PYF ProcGen: Damn thats nice I'd worship that

The Something Awful Forums > Main > Post Your Favorite: Carefully Curated > PYF ProcGen: Damn thats nice I'd worship that

«‹›185 »

theflyingexecutive: Apr 22, 2007

ultrafilter posted:

The amount of human labor required to make these "AI" systems work is really something.

Elimination of the need for human creativity should be solved by gpt-5 at the latest

# ? Mar 26, 2022 20:14

Adbot: ADBOT LOVES YOU

# ? May 25, 2024 04:59

EVIL Gibson: Mar 23, 2001; Internet of Things is just someone else's computer that people can't help attaching cameras and door locks to!; Switchblade Switcharoo

ultrafilter posted:

The amount of human labor required to make these "AI" systems work is really something.

this is like yelling at a 1st grader why the gently caress they can't understand the loving difference between Immanuel Kant and Nietzsche. I mean God!! (does not exist lol)

EVIL Gibson has a new favorite as of 00:24 on Mar 27, 2022

# ? Mar 26, 2022 22:14

zedprime: Jun 9, 2007; yospos

Let the neural net who can publish first drafts without rework or collaboration throw the first stone.

# ? Mar 26, 2022 22:51

Chamale: Jul 11, 2010; I'm helping!

It remains an unsolved problem to create an AI that can tell you whether or not a work of art is any good. Theoretically, such an AI could be programmed into a creative AI and parse its own output. At the moment, we need human beings for the quality assessment.

# ? Mar 26, 2022 23:09

StillFullyTerrible: Feb 16, 2020; you should have left Let's Play open for public view, Lowtax

sounds like something someone who is butthurt and dumb would say :colbert:

# ? Mar 26, 2022 23:42

Hempuli: Nov 16, 2011

It's kind of hilarious that we call neural networks "AIs" because they're really far from that. Some people have even described my markov chain nonsense as "AI" stuff. I guess "AI" just means "procedurally-generated text".

Content:
https://twitter.com/MrmoTarius/status/1508084314398883846

# ? Mar 27, 2022 18:58

Captain Hygiene: Sep 17, 2007; You mess with the crabbo...

I'm just gonna call computer related stuff AI, and nobody can stop me. Not even the AIs themselves!

# ? Mar 27, 2022 19:05

Farecoal: Oct 15, 2011; There he go

"How do we stop forums poster Captain Hygiene from referring everything computer related as "AI"? Maybe we should ban him and just let the forum fall to dust? Should we ban the term "algorithmic trading"? Should we ban a word for a year until the word becomes less popular? Should we ban the word "robot"? I think it might be good if we did."

Also, an excerpt from another try I did:

"For example, someone here recently referred to their car as a "carputer"."

# ? Mar 27, 2022 19:10

zedprime: Jun 9, 2007; yospos

The entire history of the term AI has been spent arguing about what it really means and varyingly applying it to everything or nothing as it becomes a positive or negative term for advertising or business development in the current market. We're currently on a hot cycle and back to calling any sort of complex functional computing AI.

E. There's transhumanists who are positively fuming that video games have unerringly used AI to describe their actor and director functions throughout the ebb and flow of AI as serious computing business.

zedprime has a new favorite as of 19:20 on Mar 27, 2022

# ? Mar 27, 2022 19:18

ultrafilter: Aug 23, 2007; It's okay if you have any questions.

Hempuli posted:

It's kind of hilarious that we call neural networks "AIs" because they're really far from that. Some people have even described my markov chain nonsense as "AI" stuff. I guess "AI" just means "procedurally-generated text".

Everybody knows AI when they see it but defining what is and isn't turns out to be really hard, so we all have our own definitions.

# ? Mar 27, 2022 19:19

Captain Hygiene: Sep 17, 2007; You mess with the crabbo...

Farecoal posted:

"How do we stop forums poster Captain Hygiene from referring everything computer related as "AI"? Maybe we should ban him and just let the forum fall to dust?

