I was so genuinely happy after discovering generative AI stuff. All the things etc. were exactly my type.
I usually really love to dive into new things, play with things, try and break different stuff, and I still love this to this day.
But time from time, I realized that I don't really feel happy without outputs, specially if it's not just for fun and is for creating commercial-level images, posts, etc. for posting on socials or the internet. Not just noodles eating Will Smith.
So I explored into another new thing that I always avoided: "Prompt engineering", aka rocket science of English literature. I feel LLMs are so context-aware that they don't need a specific type of instruction for their specific models, at least for the latest frontier models.
But then I started wondering: what is it that prompt engineering doesn't have? What is still missing that makes my photos, flyers, ad designs, etc. not look like something a professional person had made? Years passed since 2023, when generative AI became a major thing everywhere, and I realized that prompt engineering is basically just precisely giving your instructions to an LLM.
But you still have to figure out what to give it.
The simplest way to put this is: you have to imagine or design something, at least in your mind, before you can feed it to that damn LLM.
So, you need to have the first skill: making up something worth making, or at least imagining the major thing in your mind.
Then comes the second skill: executing it and making the LLM generate that thought exactly as you imagined it.
And even after knowing prompt engineering, I still need to figure out hierarchy, proximity, style, typography, what content to show, where to align things, what to remove, etc. Without knowing these things, I would still be getting similar slop, or just slight tweaks if I kept using existing things as my major inspirational reference.
So TL;DR: I would still need to learn how to design better posters, flyers, etc. to generate better ones with AI. And the same thing gets repeated with pretty much everything else, majorly with UI/UX. I could never be able to design and develop better software without knowing which things would be best for me in which case. AI agents/assistants, in my experience, are very much designed to do what they are asked. They don't really bother with the necessity behind the request. There are tons of libraries, ways, technologies, stacks, DBs, daily-trending "open source" repos and alternatives to each other.
In such a landscape, there is no better way to use generative AI without learning, diving into, and manually exploring what's actually better and why. Not just in terms of technologies, but as a stakeholder, a theoretical expert, or simply as a person who feels he's well aware of that domain, at least to himself.
And that's why I still think taste matters.
When everyone has the power to generate almost anything at their fingertips, I think the thing that will make you different is not just your ability to generate something. It's knowing what is actually worth generating in the first place, and being able to tell when the thing you generated is actually good.
Maybe that's what I was missing all along...