I slept fitfully last night. Waking up and checking the timeline, it seemed like I wasn’t alone. There was an obvious culprit: the release yesterday of GPT-4, the latest large language model from OpenAI had erupted onto the timeline, bringing unimaginable utility to our fingertips without warning.

I had been sitting on the sidelines of AI discourse, content to let it happen but wary of getting caught up in a hype bubble. There have been a few booms and busts over the years and it remained to be seen if AI chatbots were the vanguard of the future or somewhere closer to a fad. One session with GPT-4 has convinced me of its —and surely its successors’— disruptive potential.

This site has a modern look but I interface with it as a codebase, and each new post is a markdown file in a certain folder with some frontmatter metadata at the top. The header image lives in another folder and needs to be pointed to in the frontmatter, and after I find the image I need to move it to the right directory. These are all simple things but there’s a lot of steps and sometimes I mess them up. This is the sort of thing a script might be able to help with.

I asked GPT to generate a groovy script that would, given a filename slug, generate a markdown file with that name in the correct location, with the frontmatter template. And then it did.

In all of 10 seconds I had a script that is immensely useful to me (it’s the reason this post exists now vs when I would have had time to make it later) and that would have taken a fair amount of syntax and library research for me to write on my own.

I went back and forth, asking for adjustments that would get the behavior I was looking for, and each time the model spit out a new script that incorporated the changes I asked for in plain english.

When I was ready to run the script I got an error. I just copy pasted the error message into the chat window and the model apologized for using the same variable name twice and spit out a new version of the script that worked. I made the changes by hand, but later I discovered buggy behavior and manually traced it back to me not making the change in all the places I should have. I should have trusted the model!

Now, after maybe 5 minutes of work, I have reduced the amount of copy/pasting, file dragging, and context-switching required to write a new post. This is likely to increase my throughput since the context switching required to get a new post set up is a big friction point to writing at all.

Machines can do this now. This is not a group of early adopters thinking about what things will look like if trends continue, this is here now. I wrote a silly little script in 5 minutes, next I’m going to see if I can write a full-stack app in an hour. I feel like I have to grapple with this right now because it’s going to change everything and I’m mostly terrified it’s going to make me unemployed again in the near future.

Like what happens now. What do we do with all this surplus product. What happens to the people who don’t need it anymore. What do the giant corporations and nation states that run the world do with this technology. How can I protect my family against this, much less get value out of it.

I’m in a position where I can use this to help me make a living, which is exciting. But thinking about the positives or negatives induces mania. I think we all need to be very careful about how we interface with this new thing. It’s going to drive some of us crazy.