18. Brought to you by AI

2023-03-22

I have been doing a lot of work with large language models as of late, and I thought that I would try a little experiment while I was getting my blog back up and running.

I wanted to develop a new version of my website that was drastically easier to build and maintain. After having redesigned this corner of the Internet quite a few times over the years I think what I wanted most was something without any fuss. I love working with the latest front-end frameworks as much as the next person, but I’ve seen things fall in and out of fashion so quickly. I wanted something more utilitarian that actually got me to write, rather than something pretty but delicate requiring effort for frequent updates.

So, that’s why you’re reading this on one of the most basic HTML pages you’ll probably visit this month.

I thought it would be a good project to have ChatGPT write this site for me from start to finish. So, with yours truly providing direction and prompting throughout the process, this entire website was coded by ChatGPT. My role was mainly to copy and paste code it provided me into a Repl, run it, and then provide feedback or the next task on to ChatGPT. [1]

Funny enough, the longest slog involved was converting my old blog posts, in Markdown .mdx files, to .html. ChatGPT helped considerably here as well, with my dumping big text blobs into it and simply stating, “convert mdx to html.”

The whole process took me about four hours from start to finish and was drastically easier than my sitting down to work on all this myself.

[1] In terms of models, I found GPT-4 was much better at structuring the project and understanding my goals. That said, once there was a framework in place, GPT-3.5 did almost the same quality of work and at a much lower latency. In the .mdx to .html conversion task this was pretty critical.