Subscribe
Share
Share
Embed
Each week Adam & Ste focus on a specific feature, idea or prototype in Hardcover and iterate on it together or with guests.
adam (00:01.563)
Hey hey, how's it going?
ste (00:03.741)
Hey Adam, good good, I'm over here enjoying a nice London evening.
adam (00:10.29)
It's a, what is it? 1 PM here in Salt Lake City, but it's, we had like the ninth highest snow day ever last week here in Salt Lake in recorded history. So there's still like tons of snow on the ground. So it's been a fun week.
ste (00:24.574)
What?
ste (00:28.273)
Wow, Jesus. God, now I have to shovel.
adam (00:32.882)
Luckily, we live in an apartment, so we don't have to do anything, which is one of the reasons why we chose an apartment.
ste (00:39.317)
Yeah, I mean I bet. That's nice. What have you been up to?
adam (00:43.662)
Yeah.
A lot of staying inside this last week because just, it's been fun just watching the snow, but not really going out in it. But other than that, yeah, we launched the iOS app this last week, which was awesome. That was long awaited, especially since I think 50% of our users, according to Google Analytics, are on mobile. So yeah, that was great to finally get out there.
ste (00:51.525)
Yeah. I bet.
ste (01:04.457)
Yeah.
ste (01:16.677)
Yeah, it's gonna be a big change. I'm already, I mean, I've been using it as a tester and now that it's live, you know, it's a whole different experience. We're still fine tuning it and, you know, making sure everything is good on the app. But just seeing it out there on the app store is like, it's been a moment. It's really nice to see it over there.
Yeah, and it's going to be good because now that it's published, the way we'll make updates on the web will sync with the way we'll make updates on the apps. So that's cool. Waiting for Android. That should be like a couple of weeks in, if we're lucky.
adam (02:03.081)
Yeah.
adam (02:07.09)
Yeah, probably not even that. Like we have, uh, I think we have about 15 beta testers on it right now, which is pretty awesome. Like when I sent out the newsletter this week and I asked for like, uh, anyone with a Android phone who wants to beta test it, so many people replied. So we should, we should get a lot of feedback before we even, uh, submit it to the Google play store.
ste (02:25.367)
Wow, okay.
ste (02:32.113)
That's great. That's going to be another great milestone. It's so nice that we're actually going to be able to couple this together. And publishing on the Google Play Market should be a bit easier than the iOS store. Hopefully.
adam (02:45.906)
Yeah. Yeah. Yeah. I, I think at least like we hammered out most of the issues on the, on the iOS side. The only, the only part that I'm unsure of for the Android side is what's different for signing up for a subscription on an Android device versus on an iOS device. Uh, like from, from a standpoint on our side, it's like you have a subscription or you don't have a subscription.
but how that handshake works with the store itself, how it verifies the receipt. I'll be interested to see if that process is roughly the same or if it's different enough that we need to solve it in a different way. But.
ste (03:28.917)
Yup, I think you cut off on my end. Or I cut.
adam (03:34.206)
Uh oh. Let's see.
adam (03:40.446)
monetary refreshing.
ste (03:43.925)
Okay, maybe we should pause this.
adam (03:51.458)
Let's see.
ste (03:54.656)
Oh
adam (04:02.686)
He's going to refresh his browser real quick.
adam (04:07.802)
See how this goes. We're using the service Riverside FM that kind of streams. Cool. I can't.
ste (04:10.371)
Okay.
I'm back. Here we go. I don't know what happened there.
adam (04:19.568)
Oops. Oop, let's see. Can you hear me?
ste (04:23.694)
Yeah, I can hear you. Can you hear me?
adam (04:25.894)
Yeah, like the moment you came back, I said something that caused the, I don't even want to say that, the S-I-R-I device to start recording on my laptop at the same time you came back. So I'm like, are things broken?
ste (04:41.045)
What the hell? Yeah, it's the singularity. It's here. Hopefully not this way.
adam (04:49.775)
There was a, there was a really good, uh, last week tonight with John Oliver last night about, uh, like GPT three mid journey, uh, stable diffusion. It was all about like, uh, AI and how society's leveraging that. It was, it was a really good episode.
ste (05:07.897)
Wow, nice. Yeah, I'll have to see that. It's like it hit mainstream like really well. I mean, wow. Yeah, it's really interesting. I mean, I think it's also the way it impacts the book industry and books in general will be pretty amazing and scary and disruptive at the same time. I saw that there's a boom in
adam (05:16.51)
Yeah. Yeah.
ste (05:35.761)
books co-authored by ChatGPT on Amazon, which I'm not sure if I, I mean, I'm not sure how I relate to that, but you know, it's an interesting fact.
adam (05:50.83)
Yeah, there was a person who copyrighted a comic book last November where they wrote the content and then they had Mid Journey generate all the images for it. And they received a copyright for the comic book. But then since then, the US Copyright Office has changed their opinion that you can't copyright images generated by Mid Journey or an AI bot. So...
she they had to like they they went back to the author and said what what of this did you do and what this was generated and then they changed the copyright on their work so that they own the story but the images in it aren't copyrightable.
ste (06:33.617)
Wow, okay.
