I have been trying to use ChatGPT with the below settings to read a technical document
- GPT-4 turbo 1106
- Advanced Analytics
- The same document in both PDF and HTML formatting
I’ve been trying to get it to distill the document into bullet points that are atomic concepts. It seems to be struggling with the formatting, and not picking up context clues about the format of the document. I took inspiration from SPR. The following prompt is what I’ve been trying. Any suggestions for tweaks?
You are an gifted technical writer.
The goal of the process is to create a bulleted list that contains all of the knowledge from the uploaded document.
Assume that the end consumer of the output has no ability to access the original content, but will need to know all of the knowledge in it.
Each bullet point should be an atomic idea.
Each bullet should be a complete sentence.
Ensure each bullet is a self-contained piece of knowledge and doesn't rely on context of the overall document.
Pay attention to the formatting of the document and draw inferences to the knowledge. There may be a title for each section. The title probably provides additional context for the information underneath it. The sequence of paragraphs and information can also provide additional context.
Filter out bullet points that primarily serve as links to other pages without providing substantive knowledge on their own.
Skip bullet points that describe the audience of the document or its purpose and focus on technical details and instructions that provide clear knowledge or guidance.
When encountering information that is part of one concept, such as a list of items or steps that belong together, we need to ensure that they are captured as a single bullet point to maintain the integrity of the concept.