This is 100% learning off of fanfic sites. Like every one of those similies and metaphors are ones i could swear i read in a fic...probably more than once.
You guys really helped me. I am doing an MA in Creative Writing. I broke my hip a couple of months ago and have been really enjoying your videos since I got out of the hospital. My inspiration for practicing walking was walking to the computer to watch more of your videos. At one point, I used 'for Lightlark' as a battle cry as I was enjoying that video. I'm writing fantasy short stories for my dissertation and you all taught me a lot about worldbuilding, which I really struggle with. So I just wanted to say thank you and I'm loving this video.
I’m so glad we could help you out during your recovery! And for your dissertation. “For Lightlark” is an EXCELLENT battle cry! Good luck with your healing and the dreaded dissertation! I’m rooting for you! - Maria
I think most of its fiction writing is modeled from scraped fanfiction so that explains the stylistic "choices." It's pulling fanfic patterns and plugging in the prompt details.
This is literally amazing. Not only do I love your criticisms when analyzing stories, but the AI's stories are funny as hell. You should do this again if you want to!
what was funniest to me about this experiment is that when I was a younger writer, I wanted to seem older and cool so 90% of my stories were indeed set in bars either at the beginning, totally, or for a large portion of the story. so that part was eerie! Something i've found is that if you ask chatgpt to write a story that is much longer, it will constantly try to conclude the story wherever the character limit makes it stop writing, just like it did here. just to see if it could, i wanted it to write me a longer short story, but that's too long so i asked for it chapter by chapter to get around the limit. but it tried to conclude the entire plot over and over. as far as writing goes, it's most useful for technical/business writing. i use it to help know what keywords i need in product descriptions for example, and i can't say it's made business any different but i feel like my pages look more professional when i have a few paragraphs instead of it costs this much and i made it with this buy it please lol.
So... I just had to ask ChatGPT to write a story about you guys... Once upon a time in a cozy, dimly lit living room, three friends gathered for a night of laughter, snacks, and a little friendly mischief. Will, Maria, and Katie had been inseparable since their college days, and now, in their late twenties, they still found joy in each other's company. They had come up with a plan for the evening that was sure to provide plenty of entertainment. They had discovered the wonders of ChatGPT, a cutting-edge AI that could generate all sorts of content. Tonight, their goal was to see how wild, funny, and absurd they could get by challenging ChatGPT to create the craziest fictional stories. Sitting in a circle on plush couches, the friends had their laptops at the ready. Will, the imaginative and adventurous one, kicked off the challenge. "Alright, let's start with something absurd," he suggested. "How about we ask ChatGPT to create a story about a penguin who becomes a detective in a tropical island paradise?" Maria, with her wicked sense of humor, eagerly typed out the prompt and hit enter. The AI's response was quick, and it painted a vivid picture of a tuxedo-wearing penguin named Percy, solving mysteries under the swaying palm trees of the paradise known as "Palm Beach." Katie, the one with a love for science fiction, couldn't resist the temptation to crank up the weirdness factor. "Now, let's ask ChatGPT for a story about a time-traveling toilet," she said, her eyes dancing with mischief. The prompt went in, and the AI spun a tale of a magical toilet that transported its users to different eras in history. They had to navigate through hilarious situations, like narrowly escaping a medieval jousting tournament while still in their bathrobe. The laughter in the room grew louder with each new story. They continued their whimsical prompts: a pizza that could speak seven languages, a superhero hamster who saved the world from an evil cheese-loving villain, and a professional sumo wrestler who moonlighted as a ballet dancer. As the night progressed, their challenges to ChatGPT became even more outrageous, and their laughter turned into full-blown bellyaches. The stories generated by the AI were a testament to its creative capabilities, and the friends marveled at its ability to turn their wildest ideas into entertaining narratives. Hours passed, and eventually, they all grew tired, but their spirits were higher than ever. Maria summed up the evening with a contented sigh. "Who needs to read books when we can create our own wild stories with ChatGPT?" Will, Katie, and Maria exchanged a knowing look. Their friendship had always been about adventure, laughter, and the sheer joy of being together. Whether it was through their own creativity or the clever AI at their fingertips, they knew they'd always find a way to infuse their lives with a little bit of the extraordinary. So, with fond memories of a night filled with imagination, they bid each other goodnight, eager to see where their next adventure, or story, would take them.
Out of curiosity, were all of these prompts in the same chat? If so, that might explain some of the recurring themes since the old conversations would still be in its context window. Also, if you do this again you should try with GPT4, I find it's a bit better than GPT3.5
Up to now, AI generates content by squeezing all existing content through the details you enter. The "voice" of an AI story is an average of all content that fits those details. When the AI generates a story without humans entering any criteria, that will be the AI's *own* voice. That'll also indicate that the AI has moved a step closer to self-awareness.
