I am subscribed and this is gpt4. So yes, I paid for this. But I am going to cancel.
Simple scripts need too much fixing and this should be very easy.
I finally know whag I want my next tattoo to be, but Gpt4 literally cannot draw it.
I think the model itself has no clue how many teeth it's drawing. Like it's not adding the teeth one by one, rather it's applying a "teethiness" texture around the rim of the wheel, if that makes sense?
but this is so bad. I am an esl moron, so I wondered if it should be gear or cog, if teeth is the right term or it should count notches, I tried different numbers, but if it got the right amount it was by accident.
I tried 8, 12 and 14.
It made 8, but I think that was an accident.
it failed at 12 and 14, which are even.
I asked for 5 I get 8.
I'm sorry, you were going to get a fricking ChatGPT generated tattoo?
You deserve all the pain you're getting
I have a very clear design in mind. simple forms. but I am shit at drawing myself.
So I wanted gpt to give me a basic outline of what I want so I can take it to the tattoo shop and talk with the tattoo artist about it.
I did not let gpt come up with the idea for the tattoo or anything.
We know these algorithms are still being trained and assisted in real time by armies of third worlders. Perhaps the same people who write the code at Boeing are helping the AI?
but this is so bad. I am an esl moron, so I wondered if it should be gear or cog, if teeth is the right term or it should count notches, I tried different numbers, but if it got the right amount it was by accident.
If it can count at all, then it should be able to do any number of teeth.
2 months ago
Anonymous
You're hitting the limits of what is possible with AI. An algorithm has been trained to recognize pictures of gears and trains and motorbikes by an army of monkies both located in the third world, and in the first world with captcha. But it doesn't understand the purpose of a gear. It can look up a definition made from a mish mash of the inputs, but sometimes it will lie to you, or get a fundamental thing wrong. It's all just software that predicts what an AI is supposed to do, then a Pajeet clicked the "wrong" or "right" button. That's most modern AI that's being openly hyped. Of course there are more legitimate uses like, analyzing stock information, molecules, etc. and this software is being created for professional users. It's a different kind of thing from normal AI like Chat GPT or the other alikes.
This is just one of the things diffusion models are uniquely bad at and you should probably give up.
Think of how hard it was to get these to produce hands with 5 fingers each. And that's something that's constant, hands virtually always have 5 fingers. To top it off, we have millions of pictures of hands, and it was STILL hard to get these models to generate things that didn't look horrific.
Now you're asking it to generate gears with a specific number of teeth (and not even a common number, a weird number like 9!). These things are terrible at counting and the sparsity of training data makes your case pretty much hopeless.
Here's what happens when I try Midjourney, same failure cases.
It's not bad, it's just that you're overestimating current tech. Proper general cognition is quite a way off.
Protip: always take tech hype with a grain of salt.
I am very cautious. I have seen it do great stuff with generating text (easy since fiction), merging images and even find information on org charts (pdf).
Even if you go far beyond the hypsters, even in the middle band people are talking about how greet AI is at certain things. But whenever I try things that should be reasonably simple, it fails badly.
How are universities and programmers scared of this?
>How are universities and programmers scared of this?
It's a glue that will eventually absorb all the algorithms known to humanity. Don't cite me on that, just schizoposting.
It doesn't know what any of those words mean. Try something like: cog, design, china, alibaba, green background, medium gear, metric, wholesale, thanks for the Reddit gold, DEI, raytheon
that's because it doesn't "know" what "9" is
in fact, it doesn't know ANYTHING
it has no ideas it has no concepts it doesn't even have object permanence
it, for lack of a better words, knows shit about frick
you are talking to a glorified markov chain
>Is it just bad?
Yes. Other things I've seen people post as impossible is >2 white wines and 3 reds on a table >an arrow pointing left >a watch set to any time other than 10:10
There's a bunch more shit that it can't do.
>30
>46
You didn't pay for this, did you?
I am subscribed and this is gpt4. So yes, I paid for this. But I am going to cancel.
Simple scripts need too much fixing and this should be very easy.
I finally know whag I want my next tattoo to be, but Gpt4 literally cannot draw it.
I think the model itself has no clue how many teeth it's drawing. Like it's not adding the teeth one by one, rather it's applying a "teethiness" texture around the rim of the wheel, if that makes sense?
I'm sorry, you were going to get a fricking ChatGPT generated tattoo?
You deserve all the pain you're getting
I have a very clear design in mind. simple forms. but I am shit at drawing myself.
