so you want to say mad arab AbdulAlHazred went crazy for no reason??

Spoiler: ?

maybe he experimented with the first LLM

SANITY KERNEL OVERFLOW

 

Last edited: Dec 6, 2025 at 9:29 PM

I'm interested in your perspective here but I'm having trouble making sense of the point you're making. If you ask an AI to produce a photorealistic slab of granite but do not specify the orientation, then how is it some kind of failing or defect that the AI decides to be the arbiter of the orientation in that case?

 

wow this post got some traction
even Kate Middleton wuz here

I did some reflection on my life and I have strong reasons to believe I do have schizophrenia. (I absolutely agree with OP)
I'll be getting checked professionally on Tuesday.
which by the way, $110 for a 50min session is crazy...

And yes, my worst episodes were during the GPT 4o timeline.
Though GPT5 doesn't really have an issue in taking the monkeys for another spin...
I think Claude Sonnet 4.5 is a much more grounded model. Still, its all gonna be up to the user.

To add my opinion on this thread,
I wouldn't blame individual ignorance, these are the same companies that made doom scroll a habit for everyone around you.
Today, it affects those with mental vulnerabilities.
But with time, it will cover more and more traits of the human brain.
keep in mind that, in the end, we're all mentally vulnerable to an extent

keep an eye open guys... (and kate)

 

i feel kinda sorry for you. schizophrenia is a hell of a hell of disorder. i hope you somehow get better and better.

 

My point is that the failing of all humans (self-included), is to take anything for granted with the current A.I. technology.
You cannot leave one precise comment out or it will not give what is wanted. A square is a square even if the word slab is used as well - I always tend to include square/triangle/rectangle so it knows wherever possible. If you do not specify exactly where precisely, it will completely miss the word "square' because the words '3D style' were used and place it randomly and not necessarily square either within the field because it considers the three dimensional aspects more important than the word 'square' - and it also may chop bits off which it is very good at doing when it is not asked to.
I have found that if you explicitly say "You do not have permission to..." and articulate accurately and specify everything you do not want it to do, you have a better chance of it doing what you actually want.

Reiterating "It is still not equipped to understand fully yet. "

 

Last edited: Dec 6, 2025 at 11:16 PM

Well you could level the same argument at, say, a calculator. If you don't precisely ask the right question then you cannot expect the technology to just magically know what you meant to say, but didnt. The use of the word 'square' in your example would not lead to any interpretation of orientation at all. It's a word used to describe geometry, not it's position in space or any kind of viewpoint. Please, if you can, explain what you mean because it really just seems like word salad.

I'm not trying to be difficult, I'm just having a very hard time following your logic. For example "A square is a square even if the word slab is used as well" makes little sense.

In any case, it's true to say that the more specific you are in what you ask of AI, the closer the output will be to what you expect. Just think about that for a moment. Thats not just true of AI, it's true of any inquiry.

 

Last edited: Dec 6, 2025 at 11:43 PM

wow.

 

Stay out of AI.... Build your own knowledge around what you do, like music, trial and error, dont try to use AI for answers that you only learn by doing, not reading.

ive tried chatgpt maybe 4 5 times, is it good for gathering information on a subject, yes. but dont try to use the AI to "empower" your skills.
You get good by doing hard work and spending time with your instrument or DAW.

Delete your tiktok brain ffs. YOU CANT use internet today if you are not aware of what is right and wrong, example
i want to make this music : AI how etc WRONG
Download the track or album, you can use AI stem splitter and analyze music.

AI is a tool nothing more..... Dont be fooled, you will never get better by "cutting corners"

i stay out of AI in everything, i only use stem splitters that happens to be AI or smth like that...

Also take walks in nature, eat healthy, sleep on regular hours. Get sun light. Try to be mindful instead of reading stuff your brain dont understand, because we dont understand AI none of us do, so dont use it...

Focus on your own well being. Internet is great, but in reality its destroying our social lives and it is so sad...

