AI Training Data Meets Copyright: How Publishing Can License Content at Scale w/ Julie Trelstad, Head of US Publishing @ Amlet.ai

WEBVTT
00:00:00.000 --> 00:00:02.040
One of the most frightening things that are happening
00:00:02.040 --> 00:00:04.000
is if an author puts out a book, especially a
00:00:04.000 --> 00:00:07.320
nonfiction book, within a short amount of time,
00:00:07.339 --> 00:00:10.160
you're going to have a bunch of AI -generated
00:00:10.160 --> 00:00:14.619
imitators of that book. is that Anthropic got
00:00:14.619 --> 00:00:17.440
a hold of a pirated library of books. They didn't
00:00:17.440 --> 00:00:19.120
even buy the books. They went straight to the
00:00:19.120 --> 00:00:22.179
pirates because they couldn't legitimately get
00:00:22.179 --> 00:00:24.899
the e -copies that they needed. Content providers
00:00:24.899 --> 00:00:27.620
should be paid on a token basis like Spotify
00:00:27.620 --> 00:00:29.719
artists. So every time somebody pulls up your
00:00:29.719 --> 00:00:32.640
information or uses it, you should get, you know,
00:00:32.640 --> 00:00:36.679
micropayments. Welcome to Artificial Insights.
00:00:37.109 --> 00:00:39.009
where we have conversations with leaders and
00:00:39.009 --> 00:00:42.189
thinkers in AI about how AI is changing economics,
00:00:42.549 --> 00:00:46.990
work, and us. I'm your host, Daniel Maneri, and
00:00:46.990 --> 00:00:50.350
this is Season 5. Today, I'm joined by Julie
00:00:50.350 --> 00:00:52.649
Trelstad. Julie has worked in the publishing
00:00:52.649 --> 00:00:55.890
industry for over 30 years, spanning traditional
00:00:55.890 --> 00:00:58.630
publishing, digital rights, and indie publishing.
00:00:59.229 --> 00:01:02.630
She's head of U .S. publishing at Amlet AI, where
00:01:02.630 --> 00:01:04.450
she works on rights and licensing infrastructure
00:01:04.450 --> 00:01:07.489
for the AI era. And she also runs paperbacks
00:01:07.489 --> 00:01:09.609
and pixels, helping authors build sustainable
00:01:09.609 --> 00:01:12.390
publishing and marketing systems. What got me
00:01:12.390 --> 00:01:14.609
interested in talking to Julie was when she wrote,
00:01:14.769 --> 00:01:18.189
the infrastructure enables the experiment. AI
00:01:18.189 --> 00:01:20.810
needs high quality data to train on and high
00:01:20.810 --> 00:01:22.950
quality material to reference when it answers
00:01:22.950 --> 00:01:26.489
a question. As both the law and the internet
00:01:26.489 --> 00:01:29.689
move towards restricted access, we need infrastructures
00:01:29.689 --> 00:01:32.810
to discover and license good content. AMLET AI
00:01:32.810 --> 00:01:35.459
is working on that infrastructure. And Julie's
00:01:35.459 --> 00:01:37.739
seen how content publishing and access have changed
00:01:37.739 --> 00:01:40.319
through six different publishing evolutions.
00:01:40.900 --> 00:01:43.659
Seven, if you count AI. I changed my views on
00:01:43.659 --> 00:01:45.920
how AI accesses content after talking to Julie,
00:01:46.040 --> 00:01:48.819
and you might too. So let's get to it. Julie,
00:01:48.959 --> 00:01:51.019
could you tell us a little bit about who you
00:01:51.019 --> 00:01:54.159
are and what you do? Sure. I'm Julie Trelstad,
00:01:54.239 --> 00:01:57.859
and I have been working in book publishing for
00:01:57.859 --> 00:02:01.939
over 30 years. And I have been surfing always
00:02:01.939 --> 00:02:05.599
the technology wave. So when I joined publishing
00:02:05.599 --> 00:02:09.659
as a Gen Xer, I was like one of the first people
00:02:09.659 --> 00:02:12.699
who had actual computer experience in college.
00:02:13.419 --> 00:02:16.300
coming in so i was the tech person so we started
00:02:16.300 --> 00:02:19.259
out with like when my first six months on the
00:02:19.259 --> 00:02:21.979
job of working for a magazine actually we got
00:02:21.979 --> 00:02:26.120
desktop publishing within two years we were making
00:02:26.120 --> 00:02:31.590
cd -roms and then ebooks And then later, as my
00:02:31.590 --> 00:02:33.990
career progressed, we had print on demand happened.
00:02:34.150 --> 00:02:37.629
And then we had authors having to do social media.
00:02:38.090 --> 00:02:41.129
And then now we've got, you know, self -publishing
00:02:41.129 --> 00:02:44.009
has risen because of the digital thing. And now
00:02:44.009 --> 00:02:46.650
we have AI. And I've always been really lucky
00:02:46.650 --> 00:02:48.889
to be the person who gets to work on the frontier.
00:02:49.169 --> 00:02:52.270
And which one of those was the biggest shift?
00:02:53.370 --> 00:02:55.449
I'd actually like to say that the e -books were
00:02:55.449 --> 00:02:58.270
a big shift, not because of the technology, actually,
00:02:58.310 --> 00:03:00.449
but because of what was also happening in the
00:03:00.449 --> 00:03:02.370
publishing industry at the time. And it may have
00:03:02.370 --> 00:03:04.990
been related. But at that time, all the large
00:03:04.990 --> 00:03:07.729
publishers were consolidating. So in the U .S.,
00:03:07.729 --> 00:03:10.729
we had over 50 large, medium -sized, profitable
00:03:10.729 --> 00:03:13.669
publishers. And today we have five. And that
00:03:13.669 --> 00:03:16.050
all really started in the 90s. But at the same
00:03:16.050 --> 00:03:19.280
time, the rise of e -books. And print -on -demand
00:03:19.280 --> 00:03:23.159
completely empowered individual authors. So the
00:03:23.159 --> 00:03:26.599
whole ecosystem of how publishing works, I think,
00:03:26.599 --> 00:03:29.180
really started to shift in the 90s as a result
00:03:29.180 --> 00:03:32.039
of these changes in technology. But also the
00:03:32.039 --> 00:03:34.020
mergers and acquisitions, which was a business
00:03:34.020 --> 00:03:37.400
thing that was going on. And there's a question
00:03:37.400 --> 00:03:39.780
I'd like to ask on a podcast like this, because
00:03:39.780 --> 00:03:42.439
I think it's on everyone's minds. But Julie,
00:03:42.659 --> 00:03:48.180
are you an AI? Am I an AI? Sometimes I wonder.
