30 Mar 2025

Some brief thoughts on generative AI

(I wrote this as a thread on Bluesky and I’m copying and pasting it here, verbatim.)

I’ve drawn a hard mental line between generative AI for words + code and generative AI for images. For the former, I know whatever I generate I can fully understand before reworking or destroying and restarting the outputs. Or if it’s for a low-level/personal use, I’ll make cosmetic changes.

I’ve never generated an image and I never will. While you can replace words in a paragraph or a function, I couldn’t edit an image with the same level freedom. I’d have to reprompt or start hacking away at pixels of lines. That’s a skill issue but then that’d be the whole point of using AI!

I’ve not mentioned the anti-labour aspect of generative AI for images and everyone can debate what is and isn’t “fairer use”. But the proliferation of it comes from people thinking that they can produce art at the same level in seconds when they might as well just trace the originals.

What I’d love is for people to scrutinize images more and be able to point out AI-generated multimedia. I hate that I’ve had to grow this brain muscle as it has negatively affected how I browse the Web. But it’s unfortunately necessary. It’s cringe go share AI art and not know you’ve done it.

On AI code: there’s a murky area that this “vibes coding” stuff brought up. I code a fair bit and while I’m still learning, I sometimes have an internal crisis of wondering how much I could generate and use as-is and what I’d need to understand fully to maybe rewrite? Because that’s not my code lol

Would have I written this if I knew more than I do now? Would I have known enough to think of the same approach or maybe a better one? Vibes coding is the opposite of all of that thinking but I’m also not in that stage of learning where I could ignore it all in the first place.

I guess it all boils down to: how much of this is me and how much of this is a mix of other people, distilled into floats and arrays.