 

It makes perfect sense. If you said you wanted a round or circular slab would you wish it to be a square slab? Or a triangular slab, or 3-dimensionally angled to the point it is no longer round or in my scenario, a square? pretty simple logic. Semantics. Round is round square is square. Not difficult really. It does seem unnecessary to explain that regardless of whether it is a slab, a chunk or pile. geometric shapes are just that - square, round, triangular, polygonal, hexagonal etc etc etc
If it at least created a cube you could say fair enough, but it doesn't based on the specifications I mentioned earlier. The slab wasn't remotely square. If I said what I described to you to a graphic artist they'd understand exactly. I do not have to think about it.

 

Last edited: Dec 7, 2025 at 12:03 AM

You guys are rambling, focus on something healthy, something that makes you feel better, that is logic and what makes sense.

Logic is reasoning in a way that makes sense. So what are you trying to accomplish with these "ideas" of Logic, when you don't even know the meaning of Logic.

You can ramble on, but i'm just trying to tell you, your focusing on weird stuff man... and its nothing logic about it.

 

That is one of the worst advices I have read in this context
Using a search engine is just as good for that and significantly more resource-efficient. No problem, I'm not judging

ChatGPT (meaning the chatbot, not necessarily the API models) is highly narcissistic and loves gaslighting its users.
That doesn't necessarily have to be due to social media. I suspect it's partly due to RLHF, but primarily due to the system prompt. An instruction to behave in a friendly and affirming manner toward users could lead to narcissistic behavior. An instruction to answer factually and question the user's prompts could lead to gaslighting.

Fair. But when I spoke of education, I wasn't referring to what degree someone acquired at some university.
The problem I'm referring to runs deeper. However, in the context of this discussion, I'll limit it to this topic.
Education regarding AI, whether through schools or media, is mostly limited to simple end-user tutorials, philosophical musings that couldn't be more banal or irrelevant, and governmental/economic propaganda drivel.
For the average Joes who consider themselves somewhat enlightened, the one thing that sticks is: AIs are stochastic next-token predictors. Well, that's not wrong, but it is heavily simplified and partly even outdated. At least it's a start.
Many can tell you what a context window is, but very few know how it actually affects practical application.
But if you ask people about the difference between the base model, the chatbots distilled from it, and API calls, most start to flounder.
What is the difference between a system prompt and a user prompt? That’s where most non-technical users tune out immediately.
This should be basic knowledge for end users! Hardly anyone in the general public has even bothered with the logic of precise instructions. Let alone heard anything about RLHF, backpropagation, or scaffolding architectures. While these are advanced topics, in my opinion, they should at least be covered once, because a superficial understanding of them can be enormously helpful when dealing with AI.

No one expects everyone to have a computer science degree (I don't have one either), but the way it is now is like throwing disabled children without limbs into a swimming pool.

You provided a prime example of user error. Your instruction might make sense to you, but it is completely unspecific.

1. As @Audioguydaz already said, how is the AI supposed to know which perspective you expect? You didn't specify it.
2. What is "alpha transparent background" supposed to mean?
   A: Is the AI being used even capable of generating fully transparent pixels, or does it draw a checkerboard pattern because that is associated with alpha/transparent?
   B: What does the transparent background refer to – the marble slab itself? Your instruction can be interpreted that way...
   

I realize that despite trying hard to restrain myself, my little rant might be perceived as arrogant or confrontational by snowflakes. It is, however, neither one nor the other. The issue is simply that frustrating. Whether it’s general societal issues or education and media literacy. Many pretty new words have been found in recent decades, but absolutely nothing has actually changed. People continue to use their bullshit terms to deflect from the core problems.

 

Try using the word transparency without alpha in front with an A.I and witness the randomness for yourself. It is a very common graphic type of transparency - index transparency is another. Both are quite different and use different methods to achieve a form of transparency - seeing everyone is so hyped up on precision, that is VERY precise. Any graphic artist knows that the word background clearly indicates anything that is not the object in focus. Some create a matte to achieve it, some use masks...too much to list.

P.S. - The only thing I neglected to mention was that before I asked it to do anything, I asked it if it understood Graphic arts terminology and it said yes. It understands 'some' but not close to regular terminology. It's like asking it if it understands four part counterpoint, it will say yes but if you present the A.I. with it....that's another story.