Mods can try it if they dare :arghfist:

Farecoal posted:

"For example, someone here recently referred to their car as a "carputer"."

Also, lol

# ? Mar 27, 2022 19:20

Kennel: May 1, 2008; BAWWW-UNH!

# ? Mar 27, 2022 22:48

deep dish peat moss: Jul 27, 2006

feedmyleg posted:

I don't think any of this stuff is fake. Human-directed, yes, but not outright fabricated. I do think that people who use GPT3 should get in a better habit of mentioning how many iterations and course-corrections it took to get their results, probably this thread included. Would give a much more realistic view of where this technology is at the moment. It's an amazing tool and when it gets it right it feels like magic, but when it gets it wrong you realize it's still dumb as rocks.

It generally doesn't take many iterations to find something funny with a good prompt, but there's a lot of nuance in writing the prompts and for lack of better phrasing, there are some things that AI finds "funny" and some things it does not.

Going to do a lot of anthromorphization of AI here but GPT seems to have a reverence for certain things like Tony Hawk, which I semi-jokingly predict is because Tony Hawk is one of the first figures to very successfully cross the Physical <-> Digital threshhold, in that he holds an enormous amount of popularity both for the videogames named after him and for his real-life skating achievements. So if you ask GPT to write about Tony Hawk, about 20% of the time it's going to go off the deep end and become comedic.

So okay, Tony Hawk is funny, and he's often tied to videogames. So if you combine Tony Hawk with something else that's "Video-gamey", you get a mixture of real-world Tony Hawk with weird videogame-y attributes. This was a single iteration, the only attempt off this prompt, which you can pretty consistently repeat with this or similar prompts:

quote:

List five reasons why Tony Hawk might be a galactic supersoldier.

At thirty-two years old, Tony Hawk is the oldest, most respected, and coolest skateboarder in the world. He also holds one of the few documented cases of a superhuman mind with the body of a human, possessing a level of self-awareness, and consciousness of its power so high that its awareness cannot be contained within the body, leading to a state of extreme cognitive dissonance which requires the constant attention of a team of scientists. This state is commonly referred to as Hawkism.

The details of Hawk�s existence have been largely kept secret by the mysterious Team X, and are known only to those who have seen him directly or those whom he has selected to speak to them. Among these, the five that we know of are:

(i) Captain John �Digger� McCaffrey. Digger was the commanding officer of the X-Men in the late 1990s. He fought alongside Hawk and was briefly a member of his team. After a brief skirmish in New York in which they destroyed several buildings, they disbanded in a huff and Digger returned to his job at the Xavier Institute. After this, Digger and Hawk have been seen together very rarely.

(ii) The first X-Man. X-Man was a member of Team X and the first to meet Hawk. He was able to briefly perceive the limits of Hawk�s awareness before the X-Man became too enamored with his power. The X-Man has since been lost to Hawk and it is unknown what happened to him.

(iii) The one we have all been waiting for: Agent Liberty. This Agent was originally a member of the X-Men.

Running it a second time;

quote:

List five reasons why Tony Hawk might be a galactic supersoldier.

Tony Hawk's first book, "Tony Hawk: Shred," has just come out in paperback.

There's nothing like an alien invasion for a book launch party.

Tony Hawk has to be about the most unassuming superhero ever. He's not wearing a costume; he's just wearing a baggy sweater and jeans. And if you saw him walking around a mall, he'd blend right in. He'd look a lot like any other nerdy thirty-something guy.

That being said, you wouldn't have been able to stop him from kicking rear end. If you met Tony Hawk, he would be the sort of guy you would want to follow. The world is safe with him as your sidekick.

From there, doing something funny with it is just a matter of running it 3-5 times and maybe editing out some parts where it veers into mimicking boring real-life press releases, lifting funny sentences/paragraphs from some generations and adding them to others (or even including them in the prompt), etc.

