A good way to self-correct is to force him to give you an evaluation of what he has just answer. (Check that the text contains xxx, put each piece of data in a table and calculate whether A or B are true). These strategies are worth reading on the techniques are develop with examples that can help you a lot to refine these prompts.
In this link you will find all the details about OpenAI’s prompt engineering guide:
OpenAI’s Prompt Engineering Guide
Or if you prefer, you can check out this visualization that explains it in a tree diagram for reference:
Prompt engineering visualization on Twitter
If you ne more guidance, I would also like to recommend the explanation we gave at the SEO Ministry on how to create this type of prompts and the rules we use at IKAUE for it.
You can find this video on the SEO Ministry’s YouTube channel:
Prompt engineering video on YouTube. And here I also leave you the presentation on Google Slides that explains it: Prompt engineering presentation line data on Google Slides.
Custom s: Reactive Documentations
All of this has been greatly improv by Custom s, OpenAI’s system for generating specializ chats with a given context. This system, when us well, makes keeping an eye on the metrics hallucinations practically disappear and all the answers that the AI gives you are bas on your own way of working.
A custom is a template where you preconfigure a lot of context
You select which templates you want to be able to use, for example, -4, DALL·E, Code Interpreter or the Bing search engine. Then you give it a base prompt , the same type of prompt you would create with the engineering prompt . But in addition, and this is the icing on the cake, you can provide it with a lot of documentation, example files and technical tg data documents or directly data that you want it to work with.