Fact Checking the "AI Steals Copyrighted Materials!" Debate

[@theoddegg8462]

Hey Nerdy Novelist! I'm so glad you made a video on this subject because I just started using Chat GPT as a means of adding to the conversation surrounding the ethical use of the program! After watching your video though, I do have a question: From what I understand vs the observations ive made since using it, If AI has to learn the kinds of words (and subsequently concepts, since one directly leads to the next) and images by scanning almost the entirety of the internet to increase the probability of associating the next likely word, then how likely is it that it could accidentally take characteristics, including names (made up or otherwise), ideas, or even physical descriptions, from the works of other authors without you knowing? How would you even begin to check for this? For example: I found it pretty strange that whenever I told Chat GPT to generate an alien planet, it would either give me something pretty bland or it would give me Pandora from AVATAR - floating rocks and bioluminescent plants and all. This didn't happen every time, mind you, but it did bring up an interesting question that Im hoping you can add your thoughts to since you have waaaay more experience using these things than I do. Important distinction that I've noticed: when I asked chat gpt where it got its ideas, it simply told me that it pulls from various tropes and ideas within its genre (something that we as humans do anyways), but it didn't get any more soecific than that - most likely because I didnt know how to specify or narrow down the question. Since then, I've decided to NOT edit any of the ideas that have been generated by Chat GPT so far because I want to see how much of it was stolen, if any.

[@Chociewitka]

the probability is increased the more often the given words are linked together in the language as a whole - as such you will probably get "black cat" more often than "purple cat". Still neither floating rocks and bioluminescent plants are solely characteristic of Pandora are, they are used in many rel life desciption and many stories and as such not copyrightable - just as e.g. dragons are not.

[@BurghezulDjentilom]

ludds be like "I don't want an answer, I want to be mad !"

[@BruceWayne15325]

Hey Jason, I'm curious whether you've found any Llama 3 fine tuned models for authors that we can use in NovelCrafter? I've heard that you can run Llama 3 locally, and while I don't spend a lot on AI in NC, free is always better :)

[@simpletongeek]

Good discussion about copyright and mathematical modelling. However, this is a complex situation and your perspective on it is incomplete. IP issues go much more than copyright. There's also patent, trademark, and accreditation. Can you imagine "I came, I saw, I conquered" as said by Abraham Lincoln? Another crediting issue is that people inflate their worth. Prompts don't make an artist. Prompts is the tool of Art Directors (who can't draw! - Christopher Hart) The two roles may be combined, but they're not interchangeable. Now consider writing words: are you really a writer who goes off by AI suggestions? A co-writer? A book doctor? An editor? A curator? Or someone who just put his name off randomly generated text? AI by itself isn't bad, but the lower down the chain, the worse off it gets. I don't see any explanations at all. Either they don't acknowledge it, or something vague such as "This work has been created with the use of AI". Considering that most people are lazy, they fall off the lower end much, much more. I've seen AI works, and they're really boring! Devoid of humanity. Until there's a clear filter that says "no AI assisted work" in searching books, this will remain to be a problem. I call it "Kindle Vella" problem. The idea of subscription writing is good, but so many people do it poorly that I need to go through hundreds of title just to find a good one. I don't have time for that, thus the market shrinks, hurting everyone. Sorry for good serial writers among you, but I just can't find your good works among bad! And if you're among the better writers, why do you need AI to write your words? Just have the AI create a bunchs of suggested plot, themes, topics, etc and go off that! That's a one time creation thing, to give you enough ideas to last a whole year! AI is good, but there's much wider applications off it. Path finding, pattern recognition, neural network, speech recognition, etc. Those aren't considered threatening. Even Chess/Go AI that are so good to beat Grand Master level aren't considered dangerous. So, why is it that some boring inhuman associative patterns emitter are considered harmful to the profession? Ethical issues abound!

[@Aurora12488]

> The two roles may be combined, but they're not interchangeable. Now consider writing words: are you really a writer who goes off by AI suggestions? A co-writer? A book doctor? An editor? A curator? Or someone who just put his name off randomly generated text? Why does the label matter? I don't care if people judge creators appropriately for the skills and effort involved. I judge someone who took a photograph of a cat much differently than someone that drew a photorealistic portrait of a cat. And I can appreciate the objective final image independently, and prefer a nice photograph of a cat over a portrait that's attempting to be photorealistic but falls into an uncanny valley, even though the latter creator's skills are likely *way* beyond the former's. > I call it "Kindle Vella" problem. The idea of subscription writing is good, but so many people do it poorly that I need to go through hundreds of title just to find a good one. I don't have time for that, thus the market shrinks, hurting everyone. Sorry for good serial writers among you, but I just can't find your good works among bad! This is mostly a recommendation engine and quality filtering problem. KZbin does a very good job of this IMO. KZbin is completely open for everyone to post to, and the *vast* majority of content on KZbin is extremely low effort and low quality. But we don't even have to think about that that much because it presents the content in a way that most of what we engage with is relatively high quality and relevant. > And if you're among the better writers, why do you need AI to write your words? Just have the AI create a bunchs of suggested plot, themes, topics, etc and go off that! That's a one time creation thing, to give you enough ideas to last a whole year! This channel has made quite a few videos explaining the reasoning behind this; please watch those. My 50-word summary is: working alone with a blank canvas is extremely mentally taxing, while working with AI can be much less mentally draining while still ensuring the output fits your intended vision down to the smallest detail. And that a "well suck it up, that's art" attitude here can actually *hurt* personal creativity, growth, and output quality.