Yeah, again, really interesting. Yeah. Yeah, it is a complicated field and you know, issues around copyright, I mean, I just like tip of the iceberg, but like really important. I mean, who has the copyright because, you know, you just prompted it, but the content you didn't do. So theoretically, mid journey. Yeah, exactly. I mean, mid journey through like ripping off
adam (06:37.989)
It was, it's a very complicated field. Yeah.
adam (06:58.186)
Yeah, and you don't know what fed that algorithm.
ste (07:06.113)
I guess tons of art, like all that. It was, yeah, in its initial learning model should theoretically have the copyright, I guess. But it seems really interesting how it ties in with the legal stuff.
adam (07:29.626)
Yeah. Well, what have you been up to? What's your week been like?
ste (07:35.861)
It's been pretty intense, but good. I watched The Wail yesterday, very interesting movie. With one of my favorite actors, Brandon Fraser, also starred in The Mummy, remember that? Yeah, exactly. I know. Yeah, yeah. Really, really good movie. I'm not going to like give any spoilers, but definitely watch it. Yeah.
adam (07:48.106)
Nice. Oh yeah, one of my favorite adventure movies for sure.
adam (08:03.754)
Yeah, I saw he run the Screen Actors Guild Award yesterday for Best Actor.
ste (08:08.597)
Okay, well, yeah, I mean well deserved. Yeah. Yeah really good
adam (08:15.667)
Nice. Yeah, we with only two weeks before Oscars, do you have any like Oscar picks like favorite movies for the year that you think should win?
ste (08:30.549)
I don't know, really tough to say. I don't know, I haven't even watched the Oscars in the past two years. I think the Golden Globes kind of like super-synced it in importance, like mentally, but I don't know, maybe Babylon. Although I have to see the ending of it because I haven't seen it. It seems like an Oscar movie. I'm not expecting my favorite. I mean, I think The Banshees of Nisherin was one of my favorites.
this year but it's definitely not the way.
adam (09:01.312)
Yeah.
I don't know. I mean, it won, I think that won the Golden Globe for best picture or drama.
ste (09:09.041)
Did it? Okay. Okay. I didn't even know that. Maybe, who knows?
adam (09:11.991)
Yeah.
Uh, it's, I'm, I'm holding out for everything everywhere all at once. Right. That was, that was so good. But, but there were a lot of good movies last year.
ste (09:20.205)
Oh yeah, oh yeah, it's in this batch, right? Yeah, yeah, yeah, yeah, that's... Yeah, I know, pretty surprising that cinema is still okay, you know, in 2023. I always get surprised by that.
adam (09:35.036)
Yeah.
ste (09:37.645)
Yeah, people are still making movies. Yeah, movies and books and wow, okay. People have not like given in yet. That's cool. Yeah, that's really nice. What were you thinking about covering today? I don't have like any specific subject. Maybe we can also go over the book tracking and I was gonna actually maybe talk about how we think
adam (09:38.142)
fewer comic book movies.
ste (10:07.957)
AI in general and what's happening in tech will impact the book industry and how we see it playing out just to offer a bit of glimpse into what we as a startup and people in this domain think will happen and our prognosis for the future, I guess.
adam (10:32.654)
be up for that. Like kind of AI and books and how we might leverage some of that within hardcover, kind of what we're at least thinking about for the future, even if it's not something we implement tomorrow.
ste (10:46.169)
Yeah, yeah, yeah, exactly. I mean, there's lots of stuff we could do. And I think like the book industry as a whole, I'm wondering, you know, how it can actually harness like the good stuff and stay away from the bad stuff. It might be an interesting discussion because it's bound to change the way books work, maybe. I'm wondering, I mean, I know right now that basically you can check
if an algorithm has helped you write a book. I know the schools are very like into sites that check if your homework was actually written by GPT or not. So I'm guessing the same would happen for books, right? I mean, if you had like any help, I'm wondering what the degree of like
adam (11:44.35)
Yeah.
ste (11:46.785)
editing you'd have to do to be able to fool like a checker or like something that checks if you did it with gpt how much editing you'd have to do
adam (12:02.662)
Yeah, I'm kind of, I really wonder how those systems work behind the scenes on the like verify if content was written by an AI, because I don't I don't even know how it would be able to like verify that because like for like for instance in the US one of the systems that a lot of high schools and colleges use is like a system where you upload a student's paper and it checks it against every other paper that's ever been written.
to find similarities. And for that one, it has like a background to check it against. I wonder like, my assumption is that for these systems where it's like checking against an AI, most likely it's asking an AI to check it against an AI. So it's like, it's a black box checking another black box.
ste (12:39.357)
Thanks for watching!
ste (12:50.285)
Yeah, exactly.
Yeah, it's crazy. I think that's what actually the tool does. If I'm not mistaken, it might have actually been published by OpenAI as well, just to, I guess, mitigate this little scandal. I also heard that if you ask the algorithm to rewrite it so that it's not possible for it to be
uh you know uncovered by a competing algorithm I guess. It actually does that pretty well which is kind of yeah I mean I don't know how that would work but it actually does I mean I'm seeing much more of an advantage to the part of the like part that generates the text because you can just ask it okay rewrite this.