ChatGPT repeats things and can't plan to an overall structure because: 1) in simplified terms, it's a giant autocomplete function. Cal Newport and Ted Chiang have both written very good articles in the New Yorker explaining how Large Language Models (LLMs) "write" by predicting the most likely word to come based on the frequency of their occurrence in proximity to other words. 2) none of these generative AI models actually understand their inputs or prompts, let alone their outputs. Gary Marcus has done an excellent post on Medium illustrating how short it falls using prompts like "draw a bicycle and make the parts that touch the ground orange". 3) Yejin Choi did a TED Talk that explains why generative AI is not actually learning in the same way a child learns. But to the point at 47:43, I do personally agree with this because the majority of writing craft can be broken down into structures and frameworks. And just as the average layperson already has trouble distinguishing human created vs AI created art, the average reader will not be able to distinguish human written and AI written prose. 49:20 - Self-pub authors are divided on this, but there are a number already using it (and some who have been using it for quite a while) in many applications, from creating outlines, writing blurbs, writing ad copy, and getting it to actual prose that ends up in published books, in chunks of anywhere from a single line of description to entire rough drafts that they then work over in line edits for continuity and polish. Those who aren't using it are generally refraining because of their objections to the unethical way the training data has been sourced and/or because at this point, it's still faster for them to write their books. But some are turning to it to do things like create rough foreign language translations and then getting a human translator to review it in order to do foreign editions. 52:29 - this (and much more, like the ability to isolate specific areas of the image and regenerate and the new zoom out feature on Midjourney 5.2) is already possible. Top tier cover design firm, Damonza, put out their AI statement recently and it says this is how they will generate the different elements of their cover art. I have lots and lots of questions on this that I hope to cover in an upcoming Q&A with them. 54:52 - this is the key takeaway, imo. With many things, writing and other creative endeavors in particular, there's no substitute for learning by doing. Using a thesaurus, or a spelling/grammar check tool, or even using a generative AI LLM that's been ethically trained (on a corpus of material where the rightsholders opted in and were appropriately remunerated for it) to bang out an outline or random character profiles or to turn "a forest" into a more descriptive passage is fine to a point. But I think that point of making it too easy, when you get to the point of just hitting F5 over and over until it churns out something you want instead of sitting there, stringing words together until you can get it to finally click, something is lost. (Also if you can't be bothered to write the line or descriptive passage, then it can't possibly be crucial to the narrative and you should probably cut it anyway, or rewrite it so that it is narratively significant.)
"...in the style of Hemmingway." First, I'm, by no means, a Hemmingway expert. That being said, the "style" of Hemmingway would not have included Faries. Second, Hemmingway's writing was simple, direct, and somewhat sparse. This reads like a Twilight fan writing what they *think* Hemmingway would write. I'm a prog rock fan, and a huge fan of the band Yes. In some Facebook group, someone decided to ask ChatGPT to write lyrics for a song in the style of Yes. Then, the individual proceeded to ladle fawning praise over what ChatGPT produced. While the lyrics were decent - they made sense and they rhymed - there was one major problem with them: they weren't something Yes would produce. Jon Anderson typically wrote the lyrics for the vast majority of Yes's catalog. Generally, when he wrote lyrics, the words he used weren't select based on whether they made any sort of sense in the song. Jon based them on the *sound* they made when he sang them. In short, most of Yes's lyrics are pretty abstract if not bordering on gibberish. There are some notable exceptions to this, however. For a time, the members of The Buggles (they wrote Video Killed the Radio Star - the first song ever played on MTV) to produce the Drama album. They'd reform to create Fly From Here. And, recently, Jon Anderson was replaced with Jon Davidson, whose style of lyrics is somewhat fanciful. However, the VAST majority of Yes songs had lyrics penned or edited by Anderson. And they generally made absolutely no sense. So, something as lucid and understandable as what ChatGPT created couldn't really be what a Yes song would contain. Don't be surprised that ChatGPT found a story template somewhere and filled it in with other "stock" stuff from its models.
I once asked Chat-GPT to write a story in the style of Raymond Chandler. It literally started the story with the phrase, "It was a dark and stormy night..."
I've found that every AI story, regardless of prompt or premise, seems to have a "happy ending". This is especially noticeable (and bothersome) in horror stories lol
It helps a lot to be more specific. I asked it not no write a happy ending once, and it worked. If it doesn't work the first time you ask for something, keep insisting.
So long story short: no AI won't replace human writers - it's predictive text. it's just copying other things and it cannot "create" anything new - it's telling you what it thinks you want to hear.
Funny enough, Chat GPT is much better at wrting story-length poems than it is just writing stories. When I give it a story prompt and tell it to write in the style of Tennyson, Dante, and Milton, the story is still going to be super simple and cliche, but the prose itself is significantly better. I think there's something about the structure of poetry (rhyme schemes, length of sentences, etc.) that makes it easier for the AI to produce something stronger.
Also, an actual comment: maybe you're interested in reading His Dark Materials trilogy + there is a movie and tv adaptation so a lot of things to discuss, maybe you'll enjoy them :)
I don’t understand what William means when he says something is abstract e.g. “his voice filled with reverence.” Could someone explain it to me or give an example of how to fix the kind of sentences he points out? It’s driving me crazy and making me paranoid that it’s in my own writing and I don’t know it 😅
It's so interesting seeing the way it puts out some sentences that sound good on the surface until you think abt them for a second and realize they make no sense, like fully only nice noises and no substance. Seems like a lot of AIs right now is like passingly okay sentence to sentence until you examine it for what it's actually saying and what it's saying is nothing and/or actively wrong when it comes to facts lmaooo
The thing about AI is that it cannot, by definition, make something 'new'. It cannot 'create', but instead produces amalgams based off the creative works in it's data set, which let's not forget are gotten largely against copyright and without permission. This is, for instance, why it's terrible if you try to prompt it to make something uncommon (I saw an artist once show this by prompting Stable Diffusion with a description he was hired for in commission, and compared the resulting images to the *actual* artwork he'd made for the commissioner and they were impossibly different and what the AI spat out wasn't even close to what the client had him draw. AIs can only handle common enough 'stuff' that it would have had a large enough data set to draw on for generation. The prompt was something about dwarves riding black unicorns into battle with the dwarves wearing steel armored shoes with bunny ears (and some other stuff) btw).