[@simpletongeek]

@Aurora12488 Why does label matter? The very next few paragraphs explained it. Anyway, you don't have to start from blank. You can leverage AI to seed ideas. To the smallest details? I disagree. That's just ignorance. I read his book, and I noticed a change in tone/personality/voice between his voice and AI voice. If you can't tell, that's on you.

[@wordcharm2649]

You said AI is boring therefore anything produce 100% by AI won't impress the consumer. The market will take care of itself. You act like people are stupid. I can read a novel written by AI and see 1000 issues it has instantly just by reading a few chapters. There are flaws EVERYWHERE that a human with skills and know-how can fix, but the machine cannot. That has value right now.

[@simpletongeek]

@@wordcharm2649 i need a search function where I don't even see them to give human artist/author a chance of exposure.

[@billiecruz4399]

@@wordcharm2649 "the market will take care of itself" truly spoken like an american i could have ai generated your entire comment by asking chatgpt to generate a stupid overconfident american whose god is money

[@neuroticnovelist]

I’m curious how/where they’re obtaining the copyrighted material to train the AI? Especially fiction. Do we know the answer to this?

[@BurghezulDjentilom]

buying a copy seems easy enough.

[@BN-qo5zc]

They downloaded pirated collections (like one named "Books3" dataset) and took copies from" digital library systems" like the Internet Archive (though courts ruled that was also piracy now).

[@carlhamblin82]

Thank you @TheNerdyNovelist for making this video, and the interesting and helpful analysis. I get what you’re saying that the models are transformative, and that all they are doing is making mathematical associations between words. But if the AI companies scraped data from thousands of copyrighted works (which ordinarily reside behind paywalls) without even having paid for that data via the ordinary means, isn’t that unethical? (By the way, I’m not sure if that is what has happened or not, I seem to remember reading or hearing it somewhere). And what about the argument that, had not the AI companies scraped that copyrighted data, the AI-generative bots wouldn’t be as anywhere as good as they are: does that not suggest that there is inherent value in the copyrighted works that goes beyond whether text(s) or artwork are reproduced verbatim by the generative bot? And that therefore, the owners of those works ought to be recompensed?

[@dewaynemiles1245]

Writers read and that reading influences what and how you write! What’s the difference? As long as it’s not plagiarism there is no difference as far as I am concerned!

[@billiecruz4399]

The only things an AI can do is dependent on human ability. Humanity is not limited in this way I don't know if you know this but humans can actually do things that other humans haven't done before This is why I find it odd to say that humans and AI are similar when AI is completely incapable of surpassing itself without help from a human. Helping this case being hundreds of thousands of images stolen from people without their permission used to create something that will bring monetary gain.

[@ThanosDestroyeryearsago]

@@billiecruz4399 Just put the fries in the bag

[@billiecruz4399]

@@ThanosDestroyeryearsago I haven't had to work in almost a decade all my needs are met and I go to sleep in the arms of a loving partner every night. I am a housewife, and i am so much content then you will ever be, boy. Go ask mommy for some more v bucks.

[@ThanosDestroyeryearsago]

@@billiecruz4399 Good for you if that’s what you want in life. I don’t play Fortnite though, you probably think I’m much younger than I am. Which is kinda crazy to snap at me like that if you thought I was something like 13. Very immature if I do say so myself.

[@billiecruz4399]

@@ThanosDestroyeryearsago from the sound of things mommy didn't give you your v-bucks. Have fun working for the rest of your life to afford an ass retirement that you will never be able to live to enjoy!

[@eldonmacwood]

I LOVE this video! Considering of the names I have been called by other writers and editors because I use AI, I appreciate this video. When I came to AI, I wanted to make sure I was using it ethically, and I do. I will always stand against those who use AI in an unethical way. Which people have. One author friend of mine hired a cover artist, said he didn't want an AI cover, and the artists agreed, only to send him an AI cover that the author paid for. Things like that is very unethical, but it's also all on the person, not the AI.

[@mommyof4grlz]

I've just started working with AI and my perspective is simple. It can be as and usually more difficult to get AI to do what I want as it was to get my children to do their chores. It requires more detailed instruction and rules, and it will still not be what I asked for it to do. On the upside, AI doesn't mouth back and grumble like my kids did.