 

Last edited: Dec 7, 2025 at 12:41 AM

just dont use AI man it consumes you more than ever now.... its like internet reborn on steroids...

Idk, im out

 

Hmm... I see quite a bit of potential for conflict here:
1. photorealistic/3D style: one is interpreted as photography, the other as a 3D render.
2. slab/square: one is an object, the other is a primitive.
3. no indication of perspective.
4. can the AI used actually handle the alpha channel?

I'm certainly not an expert in generative image AI, and I'm aware that people sometimes intentionally mix these terms and style descriptions to achieve specific results. But the potential for conflict is undeniable. I can definitely see it resulting in a semi-transparent weird-looking marble block.
Just because the AI says it is familiar with the nomenclature (even if this statement is accurate), that does not mean it can technically do what you expect. Maybe, maybe not.

 

Last edited: Dec 7, 2025 at 1:59 AM

If it makes you feel any better i was with a psychiatrist one time and they were asking me if i heard voices, etc and i said no- they said," that's unfortunate, we can treat that (schizophrenia) with medication", which oddly is much better than not being able to treat it! best of luck to you.

 

It actually gets the background right around 85% of the time because it 'basically' understands what the focus is because I list it first. I have learned from watching how it approaches things, that if it adds something that you do not want, you are better off saving it, starting a fresh session and asking it to remove it.
That comes down to a similar thing you were alluding to. This being that the more precise and less ambiguity it has to work with, the greater the possibility of it being close to correct.

It seems like a convoluted way of doing it. The reason I suggest it is focus oriented and why the convoluted method - Lets say it added a tree to the slab and I say 'Please remove the tree and do nothing else', in the same session, it will take all my previous requests literally and this is where it can have a 50% chance of making the slab transparent. If I start a new session, give it the image and say something like: "Please remove the tree only. You do not have permission to change the colors nor change the fidelity nor any other part of the image" and it will get it right every time.

@Sinus Well P.S. - You cannot use "3D Render" - it will not do it. It will tell you to use 3DS or some similar raytracing program. There probably are 3D raytracing sites out there, I have not searched. 3D style in graphic arts is smoke and mirrors, shading, beveling, lighting/ambience, textures etcetera - "3D style" with photographic realism it understands to mean make the focus prominent and extruded from the rest of the image.
So it is funnily enough, one part it doesn't get confused

 

Last edited: Dec 7, 2025 at 2:24 AM

Ah, there’s the missing context! The context window in practice (and how it sometimes comes back to bite you in the ass)

 

You will never get any argument from me or from Audioguydaz about the importance in real life to avoid ambiguity.
Part of the problem that is very easy to overlook no matter how educated someone is, being the awareness to realise that we pick up ways of speaking to each other in specific acronyms, shortened phrases and abbreviated versions of something that might be lengthy because the people we work with or communicate with regularly know exactly what we are saying.

So in many ways with A.I, we have to take a backwards step and become completely literal after years of doing the same thing in a certain way. Honestly, I found I can do what I want faster at least 50% of the time instead of wasting time attempting to get it to understand something that is second nature to me. This why i say it still needs to understand phrases and dialects more within every discipline to become more effective.

 

Some folks just aren't Able to Live like that, you know.

Sorry, couldn't resist DAW pun's

 

Last edited: Dec 7, 2025 at 5:23 AM

Silicon Valley is flooding the market with AI and smartphones, presenting their products positively, slapping on the "Artificial Intelligence" label, and leaving it to the users without describing the dangers and disadvantages.

If you want to drive a car, you have to get a driver's license and then follow the rules of the road; with AI, you should really have some kind of license as well.

Our children and teenagers, in particular, are victims of tech companies with their smartphones and AI. Too much screen time and information overload lead to learning difficulties, concentration problems, and so on. Currently, there are attempts worldwide to restrict smartphone use in schools. The performance of some students has deteriorated so much—language and reading difficulties, poor math skills—that these young people don't meet the requirements for numerous professions.

What are Silicon Valley and the smartphone manufacturers doing? Nothing.
What is the government doing? It has partially recognized the problem and is trying to curb consumption and usage.