There is also quite a few problems even outside the ethical. There's such a thing as AI Inbreeding or "model collapse". Because AI datasets *need* millions and billions of samples to train on that pretty much *necessitates* the unethical scraping of the internet (such as scraping AO3 for writing and LAION for art and other samples), and as more and more of the internet is flooded with images and writing created by AI, the models are beginning to be trained on *itself*, which can cause a model to collapse. For instance, as the number of AI art images with weird hands grows, the models that scrape the internet for training will scrape those AI images, leading to more messed up hands, etc. etc. down the line.
And *also*, because of what AI generation fundamentally is and how it lacks the human ability to separate fact from fiction or as you said, really have an overarching "wait, this doesn't make sense" reaction to a statement, there are cases *already* of AI writing specifically generating wildly false information. For instance, the advice not to trust any nonfiction books on Amazon for things like mushroom foraging unless you are *certain* it wasn't AI written or assisted, because they've literally found examples of these books *recommending tasting to help identify poisonous mushrooms* or listing things like *literal death caps* as edible. AI cannot fact-check itself for shit, and even if it could, it would likely fact-check itself *wrong* because it could be fact-checking against previous generated content that was *already wrong*. Another example; AI generation sometimes, when presented with a prompt asking about a real event, will answer with a *fictionalized version from a game or movie* because it cannot tell the difference between what is fact and what is not.
And you mentioned the similarities to fanfic, as well as another comment on here that mentions it. It is a *fact* that AI tools like ChatGPT were trained off of specifically AO3 when it was discovered that AO3 was included in the Common Crawl dataset, which ChatGPT trained on. AO3 actually has a post about it on their website. Which opens even *another* massive potential legal trouble, because by definition fanfiction is based on another person/group's IP. So even if, say, Disney forced an AI company not to train off their IP, if they trained off of scraped Disney *fanfic* or Disney *fanart*... etc. etc. (tl;dr AI in creative spheres especially is a massive ethical quagmire and AI is starting to already show massive problems with accidentally inbreeding on itself and it's datasets tend to end up with output that distills to same-ness as a result)
Question, did you guys use the same chat "conversation" to make each prompt? Becouse that could explain a lot of the overlapping details. Within one stream of conversation it will remember the previous prompts and incorporate them thinking that you are just tinkering with the concept. If I want to get something completely diffrent i just open a new conversation for it.
I put this same prompt into GPT-4 and told it to make the prose as purple as possible... Beneath the ominous, violet sky, the air cradled an intoxicating blend of dusky jasmine and blood-tinged iron which only spelled one thing: fairie land. Amongst the dappled fawns cavorting amidst the violet foxgloves, there danced a figure of devastating beauty and implacable power: Rhysand. His eyes glinted with a preternatural azure, hypnotic as they were dangerous; they were cerulean flames flickering under shaggy ebon lashes, beckoning the unwary into his bedazzling snare. His virile frame exuded a mystifying allure, molten darkness moulded in a silhouette of celestial sin. Behind an ancient oak, a pair of ember-eyes watched Rhysand’s frolic with a blend of fascination and defiant suspicion. Feyre, the fiery enchantress of the mortal realm, with her hair cascading in rusty tendrils kissed by the copper sunset, held whispered tales of the mysterious fairie. A woman wrapped in beauty and battle scars, her strength was an ominous thunderclap in the tranquillity of the fairie land, her spirit a stormy tempest that dared to defy Rhysand’s bewitching tranquillity. “Rhysand,” called she, her voice a melodious challenge twining with the lilt of falling leaves. The fairie, ensnared by her calling, moved towards her, his poise silken as lust entwined with the stormy sky, his expression painted in celestial intrigue. “Feyre, mortal maiden of audacious boldness. What brings you to my realm?” Her eyes glowed resolute, like the last smouldering coals against the arctic night. “Curiosity, Rhysand.” Her ripe mouth curved into a tantalizing and threatening smile in harmony; her voice, pure honey laced with poison ivy. A burst of mirth erupted from Rhysand, disarming as the dawn but underlined with the lethal promise of the midnight tide. “You intrigue me, Feyre, with your beguiling spirit and blistering audacity. A blazing star amidst a void.” Feyre stood her ground, her fiery gaze searing into his icy one. “I am no star, Rhysand. I am fire. And if you’re not careful,” she lifted a finger, piercing his chest with a point fiercer than a faerie arrow, “I will burn this entire world down.” A dangerous smirk unfurled upon Rhysand’s mouth, exposing the lethal undercurrent of his presence: “I’d love to see you try, Feyre.” So ignited a battle of wits and allure, a cataclysmic intertwining of fire and ice, mortal and fairie. Feyre, the formidable mortal tempest, and Rhysand, the celestial fairie abyss, threatened to consume - nay, ignite - the fairie realm in a conflagration that neither time nor magic could ever quench. A celestial dance was thus born, drawn in hues of honeyed danger and sapphire audacity, unfurling under the violet heavens.
Omg, that was the most extra sendoff 🤣 ChatGPT is really doing the most! On a serious note, there are a bunch of stylistic choices in the stories created by AI that read exactly like fanfic. And I'm not dissing fanfic here. Fanfic stories offer actual logical context for those choices and they don't seem so out of place. Actually, now that I think about it, ChatGPT is even more annoying because it's getting even more people to shit on fanfic and we don't need more of that.
21:54 "Well- heeled" isn't a nonsense AI phrase, it's an idiom used to describe aristocratic persons. Generally meaning genteel, aristocratic, well-bred, etc.
