I've been experimenting with AI tools on this blog, utilizing GPT-4 for text and Midjourney for images. I began this project just over two weeks ago. It has been a very interesting process so far and I am learning a lot - the following is a brief rundown of my experiences so far and some things I have learned.
My approach is as follows:
- I come up with an idea for an article.
- Next, I write a bullet-point list in GPT-4, outlining my main points and desired article flow (more or less). I don't worry too much about what I write, I use my own language and I don't worry about spelling mistakes.
- I ask GPT to write the bullet points as an article.
- I then review the article, cutting out content and adding points as needed without worrying about formatting (or spelling etc).
- And again, I ask GPT-4 to rewrite the modified text.
- To get to the final work I repeat steps 4 and 5, refining paragraphs and sentences, until I like what I see and it says what I wanted to say.
Using GPT-4 like this has been really helpful. Not only has it sped up my writing process, but it has also helped me refine my ideas and improve my writing (I'm not claiming any great writing skills).
During this experiment, I've produced 11 posts totaling around 22,000 words (over 2 weeks working very much part time on the content).
22,000 words for me is a lot. But what is important to me is that I have revisited three topics I had written articles on previously. For those that are interested- https://www.robotscooking.com/single-source-publishing/ , https://www.robotscooking.com/exploring-development-methods-in-software-development/ , and https://www.robotscooking.com/md/).
I believe my writing in each of these is much clearer than previous attempts due to the assistance of GPT-4. I also have found that using GPT-4 has helped keep me intellectually more honest. There is a lot to say about what I have experienced around this issue, but here is a quick illustrative example....
I have an axe to grind in open source, in that I think the culture is 'too techie'. Having started numerous open source projects, but not myself being a programmer, I really get tired of this tech-centric (developer-centric) culture. I've tried to write about this before, but I found my axe soon came out, and I started grinding away. I tried writing about the tech-centric approach of OS again using GPT-4 as an aid, and I found I wrote an entire article (the second one linked above) that had a better structure and clarity than previous attempts. However, what was fascinating is that I asked GPT-4 to write a summary and the output was not something I agreed with. I then threw away the summary and rewrote it from scratch.
However, the new summary I wrote was back to grinding the old axe and was very much out of sync with the article I had just finished writing. So I threw away my conclusion, and started with the GPT-4 version and tweaked. It was much more intellectually honest, and also at that moment I felt the need to grind that axe disappear.
It was an interesting moment for me. Not just that the article as better, but the process had helped me clarify the ideas and let go of some intellectual baggage that I believe was holding back my ability to make convincing arguments.
Now, you might read those pieces and not be impressed with the writing. However, that would be more a critique of me than GPT. GPT has definitely helped me improve, even though my writing isn't any work of literature.
I could write a lot more on this topic, as I am finding the process fascinating, and I am glad I dedicated some time to trying things out. I can definitely see the limitations, especially if you treat an LLM like a guru - to answer questions and create beautiful things on its own. LLMs are not good at this (yet?) - but as a tool to help me write better, I have found GPT-4 very useful.
© Robots Cooking, 2023, CC-BY-SA
Image ('Robots Collaborating') public domain, created by MidJourney from prompts by RC.