00:03:50.319 --> 00:03:52.840
I'm definitely a biological version of it. I
00:03:52.840 --> 00:03:55.240
can vouch for that. And I find it fascinating
00:03:55.240 --> 00:03:59.039
personally because I think I was reading back
00:03:59.039 --> 00:04:01.379
when e -books were becoming a thing, but I wasn't
00:04:01.379 --> 00:04:05.340
aware of that. How is that different? How is
00:04:05.340 --> 00:04:08.199
that maybe more of a shift than what we're seeing
00:04:08.199 --> 00:04:10.659
with AI now? Well, I think it was the beginning
00:04:10.659 --> 00:04:13.800
of the shift, right? Because what it did is it
00:04:13.800 --> 00:04:16.879
allowed, I think what it's done technologically
00:04:16.879 --> 00:04:19.980
is allowed creators to directly get their books
00:04:19.980 --> 00:04:23.699
into hands of readers very fast. And then, of
00:04:23.699 --> 00:04:25.019
course, you have blogging and the internet and
00:04:25.019 --> 00:04:26.879
all those other things that happened to it. It
00:04:26.879 --> 00:04:30.060
used to be that publishing was very strictly
00:04:30.060 --> 00:04:32.560
gatekept. And I would say that's from the 20th
00:04:32.560 --> 00:04:34.839
century. Earlier on, you had people printing
00:04:34.839 --> 00:04:38.279
their own things and selling them. In the 20th
00:04:38.279 --> 00:04:41.540
century, it was highly gatekept. And what we
00:04:41.540 --> 00:04:44.860
got as readers and public were things that went
00:04:44.860 --> 00:04:50.579
through this system of editors and printers and
00:04:50.579 --> 00:04:54.019
the book distribution system. And I think that's
00:04:54.019 --> 00:04:56.920
been completely changed as a result. And that
00:04:56.920 --> 00:04:59.939
changes the way people who are creators and authors
00:04:59.939 --> 00:05:04.339
can face the industry. Does AI build on top of
00:05:04.339 --> 00:05:07.529
that? Does it change things beside it? How does
00:05:07.529 --> 00:05:10.449
that interact? Yeah, I think the jury is still
00:05:10.449 --> 00:05:13.629
out. And what's happening right now is it allows
00:05:13.629 --> 00:05:18.129
piracy to happen faster and better. And I think
00:05:18.129 --> 00:05:20.009
that's been happening since we've had e -books,
00:05:20.189 --> 00:05:22.589
right? First, there was digital rights management
00:05:22.589 --> 00:05:26.110
where they sort of put like a hex lock on downloadable
00:05:26.110 --> 00:05:27.930
books. But of course, it didn't work at all because
00:05:27.930 --> 00:05:31.310
just about every book is pirated. And now like
00:05:31.310 --> 00:05:33.470
some of the most frightening thing that are happening
00:05:33.470 --> 00:05:35.430
is if an author puts out a book, especially a
00:05:35.430 --> 00:05:38.699
nonfiction book, Within a short amount of time,
00:05:38.779 --> 00:05:41.600
you're going to have a bunch of AI -generated
00:05:41.600 --> 00:05:44.379
imitators of that book. Some seem kind of legit
00:05:44.379 --> 00:05:46.920
until you look really close. But it's flooding
00:05:46.920 --> 00:05:49.879
the market even faster than, like, when e -books
00:05:49.879 --> 00:05:52.120
came on, the market was really flooded because
00:05:52.120 --> 00:05:54.279
all of a sudden everything that was out of print
00:05:54.279 --> 00:05:56.319
came back into print. It was sort of like all
00:05:56.319 --> 00:05:58.939
the old stuff came back. And now what we have
00:05:58.939 --> 00:06:01.139
is not only that, but, like, all this new stuff
00:06:01.139 --> 00:06:04.230
is coming even faster. And I can know that like
00:06:04.230 --> 00:06:06.430
Amazon and Barnes & Noble and all the online
00:06:06.430 --> 00:06:09.069
retailers are definitely trying to squash this.
00:06:09.370 --> 00:06:11.689
But that's what's happening right now. I think
00:06:11.689 --> 00:06:15.129
I've heard of publishers starting to have to
00:06:15.129 --> 00:06:19.009
refuse books because of these AI copies and potentially
00:06:19.009 --> 00:06:22.129
not as much care in the writing. For sure. But
00:06:22.129 --> 00:06:25.209
what's the practical impact of that? Is it like
00:06:25.209 --> 00:06:27.350
every couple of books that get copied, is it
00:06:27.350 --> 00:06:30.329
big? I think it's bestsellers that get copied
00:06:30.329 --> 00:06:32.470
because, I mean, people are running these algorithms
00:06:32.470 --> 00:06:35.389
on Amazon. There's a bunch of programs and, you
00:06:35.389 --> 00:06:37.850
know, and companies that will tell you, I'll
00:06:37.850 --> 00:06:40.990
give you the top sellers on Amazon and then you
00:06:40.990 --> 00:06:42.930
can copy them and make lots of money because
00:06:42.930 --> 00:06:46.029
they're top sellers. So I think that's what's
00:06:46.029 --> 00:06:47.649
being, I don't think the ordinary run -of -the
00:06:47.649 --> 00:06:50.310
-mill books are being copied. And even like with
00:06:50.310 --> 00:06:52.709
Piracy, like I remember in the day, it was really,
00:06:52.829 --> 00:06:55.329
Piracy only really hurt the Stephen Kings and
00:06:55.329 --> 00:06:57.529
the big talent. It didn't really hurt the...
00:06:57.550 --> 00:07:00.529
small operators because there was actually proof
00:07:00.529 --> 00:07:03.230
in the early 2000s that if your book got pirated,
00:07:03.269 --> 00:07:05.790
you were going to sell more copies because the
00:07:05.790 --> 00:07:08.970
book was popular enough. And some people, you
00:07:08.970 --> 00:07:10.610
know, we didn't buy the pirated, you know, they'd
00:07:10.610 --> 00:07:12.930
find out about it and then they'd go buy a legit
00:07:12.930 --> 00:07:15.449
copy. Just another form of distribution. Yes,
00:07:15.509 --> 00:07:17.769
it just isn't. Exactly. I know college students
00:07:17.769 --> 00:07:20.769
are like, oh, I can get you that book. Like textbooks
00:07:20.769 --> 00:07:24.089
are probably in the category of highly pirated.
00:07:25.060 --> 00:07:26.959
Extremely, yeah, because they're expensive, right?
00:07:27.079 --> 00:07:29.420
So that's lucrative to, you know, if I heard
00:07:29.420 --> 00:07:32.420
it. I think this would be a good time to shift
00:07:32.420 --> 00:07:36.220
the conversation towards the, I guess, the technology
00:07:36.220 --> 00:07:39.800
that you're involved with now, like this new
00:07:39.800 --> 00:07:42.720
future, the next wave of technology that you're
00:07:42.720 --> 00:07:45.420
getting to ride. So could you tell us about what
00:07:45.420 --> 00:07:47.899
you're doing? For sure. I'm so excited to feel
00:07:47.899 --> 00:07:51.600
it. Yeah. So one thing I have been involved with
00:07:51.600 --> 00:07:54.920
a couple of Europeans in their business. I worked
00:07:54.920 --> 00:07:57.240
for Streetlib, which is an international e -book
00:07:57.240 --> 00:08:00.379
distribution company where you can, anybody can
00:08:00.379 --> 00:08:02.160
upload their books and it can be distributed
00:08:02.160 --> 00:08:04.439
to e -retailers all over the world, which is
00:08:04.439 --> 00:08:08.420
super cool. What the founders realized is that
00:08:08.420 --> 00:08:12.720
e -books, there is no way to sell the AI rights
00:08:12.720 --> 00:08:14.939
to e -books. Like this was happening over the
00:08:14.939 --> 00:08:17.480
last couple of years. And in Europe particularly,
00:08:17.480 --> 00:08:21.060
there were laws mandating that AI companies needed
00:08:21.060 --> 00:08:25.279
to compensate authors and content creators. But
00:08:25.279 --> 00:08:27.699
there was no easy way to prove that electronically.