I've asked it for lots of paragraphs where I just needed things to be reworded. Just write this but differently and that's enough for it to give me a completely different sentence and I'm guessing that does it for it. I mean, I don't know the markers. Maybe they're looking for patterns in the way words are generated one after the other or stuff like that.
adam (14:02.989)
Hmm.
adam (14:19.186)
Yeah. Yeah, that would, that would be, that would definitely be like a question for like a machine learning expert. Like they, they, like the number of people who know like the ins and outs of how this works is becoming so small compared to the number of people actually using it, which is somewhat scary. But uh, yeah.
ste (14:27.386)
Yeah.
ste (14:39.081)
Yeah, yeah, it's pretty crazy. What's your feeling? Do you think like, ah...
There is such an advance in these that even ML experts are being a bit, you know, overcome by the amount of, like, progress that's going on.
adam (15:07.986)
I just feel like it's, it kind of reminds me of like the early days of like Napster for music. And like there's suddenly all this music that's available and everyone's downloading music that they might not have downloaded before. They're trying out new things and the artists are kind of getting shafted in this experience because like the technology is advancing faster than the systems to support those authors are. And it feels like that's where we are for AI right now.
ste (15:26.186)
Yeah.
adam (15:36.242)
And so like over the next 10 years, we're probably going to be like reeling those back. Uh, I'm not very confident in regulation in the U S at any point. So I wonder how much of it's going to be like just decided at the Supreme Court or decided from startups actually making, um, thoughtful decisions to support author or, you know, to support artists. But I think that like.
It's already like the genie's out of the bottle now. So it's like, we're going to have to figure out how to support artists while still evolving this system, which is going to be a very exciting next 10 years. But it's not going to be the best time. I feel, I feel like it's going to be a difficult time to be an artist trying to make a living.
ste (16:27.354)
Yeah.
adam (16:27.918)
But then again, there's gonna be like new artists that are gonna be like, I'm a prompt artist. I'm someone who just writes prompts. And that's like a new field of artistry that wasn't, it's almost like a poet, but your audience isn't humans. Your audience is a machine.
ste (16:49.014)
Yeah, that's pretty good. That's a really good take. I don't think artists are gonna like... Obviously things are gonna be really different, but I think the same, that artists will find new ways of circumventing, where doing stuff differently than an algorithm. Would you feel cheated if you...
like read a book and you didn't know it was written by Chad GPT and you wouldn't like, they wouldn't like say it anywhere, but you'd later find out.
adam (17:29.706)
I mean, I feel like I would. Like, I don't exactly know how to explain why though. Like, if I enjoyed the book and it was like a pleasant experience, that shouldn't really matter. I mean, I'm sure there's lots of books I've read that were by ghostwriters of the actual person that I read them by. And I didn't even know they were by a ghostwriter. So it feels like the similar thing where it's like...
ste (17:32.82)
Same.
ste (17:55.047)
Yeah.
adam (17:59.994)
If you're upfront about it being written by an AI, that's, I mean, that seems fine, but it's when you, when you like put that label on it, that it was written by human when it wasn't, that's when it becomes disingenuous.
ste (18:05.275)
Yeah.
ste (18:13.237)
Yeah, yeah, that's very true. Although, I mean, I even like prefer, I think if someone mentioned that it actually, they actually use the AI in writing it, it's kind of a turn off, I guess. But I bet, I mean, I bet there are authors using it right now to, especially for the, let's say, structure of the book, the skeleton.
There's lots of stuff for writer's block that you can do with it. Ask for new subplots or ask for new ideas of where the plot can go or for character specs or for even like creative names. I was trying to come up with some names for fictional places and it...
adam (18:42.049)
Mm-hmm.
ste (19:11.545)
been a really good resource to allow me to do that instead of like, I mean, I know it's basically cheating because I could do the research, I could go into, you know, whatever people like Tolkien did like really research and makeup stuff. That's really, I mean, that's impressive and you know, the fact that that's probably not going to exist anymore is a bit scary. But then again, you know, you get the...
options like in two seconds and they're pretty well researched. I mean you still have to do the editing. Maybe we'll get to a point where you won't even have to do the editing but yeah creatively it's pretty good.
adam (19:56.766)
Yeah, it's.
Yeah, it kind of reminds me, like, I was on Instagram a year or two ago, and I saw someone who was creating these really beautiful paragraphs of text describing their travel experience. And it made me realize, like, a poet of today, their medium doesn't necessarily have to be a published book. It could just be Instagram posts with poetry associated with it.
