|
precision posted:i think that trying to avoid porn or whatever is a fool's errrand. the porn will be there, better to let people get it out of their systems now I think it's fine to make it so that the generators don't start to suddenly rant about the ZOG when you ask them to generate a recipe for a unicorn egg omelette, but It's fine if you're brainstorming for a movie script or something and are trying to get ideas for an rear end in a top hat racist character.
|
#
?
May 23, 2022 19:53
|
|
|
|
| # ? May 30, 2024 00:48 |
|
|
A challenger appears A blue jay standing on a large basket of rainbow macarons.  A robot couple fine dining with Eiffel Tower in the background. 
|
|
#
?
May 24, 2022 00:09
|
|
|
somehow i'm able to think "this is fine" because of the way it starts with a small, low res image and all the "detail" is like... procgen detail i don't know why, but that makes it seems not even that magical to me. it's kind of just like a really advanced Mad Libs with operative terms like "wearing" i suspect this one would actually fumble pretty hard on some of the more abstract phrases. like i doubt this one could do much with "kim jon il experiencing an existential crisis"
|
|
#
?
May 24, 2022 02:03
|
|
|
furry artists are soon out of a job https://twitter.com/LitRPGforum/status/1528668105970536448 also browse this guy's feed, he's really churning dalle2 for all that it's worth, including telling it to make game assets
|
|
#
?
May 24, 2022 02:22
|
|
|
it's just the gap between expectation and experience. the cats are perfectly rad. not normal, but rad. 
|
|
#
?
May 24, 2022 02:30
|
|
|
man if i ever get an invite i'm not gonna get any work done for weeks
|
|
#
?
May 24, 2022 02:34
|
|
|
ymgve posted:furry artists are soon out of a job Found this in his feed and cracking up because I'm dumb as hell:   peps Peper Pesser PES pepi Peppse Pes Pepen Pess Pessi
|
|
#
?
May 24, 2022 02:38
|
|
|
Telephone game! psst psst. Pass it on.
|
|
#
?
May 24, 2022 02:45
|
|
|
Imagen demo has multiple for each prompt btw, so you can click on an already-selected item and get different outputs for it
|
|
#
?
May 24, 2022 02:56
|
|
|
|
|
#
?
May 24, 2022 02:57
|
|
|
"future economical replacement penises." Lol you still have one of those old human ones? Check out my fuckin' scorpion centipede one. edit: I just noticed my model came with the wolf head. It howls at climax. The Butcher fucked around with this message at 03:47 on May 24, 2022 |
|
#
?
May 24, 2022 03:43
|
|
|
ymgve posted:furry artists are soon out of a job they'll only be out of a job if horny is allowed KakerMix posted:Found this in his feed and cracking up because I'm dumb as hell: it's funny because it's like that bullshit faux scientific theoretical new pepsi logo thing that memed around for a while there, combined with the "draw style." The whole pepsi gravity well marketing bs everyone was laughing about. I bet it picked up on that. <- P > E Y+ ->  now this one is cool cause it reminds me of a weird art movement from like... the communist bloc era. There's a specific one that has a name but i can't remember it
|
|
#
?
May 24, 2022 04:05
|
|
|
lmao i typed "bart simpson wearing a che guevara t-shirt" into google image search and i got che guevara wearing a bart simpson shirt hahaha
|
|
#
?
May 24, 2022 07:51
|
|
|
It can render text into images more accurately.  It can be better at faithfully translating prompts into images. However, a horse riding an astronaut remains an unsolved problem. 
|
|
#
?
May 24, 2022 08:01
|
|
|
The Butcher posted:"future economical replacement penises." https://www.youtube.com/watch?v=aS-sDm9-H9w edit: aww, it's age restricted. it's a link to Duck Sauce - Big Bad Wolf this thing: 
Wheany fucked around with this message at 08:07 on May 24, 2022 |
|
#
?
May 24, 2022 08:04
|
|
|
https://twitter.com/LitRPGforum/status/1528781264609759233?t=AKxeBKC5Rs4_fpoiUPLmFg&s=09 This made me laugh more than it probably should have. Confusedslight fucked around with this message at 08:26 on May 24, 2022 |
|
#
?
