Many times our system will respond to your query with something like this:
In the above example, the document "How to Ask a Multiple-Choice Question" had not yet been uploaded, so there was no document specific to this subject available.
However, sometimes it is important to note that you will mostly get this response because the AI cannot determine an answer to your question based upon the way it is worded. In other words, there may actually be an answer, possibly even in the context documents displayed (since those are determined by cosine similarity vector searches), but the AI simply unable to associate your question with the returned documents.
In cases like this, we need to "rephrase" the question. Re-word it, if you will. And sometimes, "re-think" it.
For example, when the first iteration of the Real Estate Books AI was developed, I asked a real estate agent to give me a question she had recently asked of the DRE. She asked, "Can an agent take his clients with him?"
A disappointing first answer, to be sure. But, what the AI needed was more guidance, a more detailed question like, "Can a realtor take his firm's clients with him when he leaves the firm?"
Note the significant difference just by expanding on the same question.
However, this isn't the real question. When you think about it, the actual question she wanted to know was, "Are there regulations which prevent a realtor from taking his firm's clients with him when he leaves the firm?"
What makes this question a little tricky is the fact that the AI assumes this is a regulatory question when, in fact, it is an ethics question. This is why you always have to remember that the AI is not human, and is only responding to your query and the documents it receives as context to your query.
But, the point here is to demonstrate how rephrasing, re-wording or even re-thinking the question can make a huge difference in the response. In general, the more detailed your question, the better the response you will receive.
Now, re-thinking a question does not necessarily mean making major changes to it. Note the following practice exam question:
Look at the difference adding the one word "report" makes in the response:
Yes, "D" is the correct answer. Note that our AI is still trained, despite providing the correct answer, to recommend you double-check it's answer. It is a good habit to always do this.
To summarize:
- The AI assistant will sometimes respond that there is no specific document available to answer the question if it cannot determine an appropriate answer based on how the question is phrased.
- In these cases, it is important to rephrase or re-word the question to provide more context and details that will help the AI understand what information you are looking for.
- Providing more detailed questions guides the AI to give better responses. For example, expanding a vague question about whether an agent can take clients to specifically ask about regulations preventing this provides necessary context.
- Sometimes it is also helpful to re-think the question completely if the AI is answering a different question than what you intend. For instance, turning an ethics question into a regulatory question.
- Even small changes like adding a single word can change the AI's response and help it give the right answer.
- It's good practice to always double check the AI's answers, even when correct, to verify accuracy.
In summary, rephrasing, adding details, and clarifying intent helps the AI assistant understand questions better and provide more accurate responses. Checking the answers helps ensure quality responses.