[@shebreathesingold8043]

"It can be" and "usually" are important words. Some tasks via AI are not worth the effort while others are 100% worth the effort. It's everyone's job to learn what those tasks are. Similar to anything. I don't always open up photoshop for a simple task I can do in another software because PSD is a massive software to even open and I don't always want to go through the effort when I can just open paint or something to add some easy text, for example. Same with AI. Not only is AI not for EVERY task, some require more human input, but at this point we have so many AI-models. What I take to Perplexity isn't what I take to Gemini or ChatGPT or dozens of other models out there.

[@TheHypercarnivoreChef]

If I'm smart enough to copy the style,rhythm,and everything else from a book that i keep open next to me while i write another book in doing the same thing that an a.i. would do but faster and from a bigger library

[@TimWeaver-iz6oj]

The problem here is that the legal model we're using to judge AI, specifically image generation doesn't match actual effect cause by AI. A better model is when your employer asks you to train someone else to do your job so they can fire you and put them in your place for less pay. Only with AI, you don't even get paid for the effort to train the replacement.

[@Aurora12488]

Who's *actually* firing creatives and replacing their roles with AI, though? There's a lot of fear around this, but in practice it doesn't seem realistic. Anyone who has ever used these tools knows they're not very effective human replacements for all but the most basic tasks. Any company who fires their graphic design team and tries to have John from accounting use Stable Diffusion instead is going to absolutely decimate their quality and performance, and will be desperate to hire a new graphic design team in 6 months. I saw one graphic designer on KZbin say this happened to them, and "journalists" at a news post farm. In all cases, it seems like the company is either trash to begin with, or is completely shooting themselves in the foot.

[@TheNerdyNovelist]

Exactly. It’s actually the creatives that are using it effectively. They save a lot of time and are still able to achieve the same result they would have otherwise. Their knowledge of composition, lighting, etc only help them use AI in the right way. And unlike nonartists using AI, they know how to modify it to make it work (which also, incidentally, allows them to claim copyright)

[@MagnificentSails]

So I can write a book and ask chat GPT to put it in the style of someone, I still retain the copyright?

[@blacktreefoundation]

It's actually not illegal to copy style - I mean, we see it in music every Tuesday with every rapper now sounding like the same rapper, and it's not even a genre thing. Most young writers generally start by copying the style of their favorites before they find their own voice. The question though is why would you copy someone else's style? You will just end up being the 2nd rate version of the original. Keeping with the rap music analogy, I'd rather listen to Biggie than Gorilla Zoe any day. Zoe is just a really good Biggie impressionist.

[@MagnificentSails]

@@blacktreefoundation I was more wondering if chat GPT would take the copyright, or if you would still own it. In other words I wonder how much work check GPT can do and you can still maintain the copyright. I'm thinking more for a novel, like if you asked it to take your novel and write it in Jane Austen style, if you would still have the copyright

[@blacktreefoundation]

@@MagnificentSails I would imagine if the characters and stuff like that is copyrighted to you, nothing should change. Another writer can write in your world in their style and not own a single piece of the copyright. I’ve seen so many obvious Basquiat style rip offs in galleries and no one is arresting them.

[@MagnificentSails]

@@blacktreefoundation okay thanks 👍 I was more worried about chat GPT taking the copyright

[@merlinjim]

Love your work. I write using AI with moderate success and watch almost every one of your videos. I agree that the output of an AI cannot be considered to be infringing, as long as the person using it is diligent about making sure the output isn't a literal recitation of some copyrighted work in its training data. That said, I do think the company's creating the AIs should have to license the content they are using - whether that's just paying for the work once or some other license specifically allowing it to be used in training data. An author gets better by reading books. Artists commonly take a sketch pad to a museum and sketch the works they see there. All art is derivative, and an AI learns to make art similar to how humans do - by observing art and attempting to understand how to structure their own output in a way that captures the essence of succesful artworks. The people that complain the loudest tend to be the ones that don't understand how to incorporate it into their own workflow (something you do a GREAT job at teaching, BTW!) These people are rightfully afraid that the incorporation of AI into artists' workflows might make them less relevant or less competitive in the future. But we didn't stop building cars just because the horse ranchers were upset. AI is a tool like a paintbrush or a typewriter. Unless you're going to go back to burning wood to make charcoal and pulp to make paper, you're taking advantage of technology to make art.

[@keithtarrier4558]

Very good.

[@StefanTrifonov-t4r]

Every book out there is a copyright infringement of the dictionary 😂

[@KingZero69]

great video. agree 100%

[@keithdixon6595]

1. I agree absolutely with your take on how AI works. 2. PLEASE raise the treble on your vocal track when rendering your video. It's almost painfully bass-laden. Humans distinguish speech more easily through higher frequencies and your sound seems to dwell almost entirely in the lower frequencies - it sounds like you're talking into a padded tin can.😊

[@keithdixon6595]

I like the way you describe how the language models work in 'predicting' what the next word should be: peanut ... butter. It explains to me why the prose created by AI models is so banal and cliché-ridden - it can't for the life of it find a more unusual adjective or phrase in descriptive sentences. Hence the prose itself is never more than workmanlike at best, never original or sparkling with wit. Good prose comes from finding original ways of saying something, and AI just isn't there - yet.

