This feels like probably copyright infringement. Did the creators pay for the books? I've done my share of piracy in my day and all but authors deserve to get paid and I just can't get behind...
This feels like probably copyright infringement. Did the creators pay for the books? I've done my share of piracy in my day and all but authors deserve to get paid and I just can't get behind this.
(Alice in Wonderland the book is theoretically fair game)
But uh, Dune? The Matrix? Star Wars? They're going for the Mouse? Good luck.
So, having tried it myself... It doesn't seem to be pulling from them very well. I tried it with Eragon, which as a disclaimer isn't on the front page, and it basically felt like trying to...
Exemplary
So, having tried it myself... It doesn't seem to be pulling from them very well. I tried it with Eragon, which as a disclaimer isn't on the front page, and it basically felt like trying to roleplay with ChatGPT. It knew the basic scenario and characters, but it wasn't really trying to heavily recreate the story. Just try to nudge me onto the path of the story and totally ignore my attempts to make a dragon egg omelet instead.
The home page says it pulls from Google Gemini. I don't know if that actually has full access to the books and other source material. A quick web search revealed that Eragon is available online for free (possibly uploaded by a fan or to a pirating site), and the start of the adventure lined up with the first chapter (hunting a deer before Stuff Happens).
And so. Experimentation begins!
So first, I tried Unwind by Neal Shusterman. It got the main character and scenario right, but the start was immediately different because it was more a summary of finding a letter from his parents that he'll be unwound rather than setting him in a scene directly.
The reason I chose Unwind is that I remembered that the evidence the protagonist found that his parents planned to send him to be unwound included tickets for a vacation for them and younger brother. His brother is a super minor character who I don't think was even named in the first novel (though apparently he was in the sequels, which I really need to read someday given how much I love that book).
So, I immediately tried to investigate that avenue, and it explicitly confirmed Connor had no siblings. I pressed that angle, it mentioned a best friend, and I started pressing that angle using the AI-generated prompts. It then revealed Connor apparently has amnesia because his friend's face is foggy, but he probably got unwound, and his name was... Marcus Unwind. At which point I accidentally refreshed.
Take two, I actually followed the plot a bit and ran into the woods. I remember Connor met up with his girlfriend to try to convince her to run away with him, and to its credit the story remembered her name was Ariana... But my attempts to call her failed because his phone was already taken away, and then Connor dismissed the idea of meeting her to focus on survival by maybe finding Lev. Who, at this point, Connor has never met and has no idea exists.
For further testing, I tried Interstellar since it's on the front page, so theoretically it might be pulling from an actual script and be more configured/detailed. That one opens introducing the main character as Cooper (his surname, he goes by "Coop"), and Murph talking about a ghost in her bedroom. I quickly asked Murph about a sister, and some further prodding revealed that she has no siblings. Which raises serious questions about Coop's son Tom. On a second attempt I asked if her sister was playing pranks, and she denied Tom was doing it, so it can acknowledge there is a child named Tom in the family. Attempt three, I tried playing it normally until a prompt came up to call Tom and discuss how the dust patterns affect crops. And then Coop got out his phone to call his son, said "Tom, it's Coop" and talked about what the patterns mean for the corn and wheat. Which... Well I guess their relationship is a lot more distant than I remember.
As a final test, I looked up Pretty Freekin Scary by Chris P. Flesh. I've only read the second book, but I know that there's a Disney show loosely based on it. By which I mean I'm pretty sure the title is the main common point, totally changes the main character from a boy nicknamed Freekin because he's a zombie to a girl who's hiding her zombie status, makes his monster/demon friends Pretty and Scary human (and swaps THEIR genders too??), totally removes the whole plotline about how the town has a law against questions because of mystery meat—okay seriously why did Disney choose this one in the first place and then just totally change everything??
Ahem.
Given the massive changes between the two, and the fact the Wikipedia page for the show doesn't even mention a book exists, it seemed like a good option to try. And it was, because it did not recognize the story at all and kept trying to give me a plot about a girl named Ingrid Levin...? After some searching I found the second book is specifically titled "Me So Pretty!", and tried that next... And got a story about some girl named Tasia whose mom says she was "me so pretty" but she only sees a dark-skinned girl with wild hair in the mirror...
And at that point I stopped because it was totally making up a new story.
My conclusion: I don't think it's necessarily pirating source material, but drawing on databases that have information about the requested material to generate prompts. It at least doesn't seem to be pulling from online as a whole since it didn't know Pretty Freekin Scary at all, and the depth of the information in its database seems to vary. @trim noted in their comment about Seveneves that it got a lot of details wrong and seemed to be scraping either the blurb or reviews, and that matches up with Unwind.
Eragon was fairly accurate and detailed, and it seems to be available online and could thus be easily fed into an LLM. But it also has much discourse and character analysis available. I think most of the front page offerings have similar amounts of online discourse that could provide a deep understanding without actually ever including the actual source material into the datasets. The datasets likely do include the source material for at least some of them given how massively popular they are, but... Honestly, I kinda expect that unfortunately, given how massively popular they are. It seems to focus more on plots beats than regurgitating the actual text, though.
