I’m looking for help triggering the action correctly.
The chatbot is evaluating the responce as a confidence level, and if it is low enough I want the user input to be sent to the slack action so I can become aware of it and add to the known answers PDF file and update the chat bot knowledge.
I’m not able to get it to send to slack - not able to trigger the action.
I have a rule that tells it to use the action - and when the confidence level is below 80% the chat tells the user it is sending it to slack to improve knowledge (will probably remove that later as the user doesn’t need to know).
RULE
If the confidence of your response is below 80% use the action to send a message to slack including the user question.
Also, in the action the trigger prompt is:
When a response is provided with a confidence level below 80% use this action to send the user question to Slack.
I have tested with the default trigger prompt and I am getting input to slack, so the connection is working, I’m just not getting it triggered when the confidence level is low enough.
This needs to be done entirely through prompting. The confidence level is a good idea, but just recognize the AI will be making up a confidence level. These are language models, so there is not internal confidence level calculator. When I write prompts for actions like these I use phrasing like “If you aren’t sure about the answer”. Stuff like that.
That being said, the confidence level idea can work. Just recognize it’s made-up and the AI is making up its own rules!
Thank you @admin_mike – I’m starting to get the hang of this better and have it working now… thank you!!
Rule:
Do NOT MAKE UP answers. If you truly don’t know the answer to a question, tell the user you don’t know, use the action to send a message to slack with the question, and encourage them to reach out to our support team.
Actions trigger prompt:
When you don’t know the answer, use this action to send the most recent user input to slack.
Hi @cfdynamics, there is an action in Pickaxe that integrates with the Pinecone assistant. You can use one of the functions in Pinecone to evaluate the answers: