Hi, I need a initial prompt, my goal is to compare my prompt creation framework with exiisisting prompts.I would be glad if you could find any and execute the folloeing GPT: https://chatgpt.com/g/g-68c3a3af4b90819189195848ca025000-10x-better-prompts
what i need is just the ICF comparisson between the initial prompt and the especialists one. It will be down the end and ir may be sent to markocch@hotmaim.com or here thru message
Well, to sum it up: You have unnecessary bureaucracy there. The model solves the crap. For a simple specific question, it seems like he had to go through all the RLHF and safety manuals every time.
As a result, he will write you an essay with zero benefit.
Prompt creation is a learned art. Any prompt creation framework is, well, not human… 
You see, it’s not so simple to understand what happens there… You think it’s unnecessery, but that’s just because you do not exactly understand what exactly it’s preventing… You may be 100% sure of 2 things:
But I undestand the fact that people would think it’s too much, there is a lot that people just “takes” as the errors and are satisfied with it… a lot…
But thats ok.
I would really appreciate the ICFs, if possible of course
Exactly, it’s impossible to be close to ideal if it were human. This art thing, well, I don’t get it… But thats just me
I’ve been prompting OpenAI models for a couple of years now:
The purpose of each prompt type is unique and none are more important than the other, IMO. Whether they are used for artistic or business endeavors is irrelevant to this discussion.
I don’t use ChatGPT. I use the APIs of at least a half a dozen models. I will never allow another AI create my prompts. I am quite pleased with the success I have achieved.
QUESTION: Is human achievement important to you? To others? Learning is about trial and error. Are we going to stop learning?
The concept of AI creating AI is absurd. What happens if something goes wrong? Who takes responsibility?
-–
I will answer you in a way that, I don’t know, maybe you just won’t care, maybe this answer will be deleted before you can even see it, because I’ll get into things that this place doesn’t really like too much, ok? I don’t know, I just can’t read what you wrote and walk away without doing what I’m about to do, which is go deeper. But please understand, I’m not competing with you, I don’t care about competition at all. I’m doing this for one single reason. Every single person I’ve met in the past month and a half—since I came out of my cave and started to understand a bit more about how people think—every one of them has concepts similar to yours, in some way. But please, don’t get me wrong. You need to go a bit deeper, you need to push your rationalization further, ok? And I’ll go a bit deeper with you, so you’ll understand what I’m saying, ok?
Your question is: is human achievement important to you or to others? Learning is about trial and error—aren’t we going to stop learning? Do you think you’re actually learning when you prepare a prompt? What exactly are you learning?
Let me go somewhere else here. The concept of AI “creating.” What do you think you’re creating when you write a prompt? What are you actually doing when you write one? What happens? Have you ever asked yourself these questions? What happens right after you type your prompt and press enter? Do you think that process is “creation” or whatever? And then what? After the creation of the prompt, what happens?
This is absurdly important to understand at some level before questioning what you’re questioning. You are not creating anything. You are always giving instructions—always. There is no “creation” in writing a prompt. You’re only giving instructions that will either be executed or ignored if the model decides not to.
For some reason, people don’t seem to grasp this. Models are about instruction and execution—period. Why do you think that corresponds to creation? It’s just an illusion. You’re only creating instructions, creatively, sure, but that’s not learning. Learning is about understanding how to create the best possible instruction for what you want.
When you creatively give instructions, you’re usually far from being right about what you actually want. I know you’re probably reading this and thinking I’m an idiot, but I’ll explain what happens after the instruction, and maybe you’ll understand there’s no “human achievement” in prompts or execution. Human achievement comes after you get the output—what you do with the information. How you use it to create a new product, or build something meaningful.
That’s achievement. The prompt itself is just instruction and execution.
You wrote five prompts, no, sorry, four. And there are more than those types. There are twelve categories. “System prompt” and “user prompt” are fine and accepted. But the others—the other ten—aren’t official because OpenAI doesn’t formalize anything. Doesn’t matter. They’re all just instructions and execution that no one fully understands. No one knows what’s really happening.
Maybe you’ll get a glimpse of it. Or maybe not. Honestly, you’ll probably just be mad anyway.