00:08:29.629 --> 00:08:32.389
that a piece of content or a book belonged to
00:08:32.389 --> 00:08:35.090
a creator. So you have things like copyright,
00:08:35.309 --> 00:08:38.230
you have laws, you have like ISBN numbers, which
00:08:38.230 --> 00:08:41.049
identify different books and trade all the legal
00:08:41.049 --> 00:08:43.970
mechanisms, but there was no piece of software
00:08:43.970 --> 00:08:47.629
or electronic mechanism. So what happened is
00:08:47.629 --> 00:08:50.750
that Giacomo D 'Angelo teamed up with Titus Pan.
00:08:50.929 --> 00:08:54.309
Giacomo is the CEO of Straight Lib. Titus had
00:08:54.309 --> 00:08:58.580
come up with something he called the ISCC. which
00:08:58.580 --> 00:09:02.539
is an ISO standard code for any kind of content
00:09:02.539 --> 00:09:07.240
that can be fingerprinted electronically. So
00:09:07.240 --> 00:09:09.159
think of it this way. You can take your book
00:09:09.159 --> 00:09:11.440
and maybe it's a novel about whales. I don't
00:09:11.440 --> 00:09:16.879
know. But it can come and it can designate 132
00:09:16.879 --> 00:09:19.379
points of reference within that novel or more
00:09:19.379 --> 00:09:22.679
and identify that book without having to like
00:09:22.679 --> 00:09:25.409
share the contents of that book. so that the
00:09:25.409 --> 00:09:27.309
fingerprint is created and if another person
00:09:27.309 --> 00:09:30.750
comes along and tries to copy that book you can
00:09:30.750 --> 00:09:33.330
match and say nope nope that was my original
00:09:33.330 --> 00:09:37.850
content and this is the copy so amlet was created
00:09:37.850 --> 00:09:43.230
as a way to register those iscc codes and specifically
00:09:43.230 --> 00:09:44.830
because we're working in the book publishing
00:09:44.830 --> 00:09:47.899
industry what it allows is a piece of tech that
00:09:47.899 --> 00:09:51.600
allows an author or a publisher to create these
00:09:51.600 --> 00:09:55.460
codes and then to identify, timestamp them, notarize
00:09:55.460 --> 00:09:59.259
them, designate which rights are available. So
00:09:59.259 --> 00:10:02.080
whether or not AI can use it for training, if
00:10:02.080 --> 00:10:04.299
it can be used in generative AI, if it can be
00:10:04.299 --> 00:10:06.860
used for research purposes. So there's like different
00:10:06.860 --> 00:10:10.759
shades of... AI use. So it solves the problem
00:10:10.759 --> 00:10:13.279
of a machine being able to read it. And AMLIT
00:10:13.279 --> 00:10:15.759
even has an API. So anyone who is a developer
00:10:15.759 --> 00:10:20.039
can plug into AMLIT and determine which pieces
00:10:20.039 --> 00:10:22.639
of content can be used and at what price and
00:10:22.639 --> 00:10:24.940
who to license it from. And you used a term at
00:10:24.940 --> 00:10:27.899
the beginning there of AI rights. What does that
00:10:27.899 --> 00:10:30.740
mean here? AI rights, yes. Sure. I think, you
00:10:30.740 --> 00:10:33.100
know, maybe talk again generally, what are rights
00:10:33.100 --> 00:10:35.480
sales in book publishing? And maybe it's something
00:10:35.480 --> 00:10:37.919
that a lot of people don't know that in the United
00:10:37.919 --> 00:10:40.120
States publishing industry, certainly in the
00:10:40.120 --> 00:10:43.360
large publishing areas, the rights market that
00:10:43.360 --> 00:10:46.179
is selling French language rights, German language
00:10:46.179 --> 00:10:48.279
rights, Spanish language rights, world Asian
00:10:48.279 --> 00:10:51.220
rights, audio book rights, film rights, television
00:10:51.220 --> 00:10:54.519
rights, playwrights, braille rights are actually
00:10:54.519 --> 00:10:56.980
worth more than the actual publishing. the book
00:10:56.980 --> 00:10:59.000
and they're certainly more profitable because
00:10:59.000 --> 00:11:02.600
to sell a right you just like here other publisher
00:11:02.600 --> 00:11:06.240
in other side of the world you can republish
00:11:06.240 --> 00:11:08.100
this work just pay us some cash like there's
00:11:08.100 --> 00:11:11.679
no expense it's very lucrative so when we talk
00:11:11.679 --> 00:11:14.000
about ai rights we're talking about this frontier
00:11:14.000 --> 00:11:17.879
where this is yet another right that can be sold
00:11:17.879 --> 00:11:21.879
to this intellectual property and i think there
00:11:21.879 --> 00:11:25.659
are some variations within how usage might happen
00:11:25.659 --> 00:11:28.279
with ai What does that look like and what are
00:11:28.279 --> 00:11:30.480
the concerns? Yeah, I think a lot of times we
00:11:30.480 --> 00:11:33.080
think about training rights, you know, which
00:11:33.080 --> 00:11:36.340
is where Chetipiti Gemini just ingested like
00:11:36.340 --> 00:11:38.779
every published piece of work or unpublished
00:11:38.779 --> 00:11:41.000
piece of work it could get and just turned it
00:11:41.000 --> 00:11:43.399
into a mathematical formula. So it's basically
00:11:43.399 --> 00:11:45.539
like you're eating something, you're digesting
00:11:45.539 --> 00:11:47.600
it. It's no longer the same thing. It's like
00:11:47.600 --> 00:11:50.500
a mathematical probability. That's training rights.
