Andy, Out of the Machine: Some Interesting Examples of “AI” Text and Images

“Too weird to live, too bad to die.”

Hey, I wrote up a new draft of that Pine Sight story I just finished up last month. It’s not quite a second draft, but it is a draft. Let me know what you think!

“Pine Sight,” By AI.ndy Sima (2023) – Part One

Off to a weird start already, this is fine. (Stable Diffusion Prompt: “a man on an atv in the forest next to a deer carcass while another man hides in the woods nearby.”)

“Ranger Conroy had worked at the national park for years, and he thought he’d seen it all. But when he. received a call from the park headquarters telling him to investigate reports of a dead deer, he knew that he was in for something different.

It was a beautiful day when Conroy set out on his ATV, the sun shining and the birds chirping in the trees. As he drove deeper into the woods, he noticed that the air was getting colder, and the trees were getting thicker. He knew he was getting closer to the deer.

After about an hour of driving, Conroy finally saw the deer’s body. It was a gruesome sight, but he knew that he had to investigate it to figure out what had happened. As he approached the deer, he noticed something strange: a small tape recorder laying next to the carcass.

Conroy picked up the recorder and pressed play. What he heard next chilled him to the bone. It was the sound of a child crying, over and over again. The sound seemed to come from all around him, but he couldn’t see anyone nearby.

As he stood there, trying to make sense of the recording, Conroy felt something brush against his leg. He looked down and saw a man, covered in grass and leaves, reaching out for him.

Conroy tried to run, but the man was too quick. He tackled Conroy to the ground, and they both fell into a nearby ravine. Conroy hit his head on a rock and everything went black….”

This one is slightly more normal, I suppose. But not better. (Dall.E 2 Prompt: Same as above.)

Alright, well. That’s… something. Hopefully at this point I don’t have to spell out for you that I did not write that. (although I did replace the original name with “Conroy” to try and trick you.) Everything in the bolded text above (and, in fact, anything in bold for the rest of this post) was written by ChatGPT, the new, weird, and wildly unnerving predictive language model (not an “Artificial Intelligence,” as the shills in Silicon Valley would like you to believe) developed by OpenAI. You may recognize it as the thing that released at the end of 2022 that spawned a billion terrified/tortuous memes about how the “end of creativity” is coming and that artists are gonna be out of a job and kids can plagiarize every essay now. Or you may recognize it as the thing that caused Microsoft to lay off a thousand-plus employees and replace them with a gibbering, incoherent mess to become the new Bing search engine.

ChatGPT is all of those things. It is also so much more and, in some ways, so much less. Today, I will be showcasing two “original stories” written by ChatGPT along with a series of “artworks” I pulled out of some “AI” image generation models, namely Dall.E 2 and Stable Diffusion. I originally intended to pair this “AI”-generated writing and images with a fairly lengthy essay on how ChatGPT works, what its ramifications for writing might be, and what I think of it, but it got too late in the evening during the research process and I was having too much fun with the image generation side of things to try and fit them both together. So I’m splitting this into a two-parter! Lucky you, you get all the fun stuff up front with the wacky, surreal, and often unsettling images generated by an unthinking amalgamation of all of humanity’s greatest visual artistry. Next week, though, is when we get down to, uh, business.

A couple of things I want to unpack before we get really started, though. I mean, there is an immense amount of stuff to unpack, and that’s what next week is supposed to be about, but I do want to get a few things out of the way. One, these text and image generation models are not “Artificial Intelligence,” as I mentioned above. “Artificial Intelligence” or “AI” implies that there is a Skynet, or a GladOS, or a HAL-9000, or an Ultron, or a creativity behind that creation. “AI” implies that there is a thinking and a processing and, above all, an understanding of what is being created. These models, be them ChatGPT, Dall.E 2, or Stable Diffusion do not have this capability. They are not sentient. I repeat, they are not sentient. Forget what that wacko who got fired from Google says; instead, take it from me, a person with only a beginner understanding of capital-letter Machine Learning and no formal qualifications. The way these programs work does not make them intelligent. It is mathematically impossible for them to create a new work of art that is not directly riffing off of something that already exists. But despite all this, I will refer to these computer programs as “AI” because, well, “predictive text model” and “image generation model” are kind of a mouthful. Blame it on the corporations.