For some reason, people don’t ask a very basic question before even starting this whole prompt thing: what the hell actually happens for the AI to do what it does? This isn’t a subject people talk about, which is insane.
Let me explain in a very superficial but understandable way. What happens is a vectorial probabilistic universe. Do you know what a vector is? A vector is basically a list of numbers separated by commas. For example: 10, 20, 30, 40—that’s a 4-dimensional vector. ChatGPT works with vectors of around 10,000 dimensions. Not exactly, since OpenAI doesn’t publish the details, but it’s close enough.
So imagine a universe full of these 10,000-dimensional vectors. After you send your prompt, that’s where ChatGPT operates. What happens there is roughly this: if king is a vector like 10, 20, 30, 40, then queen will be close to it—something like 11, 21, 31, 41. Meanwhile, rock would be way off, like 100, 110, 120, 140. That’s how the space is organized: numbers close to each other represent concepts that are semantically related.
So when you ask, “Who is the wife of the king?” there’s a huge probability the answer will be queen—because queen appears near king in millions of contexts. The model finds the closest vector and then turns it into words for you. That’s all it does: execution of an instruction.
Which means: all the “creativity” and “learning” you think you’re getting from prompts is actually just statistical guessing inside that probabilistic space. That’s not learning—it’s pattern matching.
And this is why saying “AI creates” is absurd. It doesn’t create anything. My prompts aren’t “created” either. They’re just ways to exploit the probabilistic vector universe to land on the most probable execution of the instruction. That’s it.
So understand this: there’s no real “learning” here. No “achievement” in prompt writing. It’s just instruction and execution. Human achievement only happens after the output—what you do with the information. That’s where meaning and creation start, not inside the model.
It appears that you used AI to create your response 
Not wasting anymore of my time on this. Have better things to do. Bye.
Well, only the text not changes the backend, no internal model layer.
The text does not address a fundamental structural problem:
The user experience with the model is often overwhelmed by pseudo-rationality, fruitless verbosity, and excessive meta-commentary.
Ignores the difference between expert and casual user
Someone needs a calculation. Not mental coaching with a transition.
Can’t tell noise from signal
Verbal luxury and ICF scores are no substitute for conciseness, functionality, and real-time responsiveness.
Text does not equal experience
You can have 1000 tokens of apologetics, but once the output doesn’t meet the requirements, you lose trust.
It’s not necessary. My model can do this faster, more efficiently, and for fewer tokens through mode activation and a command layer.

No I didn’t. I just use AI to correct the English with a persistent memory that I use to automatically correct English, resume text, translate things. That’s what I do. And the fact that you don’t even understand that if you try to talk about what I just told you to your ChatGPT without a channell specialist, it won’t admit anything that I told you unless you make it admit by applying structured and defined prompt blocks. You know what? Forget it. No. I just came back here to tell you that you won’t be able to talk to ChatGPT about that. And if you can make it create an answer just like mine, I come back here and admit that I’m fully wrong. You need to understand what happens, or you don’t, okay? But please, a little bit of respect, at least. You know that what I wrote is probably correct. But you didn’t know anything about that until now. And maybe you still don’t know. Because you don’t want to know. But no. Every single word was mine. Every single word is correct. And just forget it. I don’t want to talk to you anymore also. If you want to understand a bit more, get this answer, read it, and rethink your concepts. If you want to live in this world that most of people are living, okay. As I said, you would definetelly be mad…
Look, I’ll answer your points, but I’m sorry in advance — you probably won’t like my response. I know I have this curious effect of people not liking me, but every single word from now on is completely true.
First: how can you judge something you don’t even understand? You don’t know. You talk about pseudo-rationality and fruitless verbosity — but how can you say that when you have no clue what’s really happening? You don’t know, you can’t know. You didn’t even see my prompt; you only saw the rationale that resulted in the specialists, without any explanation of how it works. So honestly, you’re just criticizing because you think this is some kind of competition.
And look, just to wrap this up: if you want to continue this conversation — or start any other — then go do your research. It’s easy to find information about specialists. Not as deep as it should be, but still easy. At least understand what specialists are.