May 24, 2022 08:09
|
|
|
SniperWoreConverse posted:
 Confusedslight posted:https://twitter.com/LitRPGforum/status/1528781264609759233?t=AKxeBKC5Rs4_fpoiUPLmFg&s=09 
|
|
#
?
May 24, 2022 10:19
|
|
|
If anyone has access to one of the newer models, I'd be interested in seeing what they make of the prompt "a geometric proof of Pythagoras' theorem" Mainly because these things seem to struggle with straight lines.
|
|
#
?
May 24, 2022 10:57
|
|
|
strange feelings re Daisy posted:This made me prompt latent diffusion with "Soviet Pepsi can": Never thirsty again under the watchful Soviet Pepsi eye
|
|
#
?
May 24, 2022 11:41
|
|
|
Brutal Garcon posted:If anyone has access to one of the newer models, I'd be interested in seeing what they make of the prompt "a geometric proof of Pythagoras' theorem" Here's a baseline for you using the latent diffusion model: 
|
|
#
?
May 24, 2022 12:09
|
|
|
Confusedslight posted:https://twitter.com/LitRPGforum/status/1528781264609759233?t=AKxeBKC5Rs4_fpoiUPLmFg&s=09 How the hell could it possibly interpret the prompt correctly here? It picks up on the relationship between not going home and sadness.
|
|
#
?
May 24, 2022 15:51
|
|
|
There Bias Two posted:How the hell could it possibly interpret the prompt correctly here? It picks up on the relationship between not going home and sadness. Broken and cannot have negative emotional connotations as words. A restaurant review parsing AI failed catastrophically because it was trained on the internet and assumed any review with the word 'Mexican' was negative
|
|
#
?
May 24, 2022 16:28
|
|
|
Wheany posted:Here's a baseline for you using the latent diffusion model: Dall-e mini actually makes a better attempt here:  These at least look like they're showing something about triangles, some of which might be right-angled.
|
|
#
?
May 24, 2022 16:56
|
|
|
the "horse riding an astronaut" thing seems to be a clear case of syntax error? that's disappointing. can't you tell it to always read left to right?
|
|
#
?
May 24, 2022 17:20
|
|
|
precision posted:the "horse riding an astronaut" thing seems to be a clear case of syntax error? that's disappointing. can't you tell it to always read left to right? Processing language is fantastically complex
|
|
#
?
May 24, 2022 19:40
|
|
|
Tunicate posted:Processing language is fantastically complex yes i know but it doesn't seem like it should be complex to include a parser that knows that "noun verbing noun" doesn't work both ways
|
|
#
?
May 24, 2022 19:45
|
|
|
IF NOUN1 VERB NOUN2 THEN DO NOT OUTPUT NOUN2 VERB NOUN1 see, i did it
|
|
#
?
May 24, 2022 19:46
|
|
|
precision posted:IF NOUN1 VERB NOUN2 THEN DO NOT OUTPUT NOUN2 VERB NOUN1 but this isn't a bunch of IF statements, it is putting the sentence into a neural network that then outputs a set of numbers, and then that set of numbers is fed into the image generator if it was that easy, why would both openAI and google fail to fix it
|
|
#
?
May 24, 2022 19:48
|
|
|
Imagine having a lava lamp type display device that's constantly crunching a flowing AI image set that morphs into the next image and it's really loving trippy.
|
|
#
?
May 24, 2022 20:26
|
|
|
ymgve posted:but this isn't a bunch of IF statements, it is putting the sentence into a neural network that then outputs a set of numbers, and then that set of numbers is fed into the image generator that's what i want to know! i'm genuinely curious! i actually want to know why it wouldn't work to include a line of code like the one i wrote. as, you know, one of the failsafe checks. surely it has those? things like "IF RACISM THEN DON'T DO IT"? so why wouldn't it work to do a thing that's like "if you have a query that is ONLY "noun verbing noun", then ONLY show results that follow that order of operations, EXPLICITLY DO NOT SHOW any results that go the other way round" i'm not like, being a dick, i'm asking because i want to know! 
|
|
#
?