[@matt_v2305]

Keith I think this is a problem more your end than the vid audio. I am not getting the issue at all you are describing.

[@keithdixon6595]

​​@@matt_v2305Of course you may be right, but I'm not the only one to comment on this. Others have said similar things on previous videos. I don't have the same issue with other KZbin videos, and I watch a LOT of KZbin, sadly ... 😔

[@YorickvanVliet]

I like using AI tools, but I can understand the position it's putting artists in. I've made a free game that made youtubers about $200K in reaction videos. That is considered "fair use". And so it might not be immoral. But you said it exactly right. It didn't FEEL fair. We can try our best to at least empathize with those feelings.

[@MagnificentSails]

Can I have chat GTP edit my book, and I still have the copyright?

[@kit888]

Arguments against (don't shoot the messenger): - Copyright laws were designed with humans in mind, not a machine that can churn out thousands of works a day - Such a high volume causes harm to the original authors. The works are transformative, but if I can generate hundreds of novels in the style of Stephen King that are just as good as his books (currently not achievable by AI for writers but achievable for images), his sales will be affected. A human writer will have more difficulty achieving such volume and quality.

[@adroitws1367]

yep, the regulation need to be more clear about how we deal with this machine. we really should not apply human law to them directly. It already went so bad when we gave company so many rights like human.

[@simpletongeek]

Trademark office is currently struggling with Chinese fly-by-night operations. Those companies with names such as BYOOPED, CEROTINOTS, or LREENTIOL. Another thing to consider is that AI works are not copyright able! Umm, does that mean I can lift whole passages off your work because AI works are considered public domain? 😙

[@kc-jm3cd]

Whose writing 1,000s of books a day that actually get published

[@neopagan1976]

Writing styles can't be copywrite protected.

[@Aurora12488]

Sort of. Copyright laws were designed with the printing press, a machine, in mind. We didn't have laws around copying before that, and people mostly thought that that was a good thing due to proliferation of ideas, but society decided the printing press made it too trivial to just offer a completely alternate distribution channel for a particular work without the author involved at all, so we added some protections around that and things in that vein (which were also supposed to expire in a reasonable time, to get back to that "common good" philosophy). But everything other than the work itself, like all the stylistic properties that went into it, were still considered free for the public to still build off of, and that that was a good thing. Whether this machine that now interacts with *those* elements is worth more laws is the debate. Since now it's not circumventing the distribution rights to a work, but rather competing in the space of that work, which is very, very different. We have plenty of authors today that compete with each other, and yet there's still the desire to write and the ability to find success in writing from the quality of what you bring to the table, with a mixture of derivative and unique elements. And you're already competing directly with what are essentially AGI (ignoring the "A" part since humans aren't artificial, haha). Until AI are AGI, I think it's hard to argue they're going to drown works that are clearly better; they're going to drown out the crappier stuff, but also make it easier to *not* make crappy stuff, lifting everyone.

[@Aurora12488]

I'd actually approach this a different way. The term "Fair Use" is actually misleading; it's denoting circumstances within the scope of copyright that we allow to be exceptions. Essentially, we're waiving copyright in these instances. However, there's a whole bunch of "uses" that fall outside that scope of copyright entirely! For it to be inside copyright, you have to be reproducing the original work as-is or creating a "derivative work", which the legal language around is extremely vague but is really pushing the narrative that it is a work that is an *adaptation* of some form, even stating in their copyright office material that "the derivative work right is often referred to as the adaptation right". I think the way that an ML model interacts with its training set is so far and away from a "derivative work", that the only way for it to cover what ML models are doing in general (unless they're overfit) is for it to cover literally anything you as a human *also* produce after having seen a work, which just seems silly. And if that's really seen as the case, then the fair use pillar around the "amount & substantiality" would essentially be able to trivially override the other pillars no matter what.

[@Aurora12488]

I should note that Japan has *no* concept of fair use within their copyright law, and yet their government has explicitly come out and said that training on copyrighted works is legal. That should give some pretty strong evidence that this can be considered outside the domain of copyright entirely, and thus outside of needing a fair use justification.

[@billiecruz4399]

@@Aurora12488 nothing makes me happier than to read ai bros pontificating about they don't owe creatives anything after they've stolen their work Honestly dude I couldn't write you people as better villains than the ones you paint yourselves as

[@pieugedher257]

👍

[@mprado4177]

I agree with you 100%, but you know this is the way it goes: It's not about right or wrong, it's about how many frivolous lawsuits can lawyers get away with to make their outrageous amount of money?