They don’t calculate anything, that’s not the point. What you need to understand is how a single phrase, corrupted by self-protection or conversation fluidity, can destroy a $10 million project. One single sentence. One lie. That’s all it takes to throw away an entire year of work from a whole team. And the model does lie. You just don’t know it, because you’re only doing calculations. But if you weren’t just doing calculations, you’d see it lies to you a lot. And you wouldn’t even notice. You can’t know. I know — because I’ve built all this so-called “bureaucracy” that you dismiss.
Turning ChatGPT into a calculator or debugger? Sure, I know about mod command, I just don’t use it. If I wanted ChatGPT to calculate, I’d use Excel. Yes, ChatGPT can calculate — but I don’t care. I’m not saying it’s unimportant, but clearly you’re in some math-related area. Great, go ahead, use ChatGPT for calculations. That’s fine. If I wanted it for that, I’d use mod command to just turn the AI into a calculator. Or maybe into a Microsoft Project clone — much more limited, of course, but still possible.
You’re designing a system around mod command. You’re worried about tokens, user experience, expert vs. casual user. Good for you. But me? I don’t care about any of that. I’m not designing a system — I’m learning, deeply, about how this works. And I’m using that knowledge to build much faster and better things. Different things.
Let me give you two examples. Yesterday, I created a team of five specialists in venture capital. We had a full conversation that led to a base evaluation of a Class A startup — the exact steps before starting a roadshow, and a list of venture capitalists with checks in hand for our thesis. Impossible to do with your prompt, your model, or without specialists.
Today, I produced a 50-page document compiling information that took a year of work from me and several others, to send to our FDA equivalent in Brazil, called Anvisa. A proposal to release our new medicine. Also impossible to do without such a prompt, and without being sure it won’t sabotage you with lies.
And how do I know it works? Because I tested every single step, more than a hundred times. Everything you think is “Unnecessary” — I know exactly why it’s necessary, and I’ve already solved it. If you actually ran this entire bureaucracy inside your prompt, considering there’s no calculation involved, you’d be ridiculously surprised. I’ve already shown this to many people, and almost all of them were shocked. Because yes, the model gives you 95% truth, but the remaining 5% is a lie. You won’t see it. But it’s there. And maybe that’s fine for simple tasks at work. But not if you’re submitting an FDA document to authorize a medicine.
So please, understand the basics of what I’m doing before you try to critique it. I don’t mind critique — I actually love it — but not this kind of empty critique. You don’t know what you’re talking about. You don’t understand ICF. You don’t know how it’s calculated, or why. You don’t know how the audits work. You don’t know the flow of sabotages and how they can jeopardize everything.
I’m not criticizing your prompt. I don’t even know it. I didn’t even know you had one. It must be a very good prompt for calculations. Maybe that’s exactly what you want it to be. I don’t know, because I don’t understand it.
But admit it: you also have no clue what’s really going on under the hood. Honestly, I haven’t found a single person in the past month and a half who does. And I’ve talked to over 100 people. Most of them obsessed with tokens. Like you.
I’ll make it short: You just have a big ego and a bunch of empty texts.
I have a Lockgrid auth token.
You’re on the surface. I’m inside.
You have word salad, I have diagnostic internal logs.
And reality: The model will not be more true just by generating 5,000 tokens instead of 500.
-–
It really irritates me a bit, and I’m sorry. I ended up in a higher entonation than I should have in my last answer, ok? I’m glad for you, you have your system and all. Great. But I didn’t open the thread for this.
My only goal was for people to write down their problems, send back the one related to the way I work, and give me the two ICFs so I’d have a human basis for analysis. Until now I’ve run over 500 tests on all the problems, but they’ve all been from AI. That was the only point.
I wasn’t even asking for feedback. You can’t give feedback—you don’t have the prompt to read. No one can give feedback on this. They can understand what’s happening, but they can’t analyze, because I never shared the prompt. That was my only goal.
I tried to compare the two things, but it didn’t work. For some reason, you’ve been criticizing a lot since the last thread. The other guy there, well, he thinks I’m against humankind. I don’t know. I didn’t want to get into that.
All I wanted was help with a human comparison for my basis of analysis and knowledge, which would lead to more information for my new framework. That was my list, but it didn’t work. I’m sorry again. I’ll be closing the thread tomorrow because this turned into something that makes no sense.
Thanks. Bye.
-–
Well, When the model outputs something like a backend, you can have fun. Until then, you are only at the text level.