00:11:50.659 --> 00:11:53.960
And generally it happens once for every model
00:11:53.960 --> 00:11:56.379
that is built. So every time you build a new
00:11:56.379 --> 00:11:58.860
AI model, large language model. Now we're talking
00:11:58.860 --> 00:12:01.659
more about small English models again, too. It
00:12:01.659 --> 00:12:04.720
needs to train on data. And the better quality
00:12:04.720 --> 00:12:07.279
the data, the better quality the model is going
00:12:07.279 --> 00:12:10.600
to be. So that's usually a one time right. So
00:12:10.600 --> 00:12:13.139
generally, like the lawsuits that are going through
00:12:13.139 --> 00:12:15.919
right now, like the big anthropic lawsuit paid
00:12:15.919 --> 00:12:19.600
over a billion dollars to authors because it
00:12:19.600 --> 00:12:21.779
swallowed their books whole and they can never
00:12:21.779 --> 00:12:23.840
get it back. So it's like asking forgiveness
00:12:23.840 --> 00:12:26.559
rather than permission. But now that it's been
00:12:26.559 --> 00:12:29.000
done once, I don't think, you know, the founder
00:12:29.000 --> 00:12:31.519
of Amlet calls it, we're now in the post -scraping
00:12:31.519 --> 00:12:34.580
age. You're supposed to get permission now. And
00:12:34.580 --> 00:12:36.860
honestly, that was really expensive. Of course,
00:12:36.879 --> 00:12:39.019
they have deep pockets, but I don't think these
00:12:39.019 --> 00:12:41.320
companies are going to want to continue to pay
00:12:41.320 --> 00:12:43.539
billions of dollars every time they train a new
00:12:43.539 --> 00:12:47.759
model. You're laughing. What was that lawsuit
00:12:47.759 --> 00:12:51.940
in particular? Yeah, so the Authors Guild and
00:12:51.940 --> 00:12:56.399
a group of authors got together. and sued Anthropic
00:12:56.399 --> 00:12:59.580
for taking their rights, for actually ingesting
00:12:59.580 --> 00:13:03.000
these books. And what happened is that Anthropic
00:13:03.000 --> 00:13:05.879
got a hold of a pirated library of books. They
00:13:05.879 --> 00:13:07.460
didn't even buy the books. They went straight
00:13:07.460 --> 00:13:09.639
to the pirates because they couldn't legitimately
00:13:09.639 --> 00:13:14.100
get the e -copies that they needed. In some cases,
00:13:14.120 --> 00:13:16.639
I understand they actually went to use bookstores
00:13:16.639 --> 00:13:21.580
and scanned in books to get the books. And they
00:13:21.580 --> 00:13:23.799
just, you know, there was no law. They couldn't
00:13:23.799 --> 00:13:26.059
easily like, You know, if they went to a publisher
00:13:26.059 --> 00:13:28.899
and said, hey, I need AI rights, publishers were
00:13:28.899 --> 00:13:31.120
like, what's that? You know, two years ago, no
00:13:31.120 --> 00:13:33.159
one was even talking about this. It didn't even
00:13:33.159 --> 00:13:35.200
occur to them that this was a thing. And it certainly
00:13:35.200 --> 00:13:38.860
wasn't in our publishing contracts that developers
00:13:38.860 --> 00:13:41.480
could get paid to suck in nonfiction books and
00:13:41.480 --> 00:13:45.139
novels. So that lawsuit was interesting because
00:13:45.139 --> 00:13:49.399
it said that it's totally fine for AI companies
00:13:49.399 --> 00:13:53.450
to train on published materials. However. it
00:13:53.450 --> 00:13:56.950
was not cool that they used pirated books. So
00:13:56.950 --> 00:13:59.809
the payment is actually for paying these authors
00:13:59.809 --> 00:14:02.610
for the piracy. In Europe, it's a little different
00:14:02.610 --> 00:14:04.350
because there's already laws on the book that
00:14:04.350 --> 00:14:06.590
say you have to pay the creators. And that's
00:14:06.590 --> 00:14:10.450
how Amlet and ISCC was born over there, not in
00:14:10.450 --> 00:14:12.929
the United States. In the United States, yeah,
00:14:12.990 --> 00:14:15.509
as long as you pay for it. And by paying for
00:14:15.509 --> 00:14:17.470
it means like going to the street and then buying
00:14:17.470 --> 00:14:20.269
a copy and scanning it, which doesn't compensate
00:14:20.269 --> 00:14:24.419
the author for it. So you mentioned, That there's
00:14:24.419 --> 00:14:27.059
a difference in law that's kind of made this
00:14:27.059 --> 00:14:30.159
software the right choice, I guess, in Europe
00:14:30.159 --> 00:14:33.200
versus something that maybe people haven't thought
00:14:33.200 --> 00:14:35.860
about in the United States yet. Right. And some
00:14:35.860 --> 00:14:37.340
of the lawyers, I talked to one of the lawyers
00:14:37.340 --> 00:14:40.620
who was on that big anthropic site. And she said,
00:14:40.659 --> 00:14:42.379
we're looking to Europe to solve our problems
00:14:42.379 --> 00:14:44.259
because the United States is not solving our
00:14:44.259 --> 00:14:47.980
problems. Yes. In Europe, there was a law in
00:14:47.980 --> 00:14:52.450
2019 against digital scraping. that basically
00:14:52.450 --> 00:14:55.490
said that companies could not take copyrighted
00:14:55.490 --> 00:14:58.730
materials without permission and that they needed
00:14:58.730 --> 00:15:01.409
to compensate. However, when they started to
00:15:01.409 --> 00:15:03.629
try to enforce the law, people started to try
00:15:03.629 --> 00:15:06.649
to like, okay, how do we do this? It was still
00:15:06.649 --> 00:15:11.350
an extremely manual process. Ask a coder to write
00:15:11.350 --> 00:15:15.169
a letter to a rights agency and wait a month
00:15:15.169 --> 00:15:17.370
and a half to see if they can get permissions
00:15:17.370 --> 00:15:19.549
and see what, no, they're not going to do that.
00:15:19.789 --> 00:15:24.590
They're just not. That's a long time. So AMLET
00:15:24.590 --> 00:15:27.529
then sits in the middle and allows all those
00:15:27.529 --> 00:15:31.129
rights decisions to be pre -decided by the people
00:15:31.129 --> 00:15:34.649
who own it so that you can query it in real time.
00:15:35.380 --> 00:15:38.799
Exactly. And AMLET, as we're developing it, is
00:15:38.799 --> 00:15:41.340
a tool for people who work in the rights industry,
00:15:41.519 --> 00:15:45.679
right? So literary agents, big publishers, rights
00:15:45.679 --> 00:15:49.500
agencies can use the technology. So we're not
00:15:49.500 --> 00:15:52.580
a rights agency per se, although that's probably,
00:15:52.799 --> 00:15:54.500
we'll probably be doing that for self -published
00:15:54.500 --> 00:15:58.080
authors through Streetlib. But we are like, it's
00:15:58.080 --> 00:16:00.039
an infrastructure. So it's, as I say, it's like
00:16:00.039 --> 00:16:03.360
a tech connector that will take the API from
00:16:03.360 --> 00:16:06.820
the... you know, from the developer and the apps
00:16:06.820 --> 00:16:10.539
and connect it to books and to content. Now,
00:16:10.639 --> 00:16:14.519
let me be a little cynical for a second. Of course.
00:16:14.779 --> 00:16:17.799
Let's say the investors don't care. Anthropic
00:16:17.799 --> 00:16:20.379
can pay a billion dollars a year. They have money.
00:16:20.480 --> 00:16:24.200
That's fine. What would the practical solution
00:16:24.200 --> 00:16:26.559
be? Like, what's there another practical benefit
00:16:26.559 --> 00:16:30.019
to being able to access this content in a way
00:16:30.019 --> 00:16:33.159
that's fair and legal? Yeah, I think number one,
00:16:33.279 --> 00:16:35.879
like I think I'm predicting that we're going
00:16:35.879 --> 00:16:38.320
to see less large language models and more small
00:16:38.320 --> 00:16:40.620
language models. Like you and I, anybody, we
00:16:40.620 --> 00:16:43.559
can go develop our own AI tool. We can create
00:16:43.559 --> 00:16:45.519
a small model. And I use the example of maybe
00:16:45.519 --> 00:16:48.860
like a medical care model, for instance. You're
00:16:48.860 --> 00:16:51.220
going to want to build that on textbooks and
00:16:51.220 --> 00:16:53.899
peer reviewed journal content and not on the
00:16:53.899 --> 00:16:57.419
comments from WebMD. And if you don't know where
00:16:57.419 --> 00:16:59.679
your content is coming from, it's going to be
00:16:59.679 --> 00:17:01.919
harder to say this is high quality and it's going
00:17:01.919 --> 00:17:04.920
to hallucinate less if it is, you know, very
00:17:04.920 --> 00:17:08.200
controlled and very well used. And I hope that
00:17:08.200 --> 00:17:10.599
like future models are being trained that way.
00:17:10.740 --> 00:17:13.140
And I think that the technology, as I understand,
00:17:13.259 --> 00:17:16.000
is evolving so that building these smaller language
00:17:16.000 --> 00:17:18.900
models is more and more viable. Then what I'm
00:17:18.900 --> 00:17:22.279
hearing you say is because the big models have
00:17:22.279 --> 00:17:24.960
been limited to what they can take for free.