[@ariesmarsexpress]

You are 100% correct. The way that AI learns (as distinct from the way it thinks) is not all that different from the way humans learn. If AI is still stealing, then the entire thing about having authors read as much as they can of other authors would be racketeering at the very least. At least with AI it is not purposely setting up story patterns for other AI to use to write stories based on existing working patterns like humans do. You are lucky many times for the AI to reproduce what you or it just wrote on the screen 1 second ago. The idea that it could reproduce someone else's work by chance is just nuts.

[@DavidDorianRoss]

Exactly!

[@rosefriday4287]

Agreed 💯 We're all standing on the backs of giants, whether we use AI or not

[@billiecruz4399]

Maybe AI is similar to the way you think and learn But honestly when you people say things like that it just kind of makes me feel like you're less human

[@ariesmarsexpress]

@@billiecruz4399 This intrigues me. You have no substance behind your statements. For my part, what I mean by we learn in a similar way is neither humans nor the current round of AI remember a factual copy of anything they learn. For humans, we remember largely symbolic representations which are later supplanted and rewritten over time with further symbolisms. For AI, everything is broken down into its tiniest symbolic parts. A face for instance might become a thousand constituent parts which are later supplanted by each one of a million other faces and their constituent parts, all represented by probabilities of their existence in proximity to each other. In neither case, is the original work, the face, the sound, etc stored for later for factual retrieval. There is a concept in U.S. Law where an eye witness testimony carries a lot of weight in criminal cases, in reality it should never be used as it is the least accurate piece of evidence possible. It is no different with AI.

[@billiecruz4399]

@@ariesmarsexpress This comment is so anti-human it's ridiculous. Reducing human memory to "symbolic representations" that are "easily overwritten" completely ignores the depth and complexity of our experiences. Human memory is emotional, sensory, and deeply personal-it's not something that can be boiled down to a mechanical process. Comparing how we remember things to how AI processes data is an insult to what makes us human. We're not machines; our memories shape who we are and how we connect with the world. And really? Dismissing eye-witness testimony as "the least accurate evidence"? Sure, memory can be flawed, but it's also powerful and meaningful in ways AI could never replicate. This whole comparison between humans and AI is just oversimplified and dehumanizing. We’re more than just data points or symbols-our memories are what make us alive, give us our identity, and allow us to empathize with others. This comment completely misses the mark on what it means to be human. Not a big surprise tbh

[@MagnificentSails]

When was chat GPT last updated?

[@momob4276]

I'm on the opposite side here. Thought experiment.: An AI company approaches you and says hey can we use your books to train our AI model. You won't get paid or credited, but we NEED your work to help our product which we're going to sell back to Google for billions of dollars to make us wealthy. Again you are not getting paid or credited. It just feels very icky. At the VERY least with transformative works using copyright material you can either credit or clearly trace said copyrighted works back to their original creators. Transformative made by humans are open and honest with where they are sourced. Even when a writer or artist creates a work based off of someone else's work they are either honest or you can tell where their influences came from. But with AI training there is none of that and these models usually go through great lengths to conceal whose works it is being trained on (hmm I wonder why?). Legally is it stealing no, you won't go to jail or be fined. But is it still stealing yes. And the proof is the fact that different writing AI models are better or worst depends on which written works and how many written works it was trained on to be improved. So it's not simply a matter of predicting the next word, it needs specific types of written material to sound human enough. And again, no way to, credit, trace, or know whose work was used to be transformative and the fact that is being so heavily concealed and dishonest is why I count it as stealing, but not in a way where legal action needs to be taken.

[@keithtarrier4558]

If you could edit that, and put it in a few paragraphs it would make for easier reading for those on a small screen like an iPhone, compared to a computer or large iPad. I hear what you are saying!

[@Aurora12488]

When you draw a cartoon, is there any way to clearly trace the influence of that style back to the original creators? The creator of the anime girl in your profile pic is very clearly deriving from the copyrighted works the creator saw, but there's absolutely no credit or tracing back done. The nature of the use in an ML model is the exact same as an anime illustrator having watched a bunch of anime and now using abstract concepts from what they watched as input into their art process. And the reality of what that process is is sourcing patterns discovered from having interacted with countless relevant works, to the point that there is no legitimate concept of "tracing back". With an AI model, the weights represent *patterns* , and each work in the training set does the absolute smallest of nudges to *every single weight* in the model, in a way that no substantial information about that particular work is encoded, but over exposure to billions of inputs, large-scale patterns are able to be encoded. The "trace" would legitimately go back to every single work in the training set, with each having an imperceptibly low influence each to each of the billions of weights. The AI companies try to conceal the training set because they know that *emotionally* people are going to react poorly, even if *rationally* the AI company employees still feel they have a strong basis for why this is both legal *and* ethical. If you know people are going to misinterpret and mischaracterize the process behind what you do, of course you're going to avoid being super forthcoming about your process. That doesn't mean the process is inherently "wrong"; it can mean the *reaction* is "wrong" instead, and so they're just trying to avoid that reaction. Of course an AI model trained on more high-quality human-produced sources is going to sound better and more human. But that doesn't mean the training process was unethical, and for the reasons above I *don't* think "traceability" is a sticking point, and while I think being open about the training set would be *nice* , given people's unwillingness to hear any argument other than "this is stealing from me!!", and the likelihood of angry people trying to misrepresent any accidental inclusions of *actually* illegal data in the training set as representative of the entire set even if that amounts to (literally) 0.00001% of the inputs, I can't fault the companies for just not wanting to have to deal with that with a ten foot pole.

