eastern-cyan•2y ago
Avoiding a part of response
I am getting a response element that is not appropriate.
When ask a the following question:
'What are the recommendations for product X'
the response starts with:
'Yes, that is correct.'
This is anon-sequitur. Any ideas on how to suppress it? I've tried telling the instructions not to say but..to no avail.
Thanks
4 Replies
To suppress the non-sequitur response you're encountering, it seems like the issue might be related to how your assistant's responses are configured, particularly around handling unmatched inputs or the initial response setup. Here are a few steps and considerations that might help:
- Check your assistant's No Match and No Reply configurations. These settings allow you to guide users to an alternative conversation path when their input isn't recognized or if they don't respond.
- Within the Capture step, you can add reprompt responses or configure a fallback path for unmatched inputs. This ensures that users are guided correctly even when their input isn't as expected.
- Consider adjusting the Temperature setting if you're using AI-generated responses. A lower temperature results in more precise responses, which might help in avoiding unrelated initial responses.
- Review any Actions you have configured in your assistant. Actions allow for complex logic and navigation, which could inadvertently lead to the issue if not set up correctly.
By fine-tuning these aspects, you should be able to control the flow of your conversation more effectively and avoid receiving inappropriate initial responses.
eastern-cyanOP•2y ago
The bulk of the response is correct. It just starts with the unwanted statement.
Are you adding a validation check before generating the answer?
Looks like the 'Yes, it's correct' is added from another LLM pass no?
eastern-cyanOP•2y ago
I had a setAI step that seemed to be the culprit. Thanks for your help NiKo.