How I improve ChatGPT results by going in a second time
As an engineer and content creator, I’ve been experimenting with using AI for content creation. This is a journey that has been exciting, but not without its challenges. One issue I’ve encountered is that my unstructured rambling while using voice to text results in unstructured inputs for the AI, leading to less than ideal outputs. This is a testament to the age-old saying, ‘garbage in, garbage out’. So I thought, why not put the human (me!) back in the middle and try another cycle? This is my new plan.
Firstly, I’m doing the same thing that I had already been doing, but with a twist. Instead of asking the AI for a full output, I now request for an outline in Logseq format. With an outline in Logseq, I can easily organize and format the results, grouping them together, moving things up and down, and linking to already known information.
An added advantage of this approach is that I can remove or add sections as necessary. While I seldom need to add an entire section, the flexibility is there. I can also link my notes and do fact-checking in the end result.
The ultimate goal is to create a coherent story and plan from the outline. Once I have this, I can start recording a second time, this time following the outline. By giving the AI a more structured approach, I hope to end up with better structured results that can be used in blogs, tweets, or video scripts.
At this point, I’m not sure if this approach will work. I’m currently running an experiment and that means that in this blog post, I can only share the result at the end, which I’ll probably have to do the old fashioned way, by hand. But, as with all experiments, the journey is just as important as the result.
The end result is great, I need to do far less editing and the story as a result is shorter and to the point. I also like how adding a step is saving me work and helps me get a better result. As a bonus I spend more time in Logseq, always a plus.