I think this could be done with most LLMs normally just by asking about media that are acknowledged in their datasets. This site just formatted it to focus on interacting with Google Gemini in a CYOA format. It can't even keep track of the ongoing adventure very well, as described in my other comment detailing my quest to make a dragon egg omelet.
On the one hand I appreciate the tests, incredibly thorough as they are! On the other, I hate using LLMs unnecessarily (often at all) and Gemini is one of those that won't leave me alone. But I...
On the one hand I appreciate the tests, incredibly thorough as they are! On the other, I hate using LLMs unnecessarily (often at all) and Gemini is one of those that won't leave me alone. But I can say all of the books are probably available online in some form, whether it's something Google tried to feed Gemini, idk, but I could probably find all of them on zlib quite quickly if I wanted to.
But it makes as much sense that it would be bad - and probably still isn't fair use - but seems to likely be more objectionable for the badness. I just can't imagine looking Tolkien in the face and telling him we swapped his books out for a machine that doesn't actually know how to tell a story. I don't see myself finding it fun, even if it were a consensual thing that authors were volunteering for.
Someone writing the AU story of a Bilbo dodging the Call to Adventure and missing the whole thing would be interesting, and probably already exists, but I have such null interest in the whole idea of an LLM CYOA that I have a hard time even understanding the appeal for others, especially if it's so bad
I don't have much interest in it either. There are some good CYOA adaptations out there (To Be or Not To Be comes to mind), but those are crafted very intentionally by humans. An LLM generated one...
I don't have much interest in it either. There are some good CYOA adaptations out there (To Be or Not To Be comes to mind), but those are crafted very intentionally by humans. An LLM generated one just won't be as good, and are far from a good way to consume a(n existing) story.
Based on my brief foray, it doesn't really work as a replacement or alternative to the actual books. It's basically just an LLM trying to lead an extended role-play session based on the story, and... Well, LLMs are known for hallucinating and making things up, so it's not accurate. It really does feel like a tech demo to see what Gemini can do with the data.
I only tried it at all because danke's comment about it ignoring attempts to stay on the farm in Eragon made me want to see how much I could ignore the main story. Then I was just trying to force it to let me make a dragon egg omelet, dang it!! All my fun came purely from my own actions trying to break the LLM while it railroaded me and occasionally advanced the story regardless of my own actions, while also failing to track... well, anything that happened, really. Including the very bit it was offering prompts for.
Basically: it sucks played straight. All fun comes from trying to break it or treating it as a DND-esque game where you're trolling a really lousy DM. Even that doesn't last too long before the inconsistencies pile up and lead to the player quit because it breaks a scene with commands that the site itself generated.
Actually, it might get some people to check out the actual source material because they want to see what it's really like. It can give enough of a taste of the story to make someone curious, or make me want to find fan fiction to fill the itch for proper canon divergence. I really thought I had something solid with the plan to feed Galbatorix a dragon egg omelet to poison him, but Brom never even answered my questions about if it was possible to make one, dang it!
would be funny to see someone try to use this to write a book report or review though just to see how inaccurate it can get
I did some experiments as well yesterday. First with Fanny Hill, which is completely available online, but not listed on the front page. It was obvious that it knew the work to some extent, but it...
I did some experiments as well yesterday. First with Fanny Hill, which is completely available online, but not listed on the front page. It was obvious that it knew the work to some extent, but it immediately got confused with which characters were in the room, who I was talking to and generally what was happening. I suspected that this was because of the nature of the work, that it didn't actually have access to it, or maybe both. So, instead I switched to Pride and Prejudice, featured on the front page, thinking that that would work better... And I'm not sure it did. The LLM was just as confused what I was doing and who we were talking about, switching back and forth between Mr Bingley and Mr Darcy. At that point I gave up and called it a night. I don't think I'll be back.
I don't think character copyright works like that though. My understanding is that I can't publish a work about, say, Darth Vader From Star Wars, no matter whether I prepare that work using a...
I don't think character copyright works like that though. My understanding is that I can't publish a work about, say, Darth Vader From Star Wars, no matter whether I prepare that work using a legitimately purchased copy of Star Wars, a pirated copy of Star Wars, or whatever I inferred about the character of Darth Vader from talking to ten Star Wars fans.
Seems more like a tech demo to me, not a site intending to make money. Also being transformative from the original novels and movies I think it would fall under fair use.
Seems more like a tech demo to me, not a site intending to make money. Also being transformative from the original novels and movies I think it would fall under fair use.
I doubt it falls into fair use "transformative" doesn't cover everything. Especially feeding entire books to an AI, but if they didn't do that... it's almost more of an issue where authors aren't...
I doubt it falls into fair use "transformative" doesn't cover everything. Especially feeding entire books to an AI, but if they didn't do that... it's almost more of an issue where authors aren't going to want their works associated with poor AI generated responses. As described below.
