Control specific chatCompletion responses

Hey Community!
I’m using chatCompletion API, my system message include several instructions and a pretty big context of data.
Now their are scenarios which the completion return specific sentences like “Moving to the next topic…” or “Thanks for sharing…” and I want it to prevent saying those specific things.
How can I handle those situations? I thought about prompt engineering but it feels too specific for me to declare specific sentences in the system message, what is the best practice to handle this issue?

Those hallmark catch phrases are part of the tuning of the AI, which is why they are so often repeated. It didn’t come up with the phrase “I apologize for the confusion” on its own for when you tell it that it can’t code.

That makes them hard to suppress, as the meaning will still be the desired output. “Don’t say ‘as an AI language model’” only gets you “as a language model” and other ways the chat-tuned engine will work around your prohibitions.

You can find out if they fit a particular category and make it skip that entire talk. “Use curt, brief, non-polite, robotic language” “Don’t write transitions, instead just move directly to generating the next topic”, etc. to change the whole style.

Thanks for the response!
Tried to add your suggestion as prompt engineering but it didn’t helped.
Any other suggestions?

Hello, I’m using chatCompletion - I want to remove specific expressions from the assistant response.

For example I want to remove ‘Thanks for sharing…’ look alike expressions, few examples:
Example 1:
If this is my assistant result:
‘Thanks for sharing.Please tell me more about yourself’
I want the function to transfer it to:
‘Please tell me more about yourself’

Example 2:
If this is my assistant result:
‘Thank you for sharing this with me, can you continue talk about yourself?’
I want the function to transfer it to:
‘Can you continue talk about yourself?’

I was trying to write several functions but without success - I did managed to write aggressive functions that worked but it was harming my whole conversation prompts so i’m looking for a good function that will work only in those specific cases and wont affect prompts that not related to the expressions I want to remove.