I won't lie, I've had chatgpt write me countless fanfiction where i get to live out my dream of being romanced by the greek god of war Ares. It did the job but got oddly obsessed with us going to the market as a main plot. But hey, i can't say i wasn't entertained.
All this talk about AI makes me worried for sentient AI. Not because I think it’ll be evil or anything but because I know the half reason Chat GPT and AI art are such a big thing is because people don’t want to pay artists for their labor. Which means if/when the AI is sentient it probably wont be treated well and definitely wont be payed. It’s a lose-lose scenario where human artists are losing jobs and livelihoods for capitalism’s sake but then the thing replacing them will also be made to suffer. The only ones winning are the people selling the finished products. It’s like shitty all the way through, except for this brief period at the beginning where it isn’t sentient or able to fully replace human writers in the eyes of the publishing industry. For now, we can do sillies with it.
I love you guys so much but the political opinions aren't always fairly represented. There's a huge anti capitalism message and while I agree that capitalism is flawed socialism isn't a good replacement because it refuses to account for corruption. That's my unwanted opinion. But now you may see my perspective. Again, love you guys but it the unrelated politics is very alienating. I have political channels to watch for that. I watch you guys to get away from that.
@@allbutperfect This is a book channel. It's one thing when a book is directly commentary on a political topic or holds a political theme. But going into "AI will be bad because capitalism...." That's a topic for another video. One that can be thought out and talked about on a NUANCED level. Problem with these blanket statements is we know people just take these things at face value and run with it. They don't consider the implications. There's so much discussion to be had around politics on its own and the politics of AI that it really has no place to say randomly all the time. And I LOVE Will don't get me wrong, but somehow in most videos he has a way of stuffing his very liberal political views in. Point being I couldn't directly tell you with any evidence where Katie and Maria stand in their politics. They never bring it up. Even when they've discussed books with political themes. Even though I can take a good guess there's no evidence present to back it up. And quite frankly I don't come here to know their political views it's irrelevant. I don't want to hear them espoused on one of my favorite channels I thought was safe from having to hear it on any side. There was a time not so long ago you didn't have to tiptoe around the internet to try to stay away from politics, so I treasure the places that I can find people I enjoy, and even though I know I might not agree with them politicallly, it never even gets brought up. That's a respect thats a respect and courtesy that is completely missing in this country. There once was a saying that I found very useful in navigating social atmospheres "There are three things you never talk about Politics, Religion, and your Wives." The wives part is kind of wierd and is mostly based on your crowd or lack there of. It's like basically stating "Dont talk a bad about your spouse to others" Anyways I didn't want to start some debate. I didn't appreciate the condescension in your comment particularly. I'm always open to discuss things like civil adults even when it's clear we don't agree. Sometimes we are just unaware. I used to be VERY unnecessarily politicallly charged and I still have to fight it. Sometimes I need a friend to say "Hey that's not really the discussion we are having" And in my pointing it out I wasn't trying to debate or be rude in anyway but the political comments do come off a little snide and of a place that "knows better" especially to those that don't share the same view. I'm talking in circles now. I doubt anyone is still reading. Conclusion, capitalism is flawed, but it's not contributing anything to a conversation about HOW AI CREATES SHORT STORIES. If we want to hear how capitalism sucks we go to channels to hear about how capitalism sucks not to our funloving bookclub channel.
I WANT TO BE ABUNDANTLY CLEAR HOW MUCH I ADORE WILL MARIA AND KATIE. They each remind me of friends I use to have and miss dearly. I'm not trying to bash anyone of be mean or judgey. That's another reason I don't want to even debate political subjects because online debates of ideals are completely out of context and have no depth of the person or understanding. It's very..robotic because we don't treat the opposing side like theyre human with lived experience. I know they have no idea who I am as a person and a part of me would die inside if they ended up hating me over an internet misunderstanding. Please please please know that I am speaking as much as a parasocial friend can speak from.
I really like you guys, so hearing Will go on his socialist rant really disappointed me. I never expected any of you to be commies, I never actually thought about politics when watching this channel, you were supposed to be a relaxing and fun escape from it. I don't want to stop watching the channel, because I genuinely enjoy it, but now I'll have a bitter taste in my mouth every time I do. Something was lost here, my "innocence" if you will, and now I need to mourn it. Deeply I'll grieve, because it truly was once sweet, warm and cozy, when I was still ignorant.
The ai making art , writing stories and copying drawings while the people do manual labour is not the future i was hoping for
This is 100% learning off of fanfic sites. Like every one of those similies and metaphors are ones i could swear i read in a fic...probably more than once.
"feisty" translating into "redhead" is 100% because the AI is trained on scraped fanfiction 🤣
You guys really helped me. I am doing an MA in Creative Writing. I broke my hip a couple of months ago and have been really enjoying your videos since I got out of the hospital. My inspiration for practicing walking was walking to the computer to watch more of your videos. At one point, I used 'for Lightlark' as a battle cry as I was enjoying that video. I'm writing fantasy short stories for my dissertation and you all taught me a lot about worldbuilding, which I really struggle with. So I just wanted to say thank you and I'm loving this video.
I’m so glad we could help you out during your recovery! And for your dissertation. “For Lightlark” is an EXCELLENT battle cry! Good luck with your healing and the dreaded dissertation! I’m rooting for you! - Maria
Fitz Gerald literally translates to Bastard of Gerald. Don't name your child Fitz anything
I think most of its fiction writing is modeled from scraped fanfiction so that explains the stylistic "choices." It's pulling fanfic patterns and plugging in the prompt details.