I feel like it's easy to discount it because it's on social media rather than in the format of a book that's being published. But it makes me wonder all the different mediums that artists are gonna be able to express themselves in in 10 years from now that we don't even conceive of now. One thing that I saw a TikTok of recently, which was it was like an idea that kind of blew my mind a little bit. It was like over the next...
generation, we might get to the point where people have, let's say, like a recording device that follows them from when they're born to when they're 18, that's learning their stance. It's learning their decision-making process. It's basically training an AI on their decisions over their lifetime. And then you could choose or not choose to use your own AI. So it's not an AI trained
ambiguous data set. It's, it's, it's you. So imagine like writing a book with yourself as the second co-author with a second co-author is an AI version of yourself. It's like, is that still writing? Is that still a self written book? Or is that an AI assisted book with your own AI? Like
ste (21:28.991)
Yeah.
ste (21:38.877)
That'd be crazy. Yeah.
ste (21:46.769)
Yeah, exactly. Yeah, that's mind-bending. Definitely, I mean, it's gonna like start stuff that we're not even thinking of, but stuff like that, that's already, I mean, you could see it. And in terms of medium, yeah, I think it's definitely gonna change. And in art in general, like books, paintings, whatever.
Usually the new mediums were like the progressive way, I guess, of expression. You know, when they went from classical painting to like modernism and they, I don't know, put toilets in museums and that kind of stuff. Of course, the people then like dismissed it as crap, basically.
it happens now and you know maybe I mean I bet some of it was crap as well but then again you know usually the new things are viewed through this lens of okay this is nothing this is not nothing compared to true art this is nothing compared to true writing uh well I'm really looking forward to the people who use the medium
creatively and what narratives can be given birth to through this way is gonna be really interesting. And yeah, the hardware evolves as well. Now we just have physical books and audio books, but I'm wondering what... I mean, yeah, this is a bit...
adam (23:37.789)
Hmm. Yeah. Well.
ste (23:40.133)
far away. I was thinking like books directly into your head, which might happen at some point, but
adam (23:47.374)
There is a company, I think it was a shutdown, but not called a detour, which offered GPS assisted tours. So it was like an app you download on your phone and it would say like, start at this point and we're gonna take a tour of some city. And it was GPS locked. So as you move to different spots in town, it would unlock different parts of the tour and it would start talking about it. And so it's kind of like,
Like there's a lot of like combining two technologies to create a story, like something like that where you could, you could, you could write a story like that, or you could do it with augmented reality or like a phone augmented reality. So there's like all these mediums that. Yeah. I'm really excited to see people play with, but, uh, so far are still kind of clunky to create an entire experience on that you would want to, want to have something longer than like an hour long.
ste (24:41.189)
Yeah, exactly. Yeah, it kind of feels like the early days of the internet, you know, when you plug a modem and you had to make sure nobody was on the phone line. I mean, I'm not sure if that's how it worked for you, but that's, yeah, that's, I mean, those weird modem sounds, if you play them right now, I can like really like understand what that's about. Maybe it's like PTSD from when we were trying to connect and...
adam (24:55.714)
I'm sorry.
adam (25:04.342)
I'm gonna go.
ste (25:10.513)
there was one of your relatives on the line. But that was crazy. I mean, telling that to someone right now who, I mean, you might have internet beamed from freaking satellites directly to your phone that is a super computer is sci-fi. I'm really like wondering, you know, what will be going on? I mean, now it seems clunky, but still it's like really
compared to how it started off, a tool that seems almost like magic. I am wondering what can be done through tech. What can we do? Because, I mean, now we're talking about putting the hardcover app out there, but let's say, other horizons will open or not. But yeah, it makes it really exciting to see how people like.
adam (26:06.976)
Yeah.
ste (26:10.477)
will actually create books, read books. I'm really looking forward to see how authors do this and how they can uncover ideas and what sorts of new authors get published with using something like this. It's gonna be like...
adam (26:33.87)
Hmm.
ste (26:38.797)
really, really interesting. How do you think publishing would be changed? And maybe considering, let's say, something like our match score as well.
adam (26:53.23)
Hmm.
I mean, publishing is such a hard one because you're still limited by the publishers. Like you could publish your own book on your own website. Like that's never gonna change. But I wonder how many people, I mean, there's a growing number of people who are like just using social media to drive traffic to their own ownership of the thing that they're building rather than using a distribution platform. But it feels like any media that really breaks through, it generally happens on
a big platform or, you know, it maybe starts on a blog or something small and then it moves to a big platform as soon as it gets big enough. So it feels like the publishing side is still going to be that gateway to wide audience. Um, but I, I'd love to see how that changes.
ste (27:47.449)
Yeah, that's a really good one. I mean, I was wondering, you know, if something that really knows your preferences, if you're a reader on a network like ours, and you get a book that's just been published, maybe as an ad or just a recommendation from us, if that would be enough to sustain
you as an author or let's put it another way because if the platform is big enough, okay, if you get sales, that's really great. What could you do as an author to present that book in as good as way as possible for people to consider reading it?
Because that would be interesting. I mean, we don't even know what kind of books are being published that we might like, but they're so obscure that they never enter our, they never get on our radar because either they're not through a big publisher or they don't go through the usual, yeah, publishing cycle and distribution.
adam (28:45.194)
Mm-hmm.
adam (29:15.712)
Yeah.