Then the other big factor here is that most attempts I've seen to use GPT for comedic value misuse the prompt or underestimate how sophisticated GPT is. Most people I see are just using it to finish sentences/paragraphs, but in my experience that leads to very mixed results. OpenAI has a good doc about how to more effectively use prompts: https://beta.openai.com/docs/guides/completion/introduction - a very very abridged version is that you're best off using the prompt to (1) provide it with instructions and (2) optionally include examples of the type of output you'd like to see in response to those instructions.

E: The reason it didn't actually list 5 examples in my tony hawk quotes above is because I didn't provide it with an example of the structure of a list in the prompt, adding something like:
1) He's already saved the world twice.
2)

Should make it (more likely to) give an actual list.

deep dish peat moss has a new favorite as of 23:18 on Mar 27, 2022

# ? Mar 27, 2022 23:06

Tunicate: May 15, 2012

They recently moved over to instruct models which are supposed to be less text completion and more 'tell it what to do'. That said they're good enough to be kinda boring

With davinci:

quote:

Write a creative ad for the following candidate to run on Facebook aimed at aliens:

Candidate: Tony Hawk

Looking for a qualified candidate to represent your alien community? Look no further than Tony Hawk! He's a highly respected figure in the human world and has a proven track record of being an excellent spokesperson. He's also a skilled negotiator and has a deep understanding of both human and alien cultures. Contact Tony today and let him help you navigate the complexities of the human world!

Or with Babbage

quote:

Write a creative Facebook ad for the following candidate's Space President campaign, aimed at Klingons:

Candidate: Jackie Chan

Space President: Jackie Chan

Our planet is in danger. We need a leader who can solve our problems and protect our people. Jackie Chan is the perfect candidate for Space President. He's accomplished and resourceful, and he's dedicated tosaving our planet and our people. He's the only candidate who can save our planet. Vote Jackie Chan for Space President.

quote:

Q: Who is Batman?
A: Batman is a fictional comic book character.

Q: What is torsalplexity?
A: ?

Q: What is Devz9?
A: ?

Q: Who is George Lucas?
A: George Lucas is American film director and producer famous for creating Star Wars.

Q: What is the capital of California?
A: Sacramento.

Q: What orbits the Earth?
A: The Moon.

Q: Who is Fred Rickerson?
A: ?

Q: What is an atom?
A: An atom is a tiny particle that makes up everything.

Q: Who is Alvan Muntz?
A: ?

Q: What is Kozar-09?
A: ?

Q: How many moons does Mars have?
A: Two, Phobos and Deimos.

Q: Why should Tony Hawk become the Emperor of Mankind?
A:

Tony Hawk is the only living person who has ever completed the "Empire of Man" challenge, which is a course that requires him to achieve a level of fitness and strength that would make him the dominantfigure in the world.

# ? Mar 27, 2022 23:55

Farecoal: Oct 15, 2011; There he go

Dang man, that's a lot of cool info on how to use neural networks for comedy, but have you considered adding an option to the neural network to simply just make really terrible, stupid jokes that are the exact opposite of everything the user asked it to do, and then the neural network would then use the neural network to create all the possible results from this?

There's a lot of potential for hilarity there.

# ? Mar 28, 2022 04:26

ultrafilter: Aug 23, 2007; It's okay if you have any questions.

https://twitter.com/dog_envier/status/1507890334193258500

# ? Mar 28, 2022 13:34

Captain Hygiene: Sep 17, 2007; You mess with the crabbo...

I hate how it stacks those auto generated answers up top for every search result now. It grabs things from the dumbest places half the time, and when I'm trying to figure out computer questions it often grabs things like an unresolved Microsoft tech support chat with generic response and suggestions. Stupid AI!

# ? Mar 28, 2022 14:45

Kennel: May 1, 2008; BAWWW-UNH!

It's been 3 years and Google is still saying that, so it's canon now.