This is literally amazing. Not only do I love your criticisms when analyzing stories, but the AI's stories are funny as hell. You should do this again if you want to!
Something about Katie’s film setup makes it look like she’s in a found footage horror movie and I’m here for it
what was funniest to me about this experiment is that when I was a younger writer, I wanted to seem older and cool so 90% of my stories were indeed set in bars either at the beginning, totally, or for a large portion of the story. so that part was eerie!
Something i've found is that if you ask chatgpt to write a story that is much longer, it will constantly try to conclude the story wherever the character limit makes it stop writing, just like it did here. just to see if it could, i wanted it to write me a longer short story, but that's too long so i asked for it chapter by chapter to get around the limit. but it tried to conclude the entire plot over and over.
as far as writing goes, it's most useful for technical/business writing. i use it to help know what keywords i need in product descriptions for example, and i can't say it's made business any different but i feel like my pages look more professional when i have a few paragraphs instead of it costs this much and i made it with this buy it please lol.
20:00 - "Well-heeled" as a synonym for "rich" is actually quite common in older writing. It didn't strike me as odd at all!
So... I just had to ask ChatGPT to write a story about you guys...
Once upon a time in a cozy, dimly lit living room, three friends gathered for a night of laughter, snacks, and a little friendly mischief. Will, Maria, and Katie had been inseparable since their college days, and now, in their late twenties, they still found joy in each other's company.
They had come up with a plan for the evening that was sure to provide plenty of entertainment. They had discovered the wonders of ChatGPT, a cutting-edge AI that could generate all sorts of content. Tonight, their goal was to see how wild, funny, and absurd they could get by challenging ChatGPT to create the craziest fictional stories.
Sitting in a circle on plush couches, the friends had their laptops at the ready. Will, the imaginative and adventurous one, kicked off the challenge. "Alright, let's start with something absurd," he suggested. "How about we ask ChatGPT to create a story about a penguin who becomes a detective in a tropical island paradise?"
Maria, with her wicked sense of humor, eagerly typed out the prompt and hit enter. The AI's response was quick, and it painted a vivid picture of a tuxedo-wearing penguin named Percy, solving mysteries under the swaying palm trees of the paradise known as "Palm Beach."
Katie, the one with a love for science fiction, couldn't resist the temptation to crank up the weirdness factor. "Now, let's ask ChatGPT for a story about a time-traveling toilet," she said, her eyes dancing with mischief.
The prompt went in, and the AI spun a tale of a magical toilet that transported its users to different eras in history. They had to navigate through hilarious situations, like narrowly escaping a medieval jousting tournament while still in their bathrobe.
The laughter in the room grew louder with each new story. They continued their whimsical prompts: a pizza that could speak seven languages, a superhero hamster who saved the world from an evil cheese-loving villain, and a professional sumo wrestler who moonlighted as a ballet dancer.
As the night progressed, their challenges to ChatGPT became even more outrageous, and their laughter turned into full-blown bellyaches. The stories generated by the AI were a testament to its creative capabilities, and the friends marveled at its ability to turn their wildest ideas into entertaining narratives.
Hours passed, and eventually, they all grew tired, but their spirits were higher than ever. Maria summed up the evening with a contented sigh. "Who needs to read books when we can create our own wild stories with ChatGPT?"
Will, Katie, and Maria exchanged a knowing look. Their friendship had always been about adventure, laughter, and the sheer joy of being together. Whether it was through their own creativity or the clever AI at their fingertips, they knew they'd always find a way to infuse their lives with a little bit of the extraordinary. So, with fond memories of a night filled with imagination, they bid each other goodnight, eager to see where their next adventure, or story, would take them.
21:55 well-heeled = wealthy
Out of curiosity, were all of these prompts in the same chat? If so, that might explain some of the recurring themes since the old conversations would still be in its context window. Also, if you do this again you should try with GPT4, I find it's a bit better than GPT3.5
That's a really good point.
Ah, that's what it is then. I didn't think it did that unless you specifically asked it to continue the story.
--Will
Up to now, AI generates content by squeezing all existing content through the details you enter. The "voice" of an AI story is an average of all content that fits those details. When the AI generates a story without humans entering any criteria, that will be the AI's *own* voice. That'll also indicate that the AI has moved a step closer to self-awareness.
It should NEVER get to that point.
ChatGPT repeats things and can't plan to an overall structure because:
1) in simplified terms, it's a giant autocomplete function. Cal Newport and Ted Chiang have both written very good articles in the New Yorker explaining how Large Language Models (LLMs) "write" by predicting the most likely word to come based on the frequency of their occurrence in proximity to other words.
2) none of these generative AI models actually understand their inputs or prompts, let alone their outputs. Gary Marcus has done an excellent post on Medium illustrating how short it falls using prompts like "draw a bicycle and make the parts that touch the ground orange".
3) Yejin Choi did a TED Talk that explains why generative AI is not actually learning in the same way a child learns.
But to the point at 47:43, I do personally agree with this because the majority of writing craft can be broken down into structures and frameworks. And just as the average layperson already has trouble distinguishing human created vs AI created art, the average reader will not be able to distinguish human written and AI written prose.
49:20 - Self-pub authors are divided on this, but there are a number already using it (and some who have been using it for quite a while) in many applications, from creating outlines, writing blurbs, writing ad copy, and getting it to actual prose that ends up in published books, in chunks of anywhere from a single line of description to entire rough drafts that they then work over in line edits for continuity and polish. Those who aren't using it are generally refraining because of their objections to the unethical way the training data has been sourced and/or because at this point, it's still faster for them to write their books. But some are turning to it to do things like create rough foreign language translations and then getting a human translator to review it in order to do foreign editions.
