How to Ensure Complete Content Retention in GPT-4 Text Rewriting

I’ve been experimenting with GPT-4-Turbo for rewriting tasks, specifically converting interviews into commentary articles. However, I’m facing a consistent issue: the AI tends to miss out on details from the original conversation, and the longer the dialogue, the more details are omitted.

Despite my efforts to emphasize in the prompts the need to retain all details, such as using phrases like “Be sure to keep all details”, the results haven’t been significantly better.

Of course, specifying which details to include in the prompt works, but this approach lacks the generality needed for API automation.

Does anyone have tips or ‘magic words’ that could compel GPT-4 Turbo to accurately and completely retain all details in such rewriting tasks? Any advice would be greatly appreciated!

Hi and welcome to the Developer Forum!

Bare in mind that even the 128K GPT-4-Turbo will only output 4k of tokens or about 3000 words, it is trained to be concise. Asking for output close to the 4k limit will result in truncation and/or summation.


Thank you very much! I will adjust my approach.

1 Like