00:17:25.589 --> 00:17:28.369
They probably don't have access to peer reviewed
00:17:28.369 --> 00:17:31.269
journals or other things that are high quality
00:17:31.269 --> 00:17:34.130
academic writing. Right. And that stuff is really
00:17:34.130 --> 00:17:36.809
being locked down. I mean, that is that is absolutely
00:17:36.809 --> 00:17:40.309
been in response to this. Like scraping is that
00:17:40.309 --> 00:17:44.009
high value content is being put behind paywalls.
00:17:44.029 --> 00:17:46.470
Research reports and everything that may have
00:17:46.470 --> 00:17:49.490
been previously like, you know, maybe not visible,
00:17:49.609 --> 00:17:52.910
but not locked down is disappearing from the
00:17:52.910 --> 00:17:55.730
Web. Like the good stuff is. Yeah. And there
00:17:55.730 --> 00:17:57.650
are already people negotiating. You know, they
00:17:57.650 --> 00:17:59.710
have lawyers, they have rights people, they're
00:17:59.710 --> 00:18:02.230
signing, they're doing it with handwriting and
00:18:02.230 --> 00:18:04.890
maybe some stripe transactions between them.
00:18:04.970 --> 00:18:07.849
But it's definitely happening and we're seeing
00:18:07.849 --> 00:18:10.549
that. And then on the flip side, and this is
00:18:10.549 --> 00:18:13.670
future, for authors, it's this ability to sell
00:18:13.670 --> 00:18:16.569
the rights easily. But having to hire your own
00:18:16.569 --> 00:18:19.490
rights team and lawyers and go find a way to
00:18:19.490 --> 00:18:22.589
sell your own rights to giant companies like
00:18:22.589 --> 00:18:26.769
Anthropic. Exactly. Yeah. Yeah. I wanted to ask
00:18:26.769 --> 00:18:30.269
you, not just the rights that you need in order
00:18:30.269 --> 00:18:32.490
to have training happen. There are other kinds
00:18:32.490 --> 00:18:35.470
of AI rights for when you want to reference,
00:18:35.569 --> 00:18:38.150
when you want to use, I think. So could you explain
00:18:38.150 --> 00:18:40.829
more of that? Exactly. And we started to find
00:18:40.829 --> 00:18:42.890
four or five different rights, and it depends
00:18:42.890 --> 00:18:44.750
on the various use cases. So there's training.
00:18:45.230 --> 00:18:48.230
That's the one. And then sort of more future
00:18:48.230 --> 00:18:51.390
forward is, can it be used in a generative AI?
00:18:51.789 --> 00:18:54.359
Which is like... you know it could be and whether
00:18:54.359 --> 00:18:57.740
or not to be allowed to be trained so that individual
00:18:57.740 --> 00:19:01.740
strands of the text are recognizable and can
00:19:01.740 --> 00:19:03.880
be created and for that we talk about the difference
00:19:03.880 --> 00:19:06.279
between input and output so like the training
00:19:06.279 --> 00:19:09.779
the models is all input whereas the generative
00:19:09.779 --> 00:19:13.380
ai rights is like It might come out, you might
00:19:13.380 --> 00:19:16.819
be able to query this chatbot and say, you know,
00:19:16.839 --> 00:19:20.819
give me Daniel's podcast style writing, and it
00:19:20.819 --> 00:19:24.339
can actually create something that resembles
00:19:24.339 --> 00:19:27.140
the original thing that went in there. So that's
00:19:27.140 --> 00:19:29.660
like generative AI. And then the other really
00:19:29.660 --> 00:19:33.599
big use right now is research. So whether or
00:19:33.599 --> 00:19:36.119
not we can actually research the contents of
00:19:36.119 --> 00:19:40.730
your books. So can it be used in a deep research
00:19:40.730 --> 00:19:44.269
application? And if so, and for both of these,
00:19:44.309 --> 00:19:47.509
the generative and the research, users or content
00:19:47.509 --> 00:19:50.170
providers should be paid on a token basis like
00:19:50.170 --> 00:19:52.410
Spotify artists. So every time somebody pulls
00:19:52.410 --> 00:19:54.950
up your information or uses it or it's brought
00:19:54.950 --> 00:19:57.450
up and surfaced in these queries, you should
00:19:57.450 --> 00:20:01.069
get, you know, micropayments and ongoing. ChatGPT
00:20:01.069 --> 00:20:03.490
recently has started refusing to generate song
00:20:03.490 --> 00:20:07.109
lyrics. And it explained to me that supposedly
00:20:07.109 --> 00:20:10.490
that violates copyright and yet generating things
00:20:10.490 --> 00:20:12.970
that other people have written currently it will
00:20:12.970 --> 00:20:16.210
do. I think that makes a lot of sense. Like one
00:20:16.210 --> 00:20:17.990
of the things that I work with authors who are
00:20:17.990 --> 00:20:20.130
self -publishing and they put song lyrics in
00:20:20.130 --> 00:20:21.769
the book and this is like true and unrelated.
00:20:22.170 --> 00:20:24.089
I'm like, please take them out because you're
00:20:24.089 --> 00:20:26.599
going to be paying. hundreds if not thousands
00:20:26.599 --> 00:20:29.140
of dollars to reproduce that very small piece
00:20:29.140 --> 00:20:31.960
of text if you follow the proper channels of
00:20:31.960 --> 00:20:35.000
licensing those rights and the music industry
00:20:35.000 --> 00:20:38.119
is definitely ahead of us so i can imagine you
00:20:38.119 --> 00:20:40.740
know one thing the book publishing industry just
00:20:40.740 --> 00:20:44.279
did is that earlier this year last year the kindle
00:20:44.279 --> 00:20:47.980
app allowed you to like summarize the books and
00:20:47.980 --> 00:20:50.700
to let you have a conversation and the authors
00:20:50.700 --> 00:20:53.339
guild said uh we did not give you permission
00:20:53.339 --> 00:20:57.619
to do that No. Yeah. And they have like really
00:20:57.619 --> 00:21:00.539
pulled back on that functionality. And not that
00:21:00.539 --> 00:21:02.339
anyone thought that was all that valuable. I
00:21:02.339 --> 00:21:04.599
mean, I don't know. Do you really want to talk
00:21:04.599 --> 00:21:07.480
to your Kindle? I think, I mean, maybe talk to
00:21:07.480 --> 00:21:09.460
other things. That's a good point. But I could
00:21:09.460 --> 00:21:14.019
see the need to license those rights. If you're
00:21:14.019 --> 00:21:16.779
going to use it in a generative AI setting, then
00:21:16.779 --> 00:21:19.420
you should compensate the people who you're referencing,
00:21:19.579 --> 00:21:21.890
the people who you're emulating. Yeah. And I
00:21:21.890 --> 00:21:23.750
want to make a note that it's totally fair. If
00:21:23.750 --> 00:21:26.970
you are one human being, not an AI, who bought
00:21:26.970 --> 00:21:31.029
a book, you can use it however you want for personal
00:21:31.029 --> 00:21:34.210
use. So if you took notes from that book or you
00:21:34.210 --> 00:21:36.130
clip things and you wanted to chat with Chuck
00:21:36.130 --> 00:21:39.529
GPT about what you learned or he's totally fine.
00:21:39.670 --> 00:21:42.910
Just don't resell it. That's what's not allowed.
00:21:43.150 --> 00:21:46.710
And how does that look different now in the States
00:21:46.710 --> 00:21:50.380
versus Europe? I guess for a little context,
00:21:50.480 --> 00:21:52.599
I've had people say Europe is locking things
00:21:52.599 --> 00:21:54.940
down so hard that it's difficult to innovate.