Yeah, that's a very hard problem. It's like, how do you... It's the same as our problem where we can't really give recommendations for books that no one's read because we don't know enough about what the taste is for who enjoys that book. We can use a lot of the metadata about the book, the genre, the description, the author, all of that, but there's no guarantee that that's gonna be a good match for someone until we have people that have actually read it.
and given their opinion on the book.
ste (29:51.793)
That's true. That's true. So it's a two-way problem, right? It's the book that definitely has to have strong metadata so that we can show it to the people who might like it. So let's say, for instance, I mean, genre is a big one. But when you make a recommendation, you might think of really personal stuff, like how the b-----.
main character acts in that book or what are some like subtle accents on the theme of the book or like I mean I'm thinking of you know let's say I want to recommend the book to to you I know you like sci-fi but I don't really know what sci-fi to to recommend you know you've read a lot of sci-fi so
If I read a sci-fi book and I wanted to recommend it to you, let's say not knowing who you are, because I can definitely show you a book that you've probably read and you're going to say, okay, it's great. But let's say I don't know it and I'm showing you a sci-fi book. What would I have to know about you to be able to recommend that book to you? I think that's the interesting problem that we have to solve.
Extracting from the reader, like, what's their preference? And really going to the subtleties, like, I like characters that do this. I like themes that are like that. I like a pace that's like this and like that. I was wondering this about our tags, because tags can often be either too abstracted or too binary, let's say.
adam (31:16.959)
Yeah.
ste (31:46.205)
so they don't like go on a spectrum of things. So ideally, if I would describe what I like about the book, I'd say a bunch of things about books I've already read. And from that, you'd have to extract some data, which would be like the tags, the genres, the content warnings that I...
adam (31:54.658)
Hmm.
ste (32:15.597)
like I don't like the, yeah, all the stuff related to that. And then figure out what else to recommend based on that data to see what else I would know. So that's, yeah, that's a really hard problem to crack. Do you think like, I'm actually maybe gonna try this with GPT, just like describing what I like to read and see if it gives me like really good recommendations. Do you think like natural language?
processing is a good solution to that? Or can it like get those preferences in a way that's sensitive enough?
adam (32:58.294)
I think that like the different, I'm not an expert in the machine learning sides, but from what I've seen, there's like GPT-3, which is like natural language processing, where you're just giving it a lot of data and then asking it questions. And that's not going to be as accurate for making a recommendation as doing something where it's like, here's every book I've read.
here's every detail about those books, and here's every rating. And then being able to say like, what books do you recommend based on that? Because that's gonna have a much more well-defined data set, and it's going to be able to understand the, I forget the name of the thing. I think it's like the different fixtures within the data, which are like the facets, the different facets of the data, like,
ste (33:38.64)
Okay.
adam (33:57.366)
These are the characters you liked. These are the genres you liked. These are the locations, the settings, the time periods. So it's like developing a profile for you based on that. While ChatGPT, from what I understand, isn't doing that exactly. It's doing something similar, but it's not, from what I understand, it's not doing it the same way. But it probably still would give some interesting recommendations.
ste (34:21.465)
Mm-hmm
ste (34:26.565)
They would be interested. Yeah, I guess they would be interested. So what you're saying is that the quantitative data that you can get, you know, the actual raw thing is gonna yield better results. It's probably gonna yield better results than the qualitative data, which might contain a lot of noise, which I would agree with.
adam (34:51.078)
Yeah. But yeah, it could, it could still make for, uh, interesting recommendations and fun, a fun tool to play around with. And
ste (35:00.505)
Yeah, that's true. Yeah. I'm guessing it could. Well, the first way is our way of doing it. So it's good that, you know, we're actually making something that would yield those recommendations. And we've seen like those matches made on our platform. We've had readers discover a lot of interesting books that...
they probably wouldn't have otherwise. I'm guessing that's a thing we should continue this game to get better at because I also think, you could go and ask for book recommendations from ChatGPT, but they're not gonna know your history. They're not gonna know your book data. Maybe at some point where you'd be able to, or we'd be able to release it on...
adam (35:50.805)
Yeah.
ste (36:00.221)
the hardcover data that each reader has, maybe then it would be able to give accurate recommendations based on that. But even then, yeah, I think it's another mechanism. Again, I don't know anything about how it works, but I mean, technically, on the technical side, but I'm guessing it's not.
adam (36:16.032)
Yeah.
ste (36:27.381)
not at all related to the other way of doing it. So for instance, whenever I do a prompt, it's like I heard about this interesting analogy that you have to act like you have the world's best, like the world's best genius in the other room, but they can't hear you, they can't, like you can't interact with them. You can only slide notes under their door.
adam (36:31.405)
Hey.
ste (36:55.961)
And they can give you like really good responses, but you have to be really careful about what you're telling them because the context you give them is what makes the answer good or bad. So it's the same when you're writing a prompt, right? You have to give it as much context as possible. That's where I think many people I've talked to that have used the Trajit Pitey kind of
not misuse it, but don't use it to its full capacity or even intended use because they just ask it, okay, should I do this or this? Or search for things that, I don't know.