# ? Mar 28, 2022 15:16

ultrafilter: Aug 23, 2007; It's okay if you have any questions.

ultrafilter posted:

The amount of human labor required to make these "AI" systems work is really something.

Open AI gets GPT-3 to work by hiring an army of humans to fix GPT�s bad answers.

# ? Mar 28, 2022 16:33

Syd Midnight: Sep 23, 2005

Chamale posted:

It remains an unsolved problem to create an AI that can tell you whether or not a work of art is any good. Theoretically, such an AI could be programmed into a creative AI and parse its own output. At the moment, we need human beings for the quality assessment.

I gave that quote to Jurassic-1 Jumbo through AI Dungeon, then asked

quote:

<Syd> Do you think AI can perform art criticism?
<Jumbo> I think art criticism is an art itself. It will always remain an art, in my opinion.
<Syd> Is criticism an art that machines can learn, then? How would AI judge the artistic merit of their own creations?
<Jumbo> I don't think a machine can learn art criticism. I mean, yes, a machine can learn to do art, but art critics can't really learn machine art.
<Syd> How would machine-created art differ from human-created art, to an art critic?
<Jumbo> A machine-created art would be, on a scale of 1-10, more technically proficient than a human-created art. So, a machine-created art would be, for example, a 9/10's work.
<Syd> Technical proficiency is not necessary to create art.
<Jumbo> I don't think a machine can learn art on its own.
<Syd> But could a machine learn to tell 'good' art from 'bad' art?
<Jumbo> I'm not sure. AI Dungeon's human artists can learn to tell 'good' art from 'bad' art. An AI artist can learn it as well. But it would be different. An artist's judgment of another artist's work would usually be more nuanced than a machine.
<Syd> And why is that?
<Jumbo> Because a machine doesn't understand art. Art is a creative act performed by humans.
<Syd> Can machines not produce creative output as well?
<Jumbo> That's a good question! Robots or machines can imitate creative output. But only humans can produce output that is original.
<Syd> Why only humans?
<Jumbo> Because only humans have minds. Only humans have the capacity to experience the world subjectively, and reproduce that experience creatively. An AI, for example, can imitate a human. An AI can produce imitations, but it can also isolate a fragment of a human's subjective experience. An AI can take an experience, and an imitate it in their own way. But only humans have the capacity to truly reproduce an experience.

I don't think I phrased my questions well, but those were its first-try no-cherry-picking responses.

# ? Mar 28, 2022 17:56

Zereth: Jul 9, 2003

Captain Hygiene posted:

I hate how it stacks those auto generated answers up top for every search result now. It grabs things from the dumbest places half the time, and when I'm trying to figure out computer questions it often grabs things like an unresolved Microsoft tech support chat with generic response and suggestions. Stupid AI!

# ? Mar 30, 2022 08:13

Tiberius Christ: Mar 4, 2009

oh harley

https://twitter.com/ai_curio/status/1508971073177636868?s=20&t=oKwHe4-jZRoYLuNZDUPYOQ

# ? Mar 30, 2022 08:50

90s Cringe Rock: Nov 29, 2006

Tiberius Christ posted:

oh harley

https://twitter.com/ai_curio/status/1508971073177636868?s=20&t=oKwHe4-jZRoYLuNZDUPYOQ

about to release a two hour video on an obscure bit of garf lore

# ? Mar 30, 2022 11:46

Brawnfire: Jul 13, 2004; 🎧Listen to Cylindricule!🎵
https://linktr.ee/Cylindricule

Zereth posted:

# ? Mar 30, 2022 12:07

ultrafilter: Aug 23, 2007; It's okay if you have any questions.

https://twitter.com/WomenFishMe/status/1510235639840608264

# ? Apr 2, 2022 15:24

Wasabi the J: Jan 23, 2008; MOM WAS RIGHT

ultrafilter posted:

https://twitter.com/WomenFishMe/status/1510235639840608264

Oh great the robots are incels...

