A stream of thought on where AI content is headed — and who gets to decide.
A recent tweet went viral by Greg Brockman — "Taste is the key skill." And it definitely is. Let me tell you why.
Where It Started
I've been looking at this field since 2020 — started with basic Computer Vision, I worked on CNN (Convolution Neural Networks), ImageNet, style transfer. Also worked with GANs. GANs were cool back then, and there's a site which people might have forgotten now — thispersondoesnotexist.com. It used to generate faces of people who completely don't exist. Now when I think about it, the website is damn fast, and people there look super real — more real than the stiff outputs that come from Nano Banana models in one shot. Of course, with the correct prompt, Nano Banana can generate these faces as well. But a website like thispersondoesnotexist, although super fast, doesn't have prompt control — because unlike today's image models, it doesn't have a language layer attached to it. Images there are made purely out of nxn noisy pixel images, with just a small random number seed parameter feeding into it that works as a steering wheel — but we don't have much control on that random seed param, because it wasn't trained like that. And that's what makes it not very useful to me right now. But it literally felt like AGI back then.
Anyway, GANs came, Stable Diffusion came, everyone has seen the Will Smith noodles-eating video — it was horrible, but it has somehow become a benchmark for people to judge AI video models nowadays.
What AI Slop Actually Is
But here's what I think AI Slop is. I earlier used to think that something produced with very low effort could be considered slop — because if it's low effort, we're going to see a plethora of them on the internet.
For example, almost 95% of posts I see on LinkedIn are generated by AI. I'm not denying that people shouldn't use it — they should — but at least read it once after generating. Remove the em-dashes, remove the classic "No X, no Y, but Z" construction. If you're not removing these, you haven't read your post even once after writing it — and that, for me, is AI slop.
Language and Image Models
Talking about language models — they've achieved pretty good accuracy in writing English copy, and now it's the human's task to judge the outputs and refine them. Really good things can come out of these models even in one shot, but only if you give the right input. And if your input is well thought then it definitely shows your effort.
That being said, coming to image models — almost the same analogy. By default GPT generates plasticky images. Gemini aka Nano Banana definitely generates good images because people use it without even removing watermark. This looks low-effort.
AI Video Slop
Now coming to actual AI videos — AI Slop. I find Sora to be exactly close to AI slop, brainrot — a video model purely designed to make hooky, fast videos that go instantly viral on social media. (Side note — slop content existed even before AI. TikTok pranks, irrelevant stunts just for views — anything that doesn't provide value, only exists for attention, is slop.)
But being specific to AI videos — there are videos made in one shot, not well thought out, not well controlled.
The One-Shot Problem
Think of influencers who, just to gain views, put out content saying "Hey, this was made in just one shot!" But think about it — it won't make you money.
Because if you can make it in one shot, your client can too.
What matters is the storyline and control over every frame and shot.
Redefining Low Effort
Coming back to the definition of AI Slop — media that doesn't add any value and looks low effort is slop. But things which are low effort now were very high effort in the pre-GPT era — hence there were fewer of them on the internet, and people consumed that media. Maybe it wasn't providing value, but at least it wasn't low effort. Fast forward to the post-GPT era — people who were generating content in the pre-GPT era are now generating damn good content, because they know the intricacies of it and the tools have made their power exponential. People who had ideas and knew how to use the tech to put down their thoughts also got those same exponential powers. But those who make videos with just a single line of prompt, purely to get viral — what they think is their effort is actually the video model's effort, the video model's accuracy. That content, I feel, is AI Slop.
It All Comes Back to Taste
And now when I think about it, it all comes back to where we started. Now that most people have tools that are very easy to use, it depends on your creativity and your ability to judge what looks good. It boils down to taste.