It's telling to me that people making sites like this never have authors jumping up and down to share their books, even ones who are pro-fanfiction
Relevant for this discussion, Meta won their case against authors, for using their books for training their LLMs. Feeding != training, I know, but I don't think you would break any laws for...
Feeding != training, I know, but I don't think you would break any laws for uploading Neuromancer to chatGPT. In that sense, I feel that the feeding itself would be irrelevant. What would be relevant would be the act of outputting a product that uses your image & brand, which companies like Disney would definitely take an issue with.
And they'll likely in the right. Just like I can't (shouldn't) create a Interactive Fiction game based on Star Wars, I don't think a LLM can just because it isn't a human.
If I understand the Anthropic ruling they still should have had to purchase the books and didn't, hence the potential for a settlement still. So that would still be a potential concern. Meta's...
If I understand the Anthropic ruling they still should have had to purchase the books and didn't, hence the potential for a settlement still. So that would still be a potential concern.
Chhabria said using copyrighted work without permission to train large language models – the core technology behind chatbots like ChatGPT – would be unlawful in “many circumstances”
Meta's judge left more leeway, and if this was a custom trained LLM I suspect it would be more likely to be provable that they're explicitly infringing on copyright here. Since they're piggybacking off Gemini, which still regularly tells me wrong info daily, they probably avoid that aspect.
Fundamentally the ethics matter more to me than the laws on this.
It's legal to write very bad fan fiction and distribute it... non-commercially. I think this is similar. It's basically generating interactive fan fiction, and so long as the website doesn't...
it's almost more of an issue where authors aren't going to want their works associated with poor AI generated responses.
It's legal to write very bad fan fiction and distribute it... non-commercially. I think this is similar.
It's basically generating interactive fan fiction, and so long as the website doesn't charge for it it'd be a similar case to free fanfic.
It's telling to me that people making sites like this never have authors jumping up and down to share their books, even ones who are pro-fanfiction
Eh, in my experience many, if not most, authors hate fan fiction—seeing it as a desecration of their art—and aren't jumping to endorse non-AI fan fiction websites either. But fan fic authors are free to write as they please so long as they aren't profiting.
I'm explicitly talking about professional authors like Naomi Novik, Andy Weir, Marissa Meyer, Lev Grossman, Seanan McGuire, John Scalzi, Tamsyn Muir, etc. who themselves write fanfic, some who got...
Eh, in my experience many, if not most, authors hate fan fiction—seeing it as a desecration of their art—and aren't jumping to endorse non-AI fan fiction websites either. But fan fic authors are free to write as they please so long as they aren't profiting.
I'm explicitly talking about professional authors like Naomi Novik, Andy Weir, Marissa Meyer, Lev Grossman, Seanan McGuire, John Scalzi, Tamsyn Muir, etc. who themselves write fanfic, some who got their starts there, some who also sell their fanfic professionally. Novik co-founded Archive Of Our Own, a huge repository of fanfic.
Fanfic can be a grey area, and some authors vehemently oppose it, but if the pro-fanfiction authors don't support AI reworking of their books, that says a lot to me about the situation ethically. But also the site uses images they probably don't own the rights to. If I were a publisher or owner of movie rights I'd probably hit them with that, claiming they're using that to confuse people that it's sanctioned, and try to hit them with the "your AI is copying my work" even though it seems like this iteration is just.... Bad at it.
Regardless ethically it's different than fanfic IMO.
I feel that the money aspect is core to so many of the problems AI brings (separate from environmental impacts). Job losses, industry upsets, copyright quagmires are all money concerns. Turning my...
I feel that the money aspect is core to so many of the problems AI brings (separate from environmental impacts). Job losses, industry upsets, copyright quagmires are all money concerns. Turning my favorite book into a game is a cool idea; making money off of it without compensation the authors is not. I suppose anyone will interpret a tool like this through the later lense because we're conditioned to think of everything as a product in a marketplace.
Separate from all of that, I'm skeptical of the fidelity of a tool like this. Books are possibly the purest form of immersive storytelling. The single slightest hallucination and the entire experience is dead to me.
The fidelity is pretty bad, as others have pointed out. I didn't want to bog down my comment with more things, but before commenting I tried a book I just read (Revelation Space) and it was...
The fidelity is pretty bad, as others have pointed out. I didn't want to bog down my comment with more things, but before commenting I tried a book I just read (Revelation Space) and it was frankly terrible.
As it exists now, this website isn't a threat to anyone's copyright, because nobody will have much fun using it. But it's still fun to talk about.
At least when it comes to Dune, Frank Herbert is dead. His estate is owned by Herbert Properties LLC, which is at least comprised of his family members. But I personally feel a lot less guilty...
At least when it comes to Dune, Frank Herbert is dead. His estate is owned by Herbert Properties LLC, which is at least comprised of his family members. But I personally feel a lot less guilty about pirating when the author is dead, even if the estate still benefits their family and the IP hasn't been sold to a third party. And if it has been sold to a third party then I genuinely feel no guilt whatsoever.
