Other Key Challenges

Key Takeaways

so let's talk about some of the other problems one of them is mtop ranked which means you are having the context documents but they are not appearing in the top retrieval results which leads to the model obviously not able to answer the question documents to answer the question are present in your vector database but your retrieval strategy is bad so it's not able to retrieve them not in context is another problem it means the documents with the answer present during the initial retrieval but let's say if you're applying a sophisticated retrieval process you are doing retrieval with let's say cosine similarity you are then filtering out some documents you are doing some reranking right you can have a very sophisticated retrieval strategy also but because of this you end up somehow not having the relevant documents at the top they either go to the bottom or they get removed that could happen so that's known as not in context where in the final context the relevant documents to answer the question somehow get removed another one is not extracted not extracted means the llm struggles to extract the correct answer from the provided context even if it has the answer which means to answer the question you have the information in the context retrieved but is not able to answer the question this sometimes occurs when you have context with a lot of information with a lot of noise contradicting statements and so on so there are ways in which we can tackle this too next one incorrect specificity so here the output response is too vague or it is not too detailed or specific enough it just gives you a very generic answer to your question now this can happen if you're asking very vague or generic queries obviously one way you can solve this immediately is make the llm rephrase your query in such a way that the retrieval might happen in a better way there are some other ways to tackle this also I'll talk about it shortly and of course wrong chunking or bad retrieval can lead to this problem so I have kind of grouped all of this one after the other because the solutions I will present you know can be used to tackle all of these challenges