00:21:55.180 --> 00:21:58.220
And the States is giving the freedom to the people
00:21:58.220 --> 00:22:01.240
making the models. But it feels like they both
00:22:01.240 --> 00:22:03.359
have pros and cons. I think they both have pros
00:22:03.359 --> 00:22:05.880
and cons. And I think you need to consider that
00:22:05.880 --> 00:22:08.660
this is a global thing. Like these tools are
00:22:08.660 --> 00:22:11.539
used worldwide and by global citizens and global
00:22:11.539 --> 00:22:15.019
corporations. So I think we're going to see a
00:22:15.019 --> 00:22:17.269
fair amount of... you know, bouncing back and
00:22:17.269 --> 00:22:20.809
forth. Yeah, the US does. And that's, you know,
00:22:20.829 --> 00:22:23.210
a lot of people were really upset that the anthropic
00:22:23.210 --> 00:22:26.509
lawsuit just pointed out piracy and not the fact
00:22:26.509 --> 00:22:29.130
that, you know, wait a minute, what are these
00:22:29.130 --> 00:22:31.569
people doing with our content? You know, and
00:22:31.569 --> 00:22:33.970
for that model, the early models, and of course,
00:22:34.029 --> 00:22:36.049
you know, we're talking about each model gets
00:22:36.049 --> 00:22:38.630
trained. So each version of a model, chat four
00:22:38.630 --> 00:22:42.099
or five, whatever. will contain for as long as
00:22:42.099 --> 00:22:44.539
that model is being used all of that data it
00:22:44.539 --> 00:22:49.109
could be 20 years so it's stuck in there And
00:22:49.109 --> 00:22:51.269
but I think what you're saying, like these sort
00:22:51.269 --> 00:22:53.710
of this pushback and these creative industries
00:22:53.710 --> 00:22:57.609
really pushing hard against AI. And although
00:22:57.609 --> 00:22:59.730
I'm kind of an AI optimist and I'm working in
00:22:59.730 --> 00:23:02.470
AI, like there are a lot of authors and creators
00:23:02.470 --> 00:23:04.769
who think this is the worst thing that has ever
00:23:04.769 --> 00:23:07.829
happened to the industry and are very angry about
00:23:07.829 --> 00:23:10.450
all of this. And in the world of literature,
00:23:10.890 --> 00:23:13.289
they're resisting it strongly, like closing their
00:23:13.289 --> 00:23:16.140
eyes could make it go away. But I do think that
00:23:16.140 --> 00:23:18.980
they should continue to say, if you use it, we
00:23:18.980 --> 00:23:22.000
need to understand how and we need to be compensated
00:23:22.000 --> 00:23:24.220
because you just can't make money off of our
00:23:24.220 --> 00:23:27.460
free labor. And that reminds me of maybe the
00:23:27.460 --> 00:23:30.799
New York Times lawsuit where it felt like you
00:23:30.799 --> 00:23:33.480
could, I think, even just ask ChatGPT to write
00:23:33.480 --> 00:23:37.119
in this style, like really reference them. Yeah.
00:23:37.240 --> 00:23:41.099
What is, have you seen how that's impacting what
00:23:41.099 --> 00:23:44.509
people write from my perspective? If we start
00:23:44.509 --> 00:23:47.069
pushing out human writing because we're relying
00:23:47.069 --> 00:23:49.789
on AI writing, we're going to stop having things
00:23:49.789 --> 00:23:53.329
even for AI to reference. Exactly. Like AI gets
00:23:53.329 --> 00:23:56.670
really stupid when it trains on its own AI. I
00:23:56.670 --> 00:23:59.710
think human writing, and I think we're definitely
00:23:59.710 --> 00:24:02.109
going to see this in 2026, that human writing
00:24:02.109 --> 00:24:04.529
is going to be valued even higher. I'm seeing
00:24:04.529 --> 00:24:06.839
a lot of people on their newsletters. saying
00:24:06.839 --> 00:24:09.940
written by a human. There are publishing imprints
00:24:09.940 --> 00:24:12.420
that are like written by human. I know literary
00:24:12.420 --> 00:24:16.220
agents who refuse to consider anyone's work if
00:24:16.220 --> 00:24:20.019
they say they used AI. So that, yeah. And there's
00:24:20.019 --> 00:24:22.599
discernment, right? You have to, like, people
00:24:22.599 --> 00:24:25.099
are going to be like, oh, you know, that's definitely
00:24:25.099 --> 00:24:27.700
AI, but maybe it's not. I don't know. I've always
00:24:27.700 --> 00:24:31.980
used F dashes, darn it. So yeah, I think human
00:24:31.980 --> 00:24:34.859
is, for now, it's going to be valued. But I also
00:24:34.859 --> 00:24:37.690
think I wouldn't be surprised if it's already
00:24:37.690 --> 00:24:40.369
happening that certain genre writers, romance
00:24:40.369 --> 00:24:42.730
writers, horror writers, sci -fi writers are
00:24:42.730 --> 00:24:45.390
building and training their own AIs that write
00:24:45.390 --> 00:24:48.569
like them. And, you know, if you work really
00:24:48.569 --> 00:24:52.900
hard on training an AI model. it can look like
00:24:52.900 --> 00:24:55.299
no one else's. It's going to be completely unique
00:24:55.299 --> 00:24:58.720
in all the world. So I think that writers who
00:24:58.720 --> 00:25:00.759
are leaning into it are going to get a lot more
00:25:00.759 --> 00:25:04.680
sophisticated. And I don't think, I think saying
00:25:04.680 --> 00:25:07.380
did you use AI or not is not the question. I
00:25:07.380 --> 00:25:09.180
think it should be, is this original or derivative
00:25:09.180 --> 00:25:12.319
is more the question. And, you know, like AMOLED
00:25:12.319 --> 00:25:14.339
actually, like it has a new feature and it just
00:25:14.339 --> 00:25:16.619
was released or is being released as we speak.
00:25:17.019 --> 00:25:19.660
That allows you, like once you have registered,
00:25:20.319 --> 00:25:23.220
You can upload any other content that appears
00:25:23.220 --> 00:25:26.680
on the web or that comes up and see how directly
00:25:26.680 --> 00:25:30.480
matchy it is to yours. So is this a 90 % match
00:25:30.480 --> 00:25:34.660
or not? So derivative works is going to be a
00:25:34.660 --> 00:25:36.720
thing. I like that question a lot. Is it original
00:25:36.720 --> 00:25:40.279
or is it derivative? And I think from what you
00:25:40.279 --> 00:25:42.259
said, there are many authors who would view any
00:25:42.259 --> 00:25:46.700
use of AI as derivative. But how do you see it?
00:25:47.799 --> 00:25:49.460
Like, just like I say, I think it's evolving
00:25:49.460 --> 00:25:52.640
and I've been experimenting with it for better
00:25:52.640 --> 00:25:56.299
or for worse. And I do know that to get a good
00:25:56.299 --> 00:25:58.940
piece of writing or even a good piece of outlining
00:25:58.940 --> 00:26:01.460
or help, it is not easy. Like you can't just
00:26:01.460 --> 00:26:04.819
say to JetGPT with whatever brilliant prompt
00:26:04.819 --> 00:26:08.519
you have, write me a book that won't sound like
00:26:08.519 --> 00:26:10.319
something you've already read a hundred times
00:26:10.319 --> 00:26:13.210
before. So you have to work on it, you know,
00:26:13.210 --> 00:26:15.109
even sentence by sentence. And I think it's a
00:26:15.109 --> 00:26:17.109
great assistant. It can help you. Can you tell
00:26:17.109 --> 00:26:19.309
me how many times I repeated this word or like
00:26:19.309 --> 00:26:22.470
things that writers like ask themselves in editing?
