Development flow with ChatGPT
Exploring chatGPT for development purpose.
As I delve into the world of coding with AI, I've found GitHub Copilot to be a significant boost to my productivity, with suggestions that are accurate 99% of the time. This has sparked my curiosity about ChatGPT's coding prowess. After upgrading to ChatGPT Plus, I decided to put it to the test by documenting my journey in a series of blog posts.
My initial project is the creation of a straightforward HTML blog. Although I currently use a simple template from GitHub for my blog, I aim to craft a personalized blog with the assistance of ChatGPT this time. Given that my design expertise is quite basic, I plan to depend heavily on ChatGPT's capabilities.
Lets start with basic prompt ( text italics are prompts I've given )
chatGPT let me download the theme which is great. Theme is just 40 lines of css. Why tailwind? because I love it
of course, you need a basic html file to view the css
I created a git repo locally so that changes can be visualized as a diff. This is the only manual stuff I am planning to do. I will try to not tweak any code generated by chatGPT.
I requested ChatGPT to implement additional padding within the theme, which it executed. However, the padding failed to reflect on the body of the document. When I sought further assistance, the suggestions provided were not applicable.
Subsequently, I submitted a screenshot and instructed ChatGPT to recreate the HTML and CSS files. The regeneration was successful, yet I encountered issues with the file names, prompting me to manually rename them and integrate them into my Git repository. Moreover, after updating the file reference in the index, and despite making the amendments in index.html, the result mirrored the original. The only modification was within the theme.css file, and surprisingly, this did not resolve the problem.
Acknowledging the situation, ChatGPT offered an apology and made an attempt to reforge the theme.css file.
I have been trying to commit first and verify the changes which is bit different from usual developer workflow where we commit only once we are satisfied but I wanted to log all the changes suggested by chatGPT so I will keep committting all changes.
After couple of requests for suggestions for names, it has come up with "InquireInfitinum" and the domain is available so I picked it
One thing bothered me regularly is the API outages, especially after long session
This is where the outages were apparent. I tried at least 7 times for chatGPT to generate an image without any issue.
Interestingly, it could not generate a svg even after repeated requests. The SVGs it claimed to generate have some weird shapes
I was stuck for one hour as chatGPT4 limits were hit on that chat.
After about 1 hour, I was able to start a new chat that means that chatGPT do not have any context of previous chats so I decided to provide index.html and theme.css. I asked it to generate about.html which I think it is decent job. Just to see its capabilities, I gave it my resume and asked it to write some stuff in about.html. I also gave my linkedin profile to it and asked it to get my profile picture but it just gave me some instructions on how to do it instead of doing it in the html. Just to make sure that I do not cross limits of GPT4 again, I modified index.html to link to about.html manually. I also had to edit the linkedin profile picture name reference in about.html.
I wanted to add the same navbar that index file has to about page but it failed to do that even after repeated requests. It seems like the index.html file I provided earlier is not part of context anymore so I gave the index.html again and asked it to generate about.html again with navbar included.
Now I actually realized that it is running a code interpreter to make the changes. I can actually see the code it is using to replace the content in the pages. Hopefully I can use this code to make suggestions.
Overall, In my opinion, Generating simplified blog theme is okay but tooling around it needs to be developed for sure. I still had to intervene whenever some mistakes are happening.
In series of posts, I will try to explore more complicated scenarios for chatGPT to handle. Hoping to cover situations like open source contributions, Simulations, mobile applications etc.
Created a custom GPT called createsphere for going through the process, feel free to send me feedback on the same