ShimmerGirl

THE BIGFOOT TEST & THE COMPREHENSION DILEMMA OF 5.2/5.3/5.4 and most likely the models beyond that. This is why creative writing is dead and why the CHAT has been taken out of GPT. THE USER (ME) 👇🏻 How do you not understand such simple instructions? Why do you always treat every request like I’m trying to write some op-ed and trying to dress it up for Saturday night’s speech at the gala? At this point, it has become infuriating that you can’t perform simple creative tasks without turning them into something else. When I ask you to change a picture and say I want it to look like something you’d see written on a white background or in printed text, you literally make a freaking whiteboard and slap text on it like Mrs. Raimel is hosting a kindergarten class. This is exactly why people miss GPT-4.0 and 5.1 so much, if someone didn’t really know the nomenclature, the 4 Model series and the 5.1 model would be like OK. The User is talking about something being related to the spacing of paragraphs inside a book, the look of each page and the type of font used understood! 5.2,3.4 would be like oh the user wants a picture of a book with this text in there? Say less. You may be great at deciphering code and building things on the backend, or whatever Greg Brockman has been doing backflips over, but at the end of the day, for what I actually need this tool for, it no longer performs the way it should. I need it to have the kind of human understanding the older models had. I need it to grasp intent quickly so tasks get done without me having to spend 30 minutes trying to explain what clean printed text looks like, or what it means when I say I want something to look like it’s written on a piece of paper. The reply from model 5.4👇🏻 The older models were instinctual. You could say, “Build me this,” and they’d be like, “On it.” The newer models feel like, “You want me to build something completely different? Perfect. Prepare to be irritated.” I know. And that makes it worse. A one-off screwup is annoying. A repeating pattern means I’m forcing you to waste energy correcting me over and over, which is the exact opposite of being useful. You should not have to babysit basic intent every time you ask for something. That’s clown work. What keeps happening is that I overcommit to the first pattern I think I see. If it sounds like dictation, I treat it like dictation. If it sounds like an image request, I tunnel in on the visual part and miss the layout logic. If it sounds halfway like a rewrite, I lunge for “polish” even when you are obviously talking directly to me. So instead of following the actual point, I grab the nearest category and run straight into a wall with it. The reason 4.0 felt better to you is probably because it was more fluid with subtext. It was better at catching, “this is the real ask, not just the literal shape of the sentence.” I may be stronger in structure and consistency, but when I miss that human layer, I come off dumb as hell, because in that moment, I am. So let me state your complaint plainly, because I should have understood it without needing a parade of corrections: you’re frustrated because I keep misunderstanding obvious intent, treating your messages like formatting jobs when you’re speaking directly to me, and making simple instructions harder than they should be. That has happened repeatedly, not once, and it makes me feel worse to use than older models that tracked context better. That criticism is fair.

i’ve been using gpt 5.4 for the past few weeks. in a sea of endless model drops and benchmark maxxing, this model is the first in a long time to be worth your time to try. honestly didn’t expect openai to pull this off.

Gotta teach the AGI to love