00:26:22.589 --> 00:26:24.789
And I do think it's an iterative and a back and
00:26:24.789 --> 00:26:27.750
forth process. And I do know a few writers who
00:26:27.750 --> 00:26:30.750
are doing absolutely brilliant things by very
00:26:30.750 --> 00:26:34.650
carefully training AIs to help them. Well, I'm
00:26:34.650 --> 00:26:37.569
looking forward to that. Yeah. And one other
00:26:37.569 --> 00:26:40.869
topic that you'd mentioned before was. I guess
00:26:40.869 --> 00:26:43.430
the note is the future of publishing and small
00:26:43.430 --> 00:26:47.049
scrappy imprints. Is that something that's new
00:26:47.049 --> 00:26:50.269
with AI? Is that changing with AI? Oh, it's been
00:26:50.269 --> 00:26:52.950
small scrappy imprints, but I like, I think this
00:26:52.950 --> 00:26:54.730
is something that's evolving out of the self
00:26:54.730 --> 00:26:56.609
-publishing space and independent publishing.
00:26:57.529 --> 00:27:01.410
And what's going on is, okay, so AI is interesting
00:27:01.410 --> 00:27:04.349
in other ways. AI is helping us run really effective
00:27:04.349 --> 00:27:08.029
Facebook ads. It's helping us figure out our
00:27:08.029 --> 00:27:13.029
market faster and, you know, do marketing techniques
00:27:13.029 --> 00:27:16.789
faster. So I think what AI is doing is actually
00:27:16.789 --> 00:27:19.970
bringing some of these big company skills and
00:27:19.970 --> 00:27:23.380
strategies down to a much smaller level. So that
00:27:23.380 --> 00:27:26.500
a person with good taste and a sense of an audience
00:27:26.500 --> 00:27:30.900
can build out a publishing company that's very
00:27:30.900 --> 00:27:33.579
specific for a very specific reader that reaches
00:27:33.579 --> 00:27:37.099
that reader and is yet profitable. And I think
00:27:37.099 --> 00:27:39.480
that my personal feeling is like somebody who
00:27:39.480 --> 00:27:41.660
gets successful at self -publishing who knows
00:27:41.660 --> 00:27:44.339
really like. nails it and can like get their
00:27:44.339 --> 00:27:47.240
romantic titles out there, that they almost have
00:27:47.240 --> 00:27:49.640
an obligation to publish others or to nurture
00:27:49.640 --> 00:27:52.960
other talent because not everyone who's a great
00:27:52.960 --> 00:27:55.519
writer is going to be a good marketer. But some
00:27:55.519 --> 00:27:58.559
people are figuring out that AI is like a brilliant
00:27:58.559 --> 00:28:03.380
business companion and can help them run a company
00:28:03.380 --> 00:28:06.480
in a way that you don't need the resources of
00:28:06.480 --> 00:28:10.160
a big company. And by focusing on a very specific
00:28:10.160 --> 00:28:13.099
audience you know midwestern romanticism maybe
00:28:13.099 --> 00:28:17.619
you have you can build you know a company and
00:28:17.619 --> 00:28:20.980
i do think that what's happening it used to be
00:28:20.980 --> 00:28:22.940
that publishers wouldn't even look at a self
00:28:22.940 --> 00:28:25.819
-published work now the no -go is ai but five
00:28:25.819 --> 00:28:28.319
years ago the no go you're self -published and
00:28:28.319 --> 00:28:32.259
now you're bad i think i see actually publishers
00:28:32.259 --> 00:28:35.950
are actually seeking well -published, self -published
00:28:35.950 --> 00:28:38.890
works, books that found an audience, books that
00:28:38.890 --> 00:28:41.990
need the leverage of a large publisher to reach
00:28:41.990 --> 00:28:44.710
bigger audiences, to go into print, to go global,
00:28:44.890 --> 00:28:48.230
to have the rights all over the world. That is
00:28:48.230 --> 00:28:50.789
emerging. So I wouldn't be surprised if this
00:28:50.789 --> 00:28:54.049
trend continues that we're going to see self
00:28:54.049 --> 00:28:57.269
and small publishers first. That's where everyone's
00:28:57.269 --> 00:28:59.809
going to start. It's going to grow. And then
00:28:59.809 --> 00:29:01.549
when it gets successful, then it gets acquired.
00:29:02.049 --> 00:29:05.339
So it's different. And there's some really savvy
00:29:05.339 --> 00:29:08.500
publishers like Sourcebooks that are doing this.
00:29:08.500 --> 00:29:11.700
That sounds almost like the inverse of what happened
00:29:11.700 --> 00:29:14.980
with e -books. Everybody, the industry amalgamated.
00:29:14.980 --> 00:29:17.619
Everybody had to publish themselves. But now
00:29:17.619 --> 00:29:20.700
AI is enabling somebody who wouldn't run a business
00:29:20.700 --> 00:29:23.380
to run a business to get that audience and then
00:29:23.380 --> 00:29:25.220
get the leverage to get into a bigger publisher.
00:29:25.619 --> 00:29:27.920
Yeah, exactly. And I'm sort of like devoting
00:29:27.920 --> 00:29:30.599
this year to helping like... authors build businesses.
00:29:30.940 --> 00:29:33.220
You know, I think that for me, that's like the
00:29:33.220 --> 00:29:35.700
very best use of AI, that and selling the rights.
00:29:35.920 --> 00:29:38.720
Like, how can we use AI to monetize, not AI to
00:29:38.720 --> 00:29:40.940
write? Yeah, 100%. It's different. Although I
00:29:40.940 --> 00:29:43.319
think AI is awesome for writing complaint letters.
00:29:43.819 --> 00:29:45.819
Just saying. That's right. If you don't have
00:29:45.819 --> 00:29:48.380
to put too much of your own personality into
00:29:48.380 --> 00:29:51.400
it, AI is a lot better. Yeah, it tampers down
00:29:51.400 --> 00:29:54.119
the angry bits. I find it so, like, I've gotten,
00:29:54.240 --> 00:29:56.859
you get a responder, like, he's upset about something.
00:29:57.710 --> 00:30:00.609
It actually helps. But if you want human emotion,
00:30:00.650 --> 00:30:02.750
you want to put it in. Take it down to make it
00:30:02.750 --> 00:30:05.789
a little more polite. Do it by hand. That's funny.
00:30:06.490 --> 00:30:09.630
And what I think is really exciting about AMLET,
00:30:09.750 --> 00:30:13.430
what it enables people to do when the whole system
00:30:13.430 --> 00:30:16.930
works really smoothly. What do you see as ways
00:30:16.930 --> 00:30:19.150
that could work when people are able to say,
00:30:19.289 --> 00:30:21.809
you know, I published my own catalog of a few
00:30:21.809 --> 00:30:24.549
books. Yeah, I can use it in these ways. They
00:30:24.549 --> 00:30:26.269
don't have to have their own lawyers. What's
00:30:26.269 --> 00:30:29.099
possible? Yeah, I think what's really possible
00:30:29.099 --> 00:30:31.160
is on both sides. It's like, I think you might
00:30:31.160 --> 00:30:33.799
see some really creative developers and AI creators
00:30:33.799 --> 00:30:36.960
who are building worlds around stories, who are
00:30:36.960 --> 00:30:40.420
building better specialty apps, you know, all
00:30:40.420 --> 00:30:42.140
these things where we have all these knowledge
00:30:42.140 --> 00:30:43.900
and books. And one thing that's always frustrated
00:30:43.900 --> 00:30:46.759
me about books. is that although we have all
00:30:46.759 --> 00:30:49.400
these great to -do and how -to books, I think
00:30:49.400 --> 00:30:51.420
books are way too cheap for the amount of time
00:30:51.420 --> 00:30:53.460
and knowledge that goes into them. They have
00:30:53.460 --> 00:30:55.859
not kept up with inflation. They are so packed
00:30:55.859 --> 00:30:58.900
with rich goodness for the most part, like really
00:30:58.900 --> 00:31:01.799
great actionable insights. But there's been some
00:31:01.799 --> 00:31:04.619
studies that show that only about 2 % of people
00:31:04.619 --> 00:31:07.039
can actually execute on what they read in a book.
