Cover photo

AI Fails: Careful what you ask for

AI tools are incredibly empowering but no tool is without its issues. They will always try to solve your problem, even if that problem is what you asked for rather than what you want.

For one of my earliest AI coding exercises, I was (and still am) building digital travel souvenirs. I imagined a full marketplace of sorts for artists but to start, I needed something I could use as a country specific default images. I wanted to create a flag image for each country. I took an existing script I’d co-created to generate country code images and dropped it in asking ChatGPT to modify it to allow me to generate flag images. I mentioned it could use emojis as source data if it couldn’t figure out another source. This set off multiple rounds of running the resulting script with negative results followed by tweaks to the code, recommendations to install a different font, or try another image library. Or maybe all three. Round and round for over an hour. Eventually, ChatGPT tells me that it’s going to download images from a free Flagpedia service and then manipulate those. WAIT! There's a place I can just download the images?!?

I deleted my script and downloaded an archive of images, proceeded with my project AND learned an important lesson. Be careful what you ask for. The LLM wants to help even when you don’t know what to ask for. I needed images. I didn’t need a script. There is some balance to giving it enough context to be able to solve your problem effectively but also not being so overly prescriptive that you try to have it solve the problem in the wrong way. It’s more art than science but helpful to keep in mind.

Loading...
highlight
Collect this post to permanently own it.
Progress Over Perfection logo
Subscribe to Progress Over Perfection and never miss a post.
#ai