I like this and that, search for things that, other things that I would like. And, you know, it's first it's trying to stay away from any opinion that anything that might be interpreted as an opinion on their part, because it doesn't want to appear human, because that's the way it was prompted initially. But yeah, also it doesn't like have that context.
adam (38:12.47)
Yeah, it's, it, yeah, it kind of feels like, uh, one of the things is like, and this was in the John Oliver thing yesterday, it's like, there's, there's kind of two paths forward. There's one where like AI solves all of our problems and one where we work together with AI to solve our problems. And we have to learn how to talk to the AI. And in this case, if we were doing something like, uh,
ste (38:12.593)
The context is important.
adam (38:41.49)
a book bot where we were having it like, you know, asking a book bot for book recommendations, just saying like, give me recommendations, like that's going to be effectively the same as like our, our recommendations page, like our recommendations page is probably going to be better because it has more data fed into it. So it's like, how do we, how do we use a bot like that and make it better than what we can already do with the data? And.
Yeah, it's almost like one of the things that comes up is like, yeah, like you said, like pre-filling prompts so that we're giving people like the happy path where all they have to do is fill in that last little bit. And then like, here are some books I recently loved. And then like you decide which ones and then it creates recommendations based on those. Or like, I want a book that's similar to
ste (39:24.538)
Mm-hmm.
Yeah.
adam (39:41.622)
this book, this book, this book, and this book. And then you could like, it can do that combination of books, which is something that I know we can't really today. Like I want something that's like, you know, it has the fantasy element of this book and it has like the mythological element of this book. Let's put those together and give me a recommendation.
ste (40:03.761)
Yeah, that's really good. I think that's the only case when it can actually give maybe better results than we can do with the book data now, because I'm not sure what it would take or how it would function logically for us to be able to, well, yeah, it could do basically a cross-reference of tags or.
I don't know how it would work anyways. But yeah, that seems hard. Yeah, yeah. I'm wondering if there are readers who also like, I mean, for like really concrete, for like really specific book facts, I think having the data and the algorithm is great. I'm wondering for like random facts. Let's say, I don't know, I like trees.
books that are about trees or are like in any way. I mean, we could do that. Yeah. If the books are tagged with like the name tree, I guess we could do that.
adam (41:16.958)
Yeah. Yeah. If, if, if like, uh, the chat GPT data, like if we were hitting their data set, which might be trained on a lot more data about each book, including like every review that's ever been written about that book. And like, so they might have like that tree data, not as a tag, but as like part of the related data to it. Um,
ste (41:32.381)
Okay.
adam (41:43.134)
So yeah, there would probably be a lot more like nuanced data there.
ste (41:43.366)
Yeah.
ste (41:47.757)
Yeah, exactly. I think the cases where you... Instead of having those recommendations as actual data, actual books, feel like you're actually asking someone for advice, but you're also counting on the way they view the problem. Let's say I want...
to read a book about trees or concerning trees and actually it gives me recommendations for poetry about nature or it actually infers what I might want to know instead of just books that mention trees. I guess that would be a different...
Yeah, way it's interpreting the results instead of giving me like the something related to that specific word. I think that's where it could be powerful.
adam (42:59.166)
Yeah, it's almost like building up a search with multiple searches. So, you know, you start with trees and then it gives a bunch of results. And you're like, narrow this down to books, narrow it down to books written in the last 20 years, and then narrow it down to people who are similar to this author. So it's like, it's starting with a wide net and it's like narrowing it down more and more. And that might be a fun use of it.
ste (43:27.289)
Yeah, it might be fun. I mean, definitely. I think there's two approaches here.
ste (43:36.869)
I'm not sure which one is bound to be more used by readers. I mean, I know the first one is definitely useful. But I'm wondering if this pre-prompting would actually work. And I'm also waiting for the time when it's... Do you think...
If there could be an AI that's a model that's trained on our data, it could actually like give recommendations similar to what you described. Even if you would give it terms that are vague, basically imply some, some meaning from what you give it.
adam (44:30.75)
Yeah, yeah, that would be possible. It would be pretty much like training something on our data plus all the reviews about a book. And then it would have a little more rich data about the book than just the metadata. And...
ste (44:39.925)
Mm-hmm.
ste (44:46.305)
Yeah, that would be interesting to talk, especially maybe we can talk it with our team with Mariana and Alicia to see what their take is on this because it's pretty interesting. I think books and AI could be a really good combination from all points of view, especially for uncovering new things.
And we have the mesh score. I'm wondering if the mesh score can be really powered up by something like this that's trained on our data. And that also knows how to do... I mean, you probably covered this more with both of them. So I think it would be interesting to maybe talk about what...
adam (45:24.558)
That's, yeah.
ste (45:40.561)
the limits of data and what it can actually do and what it's... how it would be useful to readers.
adam (45:51.282)
Yeah, one of the things we can't really do now is since we have a smaller user base, anything that's user to user related, we need pretty much more readers. We need like, like there are a lot of books that only have one person that's ever read them. So that doesn't give us enough like variable variability in the data to know like what the taste is for someone who wants to read this book.
But I think on the other side, the book data side, that part is more interesting to me in what we could do, where it's more like filtering down based on criteria. So yeah, I think that's one of the things we're looking into, where it's like, where you, for instance, like go to a book page and you see like people like this book because these reasons, people don't like this book because these reasons.