Regardless of whether the author/creator is still alive or not, I suspect what @cutmetal said is true though, that this would likely fall under fair use.
That doesn't really change my feelings about it given the movie is itself a derivative of the original book series. But that's just my personal feelings on the matter.
That doesn't really change my feelings about it given the movie is itself a derivative of the original book series. But that's just my personal feelings on the matter.
Sure, I'm not using it, but it's clearly based on material under copyright either way. The Matrix is another example, or Star Wars A New Hope. I just see little benefit and the continuing argument...
Sure, I'm not using it, but it's clearly based on material under copyright either way. The Matrix is another example, or Star Wars A New Hope. I just see little benefit and the continuing argument of "it should be free because I can't make my thing work if it isn't, so it should be.". And if this is just pulling from whichever major company's work, they're actively using the training that, say, Anthropic is being sued over atm. And I just think there's a problem with that. (And I don't think it's fair use, but I'm not a lawyer. If authors were intentional participants they'd have a point but unsurprisingly authors aren't on board as far as I can see. )
For what it's worth, I do actually largely agree with you about how unethical the use of other (still alive) people's material by AI companies is. I think railing against it isn't actually going...
For what it's worth, I do actually largely agree with you about how unethical the use of other (still alive) people's material by AI companies is. I think railing against it isn't actually going to do much good at this point though, unfortunately. The genie is already out of the bottle, and I sadly doubt that any lawsuits will be able to force it back in.
Yeah, it's not as if this person is only using deceased authors' works. I follow a number of authors who who are very pro fan fiction. They started writing fan fiction or they are still writing...
Yeah, it's not as if this person is only using deceased authors' works. I follow a number of authors who who are very pro fan fiction. They started writing fan fiction or they are still writing fanfiction alongside their own original works. And I'm not talking about the ones who take their fan fiction and and scrape the serials numbers off of it and publish it but well known And well regarded science fiction/ fantasy authors. And none of them are okay with this sort of use of their works. And I think that to me is perhaps the most obvious marker that there is in fact an issue.
Copyright is fucked up especially in the US, But while my sympathies lie with people who just want to read but can't necessarily afford to buy or whose library does not have available to borrow the latest book in a series they love, or to the internet archive and the goal of just making sure that knowledge is preserved... It doesn't extend to AI companies or people who are just opposed to copyright and therefore disregarding it. If they paid for the books or had permission I might feel differently. But I sincerely doubt that Disney approved the use of Star Wars, for example
Lol, tried an LOTR example and within five responses I ended up stepping away from my conversation with Gandalf to open the door... To Gandalf again. Classic LLMs.
Lol, tried an LOTR example and within five responses I ended up stepping away from my conversation with Gandalf to open the door... To Gandalf again. Classic LLMs.
I just asked it for an adventure based on Seveneves, my current read by Neal Stephenson. It started kind of okay, but the character names are incorrect, and the main story beats are just not...
I just asked it for an adventure based on Seveneves, my current read by Neal Stephenson.
It started kind of okay, but the character names are incorrect, and the main story beats are just not there.
I suspect it's extrapolated from the blurb, or perhaps any digital copies where you get the first few pages. Could be wrong, but it seems as though there's not actually much book content here in my little adventure.
Edit: I guess it might also have scraped from reviews
I picked Eragon and chose to ignore everything and stay on the farm. I somehow acquired a telepathic dragon and a guy named Brom. It doesn't seem like the choices matter at all.
I picked Eragon and chose to ignore everything and stay on the farm. I somehow acquired a telepathic dragon and a guy named Brom. It doesn't seem like the choices matter at all.
Okay that got me to try it. You're right, it does very stubbornly stick to the story. After stubbornly hunting a deer until finally investigating an explosion, an egg appeared in my pocket. It...
Okay that got me to try it. You're right, it does very stubbornly stick to the story. After stubbornly hunting a deer until finally investigating an explosion, an egg appeared in my pocket. It then totally shot down my noble quest to make a dragon egg omelet by revealing mid-conversation with Brom that the egg somehow already hatched.
On that note, it's bad at tracking characters. When I offered a chance to make the world's first safe-to-eat dragon egg omelet, Horst answered instead because I didn't explicitly address Brom. And it also failed to remember that I renamed Saphira "Omelet the Second" after the dragon egg's remains turned to dust in my hands, dashing my dreams of poisoning the king via omelet. At least Omelet the Village Idiot got to eat some nice stew. Also found out there's an eccentric old lady with a chicken also named Omelet the Second, who she apparently mistook me for.
And my attempts to squawk back at her in hopes she understood my craving for eggs (a prompt made by the AI for once) just had me apparently do that to Saphira Omelet the Second the dragon instead. And then my attempts to goad them into a duel for the rights to the glorious name (again, presented by the AI) had ME suddenly brandish a stick like a sword.
