Avoiding repetitive language in completions

I’m working on an app where I include a bunch of information in the prompt and ask for the response to provide three recommendations based on this, including an explanation to justify each recommendation.

As the three explanations will be read sequentially, it’s important that the language used is varied and without repetition. However, despite my many attempts to clarify this in the prompt, the response still reuses language between each explanation.

Here are a few iterations of the relevant section of the prompt, written with the help of GPT-4, that I have tried without much success:

“Your entire response must use diverse vocabulary and varied structure to avoid repetitive language. Do not reuse words or phrases from one explanation in another explanation; their phrases and structure must be completely distinct from each other.”

“When writing the explanations, avoid repetitive language within each recommendation and between recommendations, and avoid using the same phrase or sentence structure more than once.”

“Employ diverse phrasing, synonyms, and varied paragraph structures to create engaging explanations. Use a mix of simple and complex sentences, use paragraph breaks, and vary the way you introduce reasons or explanations to make the text less repetitive. Try to rephrase or restructure the reasons in each recommendation so that they sound distinct from each other.”

"When writing the explanations, it is essential that you adhere to the following three rules:

  1. The language phrasing, word choice, and sentence structure of each explanation must be completely different to the other two explanations. …"

The model I’m using is text-davinci-003. Should I try a different model? Or is GPT-3.5 just incapable of this request?

1 Like