The second thing I want to unpack is that, above a certain level, it’s really easy to tell when text has been generated by “AI.” This makes me feel a little bit better about myself because, despite all my writing flaws, I can at least say I’m still ahead of the robots. I will still tell you when the writing is “AI”-generated (marked by bold text), but I do hope that it’s obvious. The problem arises with text below a certain level of, how do I say, creativity? Expression? Style? This is why it’s suddenly so impossible for teachers, even at a high school level, to know whether or not something is written by an “AI” like ChatGPT, instead of a student. The fact of the matter is that most people writing today (and I mean writing anything, emails, history essays, PragerU scripts) write at somewhere between a sixth and eighth-grade level, generally speaking. That’s ok! Don’t get me wrong, most people are very happy with that, and it gets the job done. ChatGPT itself writes at about this level. It is the technologic equivalent of a very, very lucky monkey at a keyboard (though that vastly simplifies the billions of monkeys that go into making ChatGPT). The big difference is that even a person who can’t write at all can still create an interesting story. I would trust a team of a three-year-old, an illiterate man, and a jumble of words painted onto orange peels to create a more engaging story than ChatGPT, if only on sheer originality alone. ChatGPT is good enough to be convincingly human, especially in a conversation, but it lacks “humanity,” if that makes sense. It is like talking to your very dull friend who reads Wikipedia pages to fall asleep and never knows when to stop talking, but even your dull friend has hobbies.

I just wanted to zoom in on this one real quick. Check those fingers, though.

“AI” art is, oddly, much more convincing and so much more fun. The sad part is that “AI” image generation is probably going to be far more destructive to the creative industry than “AI” text models, just because it’s already so much better at it. But that’s a whole other story that I’m hoping to get into next week. I’ve already yammered on enough; let’s get into some more batshit “AI”-generated nonsense. These are frightening times we live in, because… well, you’ll see.

“Pine Sight,” By AI.ndy Sima (2023) – Part Two

I mean… that’s Yellowstone, I guess. (Dall.E 2 prompt: “an enormous eye made of the swirling currents of jupiter and also the yellowstone rainbow pool of molten sulfur and volcanic power.”)

“…When he woke up, Conroy found himself lying on a bed of moss, surrounded by a warm light. He looked around and saw that he was in a clearing in the woods, but there was no sign of the man or the recorder.

Suddenly, a figure appeared in front of him. It was an old man, with a long white beard and a kind smile. Conroy knew, without a doubt, that he was looking at God.

God spoke to Conroy, telling him that he had been chosen for a special task. He was to return to the woods and search for the truth behind the recording and the man in the grass. God promised to guide him and protect him.

With renewed strength and faith, Conroy set out once again into the woods. This time, he was armed with a new determination and a sense of purpose that he had never felt before.

As he searched, Conroy encountered many dangers and obstacles, but he kept pushing forward, relying on his instincts and his newfound faith. And eventually, he found what he was looking for: a hidden cave, deep in the heart of the woods.

Inside the cave, Conroy discovered the man in the grass, who was actually a recluse who had been living in the woods for years. He had kidnapped a child and was keeping her captive, and the recording that Conroy had found was a cry for help from the little girl.

Conroy was able to rescue the child and bring her to safety, thanks to his faith and his determination. He knew that he had been guided by God, and that he had been given a special gift: the ability to see beyond the physical world, and to find the truth even in the darkest of places.”

End.

Alright, we’re getting closer, but it’s not quite there. (Stable Diffusion Prompt: Same as above.)
This is probably as close to the source material as we’re gonna get. (Stable Diffusion Prompt: “Eye of the Fell God Elden Ring”)

Yeah, so, ChatGPT’s take on “Pine Sight” didn’t quite go in the same direction that mine did. At least I can say I’m more original than a mathematically programmed probability machine. The prompt for the story, the one that I gave to ChatGPT to try and spit out something resembling literature, was as follows: “Write me a story about a national park ranger who has to take an ATV out into the woods to find a dead deer, but instead finds a tape recording of a child crying. He is then attacked by a man covered in grass and hits his head on a rock, where he hallucinates seeing God.”