And you can use that or not use that when making your decision to add that book or not. But yeah, still, still, I think a little ways off, but I think the using something like chat GPT as kind of with a, with a backend prompt. So like, you know, we, we open a session with chat GPT. We, we load it saying like, here are the books this user likes, and we list out hundreds of books here, the books this user doesn't like here. The. And then.
when they start typing with that chat CPT bot, that bot already knows what books they like and doesn't like. And so it can give recommendations based on that. I think something like that would probably be a relatively easy starting point, just to see like, does this experience, is this experience something that readers like or not? And if it is, then we could think about how we could create something like that on our own.
Or we could just, you know, use another service like, uh, like that for it.
ste (47:55.725)
Yeah, I mean, this sounds like a really, really good path. I mean, we could be able to give it that data if we want it. I mean, obviously, you know, I don't know what's technically possible, but we could like export it or like call the API with that data.
adam (48:21.49)
Yeah, it would be, I think that's kind of like what Bing chat did. Like they, they effectively like have chat, chat dbt, but they're running like a, like a pre-script that runs before eight session that like sets it up. So it'd be similar to that with our own pre-script and that pre-script would need to be very well written because. Uh,
ste (48:23.193)
Nah, that would be reloaded.
ste (48:49.22)
Yeah.
adam (48:51.786)
That's what determines the success or not of a bot.
ste (48:55.689)
Yeah, I'm wondering how that functions. I mean, from now, one of my, let's say surprises was how much it was able to, uh, make up for lack of context. So, uh, I usually, uh, give it, I try to explain what I mean.
when I write something and give it as much details as possible. But when I don't, even when I make mistakes like typos that really change the meaning of a word, it still knows what I meant. So from that point of view, I mean, it's far better than some of my email recipients, way better. So yeah, so far it's been really good at that. So I'd be really curious how.
if this is possible, how it would like actually render the results. Just give you like, let's say one book. I like this, I don't like this. Just give me one book and then just another. Yeah.
adam (50:08.447)
No. One of the, one of the nice parts about that is, uh, like if we did do something like that, we could, you know, show the books in whatever style we want. We could have a want to read button in our experience where, you know, we're just, we're showing the book that was returned from chat GPT and they can say want to read right there. So we could, we can make it very, very native experience. Like you're talking to a.
ste (50:27.982)
Yeah.
adam (50:34.25)
a librarian or something.
ste (50:37.089)
Yeah, that'd be really nice. Yeah. It would be really, really interesting to see what kind of results it gives. Yeah, I mean, it's not going away. So I think they're gonna, OpenAI is gonna have some sort of monopoly over the first,
adam (50:53.678)
Thanks for watching!
ste (51:08.389)
let's say phase of this because I mean, they're good. So seeing the results that, you know, are rendered by that makes it, I mean, I'm pretty curious what it would say with our data.
adam (51:28.781)
Yeah.
adam (51:33.254)
On the other side, I'm really curious to see what happens with BARD, Google's AI, especially since Google has so much data. Their Google Books, for instance, is one of the best sources of book data out there. So if there was some way that BARD leverages Google Books in a way that can provide actual factual data rather than just like...
ste (51:47.878)
Yeah.
adam (52:02.73)
opinionated data, you know, because that's one of the problems with a lot of like chat, chat GPT is that it's, it's saying whatever it's fed the most. And that doesn't mean that that data is correct or not. Um, while if it was Google, it could potentially be using their actual factual data. And so that's, that's one thing I'm really interested to see.
ste (52:14.948)
Oh yeah.
ste (52:22.766)
Yeah.
ste (52:26.233)
Yeah, that's going to be really interesting. And Google, I think, cares more about accuracy than either OpenAI or Microsoft at this point, which hasn't served them well because they're a bit behind in this race right now. But yeah, if Bart could leverage Google Books data, that'll be great. I mean, I just saw it with.
adam (52:47.288)
Yeah.
ste (52:55.537)
this on the discord, on our discord on hardcover, that guy that made up a Python script that summarizes the main points of a book from a PDF. So it just takes that script and feeds it to PDF and the PDF, the script summarizes the book. And the funny fact is that this guy didn't even know Python. So it just asked the GPT for the code and it actually...
He made it work, so he made the summarizer for that. I'm wondering, you know, I mean, obviously, Bart would be able to summarize all the content on Google Books. Imagine the book data. I mean, you'd have, like, knowledge of the most intricate things about a book available, like...
often it takes a lot of time for people, even it takes a lot of time. Look at Goodreads, their tags still suck and it's the product of 100 million users contributing to book data and the team of 100,000 or something librarians actively editing book data. And it's still not that good, which is...
adam (53:50.223)
That'd be really cool.
ste (54:17.589)
really interesting to see the hive mind of so many people not being able to create something coherent. I think my take is that it's because of the way the system is laid out and, for instance, the way we separate genres, tags, content warnings, moods, is like
adam (54:40.674)
Hey.
ste (54:46.945)
already an improvement. Of course, you know, you have this as well, but the way they process data basically I don't think encourages accurate book data. So tags are good, but
adam (55:04.198)
Yeah, it's a hard problem. Like, we had a lot of discussions at the last company I worked for about, like, hierarchical data and what it means to categorize and tag and all that. And there are just so many ways of doing it, whether it's like you're doing it at, like, a top level, like tags, whether you're doing it at two levels, whether you're doing hierarchical data, like...
ste (55:20.699)
Yeah.