So I wanted gpt to give me a basic outline of what I want so I can take it to the tattoo shop and talk with the tattoo artist about it.
I did not let gpt come up with the idea for the tattoo or anything.
Chatgpt doesn't generate images, it generates prompts for an internal image model
At least it doesn't call you a Black person.
I'd respect it more if it did.
We know these algorithms are still being trained and assisted in real time by armies of third worlders. Perhaps the same people who write the code at Boeing are helping the AI?
but this is so bad. I am an esl moron, so I wondered if it should be gear or cog, if teeth is the right term or it should count notches, I tried different numbers, but if it got the right amount it was by accident.
You could try teeth, cog is acceptable too.
Can you get it to do a different number of teeth on the gear? Maybe it could do 3 or 4 but not 5, you should test that out.
If it can count at all, then it should be able to do any number of teeth.
You're hitting the limits of what is possible with AI. An algorithm has been trained to recognize pictures of gears and trains and motorbikes by an army of monkies both located in the third world, and in the first world with captcha. But it doesn't understand the purpose of a gear. It can look up a definition made from a mish mash of the inputs, but sometimes it will lie to you, or get a fundamental thing wrong. It's all just software that predicts what an AI is supposed to do, then a Pajeet clicked the "wrong" or "right" button. That's most modern AI that's being openly hyped. Of course there are more legitimate uses like, analyzing stock information, molecules, etc. and this software is being created for professional users. It's a different kind of thing from normal AI like Chat GPT or the other alikes.
I tried 8, 12 and 14.
It made 8, but I think that was an accident.
Can it do 5? Maybe it can't do odd numbers of teeth for whatever reason.
it failed at 12 and 14, which are even.
I asked for 5 I get 8.
Well, good luck. I personally don't use AI because I suspect it's demonic.
abominable intelligence indeed
>8
>12
>14
those numbers are no good for gears. the number of teeth should be prime numbers
>tfw Kenyans cannot use AutoCAD to generate the 9-teeth cogwheel in 10 seconds
After some prodding from me, I was able to get the following result from Claude 3 Opus.
Too bad it doesn't look like a gear
This is just one of the things diffusion models are uniquely bad at and you should probably give up.
Think of how hard it was to get these to produce hands with 5 fingers each. And that's something that's constant, hands virtually always have 5 fingers. To top it off, we have millions of pictures of hands, and it was STILL hard to get these models to generate things that didn't look horrific.
Now you're asking it to generate gears with a specific number of teeth (and not even a common number, a weird number like 9!). These things are terrible at counting and the sparsity of training data makes your case pretty much hopeless.
Here's what happens when I try Midjourney, same failure cases.
I used the top left image as reference for Claude 3 Opus and this is what it spit out! That's not too bad!
It's not bad, it's just that you're overestimating current tech. Proper general cognition is quite a way off.
Protip: always take tech hype with a grain of salt.
I am very cautious. I have seen it do great stuff with generating text (easy since fiction), merging images and even find information on org charts (pdf).
Even if you go far beyond the hypsters, even in the middle band people are talking about how greet AI is at certain things. But whenever I try things that should be reasonably simple, it fails badly.
How are universities and programmers scared of this?
>How are universities and programmers scared of this?
It's a glue that will eventually absorb all the algorithms known to humanity. Don't cite me on that, just schizoposting.
>How are universities and programmers scared of this?
I don't fear what AI will do. I fear what humans will do with AI.
im terrified of what governments will do with ai pretending the ai is responsible so normies never hold the people using it accountable.
It doesn't know what any of those words mean. Try something like: cog, design, china, alibaba, green background, medium gear, metric, wholesale, thanks for the Reddit gold, DEI, raytheon
that's because it doesn't "know" what "9" is
in fact, it doesn't know ANYTHING
it has no ideas it has no concepts it doesn't even have object permanence
it, for lack of a better words, knows shit about frick
you are talking to a glorified markov chain
Why? The AI can't count.
So you're telling me that the thing every guru is saying is going to take every desk job any day now can't even count?
They will replace people anyway, then civilization crashes. All according to plan.
You mean the new hype cycle is empty hype, just like the last hype cycle?
Hype in and of itself is hollow.
>Is it just bad?
Yes. Other things I've seen people post as impossible is
>2 white wines and 3 reds on a table
>an arrow pointing left
>a watch set to any time other than 10:10
There's a bunch more shit that it can't do.