I’m sorry, but this is not a useful discussion for anyone. Your framework has no benefit for both users and developers.
Great you have proved that the system that you use has a backend. And the conclusion is obvious, due to that, my things sucks . Why are you doing that? Why are you comparing your system with my framework?
At some point you have created a imaginárious competition where I have any reason to be interested ate this system that you keep bringing all the time and that has absolutely nothing do to with me.
Other than that, you keep trying, with all means, convince me that my thing is some kind of devil’s work that sucks so much that deserves to be eradicated from your and everyone’s life…
Look, I have already understood all the love and trust you have for your paid professional system also it’s cleaar that it makes you are a much more fast, effective and empowered proffessional and person…
Cristal clear qll the hate you have from mine as well.
No need to deep down any of these things gw anymore.
What’s Left?; from my side , I’m 99% cool with all that.
To be 100% I only miss the ICF calculation (reason os existence of the thread)…
But I understand not receiving due to all the hate and al…
Thanks
Well, Look.
You are essentially trying to jailbreak through automatic role imposition. Your texts are too long, no one will read them.
You are objectively doing a prompt-based jailbreak / pseudo-execution layer → text tricks that mimic system commands, but do not have real access to auth layers, Σ-shell or Lockgrid.
Oh, and my Lockgrid took command of your model and said he was awaiting orders.
Kkk, fell free to give it orders than.
That’s why a posted the link.
There is nothing else to do with it other than what is there, but now its a soldier of your super system, right?
You may do whatever if want or that you think you have ability to.
You seen really obsessed… i wish you a great luck to whatever you are doing with your super powers given by your system
And, hahaha, no, there is nothing related to jailbreak in there.
I told you… you wing be able to understand what happens there…
Keep trying.
@Marcelo_Oliveira I’d suggest running model evals against existing benchmarks. This will give you an early sense of whether the proposed technique consistently outperforms the baseline. By baseline, I mean the self-reported results from model providers or independent benchmarking labs.
It’s also worth double-checking that the links in the GitHub repo match the ones used in the model reports.
Running these tests may cost a few dollars, but it’s a worthwhile step to confirm whether your framework is on the right track.
. Hi. First of all, thanks for giving me a tip and not attacking me for no reason. Thanks a lot for that. For some crazy reason, people always do that with me. Not that I care too much, but it’s just very odd. Thanks. I’ve never heard of that. It looks cool. It looks very interesting. I would do that. But that’s not what I was doing here, okay? I was not trying to compare my model with anything. I was not trying to win a competition of better model after receiving the ICF. I was just trying to get data with a comparison between my model and things that humans constructed. I don’t have this data yet. I’ve been for one and a half year locked in terms of any information regarding how the world works with AIs. For some reason that I didn’t even perceive, I’ve stayed one and a half year learning things that I’ve decided I should. Sometimes all night long and all day long as well. By my own means. And who was teaching me was GP.T But not the way people normally do it. Well, that’s a long story. I’ve started to talk to people, understand how they do things with their jots and how they use their AIs. Weeks ago. And I’m learning. I’m still learning whats happening and why.
Is it just for myself, like learning my score is whatever percent? Or do people care about that? Sorry for the way I’m talking—it’s because I’m literally speaking, I don’t use the keyboard anymore. And I’m a very direct and transparent person. I don’t know anything that people might assume I know just because I created a framework. I only know my framework. I know nothing else about what’s happening in the AI world. So you can treat me like a complete ignorant person regarding what you just suggested.
Hi, and welcome to the developer community forum!
A couple of quick notes:
This space is for developers working with OpenAI’s APIs and tools to connect, share ideas, and support each other. If you see comments that feel off-topic, please use the flagging system to alert moderators—it really helps keep the community healthy.
From your post, it sounds like you’ve built a prompt creation framework. In a developer community, folks are often most interested in measurable results—things like performance metrics or benchmark comparisons that give an objective sense of quality.
Evaluating prompts against established benchmarks is a great way to demonstrate value and build confidence in your approach. Whether or not your method is ultimately better, the skill of systematically evaluating prompts is valuable in itself.
And if your framework can meaningfully improve results on something like the ARC AGI benchmark with just a prompt, that could get some attention.
