When I tried to edit the information chunks in the knowledge base, it told me I needed to upgrade to the PRO plan. (I’m currently on the Gold plan.)
What matters most to me is the correct usage of the information in the knowledge base.
I’m working on an AI application that I anticipate could have thousands or even tens of thousands of monthly users, and the knowledge base mostly contains content such as books, author articles, biographies, etc.
Sometimes, the information chunks aren’t the way I want them to be. For instance, when a user asks about a particular author, the system often gives incomplete or incorrect information. However, if a user asks about a concept used by the same author, the answer is often quite good.
In this project, even a simple error or missing information could be ridiculed by thousands of people. So for me, the real problem isn’t hallucination — the main issue is that the information chunks are either cut off or not properly aligned with the user’s question. Sometimes, the system even tries to answer using unrelated chunks.
1. My first question: If I upgrade to the PRO plan, will I be able to better control or improve the length and quality of the information chunks?
2. My second question: Is there any way to configure this better under the Gold plan?
Yeah, upgrading to the Pro Plan is the most direct way to get more granular control over your KB chunks. You also get 4500 testing credits every month, more user memories per studio, unlimited studios, access to smarter LLMs, you unlock 8% take rate (vs. the 10% on the Gold plan), and much more. However, if you’re just starting out and you’re still testing the waters, it might not be the best plan for you at the moment. Maybe hold until you get at least 20-30 paying user’s then consider upgrading.
There’s a possible MVP workaround available in both the Gold Plan and the Pro, but it’s not necessarily better. Under the Configure tab, you can use what’s known as “Training Dialogue.” This lets you give your chatbot specific examples of how it should respond to user queries. By providing a few perfect question-and-answer pairs for the topics you’re having trouble with (like the author bios), you can guide the AI to give the right answer, even if the underlying knowledge chunks aren’t perfectly aligned.
Essentially, take what you want to chunk and add it to the dialogue.