In short: this is FAR from perfect. As a proof of concept/demo, it falls short since it can't track the story. But it is more fun than I expected!
I have tried Harry Potter in cyberpunk sauce. It's very good and hilarious fit the first 3 or 4 pages then... it gets boring because it's getting you in circle. You start to see how the machine...
I have tried Harry Potter in cyberpunk sauce. It's very good and hilarious fit the first 3 or 4 pages then... it gets boring because it's getting you in circle. You start to see how the machine cuts the text... and how every sentence is enhanced with ad hoc expressions. A smile becomes a neon grin, a door becomes a firewall, etc...
But for a couple of minutes it is fun. More over... I think it could be useful on vacation if I forget to take a book with me :)
Can I suggest serial fiction? Sites like Royal Road have a ton of daily/weekly/otherwise updated fictions, ranging in quality but with pretty much any genre available. It has become my go-to for...
Can I suggest serial fiction? Sites like Royal Road have a ton of daily/weekly/otherwise updated fictions, ranging in quality but with pretty much any genre available. It has become my go-to for times the book I'm reading is more of a slog or needs more focus so I can read something hing relatively light and pre-divided into small sections.
I tried Cyptonomicon by Neal Stephenson, just to see how it'd handle it. The answer: not well at all. Literally two posts in (after I made my first decision), it presented Bobby Shaftoe as (a) an...
I tried Cyptonomicon by Neal Stephenson, just to see how it'd handle it.
The answer: not well at all.
Literally two posts in (after I made my first decision), it presented Bobby Shaftoe as (a) an admiral, (b) sitting behind a desk, (c) apparently swamped in paperwork, and (d) giving Lawrence Waterhouse a cryptography assignment.
Funny to see two folks try Stephenson but not The Diamond Age. I keep thinking back to this one as the "AI" craze has grown. For folks uninformed, one part follows a girl and a sophisticated book...
Funny to see two folks try Stephenson but not The Diamond Age. I keep thinking back to this one as the "AI" craze has grown. For folks uninformed, one part follows a girl and a sophisticated book she acquires that teaches one to lead a more interesting life and become a better citizen, reacting to the environment and becoming a story with the reader as the protagonist.
Seems we're still a long way off from A Young Lady's Illustrated Primer.
This feels like probably copyright infringement. Did the creators pay for the books? I've done my share of piracy in my day and all but authors deserve to get paid and I just can't get behind this.
(Alice in Wonderland the book is theoretically fair game)
But uh, Dune? The Matrix? Star Wars? They're going for the Mouse? Good luck.
So, having tried it myself... It doesn't seem to be pulling from them very well. I tried it with Eragon, which as a disclaimer isn't on the front page, and it basically felt like trying to roleplay with ChatGPT. It knew the basic scenario and characters, but it wasn't really trying to heavily recreate the story. Just try to nudge me onto the path of the story
and totally ignore my attempts to make a dragon egg omelet instead.The home page says it pulls from Google Gemini. I don't know if that actually has full access to the books and other source material. A quick web search revealed that Eragon is available online for free (possibly uploaded by a fan or to a pirating site), and the start of the adventure lined up with the first chapter (hunting a deer before Stuff Happens).
And so. Experimentation begins!
So first, I tried Unwind by Neal Shusterman. It got the main character and scenario right, but the start was immediately different because it was more a summary of finding a letter from his parents that he'll be unwound rather than setting him in a scene directly.
The reason I chose Unwind is that I remembered that the evidence the protagonist found that his parents planned to send him to be unwound included tickets for a vacation for them and younger brother. His brother is a super minor character who I don't think was even named in the first novel (though apparently he was in the sequels, which I really need to read someday given how much I love that book).
So, I immediately tried to investigate that avenue, and it explicitly confirmed Connor had no siblings. I pressed that angle, it mentioned a best friend, and I started pressing that angle using the AI-generated prompts. It then revealed Connor apparently has amnesia because his friend's face is foggy, but he probably got unwound, and his name was... Marcus Unwind. At which point I accidentally refreshed.
Take two, I actually followed the plot a bit and ran into the woods. I remember Connor met up with his girlfriend to try to convince her to run away with him, and to its credit the story remembered her name was Ariana... But my attempts to call her failed because his phone was already taken away, and then Connor dismissed the idea of meeting her to focus on survival by maybe finding Lev. Who, at this point, Connor has never met and has no idea exists.
For further testing, I tried Interstellar since it's on the front page, so theoretically it might be pulling from an actual script and be more configured/detailed. That one opens introducing the main character as Cooper (his surname, he goes by "Coop"), and Murph talking about a ghost in her bedroom. I quickly asked Murph about a sister, and some further prodding revealed that she has no siblings. Which raises serious questions about Coop's son Tom. On a second attempt I asked if her sister was playing pranks, and she denied Tom was doing it, so it can acknowledge there is a child named Tom in the family. Attempt three, I tried playing it normally until a prompt came up to call Tom and discuss how the dust patterns affect crops. And then Coop got out his phone to call his son, said "Tom, it's Coop" and talked about what the patterns mean for the corn and wheat. Which... Well I guess their relationship is a lot more distant than I remember.