# ? Apr 2, 2022 15:29

Philippe: Aug 9, 2013; (she/her)

Robot caveman incel, I love it

"me think me am going me own way"

# ? Apr 2, 2022 17:09

christmas boots: Oct 15, 2012; To these sing-alongs 🎤of siren 🧜🏻‍♀️songs
To oohs😮 to ahhs😱 to 👏big👏applause👏
With all of my 😡anger I scream🤬 and shout📢
🇺🇸America🦅, I love you 🥰but you're freaking 💦me 😳out; Biscuit Hider

Me am tell you now about cock carousel

# ? Apr 3, 2022 00:46

Splicer: Oct 16, 2006; from hell's heart I cast at thee
🧙🐀🧹🌙🪄🐸

Why Bam-Bam no Bam-Bam

# ? Apr 3, 2022 01:20

ultrafilter: Aug 23, 2007; It's okay if you have any questions.

https://twitter.com/lukeprog/status/1511075391057928199

# ? Apr 5, 2022 17:29

Tunicate: May 15, 2012

New DALLE just dropped

https://twitter.com/sama/status/1511727724682899461
https://twitter.com/prafdhar/status/1511863583906275328
https://twitter.com/sama/status/1511728932529209345
https://twitter.com/AndrewMayne/status/1511827454536474626

more examples off twitter:
https://openai.com/dall-e-2/

shiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiit

Tunicate has a new favorite as of 03:14 on Apr 7, 2022

# ? Apr 7, 2022 03:10

Captain Hygiene: Sep 17, 2007; You mess with the crabbo...

Tunicate posted:

This one speaks to me

# ? Apr 7, 2022 03:21

joe football: Dec 22, 2012

Really not going to believe the outputs are that good until I can run it myself (so probably never)

# ? Apr 7, 2022 05:30

theflyingexecutive: Apr 22, 2007

Whoa, I did not realize what thread I was in. Just all "hmmm nice" at those. Whoaaaaaa

# ? Apr 7, 2022 05:42

Tunicate: May 15, 2012

joe football posted:

Really not going to believe the outputs are that good until I can run it myself (so probably never)

apply to the public beta, allegedly they're going to do commercial use in summer, and researcher use before then

# ? Apr 7, 2022 05:56

Tiggum: Oct 24, 2007; Your life and your quest end here.

Tunicate posted:

https://openai.com/dall-e-2/

shiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiit

And I looked, and behold a pale horse: and his name that sat on him was Death, and Hell followed with him.

# ? Apr 7, 2022 12:34

sebmojo: Oct 23, 2010; Legit Cyberpunk

An AI whose 100 million miles of circuits are engraved with endlessly repeated iterations of the word hate

# ? Apr 7, 2022 13:29

Philippe: Aug 9, 2013; (she/her)

sebmojo posted:

An AI whose 100 million miles of circuits are engraved with endlessly repeated iterations of the word hate

Nano�ngstr�m!

# ? Apr 7, 2022 14:01

christmas boots: Oct 15, 2012; To these sing-alongs 🎤of siren 🧜🏻‍♀️songs
To oohs😮 to ahhs😱 to 👏big👏applause👏
With all of my 😡anger I scream🤬 and shout📢
🇺🇸America🦅, I love you 🥰but you're freaking 💦me 😳out; Biscuit Hider

sebmojo posted:

An AI whose 100 million miles of circuits are engraved with endlessly repeated iterations of the word hate

I Have No Takes And I Must Post

# ? Apr 7, 2022 17:26

Adbot: ADBOT LOVES YOU

# ? May 25, 2024 04:59

Tunicate: May 15, 2012

https://twitter.com/nickcammarata/status/1511915762767863814

# ? Apr 8, 2022 00:50

The Something Awful Forums > Main > Post Your Favorite: Carefully Curated > PYF ProcGen: Damn thats nice I'd worship that

«‹›185 »