52:29 - this (and much more, like the ability to isolate specific areas of the image and regenerate and the new zoom out feature on Midjourney 5.2) is already possible. Top tier cover design firm, Damonza, put out their AI statement recently and it says this is how they will generate the different elements of their cover art. I have lots and lots of questions on this that I hope to cover in an upcoming Q&A with them.
54:52 - this is the key takeaway, imo. With many things, writing and other creative endeavors in particular, there's no substitute for learning by doing. Using a thesaurus, or a spelling/grammar check tool, or even using a generative AI LLM that's been ethically trained (on a corpus of material where the rightsholders opted in and were appropriately remunerated for it) to bang out an outline or random character profiles or to turn "a forest" into a more descriptive passage is fine to a point. But I think that point of making it too easy, when you get to the point of just hitting F5 over and over until it churns out something you want instead of sitting there, stringing words together until you can get it to finally click, something is lost.
(Also if you can't be bothered to write the line or descriptive passage, then it can't possibly be crucial to the narrative and you should probably cut it anyway, or rewrite it so that it is narratively significant.)
"...in the style of Hemmingway."
First, I'm, by no means, a Hemmingway expert.
That being said, the "style" of Hemmingway would not have included Faries. Second, Hemmingway's writing was simple, direct, and somewhat sparse. This reads like a Twilight fan writing what they *think* Hemmingway would write.
I'm a prog rock fan, and a huge fan of the band Yes. In some Facebook group, someone decided to ask ChatGPT to write lyrics for a song in the style of Yes. Then, the individual proceeded to ladle fawning praise over what ChatGPT produced. While the lyrics were decent - they made sense and they rhymed - there was one major problem with them: they weren't something Yes would produce.
Jon Anderson typically wrote the lyrics for the vast majority of Yes's catalog. Generally, when he wrote lyrics, the words he used weren't select based on whether they made any sort of sense in the song. Jon based them on the *sound* they made when he sang them. In short, most of Yes's lyrics are pretty abstract if not bordering on gibberish. There are some notable exceptions to this, however. For a time, the members of The Buggles (they wrote Video Killed the Radio Star - the first song ever played on MTV) to produce the Drama album. They'd reform to create Fly From Here. And, recently, Jon Anderson was replaced with Jon Davidson, whose style of lyrics is somewhat fanciful. However, the VAST majority of Yes songs had lyrics penned or edited by Anderson. And they generally made absolutely no sense.
So, something as lucid and understandable as what ChatGPT created couldn't really be what a Yes song would contain.
Don't be surprised that ChatGPT found a story template somewhere and filled it in with other "stock" stuff from its models.
I once asked Chat-GPT to write a story in the style of Raymond Chandler. It literally started the story with the phrase, "It was a dark and stormy night..."
40:13 it ends its stories like a college student realizing they're almost at the minimum expected word count and it's time to wrap this mother up 😂
Well, if Katie likes the name Fitzgerald so much, why won't she marry it, huh? Check mate, atheists!
I've found that every AI story, regardless of prompt or premise, seems to have a "happy ending". This is especially noticeable (and bothersome) in horror stories lol
It helps a lot to be more specific. I asked it not no write a happy ending once, and it worked. If it doesn't work the first time you ask for something, keep insisting.
So long story short: no AI won't replace human writers - it's predictive text. it's just copying other things and it cannot "create" anything new - it's telling you what it thinks you want to hear.
@@iloveblender8999That’s depressing as fuck.
Funny enough, Chat GPT is much better at wrting story-length poems than it is just writing stories. When I give it a story prompt and tell it to write in the style of Tennyson, Dante, and Milton, the story is still going to be super simple and cliche, but the prose itself is significantly better. I think there's something about the structure of poetry (rhyme schemes, length of sentences, etc.) that makes it easier for the AI to produce something stronger.
My new favorite hobby is making Skynet write stories with absurd premises in the style of Shakespeare.
Tell it to rewrite the Declaration of Independence in the style of Donald Trump, you won't be disappointed.
Before I start watching let me say that this is a very good idea for a video
Also, an actual comment: maybe you're interested in reading His Dark Materials trilogy + there is a movie and tv adaptation so a lot of things to discuss, maybe you'll enjoy them :)
I don’t understand what William means when he says something is abstract e.g. “his voice filled with reverence.” Could someone explain it to me or give an example of how to fix the kind of sentences he points out? It’s driving me crazy and making me paranoid that it’s in my own writing and I don’t know it 😅
It's so interesting seeing the way it puts out some sentences that sound good on the surface until you think abt them for a second and realize they make no sense, like fully only nice noises and no substance. Seems like a lot of AIs right now is like passingly okay sentence to sentence until you examine it for what it's actually saying and what it's saying is nothing and/or actively wrong when it comes to facts lmaooo
The thing about AI is that it cannot, by definition, make something 'new'. It cannot 'create', but instead produces amalgams based off the creative works in it's data set, which let's not forget are gotten largely against copyright and without permission. This is, for instance, why it's terrible if you try to prompt it to make something uncommon (I saw an artist once show this by prompting Stable Diffusion with a description he was hired for in commission, and compared the resulting images to the *actual* artwork he'd made for the commissioner and they were impossibly different and what the AI spat out wasn't even close to what the client had him draw. AIs can only handle common enough 'stuff' that it would have had a large enough data set to draw on for generation. The prompt was something about dwarves riding black unicorns into battle with the dwarves wearing steel armored shoes with bunny ears (and some other stuff) btw).