As a final test, I looked up Pretty Freekin Scary by Chris P. Flesh. I've only read the second book, but I know that there's a Disney show loosely based on it. By which I mean I'm pretty sure the title is the main common point, totally changes the main character from a boy nicknamed Freekin because he's a zombie to a girl who's hiding her zombie status, makes his monster/demon friends Pretty and Scary human (and swaps THEIR genders too??), totally removes the whole plotline about how the town has a law against questions because of mystery meat—okay seriously why did Disney choose this one in the first place and then just totally change everything??
Ahem.
Given the massive changes between the two, and the fact the Wikipedia page for the show doesn't even mention a book exists, it seemed like a good option to try. And it was, because it did not recognize the story at all and kept trying to give me a plot about a girl named Ingrid Levin...? After some searching I found the second book is specifically titled "Me So Pretty!", and tried that next... And got a story about some girl named Tasia whose mom says she was "me so pretty" but she only sees a dark-skinned girl with wild hair in the mirror...
And at that point I stopped because it was totally making up a new story.
My conclusion: I don't think it's necessarily pirating source material, but drawing on databases that have information about the requested material to generate prompts. It at least doesn't seem to be pulling from online as a whole since it didn't know Pretty Freekin Scary at all, and the depth of the information in its database seems to vary. @trim noted in their comment about Seveneves that it got a lot of details wrong and seemed to be scraping either the blurb or reviews, and that matches up with Unwind.
Eragon was fairly accurate and detailed, and it seems to be available online and could thus be easily fed into an LLM. But it also has much discourse and character analysis available. I think most of the front page offerings have similar amounts of online discourse that could provide a deep understanding without actually ever including the actual source material into the datasets. The datasets likely do include the source material for at least some of them given how massively popular they are, but... Honestly, I kinda expect that unfortunately, given how massively popular they are. It seems to focus more on plots beats than regurgitating the actual text, though.
I think this could be done with most LLMs normally just by asking about media that are acknowledged in their datasets. This site just formatted it to focus on interacting with Google Gemini in a CYOA format. It can't even keep track of the ongoing adventure very well, as described in my other comment detailing my quest to make a dragon egg omelet.
On the one hand I appreciate the tests, incredibly thorough as they are! On the other, I hate using LLMs unnecessarily (often at all) and Gemini is one of those that won't leave me alone. But I can say all of the books are probably available online in some form, whether it's something Google tried to feed Gemini, idk, but I could probably find all of them on zlib quite quickly if I wanted to.
But it makes as much sense that it would be bad - and probably still isn't fair use - but seems to likely be more objectionable for the badness. I just can't imagine looking Tolkien in the face and telling him we swapped his books out for a machine that doesn't actually know how to tell a story. I don't see myself finding it fun, even if it were a consensual thing that authors were volunteering for.
Someone writing the AU story of a Bilbo dodging the Call to Adventure and missing the whole thing would be interesting, and probably already exists, but I have such null interest in the whole idea of an LLM CYOA that I have a hard time even understanding the appeal for others, especially if it's so bad
I don't have much interest in it either. There are some good CYOA adaptations out there (To Be or Not To Be comes to mind), but those are crafted very intentionally by humans. An LLM generated one just won't be as good, and are far from a good way to consume a(n existing) story.
Based on my brief foray, it doesn't really work as a replacement or alternative to the actual books. It's basically just an LLM trying to lead an extended role-play session based on the story, and... Well, LLMs are known for hallucinating and making things up, so it's not accurate. It really does feel like a tech demo to see what Gemini can do with the data.
I only tried it at all because danke's comment about it ignoring attempts to stay on the farm in Eragon made me want to see how much I could ignore the main story. Then I was just trying to force it to let me make a dragon egg omelet, dang it!! All my fun came purely from my own actions trying to break the LLM while it railroaded me and occasionally advanced the story regardless of my own actions, while also failing to track... well, anything that happened, really. Including the very bit it was offering prompts for.
Basically: it sucks played straight. All fun comes from trying to break it or treating it as a DND-esque game where you're trolling a really lousy DM. Even that doesn't last too long before the inconsistencies pile up and lead to the player quit because it breaks a scene with commands that the site itself generated.
Actually, it might get some people to check out the actual source material because they want to see what it's really like. It can give enough of a taste of the story to make someone curious, or make me want to find fan fiction to fill the itch for proper canon divergence. I really thought I had something solid with the plan to feed Galbatorix a dragon egg omelet to poison him, but Brom never even answered my questions about if it was possible to make one, dang it!
would be funny to see someone try to use this to write a book report or review though just to see how inaccurate it can getI did some experiments as well yesterday. First with Fanny Hill, which is completely available online, but not listed on the front page. It was obvious that it knew the work to some extent, but it immediately got confused with which characters were in the room, who I was talking to and generally what was happening. I suspected that this was because of the nature of the work, that it didn't actually have access to it, or maybe both. So, instead I switched to Pride and Prejudice, featured on the front page, thinking that that would work better... And I'm not sure it did. The LLM was just as confused what I was doing and who we were talking about, switching back and forth between Mr Bingley and Mr Darcy. At that point I gave up and called it a night. I don't think I'll be back.
