Step one on this strategy is to extract the textual content from the PDF in a structured format.