There is also quite a few problems even outside the ethical. There's such a thing as AI Inbreeding or "model collapse". Because AI datasets *need* millions and billions of samples to train on that pretty much *necessitates* the unethical scraping of the internet (such as scraping AO3 for writing and LAION for art and other samples), and as more and more of the internet is flooded with images and writing created by AI, the models are beginning to be trained on *itself*, which can cause a model to collapse. For instance, as the number of AI art images with weird hands grows, the models that scrape the internet for training will scrape those AI images, leading to more messed up hands, etc. etc. down the line.
And *also*, because of what AI generation fundamentally is and how it lacks the human ability to separate fact from fiction or as you said, really have an overarching "wait, this doesn't make sense" reaction to a statement, there are cases *already* of AI writing specifically generating wildly false information. For instance, the advice not to trust any nonfiction books on Amazon for things like mushroom foraging unless you are *certain* it wasn't AI written or assisted, because they've literally found examples of these books *recommending tasting to help identify poisonous mushrooms* or listing things like *literal death caps* as edible. AI cannot fact-check itself for shit, and even if it could, it would likely fact-check itself *wrong* because it could be fact-checking against previous generated content that was *already wrong*. Another example; AI generation sometimes, when presented with a prompt asking about a real event, will answer with a *fictionalized version from a game or movie* because it cannot tell the difference between what is fact and what is not.
And you mentioned the similarities to fanfic, as well as another comment on here that mentions it. It is a *fact* that AI tools like ChatGPT were trained off of specifically AO3 when it was discovered that AO3 was included in the Common Crawl dataset, which ChatGPT trained on. AO3 actually has a post about it on their website. Which opens even *another* massive potential legal trouble, because by definition fanfiction is based on another person/group's IP. So even if, say, Disney forced an AI company not to train off their IP, if they trained off of scraped Disney *fanfic* or Disney *fanart*... etc. etc. (tl;dr AI in creative spheres especially is a massive ethical quagmire and AI is starting to already show massive problems with accidentally inbreeding on itself and it's datasets tend to end up with output that distills to same-ness as a result)
Question, did you guys use the same chat "conversation" to make each prompt? Becouse that could explain a lot of the overlapping details. Within one stream of conversation it will remember the previous prompts and incorporate them thinking that you are just tinkering with the concept. If I want to get something completely diffrent i just open a new conversation for it.
I put this same prompt into GPT-4 and told it to make the prose as purple as possible...
Beneath the ominous, violet sky, the air cradled an intoxicating blend of dusky jasmine and blood-tinged iron which only spelled one thing: fairie land. Amongst the dappled fawns cavorting amidst the violet foxgloves, there danced a figure of devastating beauty and implacable power: Rhysand. His eyes glinted with a preternatural azure, hypnotic as they were dangerous; they were cerulean flames flickering under shaggy ebon lashes, beckoning the unwary into his bedazzling snare. His virile frame exuded a mystifying allure, molten darkness moulded in a silhouette of celestial sin.
Behind an ancient oak, a pair of ember-eyes watched Rhysand’s frolic with a blend of fascination and defiant suspicion. Feyre, the fiery enchantress of the mortal realm, with her hair cascading in rusty tendrils kissed by the copper sunset, held whispered tales of the mysterious fairie. A woman wrapped in beauty and battle scars, her strength was an ominous thunderclap in the tranquillity of the fairie land, her spirit a stormy tempest that dared to defy Rhysand’s bewitching tranquillity.
“Rhysand,” called she, her voice a melodious challenge twining with the lilt of falling leaves.
The fairie, ensnared by her calling, moved towards her, his poise silken as lust entwined with the stormy sky, his expression painted in celestial intrigue. “Feyre, mortal maiden of audacious boldness. What brings you to my realm?”
Her eyes glowed resolute, like the last smouldering coals against the arctic night. “Curiosity, Rhysand.” Her ripe mouth curved into a tantalizing and threatening smile in harmony; her voice, pure honey laced with poison ivy.
A burst of mirth erupted from Rhysand, disarming as the dawn but underlined with the lethal promise of the midnight tide. “You intrigue me, Feyre, with your beguiling spirit and blistering audacity. A blazing star amidst a void.”
Feyre stood her ground, her fiery gaze searing into his icy one. “I am no star, Rhysand. I am fire. And if you’re not careful,” she lifted a finger, piercing his chest with a point fiercer than a faerie arrow, “I will burn this entire world down.”
A dangerous smirk unfurled upon Rhysand’s mouth, exposing the lethal undercurrent of his presence: “I’d love to see you try, Feyre.”
So ignited a battle of wits and allure, a cataclysmic intertwining of fire and ice, mortal and fairie. Feyre, the formidable mortal tempest, and Rhysand, the celestial fairie abyss, threatened to consume - nay, ignite - the fairie realm in a conflagration that neither time nor magic could ever quench. A celestial dance was thus born, drawn in hues of honeyed danger and sapphire audacity, unfurling under the violet heavens.
This was amazing.
Also, you can't roam with purpose? The Mongol Empire would like a word.
Omg, that was the most extra sendoff 🤣 ChatGPT is really doing the most!
On a serious note, there are a bunch of stylistic choices in the stories created by AI that read exactly like fanfic. And I'm not dissing fanfic here. Fanfic stories offer actual logical context for those choices and they don't seem so out of place. Actually, now that I think about it, ChatGPT is even more annoying because it's getting even more people to shit on fanfic and we don't need more of that.
Fitzgerald is a shit name for anyone. Maybe as a last name, but any other use is horrible.
well-heeled means rich. It's an arcane phrase, but it's a thing.