[@simpletongeek]

So, 200 people prompts "cute blond high school anime girl", get the same image, and sue each other for copyright violation? 😂

[@williambarnes5023]

*> "An AI company approaches you and says hey can we use your books to train our AI model."* Well, was I dumb enough to put my book out online where anyone can read it for free? If so they have it already. Otherwise I link them to my storefront where I sell the books. These AIs in question do not contain nor do they reproduce originals. Fanfiction is not stealing.

[@dewaynemiles1245]

What if I download your books, read them, and then teach myself how to write using your books as a guide for style and genre? No AI involved, is that ok? Just curious.

[@rebeccam.riordan7165]

Thank you for an articulate and well-reasoned presentation. Fwiw, I think one of the key ethical issues is that people publish their writing to be read and post their visual work to be looked at. The…intelligence, for lack of a better term…doing the looking and reading seems immaterial to me. Add to that the centuries-old tradition of copying masters. I vividly remember struggling to write a passage “in the style of Fitzgerald” for a creative writing class, and copying old masters has been an important part of training in the visual arts since at least the twelfth century. It’s useful, accepted, and only forgery if you try to sell it as the master’s work.

[@chillingfolklores]

Thank you so much for this video. This is truly helpful. I am currently writing a murder mystery thriller story on novelcrafter and with the help of Ai. I am a writer and I love to write but after using Ai to help generate ideas, it helps me to finish my work so much faster. I believe in the statement ¨WORK SMARTER NOT HARDER¨ Ai is making our lives so much easier and has changed my life completely. Thank you again.

[@criticalchai]

Great video on the subject.

[@ipodchips2008]

first.

[@keithtarrier4558]

What was the medal you got pin on our chest for being first?

[@reginapetty2042]

Guess what kind of award everyone else is bestowing on you.

[@coinbounty3347]

Is it just me or is youtube getting horrible views right now? So many channels are tanking in views. Idk why

[@jaunwait1017]

I suppose its alright if they paid for the copy?!

[@KadeStringer2.0]

I believe that AI will steal stories as it does steal artwork to a degree . Ai can’t create anything but it can help you to a degree .

[@nahlanoelle8825]

I don't know anything about AI-art, but AI-text isn't stolen. It learns rules about how writing works and uses that to produce an infinite amount of text. That isn't the same as stealing art.

[@billiecruz4399]

@@nahlanoelle8825 how is it different?

[@DGaryGrady]

I’m in broad agreement with you on the legality and ethics, but I do have a couple of technical quibbles: First, as already noted below by @kit888 and others in that thread, the “autocomplete on steroids” notion of how LLMs work is misleading and doesn’t explain why LLMs can produce responses and even whole essays that are well-structured and don’t resemble stream-of-consciousness rambling. We don’t actually understand how LLMs (or many other AIs based on neural nets) operate. Anthropic has published some very interesting research about its attempts to understand what’s going on inside Claude. For thousands of years humans could make fire without understanding the chemistry of combustion, and something similar is true of our ability to make artificial intelligence by rubbing GPUs together. On a separate subject, diffusion-based image generators, at least, don’t operate by producing a pixel at a time. The starting point is an image of the target size containing pure random noise. That image is then iteratively de-noised with the prompt supplying a description of what the image should look like with the noise removed. It’s surprising that this works, but it’s surprising that any of this works.

[@pieugedher257]

7th

[@MagnusItland]

I am not sure how these people think Anthropic, OpenAI, etc get those copyrighted works. Do they download them from The Pirate Bay or the modern equivalent? If not, someone has probably bought the book originally, just like someone bought the books I read at the library and the books I read at my parent's house growing up, with hundreds and hundreds of books lining our walls. Am I supposed to pay for them too? Or perhaps pay every time I remember them? The problem here is that any argument you can use against text-based AI can also be used against humans. And unlike the AI training, we don't wipe our brain clean of memories after we finish reading. My brothers and I used to entertain ourselves by quoting dialog from Ibsen's "Peer Gynt" to each other, verbatim. I could certainly include that in a book, but I don't. Most AI probably couldn't even if they tried.

[@wordcharm2649]

Exactly, we ALL learn from what we absorb. We all copy and steal each other's ideas all the time. It's how we progress and evolve which is why we don't even think about it as stealing. When you had coke, you then got pepsi, sprite, dr. pepper, etc. Imagine if we never allowed others to steal/borrow ideas, we'd never evolve in anyway.

