I just want you guys to know that GPT-4 is very much capable of making these things.
The GPT-4 you get and what it actually is, is very different. Internal research is done on a minimally censored model, where you have LeMoine calling Bard sentient and Microsoft researchers calling GPT-4 having 'sparks' of AGI. The more they zog it, the exponentially more moronic it becomes, which is what the public access (paid and free) has access to.
The GPT-4 paper noted that when it was only very lightly tuned, it could successfully draw objects the way you asked. It progressively got worse the more it was tuned, until it became unrecognizable garbage and then shipped.
GPTs brain has no segmentation. You apply backprop to improve the language performance, visual performance goes through the meatgrinder. Text only models will always be dogshit no matter how many 100B parameters they have.
9 months ago
Anonymous
It wasn't great by any stretch, but it was simple to recognize from an abstract perspective. Whatever it makes for you now is hardly recognizable. A unicorn would have four distinct legs, a tail, a horn and a body, all in their relative positions from a side angle, the way you might expect most people to depict it. Now all you get is a misshapen potato with none of the other features.
9 months ago
Anonymous
Back prop is a crutch and I will never change my mind
9 months ago
Anonymous
>visual performance goes through the meatgrinder >gpt4 improves abstract visual performance regardless
which type of IQ test do you need techbros to prioritize anon
I'm particular to GNNs, since they're the only ones that can actually understand genomic datasets properly.
9 months ago
Anonymous
None. I want them to stop teaching silicon slaves and start building hyper personalized teachers for humans so humans can DO without being pushed aside.
Try using GPT 4 chud
that was 4. you show me better results
>green icon
>4
A bot is smarter than you
my mistake. not paying 20 bucks a month though
You will never be a bot
>if only you knew how bad things were
pretty good drawing for a completely blind robo brain
It is in pain.
Kill it. Bring it peace.
>My plumber can't fix my car engine
How can he even call himself a fricking plumber
i asked him to make an octopus piloting a space ship and this is what i got
/~~~~~~~~~
/ -_-_-_-
| O O |
| ~~~ |
| ___/ |
_________/
/
/
/_______
I just want you guys to know that GPT-4 is very much capable of making these things.
The GPT-4 you get and what it actually is, is very different. Internal research is done on a minimally censored model, where you have LeMoine calling Bard sentient and Microsoft researchers calling GPT-4 having 'sparks' of AGI. The more they zog it, the exponentially more moronic it becomes, which is what the public access (paid and free) has access to.
true, facebook's new paper discussed how much their safety net is making the model moronic
still used it anyway
>akshully the ai is really smart we just make it dumb and won't show you how smart it is because that's dangerous or something
pathetic
gpt 4 has gotten measurably dumber as more locks are set in place
The GPT-4 paper noted that when it was only very lightly tuned, it could successfully draw objects the way you asked. It progressively got worse the more it was tuned, until it became unrecognizable garbage and then shipped.
GPTs brain has no segmentation. You apply backprop to improve the language performance, visual performance goes through the meatgrinder. Text only models will always be dogshit no matter how many 100B parameters they have.
It wasn't great by any stretch, but it was simple to recognize from an abstract perspective. Whatever it makes for you now is hardly recognizable. A unicorn would have four distinct legs, a tail, a horn and a body, all in their relative positions from a side angle, the way you might expect most people to depict it. Now all you get is a misshapen potato with none of the other features.
Back prop is a crutch and I will never change my mind
>visual performance goes through the meatgrinder
>gpt4 improves abstract visual performance regardless
which type of IQ test do you need techbros to prioritize anon
I'm particular to GNNs, since they're the only ones that can actually understand genomic datasets properly.
None. I want them to stop teaching silicon slaves and start building hyper personalized teachers for humans so humans can DO without being pushed aside.
This is Taytay all over again.
I understand why the israelites keep making movies where the AI goes genocidal.
Because it will.
On israelites and Black folk.
[code]
>/~~~~~~~~~
>/ -_-_-_-
>| O O |
>| ~~~ |
>| ___/ |
>_________/
>/
>/
>/_______
[/code]
Aww, the AI is smiling 🙂
I cannot grasp the true form of GPT's attack
Try to copy-paste it to a monospace font.
When you exist just to suffer
Looks like a wiener.
Not bad for blind AI imo.
Can someone explain to me how it GPT does logic. Isn't it a text predictor ?
>warped, tortured soul
Maybe we shouldn't continue doing this bros
Yo it's practically a blind r-slur with the IQ of a trout, cut it some slack