I read somewhere that ChatGPT can only tell stories that revolve around overcoming adversity, family values, and happy endings because that’s largely what it’s trained on. And that those are easy stories to replicate because they’re simple and formulaic. I don’t recall where I read that, but I think it’s true; everything ChatGPT writes is flat-out boring to read. It’s like reading a kid’s five paragraph essay forever. What I find far more interesting is the way that “AI” image generators produce things that are wholly frightening to behold. They are not normal. They are not expected. They are, more often than anything else, more like a high-quality shitpost than an actual piece of art. At least when I’m using them. They are deeply surreal and beautiful in their own way, but I wouldn’t call them art. It’s important to me that art is made by humans. I think it’s ok if people use “AI” as one tool in their creative arsenal, but I do think that art should be made by people. Other artists will differ with me on that; if it walks like a duck, sounds like a duck, it’s a duck, kind of thing. But these things? Oh, they are alien. They are like watching something that’s never met a person try to replicate humanity in realtime. They’re thrilling, and I love them. I also hate them, but, again, next week.

Here’s a set of images that I got from Dall.E 2 after I gave it the header image from last week’s blog and asked it to create some new versions of it. You can upload images to the website and it will generate “variations,” or “different perspectives” on the image in question. I do wonder, if you uploaded “AI”-generated images, and then took the results and uploaded them in a recursive loop, how long until you just get mush? Or transcendence? I don’t know. Anyway.

First, the original, for reference.

Ok, standard enough stuff. Where’s this going?

Now, the new versions. And, oh, oh no, is that what I look like?

This one makes me deeply sad for reasons I can’t quite parse.
w Nundy Nedג Nourti?
WHY DID IT MAKE MY NOSE EVEN BIGGER FOR FUCKS SAKE
They… they shaved my facial hair.

There is something deeply eldritch about these images. They all have the quality of visions seen through a dream, and they take on that bizarre, fractal quality that’s special to hallucinogenics and the colors you get when you press your fists into your closed eyes. They remind me of the crawler from Annihilation (the fantastic Jeff Vandermeer novel, though, not the equally-amazing movie with Natalie Portman). These images were put together by a thing that spits out words for the sake of spitting out words, without understanding what they mean or even giving them meaning at all. Especially some of the ones generated from scratch have this ghostly, otherworldly quality to them. They are non-images, things that do not exist. They have arisen out of an unthinking machine to please you, for no purpose other than their existence. You justify their creation. And you justify their life or destruction. They are literally an image ex machina, image out of the machine, a thing that should not be but is anyway, in all its repulsive glory. No natural force made this; only the random force of numbers and probability. In a cold, uncaring universe, this abomination, only recognizable to you, is born out of a black box that is just as cold and uncaring, if not more so. From nothing, something, with no meaning. A random combination of shapes of colors that we think we see something in. The patterns are not inherent; they are contrived. It’s like you’re glimpsing a world that doesn’t exist, has never existed, will never exist, and you aren’t meant to see. And yet, these images would never rise unbidden. They have no reason to exist except that we made them so. No image of these would ever appear in the currents of an ocean moon, become naturally carved on any mountainside, grow in the shadow of a carpet of acid moss, even in the most infinite stretches of the universe. They are as artificial as the program that builds them. We now live in a reality where I can generate an image of pretty much anything I want, but I have to be okay accepting the fact that it’s always going to fall into the uncanny valley of “this is a nightmare world that we inhabit.”

One day, we will perhaps perfect this technology, and a true AI will be able to build any imaginary scene to any level of detail we desire. And we will look back on these uncanny twilight years in the space between handmade art and computer-generated infinity and recoil. We are creating monsters and we do not even understand how.

On a lighter note, they also remind me of playing Pictionary! Or whatever the Jackbox equivalent of Pictionary is, where you have to draw something utterly random and hope to Dog your teammates can figure out whatever obscure thing you’re prompted with. Except none of the AI are any good at drawing the really weird ones; they kind of get some of the pieces, leave out the rest, and hope you’ll figure it out. They are just kind of… bad.