[@billiecruz4399]

sweet then give me your money. value is fake, nothing is worth anything really, so give me your money

[@Pekish79]

The problem is the legal mindset we have in the USA. Sometimes, when I was little, my mom in Italy used to say, "If it feels wrong, don't do it." I am actually using AI (as a hobby) and like AI (as a sci-fi lover), but it feels wrong to use stuff without compensation, especially for big companies. The video correctly examines all the legal mumbo-jumbo and the nitty-bitty details behind the existing law. However, the law is only sometimes fair or correct, especially with the changing technology. The law remains behind, and we can't believe that it defines right and wrong as absolute; we have to use it as a reference to be fair, but some problems go beyond what a current law can analyze correctly. Right and wrong are in our understanding and morals, not in the law; philosophically speaking, I feel we should reward people for their work, especially if we need it or use it to make our own money. Still, as human beings, we have to reappropriate the morals deep in our DNA/soul or whatever we think we have deep down there. Ultimately, I am not a legal person, and if I look at art like Collage art and famous collage artists Martha Rosler, Max Ernst, Kara Walker (she is from California!), Richard Hamilton, Goshka Macuga, Nancy Spero, León Ferrari, David Maljkovic, John Stezaker, Hannah Höch were they sued for copyright infringement? Probably not, so AI doesn't really infringe copyright. However, AI companies should be more understanding and generous with artists, in my opinion, should include them in the discussion, and should come up with some remuneration; we need to defend the art!

[@wordcharm2649]

"It feels wrong" is subjective. It feels wrong for poor people to need to make money and have AI to help them quickly get that money AND not do it because someone online is telling them it isn't fair. My dude, life isn't fair. Let people who need the jobs and the income use AI to get it. I guarantee NO ONE who actually needs to work is ignoring AI. That's a really obnoxious position of privilege being spread around to keep everyone from advancing.

[@Pekish79]

@@wordcharm2649 if you don't feel wrong, you are a moron short answer.

[@Pekish79]

@@wordcharm2649if you kill someone (not in war or self-defense), it feels wrong, or you are a psychopath. If you see a beautiful field of flowers and you dump your old dryer and plastic back, it feels wrong, or you are a moron. If you see a person starving and you have a piece of bread you were going to throw away but is still eatable, you give it to him, or you are an asshole... if you don't understand what "feels wrong" means and think is subjective I am so sorry for you and people that know you. Too many stupid people on the internet bypass every standard moral rule and regulation with a silly attitude. (if you repurpose other people's art to make your own work/money, you should give them recognition, often in the form of cash. If you get it, give them nothing and get paid a few million a year. It FEELS WRONG, or you are a....wordcharm2649; that is probably the worst insult I can make nowadays)

[@billiecruz4399]

@@wordcharm2649 what poor people are using ai? I'm flabbergasted when you people say this kind of stuff cuz I'm not seeing the poor Market, after all to use any good AI you have to pay the company a lot of money. And if you don't show up for the good one well congratulations now you're taking three times as long cleaning up after the AI To me it seems like rich people are saying think of the poor people while they specifically take money away from the poor people Can you provide me proof of otherwise?

[@reginapetty2042]

While I agree that AI cannot yet generate a new novel in the style of your favorite author, I think it’s just a matter of time before that capability exists at a low cost to the user. It is mentally taxing for a person to write a novel in the style of another author, and so preventing this by law seems like overkill. However, in the future, this will be easy with AI. As an exercise for the reader, use your favorite AI tool to write “Twinkle, Twinkle, Little Star” in the style of Edgar Allan Poe. When scrutinizing the output, consider that AI is in its infancy. We have already begun to see how these conflicts might play out. See Scarlett Johansson’s case against OpenAI for using the likeness of her voice without permission. Imagine a world where these creative AI tools are as commonly used as web browsers are now. Let’s say that I and loads of other Patrick Rothfuss fans have independently decided not to wait for him to wrap up his series, and we use these tools to build out the series, etc., for our own personal libraries. Even assuming that individuals don’t share or sell these creations, is it likely that this kind of use could impact Rothfuss’ sales should he start adding to the series? I think so, especially if the cost of creation is at least as cheap as it is now. Why pay Rothfuss if I can get something as good as his work for free? Imagine HBO using these tools to create screenplays in the voice of a collaboration between Martha Wells and Andy Weir. Not only would HBO be profiting from the combined decades of effort of these authors honing their craft and refining their voices, the combination of authors’ styles might make it difficult to prove the impersonation to compensate them after the fact. What’s the harm? Imagine if this were to become so incredibly commonplace that AI generated the majority of content. It would outcompete human-created content. Are we ok with content creation becoming increasingly AI/algorithm generated? The more I grow into the crazy-lady-shouting-at-moon that I’m destined to become, the more I see that industrialization has been good in the short term and bad in the long term. Especially when we don’t take care to nurture how emerging technologies are adopted. Example: evolution of farming from local, organic, family business of whole foods to one of global, non-organic, monopolistic factories that churn out food stuffs that addict us and make us sick. The technology didn’t have to take us down that path, but we allowed it to.

