Incident creation through Chat test LLM appears strngly
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
a week ago
Dear all,
I have setup a conversational assistant Twilio Agent in order to create an incident using a voice Agent.
The creation works great but wierd things happen
When testing this using chat input from Assistant Test button
This open an Agent panel where I can test my Agent either using voice or chat. For simplification I use chat mode first.
What is strange is as below :
- It ask me only 3 questions, while I am telling my LLM prompt to ask at least 5
- It did not ask me if the impact is affecting only me or all users but decide alone that it affect all users
- I ask the prompt to set the description as a summarization without Question/Answer format but It ignores it.
Here is below the prompt use :
And here below the resulting created record
As you can see from description it records it as Question/Answer format and the worst thing is that it has created 2 last question that it did not really ask me
Any idea how to improve and explain why it behaves like this ?
Regards
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
a week ago
In My view LIST of Steps Section needs to be More Clear and in Simple Short Lines.
DID You Tried Changing the LLM model and Test with same Configuration?
It Seems that LLM is Hellucinating- Instruction Line 2 and 4 Is Ambiguous - WHat is enough Information.
Instruction line 3 needs more clear and Precise word for LLM.
Instruction Line 5 - Try Adding a Sample Format too Instead of writing again a long instruction for LLM
Regards
RP
