Building this site: How We AI
Overview
I wanted to create a site to gather use cases as to how everyone is using AI at their workplace or personal lives.
How We Used AI
I essentially asked ChatGPT to help me get started. After a series of prompts it gave me some templates to work with. I also looked at the official Jekyll site, and then settled on the minima template.
While building this site, I used ChatGPT to help me with a lot of one off questions about how Jekyll worked, as I had never worked with it before. I pointed me in the right direction (most of the time) without me having to learn everything about Jekyll/GithubPages.
It shined while getting things working on localdev as it required a number of iterations and hit and trial. When I ran into errors, I just pasted those into ChatGPT and I was given very specific fixes, which reduced my time to production considerably.
Key Takeaways
- It definetely helped a lot in terms of hitting the ground running
- There were quite a lot of places where it did make mistakes though e.g. the templates it made pointed to layouts that did not exist etc
- Some things it did not warn you about. e.g. during localdev, Jekyll creates a
_sitedirectory. This should not be checked in and should be in the .gitignore. - Getting the UI right is challenging. (Thats why I went with a prebuilt theme). Getting the nuances of UI right via prompts is challenging, so that part was not very fruitful
- Every time you ask ChatGPT something, it will come up with a new answer, even if you ask the same question. Something to remeber if you are cut/copy/pasting templates etc
Conclusion
I can say conclusively that using ChatGPT helped me get this done much qucker. However, expecting that AI would get everything right and give you a ready made site to use, is quite far fetched. You should still know the problem you set out to solve and be ready to get your hands dirty by trying out different things. Think of ChatGPT as a very patient teacher. Also, by narrowing down the solution space e.g. having submissions as markdown files via PRs, helps AI in giving very relevant solutions as opposed to generic ones.