If you prompted the F. Scott Fitzgerald right after the Hemingway, it's basically going to take the Hemingway as its base and edit from there.
21:54 "Well- heeled" isn't a nonsense AI phrase, it's an idiom used to describe aristocratic persons. Generally meaning genteel, aristocratic, well-bred, etc.
Or with that final prompt written we bring our exploration of ChatGPT fantasy to a close.
I won't lie, I've had chatgpt write me countless fanfiction where i get to live out my dream of being romanced by the greek god of war Ares. It did the job but got oddly obsessed with us going to the market as a main plot. But hey, i can't say i wasn't entertained.
Drinking boba while watching this was the wrong choice😂 choked a few times from the laughs
That was fun, I'm happy to know I can write better than an AI. Hopefully that means I can survive Skynet too.😂😂😂
This was fascinating!
All this talk about AI makes me worried for sentient AI. Not because I think it’ll be evil or anything but because I know the half reason Chat GPT and AI art are such a big thing is because people don’t want to pay artists for their labor. Which means if/when the AI is sentient it probably wont be treated well and definitely wont be payed. It’s a lose-lose scenario where human artists are losing jobs and livelihoods for capitalism’s sake but then the thing replacing them will also be made to suffer. The only ones winning are the people selling the finished products.
It’s like shitty all the way through, except for this brief period at the beginning where it isn’t sentient or able to fully replace human writers in the eyes of the publishing industry. For now, we can do sillies with it.
For the first 20 minutes, I thought y'all meant "Phaedra" xD
All praise the Omnissiah
After ten minutes this hardly makes any sense... That's like taking all cliches ever and put them in one story.
No fitzgerald a boys name, ok, weird but ok,
but never a girl name.
Take darla, darla is good.
I love you guys so much but the political opinions aren't always fairly represented.
There's a huge anti capitalism message and while I agree that capitalism is flawed socialism isn't a good replacement because it refuses to account for corruption. That's my unwanted opinion. But now you may see my perspective. Again, love you guys but it the unrelated politics is very alienating. I have political channels to watch for that. I watch you guys to get away from that.
what? 😂 The politics are anything but unrelated...Hello??!
@@allbutperfect This is a book channel. It's one thing when a book is directly commentary on a political topic or holds a political theme. But going into "AI will be bad because capitalism...." That's a topic for another video. One that can be thought out and talked about on a NUANCED level. Problem with these blanket statements is we know people just take these things at face value and run with it. They don't consider the implications. There's so much discussion to be had around politics on its own and the politics of AI that it really has no place to say randomly all the time. And I LOVE Will don't get me wrong, but somehow in most videos he has a way of stuffing his very liberal political views in. Point being I couldn't directly tell you with any evidence where Katie and Maria stand in their politics. They never bring it up. Even when they've discussed books with political themes. Even though I can take a good guess there's no evidence present to back it up.
And quite frankly I don't come here to know their political views it's irrelevant. I don't want to hear them espoused on one of my favorite channels I thought was safe from having to hear it on any side. There was a time not so long ago you didn't have to tiptoe around the internet to try to stay away from politics, so I treasure the places that I can find people I enjoy, and even though I know I might not agree with them politicallly, it never even gets brought up. That's a respect thats a respect and courtesy that is completely missing in this country.
There once was a saying that I found very useful in navigating social atmospheres "There are three things you never talk about Politics, Religion, and your Wives."
The wives part is kind of wierd and is mostly based on your crowd or lack there of. It's like basically stating "Dont talk a bad about your spouse to others"
Anyways I didn't want to start some debate. I didn't appreciate the condescension in your comment particularly. I'm always open to discuss things like civil adults even when it's clear we don't agree.
Sometimes we are just unaware. I used to be VERY unnecessarily politicallly charged and I still have to fight it. Sometimes I need a friend to say "Hey that's not really the discussion we are having"
And in my pointing it out I wasn't trying to debate or be rude in anyway but the political comments do come off a little snide and of a place that "knows better" especially to those that don't share the same view.
I'm talking in circles now. I doubt anyone is still reading.
Conclusion, capitalism is flawed, but it's not contributing anything to a conversation about HOW AI CREATES SHORT STORIES.
If we want to hear how capitalism sucks we go to channels to hear about how capitalism sucks not to our funloving bookclub channel.
I WANT TO BE ABUNDANTLY CLEAR HOW MUCH I ADORE WILL MARIA AND KATIE. They each remind me of friends I use to have and miss dearly. I'm not trying to bash anyone of be mean or judgey. That's another reason I don't want to even debate political subjects because online debates of ideals are completely out of context and have no depth of the person or understanding. It's very..robotic because we don't treat the opposing side like theyre human with lived experience. I know they have no idea who I am as a person and a part of me would die inside if they ended up hating me over an internet misunderstanding. Please please please know that I am speaking as much as a parasocial friend can speak from.
Interesting how right wingers complain about safe spaces, but demand them for themselves... It's their channel and they can say what they want.
I really like you guys, so hearing Will go on his socialist rant really disappointed me. I never expected any of you to be commies, I never actually thought about politics when watching this channel, you were supposed to be a relaxing and fun escape from it. I don't want to stop watching the channel, because I genuinely enjoy it, but now I'll have a bitter taste in my mouth every time I do. Something was lost here, my "innocence" if you will, and now I need to mourn it. Deeply I'll grieve, because it truly was once sweet, warm and cozy, when I was still ignorant.
🙄
I hope this is a joke lmao 😂 if yes great copypasta