Issue with AI queries and Knowledge Base articles
# 🤝help
c
I am prompting the knowledge base with very direct article specific prompts (see attached images) and turning up no results. - All of my articles are indexed - There are multiple keywords in the article - They're all raw text files Is it possible to use ChatGPT 4 instead of 3.5 Turbo to search the knowledge articles? This is a fairly big limitation and a HUGE difference in how well the articles are evaluated, not to mention the quality of the response. Secondly, I noticed the knowledge base query does not seem to pull context from multiple articles and only returns the content from one. Is this the case? If so, that's another HUGE limitation. It is a notable difference between BotPress and TextCortex.
f
Hey there, You can select GPT-4 under the Knowledge Agent settings. It will not search in the articles, it will just be the LLM that generates the answer. It does a semantic search, then retrieves some chunks of context from the KB, which is given to the LLM to generate an answer. It should still improve the responses though!
c
@fresh-fireman-491 where do I find knowledge agent settings?
Oh
Duh
I'm stupid
f
Nono, just too fast
c
@fresh-fireman-491 under "additional context" can you incorporate prompt information as well or is that really designed for contextual understanding. For example, would it be beneficial to include a lengthy description of the product and all its features?
@fresh-fireman-491 another question - what types of content are best for the LLM? Previously we were using a RAG based platform and it was able to retrieve multiple knowledge sources then combine them into a more comprehensive response.
f
It's designed for adding context, so yeah, you can 100% add a long description, it will just increase the costs.
This is also rag. Unstructured data in docs or rich text. Structured data in Tables.

https://youtu.be/eWKnezZp9qk?si=qlGE3vm9Z-VOXFL4

3 Views