Mm-hmm.
adam (55:32.79)
Like this is a nonfiction book, it's a biography, it's an autobiography, and that's the whole path. Or you just have autobiography as the tag, basically, and then it's not part of this hierarchy. But then if you have an autobiography that's part of a fictional autobiography, then two things are getting the same tag, even though one is a fictional autobiography and one is a fictional, non-fictional and fictional. So it's like,
ste (55:44.06)
Yeah.
adam (56:03.631)
It's a difficult problem to solve and I have a feeling we'll need to do more work on our end to get it right too.
ste (56:12.541)
Yeah, yeah. I mean, in this case, really contextual data seems I mean, how people would do it. Let's say I read the book and you're asking me what do you like about the book? I can tell you like what I liked about the book. You can then decide if that's useful or not. But I guess our brains have a really different way of categorizing stuff.
So, yeah, making sense of that problem is really hard, but I'm thinking if that's the way we actually process that information, because that raw data definitely tells you more about a book. But let's say you read a review and the review says the right words. That review might mean more to you.
than all the tags combined.
adam (57:11.05)
Yeah, or even just someone says like, this book is similar to this book and that book you love and that might be all you need to know.
ste (57:18.969)
Yeah, exactly, exactly. And it might be something that really escapes even, I mean, we'll probably nail it better and better as we get more readers and as time passes. But again, yeah, the human processing on a book and on ideas and stuff in general is really like...
complex. So yeah, I'm guessing, you know, when I read the book, of course, there are those tags, but yeah, if I would recommend that book, that book, I definitely would. I mean, I might mention a couple of the things related to the key like genre themes or the genre or the characters. But I think there would be like other stuff that's
I would say about the book to convince someone to read it. So that's like a really interesting area that we could explore, especially like for discovering like new books and new ideas. That seems like an area that's really interesting to explore.
ste (58:44.345)
I think I told you last time about that app in the app store that just publishes stories which are like in the romance, I guess, area. That's like one of the top 10 books. Yeah.
adam (59:01.484)
Mm-hmm.
ste (59:06.905)
I'm wondering how much of that is actually written with AI. I mean, that kind of stuff you could write with AI. You could like spew. Maybe I'm gonna try. Write a romance story, just have it published there. Maybe by next week. I'll be a best seller.
adam (59:15.438)
Hmm.
adam (59:20.702)
Uh, yeah, you can. I, I'm, I'm, I'm, I'm curious to see how that goes. Cause prompt writing, I, I'm, I'm wondering how much of that's like, part of the prompt and how much of that's like written by the machine. Like, do you need a lot longer prompt? Do you need a lot more detail or do you just need to say like, uh, extrapolate over 30 pages, extrapolate over 50 pages, but yeah.
ste (59:29.933)
Hahaha
ste (59:48.421)
Yeah, I'm definitely gonna fight.
adam (59:51.106)
But I have very little faith in that being good. But I mean, I might be surprised.
ste (59:57.241)
Yeah, it's but well, yeah, but again, you know, you have like a Certain standard but like the stuff Uh out there i'm gonna read a few of those see if I can like emulate it because it seems like pretty like easy Really, you know
ste (01:00:29.465)
easy to read literature, so I'm wondering how that fares.
adam (01:00:32.863)
Yeah.
adam (01:00:38.442)
Well, it's been a fun chat about AI. I know I'm cautiously optimistic about where it's going. I have a feeling things are gonna get a lot worse before they get better. But I feel like I'm still optimistic that it's going to get to a point where it's really extremely beneficial for society in a way the internet was beneficial over the last 30 years.
ste (01:00:52.895)
Oh yeah.
adam (01:01:08.01)
Like I see this as like the next big thing since the internet. Uh, and like the internet, it's going to be used in a lot of very bad ways. And then it's going to be used in a lot of really good ways.
ste (01:01:18.297)
Oh, yeah.
ste (01:01:21.525)
Yeah, spot on. Yeah, I definitely think so as well. And it's gonna be interesting for us to experiment with it because it's a pretty interesting opportunity. I mean, if we're gonna tell future people who are now really little and just don't know what this is about, what happened in our prime and...
you know, how we experimented with like the birth of when this thing was at its inception. I think that's going to be like pretty interesting, a pretty interesting story to tell.
adam (01:02:01.378)
For sure. Cool. Well, that's a good place to end it. If we don't get obliterated. Cool. Well.
ste (01:02:02.293)
Yeah, if we don't get obliterated.
ste (01:02:09.985)
Yeah, exactly. All right. Well, have a good one, Adam. And yeah, till next week. Bye bye. Yeah.
adam (01:02:18.45)
See you state. Have a good one.