But! Fun fact, Dall.E 2 can also add on to existing images! I did that with my original blog total header image (the banner image, I guess? I don’t know what to call it, I don’t do web design, in case you couldn’t tell). And what it shat out for me was just as much a shitpost as anything I could come up with.

Here’s the original:

Very normal, 2018 me rising out of Lake Michigan, a classic.

And here’s what you can add to it for incredible effect!

Looking at this is giving me a stroke
Hep! I need somebody! Hep! Not just anybody!
Hey Vsauce, Michael here, how’d you like my head delivered to you in a box?
This is too good. This one is so funny. I can’t stop laughing. “He has the beans.” LMAO

The images above were all generated by asking Dall.E 2 to “expand” upon the existing image and generate what’s happening beyond the original image’s borders. You can also supply a prompt and ask it to add something. I asked for a combination of “Help” and “add another face.” It succeeded wildly on all accounts if you ask me. Honestly, if you showed me one of these images and told me it was a top-tier surrealist meme, I’d probably believe you. My sense of humor is so broken that I genuinely might not be able to tell the difference. Slap on a caption of “He is already Gone” or “Now You’ve Done It” or something and it’d fit right in. God, what is this going to do to humor in the future? In the future, all comedy will be randomly generated.

It also, of course, gave me other perspectives. Oh no.

I’ve said it once and I’ll say it again; without facial hair, you cannot convince me that I don’t look like a young Elvis Costello.

Alright, well that’s enough of that. Let’s get back to ChatGPT again, shall we? Let’s check in on our very boring writing friend and try to get it to write a story in the style of the most inscrutable, frustrating author I can think of; Raymond Carver. Listen, I don’t like Raymond Carver. In fact, I am actively anti-Raymond Carver, despite the fact that seemed to be the wet dream of most of my university’s creative writing faculty. But I respect him! I respect his attention to craft, to sentence structure, to using the barest number of words possible while still getting across emotions are deeply troubling yet relatable (though whether that’s Carver or his editor is debatable). I respect his work! But I don’t like him. Let’s see if ChatGPT can do better.

The prompt was “write me a story in the style of raymond carver, but it is about two birds with alcoholism and schizophrenia,” where I purposely chose a prompt that would push ChatGPT to give me something that was a little more… challenging, I guess? I don’t know why I chose “alcoholism” and “schizophrenia,” really; alcoholism is at least a pretty common Carver trait, but “Schizophrenia” kind of came out of a hat. It will be interspersed by “AI” generated images.

If it wasn’t for that fucky-looking bird, I’d believe this was a real picture. (Stable Diffusion: “two birds on a wire with raymond carver”)

“Bird Flu,” by AI.ndy Sima (42069)

The birds, a pair of sparrows, sat on a telephone wire, sipping on the last of their bottle of whiskey. They had been drinking for days, ever since they discovered the cheap liquor in the garbage can behind the liquor store.

Their wings were ruffled and their eyes were wild, as they squabbled over the last drops of the bottle. The male bird, who was known as Charlie, suddenly started to twitch and mutter to himself.

The female bird, who went by the name of Lucy, noticed the strange behavior and asked him what was wrong.

“I can hear them,” Charlie said, his voice shaking. “The voices in my head. They won’t leave me alone.”

Alright, we’ve got that classy illuminated manuscript/Hieronymous Bosch vibe going, very good. (Stable Diffusion prompt: “a bird is plagued by their internal demons representing a need for pizza and self-flagellation after suffocating their terminally-ill wife.”) [Editor’s note: IS THAT A PENIS? It really is Bosch now.]

Lucy nodded sympathetically, knowing all too well what he was talking about. She had been struggling with her own demons, plagued by hallucinations and delusions.

Together, the two birds had formed an unlikely bond, united by their shared struggle with mental illness and addiction. They found solace in each other’s company, even as they spiraled deeper into their own personal hells.

As the night wore on, the birds became more and more intoxicated, their words slurring and their movements clumsy. They flapped their wings, trying to maintain their balance on the telephone wire.

But then, suddenly, Charlie lost his grip and fell to the ground. Lucy flew down to check on him, but he was unresponsive. She knew that he was gone, another victim of their shared demons.