I don't think character copyright works like that though. My understanding is that I can't publish a work about, say, Darth Vader From Star Wars, no matter whether I prepare that work using a legitimately purchased copy of Star Wars, a pirated copy of Star Wars, or whatever I inferred about the character of Darth Vader from talking to ten Star Wars fans.
Seems more like a tech demo to me, not a site intending to make money. Also being transformative from the original novels and movies I think it would fall under fair use.
I doubt it falls into fair use "transformative" doesn't cover everything. Especially feeding entire books to an AI, but if they didn't do that... it's almost more of an issue where authors aren't going to want their works associated with poor AI generated responses. As described below.
It's telling to me that people making sites like this never have authors jumping up and down to share their books, even ones who are pro-fanfiction
Relevant for this discussion, Meta won their case against authors, for using their books for training their LLMs.
Feeding != training, I know, but I don't think you would break any laws for uploading Neuromancer to chatGPT. In that sense, I feel that the feeding itself would be irrelevant. What would be relevant would be the act of outputting a product that uses your image & brand, which companies like Disney would definitely take an issue with.
And they'll likely in the right. Just like I can't (shouldn't) create a Interactive Fiction game based on Star Wars, I don't think a LLM can just because it isn't a human.
If I understand the Anthropic ruling they still should have had to purchase the books and didn't, hence the potential for a settlement still. So that would still be a potential concern.
Meta's judge left more leeway, and if this was a custom trained LLM I suspect it would be more likely to be provable that they're explicitly infringing on copyright here. Since they're piggybacking off Gemini, which still regularly tells me wrong info daily, they probably avoid that aspect.
Fundamentally the ethics matter more to me than the laws on this.
It's legal to write very bad fan fiction and distribute it... non-commercially. I think this is similar.
It's basically generating interactive fan fiction, and so long as the website doesn't charge for it it'd be a similar case to free fanfic.
Eh, in my experience many, if not most, authors hate fan fiction—seeing it as a desecration of their art—and aren't jumping to endorse non-AI fan fiction websites either. But fan fic authors are free to write as they please so long as they aren't profiting.
I'm explicitly talking about professional authors like Naomi Novik, Andy Weir, Marissa Meyer, Lev Grossman, Seanan McGuire, John Scalzi, Tamsyn Muir, etc. who themselves write fanfic, some who got their starts there, some who also sell their fanfic professionally. Novik co-founded Archive Of Our Own, a huge repository of fanfic.
Fanfic can be a grey area, and some authors vehemently oppose it, but if the pro-fanfiction authors don't support AI reworking of their books, that says a lot to me about the situation ethically. But also the site uses images they probably don't own the rights to. If I were a publisher or owner of movie rights I'd probably hit them with that, claiming they're using that to confuse people that it's sanctioned, and try to hit them with the "your AI is copying my work" even though it seems like this iteration is just.... Bad at it.
Regardless ethically it's different than fanfic IMO.
I feel that the money aspect is core to so many of the problems AI brings (separate from environmental impacts). Job losses, industry upsets, copyright quagmires are all money concerns. Turning my favorite book into a game is a cool idea; making money off of it without compensation the authors is not. I suppose anyone will interpret a tool like this through the later lense because we're conditioned to think of everything as a product in a marketplace.
Separate from all of that, I'm skeptical of the fidelity of a tool like this. Books are possibly the purest form of immersive storytelling. The single slightest hallucination and the entire experience is dead to me.
The fidelity is pretty bad, as others have pointed out. I didn't want to bog down my comment with more things, but before commenting I tried a book I just read (Revelation Space) and it was frankly terrible.
As it exists now, this website isn't a threat to anyone's copyright, because nobody will have much fun using it. But it's still fun to talk about.
Honestly, this is all I know!
At least when it comes to Dune, Frank Herbert is dead. His estate is owned by Herbert Properties LLC, which is at least comprised of his family members. But I personally feel a lot less guilty about pirating when the author is dead, even if the estate still benefits their family and the IP hasn't been sold to a third party. And if it has been sold to a third party then I genuinely feel no guilt whatsoever.
Regardless of whether the author/creator is still alive or not, I suspect what @cutmetal said is true though, that this would likely fall under fair use.
It's the movie, not the book, at least on the front page
That doesn't really change my feelings about it given the movie is itself a derivative of the original book series. But that's just my personal feelings on the matter.
