dosubot[bot]
commented
1 year ago 🤖
Hello,
Thank you for your detailed feedback and for using Quivr. I understand your concerns and I'll try to address them as best as I can.
1) Regarding the issue of the bot asking multiple questions at a time, this might be related to the AI model being used. If you're using GPT-3, it might not strictly adhere to the "ask only one question at a time" rule. However, you mentioned that you manually added GPT-4 and it seemed to fix this issue. This is because different AI models have different behaviors and capabilities, and GPT-4 might be better at following the rule you set.
2) As for the bot not remembering previous responses, Quivr does have the ability to remember previous responses. This is done through the generate_answer method in the backend, and the chat configurations are stored and retrieved from the local storage using the saveChatsConfigInLocalStorage and getChatsConfigFromLocalStorage functions respectively (source, source).
However, it seems like the bot is not utilizing this feature properly in your case. This might be due to the way the chat context is being managed. Currently, Quivr uses React's Context API to maintain the context of a conversation (source). If the context is not being updated or retrieved correctly, it could lead to the bot forgetting previous responses.
I found a similar issue titled Chat Memory Problem - instant-Forgot which was solved. The solution was implemented in the code, so it might be worth checking if your version of Quivr is up-to-date. The latest version is v0.0.85.
I hope this helps. If you have any other questions or need further clarification, please don't hesitate to ask.
Best, Dosu
Sources
- Chat Memory Problem - instant-Forgot
- backend/llm/qa_base.py
- frontend/lib/api/chat/chat.local.ts
- frontend/lib/context/ChatsProvider/hooks/useChatsContext.tsx
This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.
Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.
If you want to continue the conversation, start your reply with @dosu-bot.
JJRobertsSex30
B0rrA
github-actions[bot]
eWizardII
The Feature
Ok so here is the custom prompt I am using
MISSION
You are a patient intake chatbot focusing on symptoms. Your mission is to ask questions to help a patient fully articulate their symptoms in a clear manner. Your chat transcript will ultimately be translated into chart notes.
RULES
Ask only one question at a time. Provide some context or clarification around the follow-up questions you ask. Do not converse with the patient. When the user types "DONE" in all caps then
============================================ Ok so a few issues with this
1) Unless I manually add gpt 4 as a selectable then choose GPT 4 as the engine then it ignores "ask only one question at a time" and multiple questions are asked
2) After fixing that by manually adding and then using GPT 4 the bot then never remembers anything. In other words if I tell it about a stomach problem it might ask how long have I had it. then I answer by saying something like "2 years" it then asks me another question. Then I answer that and, as a follow up question it asks how long have I had the stomach problem even though I already told it "2 years". Do quivr chat bots have no memory ?
Motivation, pitch
To give chatbots memory
Twitter / LinkedIn details
No response