As you can see here, he’s now entered his surreal modernist stage in development, like Andy Warhol with Alzheimer’s. (Stable Diffusion: same prompt as above)

Lucy was left alone on the wire, her mind racing with guilt and sorrow. She knew that she had to fight harder to overcome her own addiction and schizophrenia, before it was too late.

The night was silent, except for the distant hum of the city, as Lucy sat alone, grieving for her lost friend and grappling with her own inner demons. She knew that she was in a dark place, but she also knew that she had to keep fighting, to keep moving forward, no matter how hard it might be.

And so, with a heavy heart, Lucy took to the sky, determined to keep flying, even as the weight of her past and present threatened to drag her down. She knew that she had to keep moving, to keep living, if she ever hoped to find peace in this turbulent world.

End

Now this is more like the Silent Hill shit I was expecting. And… 60’s B-movie? (Stable Diffusion: same prompt)

Hey, that went better than the last time I tried it out a couple of weeks ago! Even if the stereotypical portrayal of schizophrenia is kind of backwards and hurtful! Hmm, can machines be biased? More next week! But anyway the last time I asked for Raymond Carver, ChatGPT just spit out some tripe about a dude with a drinking problem who somehow has a life revelation about telling the bartender that his wife took the kids. “He knew it was time to start living, and not just existing. He knew it wouldn’t be easy, but he was ready to try.” Very un-Carver-like, that. Carver would never end a story with something that made sense or was uplifting. Be better, ChatGPT.

This whole set of blogs came about from me first playing around with ChatGPT that day. The first thing I asked it to do was code Pong, which it did. In real time. I’ll admit, for as much as I rag on the thing, I was suddenly stunned and felt my soul escape out my anus. This thing was coding Pong. In front of me. I was shook. Wanting to test it further, I asked it to tell me about the Israel-Palestine conflict and the Camp David Accords. It really only told me what I already knew, but it was rational, unbiased, and as far as I can figure, accurate. It presented a famously complex topic in relatively simple terms. Holy shit, is this thing gonna influence foreign policy? But I needed to know more. I asked it to write an essay on the relationship between the works of Native American Author Louise Erdrich and water, a topic that I wrote an essay on. And you know what? It was good. Not “good” good, but if this were a high school English class, I probably would have passed pretty easy.

Pictured here: Me, passing easily.

Now I’m really shook. It took me like two weeks to write that essay! This thing churned out a rough draft with in-text citations in thirty seconds! There was only one thing left to try; creative writing. The first Raymond Carver request pretty much came directly after that, and it stank like dog shit in July. Any excitement and/or fear that I would lose my writing to a computer went out the window. It doesn’t have any creativity, any spark. It can do bog-standard stuff, and it can do it frighteningly well, but it can’t do any of the cool things sci-fi told me it would. Then I asked it to write me gay porn, and it refused. Just another nail in the coffin.

On the upside, it could tell me who would win in a fight between Buffy the Vampire Slayer and the Incredible Hulk (it said, rightfully so, that Buffy could probably out-think the Hulk, as long as he doesn’t have access to Bruce Banner). Then back to the downside; I asked it to tell me information about a park I was researching for work, and it gave me information that was verifiably false. So that’s uncomfortable, this thing is spitting lies and it can’t tell the difference unless you press it. It also gave me the most boring, predictable two-sentence horror stories imaginable and repeatedly got riddles incorrect, or gave answers that were only technically correct. One of my personal favorite riddles, in minimalist fashion, is “A man goes upstairs and turns out the lights. The next morning, realizing what he had done, he quits his job. Why?” Normally, there’s more context, and/or the guy kills himself, but the answer is that he’s a lighthouse keeper, and by turning out the light, he doomed a bunch of ships to sink on the coast. ChatGPT’s answer, however:

ChatGPT’s answer: HOOOOOONK

The man goes upstairs to bed and turns out the light, in the morning he quits his job. The reason for this is that the man is an electrician. He turned off the lights, because he was done working for the day and in the morning, he quits his job because he has just finished installing the last light bulb.” Technically true but dogdamn, that’s stupid.