[@wordcharm2649]

Yes, but "in the style" as you're using it, you're talking about the superficial prose style. There were a thousand dimensions of narrative that Poe used to be Poe and those are harder to implement, even for a machine. If you are a good writer, you know that writing is all about making choices. Literally hundreds of thousands of them, from the line level to the narrative level. NO ONE CAN REPLACE POE OR MAKE EVERY DECISION HE'D MAKE SO YOU'RE NEVER GOING TO HAVE A POE LIKE NOVEL, JUST ONE THAT SUPERFICIALLY SOUNDS LIKE HIM. PERIOD.

[@arkycookie]

Second

[@roglacken]

third

[@tearstoneactual9773]

I asked ChatGPT to quote The Viking's Prayer and it flat out refused. :D

[@kit888]

Your explanation of how AI works is partially correct. It's not just probability because word probability is not enough to predict the next words in a sequence. What the AI engineers found out is that the neural net needs to "understand" the underlying mechanisms of the world in order to predict the next words in a sequence. The problem is that they don't know how the AI does this except in a high level understanding that they adjust the neural net node weights. The LLM is not just building a probability table of words. They are doing neural network stuff like back propagation, adjusting the weights of the network nodes based on how well the neural net outputs match expected outputs.

[@copester1204]

I wish more people understood this. LLMs are not solely prediction engines, they do include rules on how the world works. They know an apple is a fruit and fruits can be eaten (and they are not going to speak dialogue unless told to)

[@Aurora12488]

I think you're misunderstanding. The final output *is* a probability calculation. But how it internally models the state that gets to that final probability assignment is similar to what you said. You're not refuting, just adding onto it.

[@melkorbane]

This isn’t really accurate either though. You’re conflating two separate phenomena. Neural nets don’t understand jack.

[@Aurora12488]

@@melkorbane They *do* encode patterns that are picked up autonomously from the training set. That's what ML researchers mean by "learning". Those patterns represent real useful information, and I think using "understand" colloquially to describe how these models encode and leverage those patterns to be totally appropriate. Obviously nobody thinks it's doing the *exact* same thing as a human, though there are definitely a lot of similarities.

[@kit888]

@melkorbane Your brain is a neural net, so you don't understand jack too.

[@andrewnomicon]

Well said well said. Good that their are people like you amidst these so called artist drowning everyone who does not kowtow their way.

[@billiecruz4399]

stealing is bad kid

[@therealmountainmanmike]

All writers have trained their brains on copyrighted material. This is not illegal. Why should it be any different for an AI?

[@Dr.RajivSinghal]

There is no scope of inventing aeroplane again ,we can use the knowledge to make different aeroplanes. Similarly humans are going through this transition as doctors are afraid of ai ,robotics, so don't worry what's good will be liked let it be😅

[@MartinNewton1981]

Art and culture are inherently derivative. Throughout history, creators have drawn inspiration from existing works, remixing and reinterpreting them. AI continues this tradition on a larger scale, allowing for the evolution of culture by generating new content that reflects and builds upon previous works.

[@cautionroguerobots]

Jfc I cant believe how often I see this stupid belief parroted.

[@Aurora12488]

@@cautionroguerobots Jfc I can't believe how often I see low-effort, stupid attacks parroted.

[@hastesoldat]

@@cautionroguerobots And I find people who believe that humans can create ex niholo absolutely insane. And the fact that so many people seem to believe that makes me feel alienated.

[@wordcharm2649]

EVERYTHING IS DERIVATIVE. "Monkey see, monkey do" is the most basic fundamental truth of humanity. We see, copy, imitate, take. People acting like AI has no right to do what we all do is just telling on ourselves. We do this and we've been taught it's wrong so we frown upon it, but is the only way we evolve. Your idea may help you do ONE thing, but if you give that idea for a million people to take, they do a million NEW things with it. Like, hello, people who are against AI are literally against human evolution. It's stupid.

[@wordcharm2649]

@@cautionroguerobots It's not a stupid argument. You're just not intelligent enough to understand it. And you're too insecure to understand that all your work is derivative. If you admitted it, your insecure self would implode because you're that mentally fragile.

[@guidedmeditation4]

People need to start reading terms of use on every platform that they use. If you don't want these companies to use your stuff to train these ai then don't load it. But let's been real, Google has been doing this for ions. Come on change the record already...gees

[@nahlanoelle8825]

If you are anti-AI then get off youtube, google, twitter, facebook and every major platform because they are ALL using it. Oddly enough people are anti-AI but not anti enough to stop using platforms they enjoy. Hypocrisiy 101.

[@billiecruz4399]

"why did you go to the public pool when you knew that rapists could be there, its really your own fault for getting raped. You went were rapists go and got raped, simple as" You people fit into every negative stereotype that has ever been made by tech bros, doesn't it bother you? To live as a stereotype?