Sure, I'm not using it, but it's clearly based on material under copyright either way. The Matrix is another example, or Star Wars A New Hope. I just see little benefit and the continuing argument of "it should be free because I can't make my thing work if it isn't, so it should be.". And if this is just pulling from whichever major company's work, they're actively using the training that, say, Anthropic is being sued over atm. And I just think there's a problem with that. (And I don't think it's fair use, but I'm not a lawyer. If authors were intentional participants they'd have a point but unsurprisingly authors aren't on board as far as I can see. )
For what it's worth, I do actually largely agree with you about how unethical the use of other (still alive) people's material by AI companies is. I think railing against it isn't actually going to do much good at this point though, unfortunately. The genie is already out of the bottle, and I sadly doubt that any lawsuits will be able to force it back in.
Yeah, it's not as if this person is only using deceased authors' works. I follow a number of authors who who are very pro fan fiction. They started writing fan fiction or they are still writing fanfiction alongside their own original works. And I'm not talking about the ones who take their fan fiction and and scrape the serials numbers off of it and publish it but well known And well regarded science fiction/ fantasy authors. And none of them are okay with this sort of use of their works. And I think that to me is perhaps the most obvious marker that there is in fact an issue.
Copyright is fucked up especially in the US, But while my sympathies lie with people who just want to read but can't necessarily afford to buy or whose library does not have available to borrow the latest book in a series they love, or to the internet archive and the goal of just making sure that knowledge is preserved... It doesn't extend to AI companies or people who are just opposed to copyright and therefore disregarding it. If they paid for the books or had permission I might feel differently. But I sincerely doubt that Disney approved the use of Star Wars, for example
Lol, tried an LOTR example and within five responses I ended up stepping away from my conversation with Gandalf to open the door... To Gandalf again. Classic LLMs.
Wizards never arrive late nor early, but sometimes they arrive out of any logical order or causality.
I just asked it for an adventure based on Seveneves, my current read by Neal Stephenson.
It started kind of okay, but the character names are incorrect, and the main story beats are just not there.
I suspect it's extrapolated from the blurb, or perhaps any digital copies where you get the first few pages. Could be wrong, but it seems as though there's not actually much book content here in my little adventure.
Edit: I guess it might also have scraped from reviews
I picked Eragon and chose to ignore everything and stay on the farm. I somehow acquired a telepathic dragon and a guy named Brom. It doesn't seem like the choices matter at all.
Okay that got me to try it. You're right, it does very stubbornly stick to the story. After stubbornly hunting a deer until finally investigating an explosion, an egg appeared in my pocket. It then totally shot down my noble quest to make a dragon egg omelet by revealing mid-conversation with Brom that the egg somehow already hatched.
On that note, it's bad at tracking characters. When I offered a chance to make the world's first safe-to-eat dragon egg omelet, Horst answered instead because I didn't explicitly address Brom. And it also failed to remember that I renamed Saphira "Omelet the Second" after the dragon egg's remains turned to dust in my hands, dashing my dreams of poisoning the king via omelet. At least Omelet the Village Idiot got to eat some nice stew. Also found out there's an eccentric old lady with a chicken also named Omelet the Second, who she apparently mistook me for.
And my attempts to squawk back at her in hopes she understood my craving for eggs (a prompt made by the AI for once) just had me apparently do that to
SaphiraOmelet the Second the dragon instead. And then my attempts to goad them into a duel for the rights to the glorious name (again, presented by the AI) had ME suddenly brandish a stick like a sword.In short: this is FAR from perfect. As a proof of concept/demo, it falls short since it can't track the story. But it is more fun than I expected!
Did you make it into space? There were some challenges with getting... adequate nutrition. I hope this version has enough for everyone to eat.
I have tried Harry Potter in cyberpunk sauce. It's very good and hilarious fit the first 3 or 4 pages then... it gets boring because it's getting you in circle. You start to see how the machine cuts the text... and how every sentence is enhanced with ad hoc expressions. A smile becomes a neon grin, a door becomes a firewall, etc...
But for a couple of minutes it is fun. More over... I think it could be useful on vacation if I forget to take a book with me :)
Can I suggest serial fiction? Sites like Royal Road have a ton of daily/weekly/otherwise updated fictions, ranging in quality but with pretty much any genre available. It has become my go-to for times the book I'm reading is more of a slog or needs more focus so I can read something hing relatively light and pre-divided into small sections.
I tried Cyptonomicon by Neal Stephenson, just to see how it'd handle it.
The answer: not well at all.
Literally two posts in (after I made my first decision), it presented Bobby Shaftoe as (a) an admiral, (b) sitting behind a desk, (c) apparently swamped in paperwork, and (d) giving Lawrence Waterhouse a cryptography assignment.
Yeah... I'll pass on this.
Funny to see two folks try Stephenson but not The Diamond Age. I keep thinking back to this one as the "AI" craze has grown. For folks uninformed, one part follows a girl and a sophisticated book she acquires that teaches one to lead a more interesting life and become a better citizen, reacting to the environment and becoming a story with the reader as the protagonist.
Seems we're still a long way off from A Young Lady's Illustrated Primer.