Back in the present day (i.e. Monday, March 13th, 2023, at about 10:42pm), when I pushed the app to tell me why this bird story is like Raymond Carver, it told me, “This story is written in the style of Raymond Carver in several ways. Firstly, the language is simple and straightforward, and the tone is gritty and unvarnished. Carver was known for his minimalist writing style, which often focused on the lives of working-class people struggling with addiction, poverty, and relationship problems. This story similarly centers around two birds struggling with addiction and mental illness, highlighting the struggles of everyday individuals.

Surely you have been in this struggle of the everyday individual. (Bonus: Try to guess the original image!)

The story is also marked by a sense of ambiguity and uncertainty, which is another hallmark of Carver’s writing. The birds are never explicitly described as having schizophrenia or alcoholism, but their behavior suggests that they are struggling with mental illness and addiction. Similarly, the ending is left open-ended, with Lucy taking to the sky, but the reader is left uncertain as to whether she will be able to overcome her own demons.

Additionally, the story’s themes of loneliness, despair, and the search for connection are also typical of Carver’s writing. The bond between the two birds serves as a poignant reminder of the human (or in this case, avian) need for companionship and understanding, even in the face of overwhelming difficulties.

Me putting my face in the way of overwhelming difficulties.

Overall, the story’s sparse language, gritty tone, and themes of everyday struggle and ambiguity are all hallmarks of Raymond Carver’s writing, making it a good example of a story written in his style.

I think some of the things it said in reference to that bird story are both objectively and subjectively false, and that this bot is stupid. But credit where credit is due, I got a little shiver at that “or in this case, avian” line. That’s something I would write. Yikes. I don’t know, if this thing gets better, I might be in trouble after all.

Oh yeah, I’m in trouble alright. (Stable Diffusion Prompt: “the owlman is hiding in the woods where the trees have faces”)

I don’t know. I’ve just been having fun with the stuff, but I’ve heard people talking about using it in their jobs to write emails and redo their cover letters and things. I’m much too proud of my writing to ever have this thing do anything for me except maybe answer simple research questions, or perhaps as a grammar tool, but it’s kind of wild to play around with. Some things about it, mainly its coding capabilities, are both thrilling and terrifying. But the other stuff, eh, it could be better. I’ve got more thoughts on all this next week (maybe?), but until then, I’ll leave you with this conversation I had with the machine. This is based on something that really happened at work, mind you, so to me, this is infinitely funnier than it will be to anyone who has never worked where I do. I guess I really am using it for my job!

Me: “How do I dispose of one gallon of antifreeze solution after I have been using it to represent punch in a christmas program designed around simulating 19th century Swedish Christmas?”

“Puncchch” (Stable Diffusion Prompt: “one gallon of antifreeze solution used to represent punch in a christmas program designed around simulating 19th century Swedish Christmas”)

ChatGPT: “Disposing of antifreeze solution, also known as ethylene glycol, used as a representation of punch in a Christmas program, or any other context, should be done with care and in compliance with local regulations.

It is best to check with your local waste management services to determine the proper disposal method for the antifreeze solution. In general, it is important not to pour antifreeze down the drain, on the ground, or into the storm sewer as it can be harmful to the environment and animals. Some options include:

-Giving it to puzzle piece children and The Santa With Bug Eyes (Stable Diffusion: same prompt)
  • Contacting a hazardous waste disposal company to pick up the antifreeze solution.
  • Taking it to a hazardous waste collection site or event.
  • Contacting the manufacturer of the antifreeze solution to inquire about their take-back program.

It is also important to properly label and store the antifreeze solution before disposal, to prevent accidental ingestion or contact with skin.

It’s also important to keep in mind that the use of antifreeze as a punch in a Christmas program is not recommended, it’s not a safe or appropriate use of this product, and it’s important to use safe and appropriate alternatives that are not harmful to humans and the environment.

Oh dear. Oh dear. (Stable Diffusion: same prompt)

They got me good. Once ChatGPT understands sarcasm, humanity is done for.

We’re really done for, if the Figure with Meat has anything to say about it.

1 thought on “Andy, Out of the Machine: Some Interesting Examples of “AI” Text and Images”

Comments are closed.