Approval issue created: CONTRIB-9293

This is an excellent plugin. It allows us to customize a bot specific to each course, with a great deal of flexibility. Support from the developer has been outstanding, and they have continued to add features and value.

Hi i succeeded in install and setup with a free account but i think i misunderstood the settings. I have uploaded 2 documents, created metadata and filters. I had supposed that filters would constrain the Assistant to use as source only the documents selected when i setup the filter in the block. But i can ask about any subject and he will allucinate using parts of the document to try to create a reasonable explanation. So, what should i do to prevent this behaviour? Couldnt he simply answer that his sources does not contain info about that question?

Hello Paula,

Great questions! Yes, with the prompt engineering feature you can give your chatbots special instructions like "If you don't know the answer, say `Hmmm, I'm not sure. Please message the Teacher.`". We've found, however, that in most cases where the chatbot has hallucinated a response, the cause has been misconfigured filters that result in the chatbot not having access to ANY data. Usually if the chatbot has access to data, even if the full answer is not contained in that data, the chatbot will provide very acceptable answers, noting lack of information where possible. It can really struggle when the filters remove all sources of information though. --So, I'd first recommend making sure your filters are working properly.

I would suggest (at least temporarily) turning on the "show sources" feature, if you don't already have it enabled. Then when you ask it a question, the chatbot's response should have a little "i" icon in the bottom right corner. Clicking that icon will expand the response to show snippets of source related to the response. If no "i" icon appears, that means the chatbot didn't find any relevant source material (often because the filters didn't give them access to it). You can use this to determine if your filters are giving your bots access to the information you expect.

If you find you are having difficulty getting the filters working (or want help with prompt engineering), please don't hesitate to request support using our website's contact forms. The plugin's comment section is a bit unwieldy compared to email or a zoom call.

Hope that help,

Thank you David, i wil go to the site an be more precise. Just to leave a short trace here: the plugin has access to the contents; It seems that on free account we do not have the possibility at the moment to activate prompt engineering; it is allucinating because it is consulting other sources as well. It is not restricting the sources to the ones uploaded. I have stressed it, of course, i am testing

Saya mencoba menggunakan plugin ini dengan tingkatan gratis. Namun kendala pada chat, ketika membuat pertanyaan. Muncul informasi chat: Sorry, there seems to be something wrong with my network connection. Please try again later.
Mohon arahannya Pak David. Terima kasih