May 24, 2022 20:26
|
|
|
is the problem that it doesn't think horses can ride other things, because horses always get ridden? someone try "horse riding an apple cart"
|
|
#
?
May 24, 2022 20:27
|
|
|
I can't be bothered to create a new thread for this, but I've been playing with some OpenAI generated story outlines, specifically trying to get it to write a plot for a psychedelic cowboy novel. Here's a couple of the best attempts, which did require a bit of nudging and some slight editing.Desert City posted:A cowboy is riding a camel through the desert in the Middle East. The sun is setting on civilization as we know it, and our hero is one of the last survivors. He's been wandering the desert for years, searching for a way to end the madness. Finally, he comes across a city that seems to be untouched by the apocalypse. He enters the city and finds that it's full of people who are living in a state of permanent euphoria. They offer him a place to stay, and he soon realizes that this is the paradise he's been looking for. The Last Man on Earth posted:In the town of Desperation, Nevada, the local sheriff is on the hunt for a missing person. When a drifter named Johnnie comes to town, the sheriff takes him in as a suspect. Johnnie tells the sheriff about his travels across the country. He also tells the sheriff about a vision he had in which the world was going to end. The sheriff is fascinated by Johnnie's stories and decides to let him go. Johnnie then sets off into the desert to find the source of his vision.
|
|
#
?
May 24, 2022 20:33
|
|
|
precision posted:is the problem that it doesn't think horses can ride other things, because horses always get ridden? someone try "horse riding an apple cart" Basically this is how it works  dalle is better because they threw more expensive ???? into the ???? areas, not because they have any special sauce logic
|
|
#
?
May 24, 2022 20:38
|
|
|
There Bias Two posted:How the hell could it possibly interpret the prompt correctly here? It picks up on the relationship between not going home and sadness. Tunicate posted:Broken and cannot have negative emotional connotations as words. It might be partially that but I think just the word realizes is enough, if you google realizes you get stuff like this: 
|
|
#
?
May 24, 2022 20:52
|
|
|
precision posted:yes i know but it doesn't seem like it should be complex to include a parser that knows that "noun verbing noun" doesn't work both ways
|
|
#
?
May 24, 2022 21:00
|
|
|
This Is the Zodiac posted:I'm an AI, and I know what "an astronaut" looks like, and I know what "horse riding" looks like, and in all the data I'm trained on when "horse" and "riding" are in the same sentence it looks like this, so... right right i get that part what i'm saying is, why wouldn't it have some kind of "override" for situations like that?
|
|
#
?
May 24, 2022 21:49
|
|
|
precision posted:right right i get that part what i'm saying is, why wouldn't it have some kind of "override" for situations like that? because it isn't being taught to comprehend language, it's a big mishmash of vector algebra optimizing for things which look *plausibly* correct
|
|
#
?
May 24, 2022 21:56
|
|
|
basically this is like those orchids which evolve to look like a sexy female bumblebee so male bumblebees try to gently caress them and spread the pollen. It's fine-tuned to look plausibly bee-like but doesn't have any understanding of bees
|
|
#
?
May 24, 2022 21:57
|
|
|
precision posted:right right i get that part what i'm saying is, why wouldn't it have some kind of "override" for situations like that?
|
|
#
?
May 24, 2022 21:58
|
|
|
|
| # ? May 30, 2024 00:48 |
|
|
precision posted:that's what i want to know! i'm genuinely curious! i actually want to know why it wouldn't work to include a line of code like the one i wrote. as, you know, one of the failsafe checks. surely it has those? things like "IF RACISM THEN DON'T DO IT"? so why wouldn't it work to do a thing that's like "if you have a query that is ONLY "noun verbing noun", then ONLY show results that follow that order of operations, EXPLICITLY DO NOT SHOW any results that go the other way round" There's no such thing as "failsafe checks" in the model, "if racism then don't do it" isn't something anyone knows how to do. The best they can manage is to look at your input and refuse to run it if they notice keywords they don't like. It's a basically a blackbox system where you have no control over what's happening inside, you can just feed it an input and get an output. The only way they know how to fix issues is by changing the training data and techniques to produce a model that behaves differently. Think of it like having a program but not the source code.
|
|
#
?
May 24, 2022 22:01
|
|