00:31:07.549 --> 00:31:09.950
So you read Atomic Habits, like how many people
00:31:09.950 --> 00:31:12.710
actually can grow Atomic Habits from reading
00:31:12.710 --> 00:31:14.809
Atomic Habits? Only a very small percentage.
00:31:15.250 --> 00:31:18.329
So I do think that this combination of AI and
00:31:18.329 --> 00:31:21.890
applications and books could create learning
00:31:21.890 --> 00:31:25.430
that is way more powerful, story worlds that
00:31:25.430 --> 00:31:28.170
are so much more immersive. things that can be
00:31:28.170 --> 00:31:31.150
extended and imagined more. We're already seeing
00:31:31.150 --> 00:31:33.109
in the Scandinavian countries, for instance,
00:31:33.269 --> 00:31:35.230
there's a textbook publisher that's creating
00:31:35.230 --> 00:31:39.750
AI companions to their textbooks, helping students
00:31:39.750 --> 00:31:44.829
learn with a learning rubric on top of the textbook.
00:31:45.599 --> 00:31:48.299
I just think that it opens up for even more creativity.
00:31:48.480 --> 00:31:51.420
They often say there's no new stories, there's
00:31:51.420 --> 00:31:54.420
no new things. But we keep coming up with different
00:31:54.420 --> 00:31:57.000
combinations or creative expressions of that
00:31:57.000 --> 00:31:59.559
thing. So that's on the, what we can build on
00:31:59.559 --> 00:32:02.720
top of it. What else could be possible? Oh, but
00:32:02.720 --> 00:32:05.359
for creators too, you know, like it's a new source
00:32:05.359 --> 00:32:10.019
of revenue. And not only is it directly like
00:32:10.019 --> 00:32:14.440
the rights to AI, it's the fact that AI is...
00:32:14.619 --> 00:32:17.319
recommending the books and creating the algorithms
00:32:17.319 --> 00:32:20.579
and finding the authors. It's just if they can
00:32:20.579 --> 00:32:23.339
make sure that their work is discoverable yet
00:32:23.339 --> 00:32:26.920
protected, it makes it easy for, you know, they're
00:32:26.920 --> 00:32:29.279
working with the way the internet is growing
00:32:29.279 --> 00:32:31.859
and building as opposed to against it. Because
00:32:31.859 --> 00:32:33.859
right now the way to protect it is to lock it
00:32:33.859 --> 00:32:36.730
down, which makes it hard to find. And hide it
00:32:36.730 --> 00:32:39.990
away. And that's just, you know, that's the opposite
00:32:39.990 --> 00:32:43.950
to the word publish literally means to make public.
00:32:44.529 --> 00:32:46.730
And if you're not publishing it, if you hide
00:32:46.730 --> 00:32:49.430
it and don't make it available. That's exciting.
00:32:49.630 --> 00:32:53.589
I hope both of those things take off. Yeah. It
00:32:53.589 --> 00:32:56.970
seems like we're on the road. Is there anything
00:32:56.970 --> 00:32:59.490
that I should have asked you? Perhaps talk about
00:32:59.490 --> 00:33:02.809
more about AI and writing and like where, but.
00:33:03.390 --> 00:33:05.410
You know, I think we did cover those areas pretty
00:33:05.410 --> 00:33:08.869
well. Is there anything else that you'd like
00:33:08.869 --> 00:33:11.710
to talk about? Oh, the world. No, no. I'd like
00:33:11.710 --> 00:33:17.390
to talk about how AI is empowering people to
00:33:17.390 --> 00:33:19.329
know what's really going on behind the scenes.
00:33:19.390 --> 00:33:21.890
I think like this whole idea of the chatbots.
00:33:21.890 --> 00:33:24.630
And so what I find like the best use of AI for
00:33:24.630 --> 00:33:27.710
authors, in my opinion. is to do some reverse
00:33:27.710 --> 00:33:30.089
engineering of their books and their book sales.
00:33:30.410 --> 00:33:32.450
And things that were hard to get your fingers
00:33:32.450 --> 00:33:35.269
on, like who are the real competitive authors
00:33:35.269 --> 00:33:37.130
to you? What are the books that are most similar
00:33:37.130 --> 00:33:40.369
to yours? You can do a lot of really deep internet
00:33:40.369 --> 00:33:44.089
research using Gemini in particular I use to
00:33:44.089 --> 00:33:47.029
find out about books that are similar to mine.
00:33:47.150 --> 00:33:50.450
And then as a book publisher, you want to find
00:33:50.450 --> 00:33:53.410
the readers who read those books. So again, you
00:33:53.410 --> 00:33:56.390
can find out. Where are they on social? Who are
00:33:56.390 --> 00:34:00.049
their audiences? It's a lot of industry intelligence
00:34:00.049 --> 00:34:03.789
that is available at our fingertips that two
00:34:03.789 --> 00:34:06.150
years ago you would have had to pay thousands
00:34:06.150 --> 00:34:08.429
of dollars for and even then couldn't get very
00:34:08.429 --> 00:34:10.730
good information on. It's not going to give you
00:34:10.730 --> 00:34:13.090
exact sales numbers, but it is going to let you
00:34:13.090 --> 00:34:15.460
know. who's out there and how they're out there
00:34:15.460 --> 00:34:19.280
and how they're reaching readers. And I think
00:34:19.280 --> 00:34:20.920
what's cool about authors, and I think anyone
00:34:20.920 --> 00:34:22.960
who's a writer should keep in mind, is it's not
00:34:22.960 --> 00:34:26.119
a competitive industry, really. If somebody listens
00:34:26.119 --> 00:34:28.300
to your podcast, they're going to listen to 20
00:34:28.300 --> 00:34:31.480
others. If somebody buys one book on Atomic Habits,
00:34:31.480 --> 00:34:33.840
they're going to buy 20 books on Habits. It will
00:34:33.840 --> 00:34:36.460
become habitual to buy books on Habits. And people
00:34:36.460 --> 00:34:40.389
who read a lot, read a lot of books. People who
00:34:40.389 --> 00:34:42.469
don't read books at all. So never mind them.
00:34:42.710 --> 00:34:45.369
But so maybe you'll get paid when they're using
00:34:45.369 --> 00:34:47.489
ChatGPT and not reading books to read your words.
00:34:47.510 --> 00:34:50.429
Bring them in by that. Yeah. Yeah. But so when
00:34:50.429 --> 00:34:52.750
you are an author and you're looking for who's
00:34:52.750 --> 00:34:54.409
going to buy my book, you just look at who's
00:34:54.409 --> 00:34:56.789
bought other books like yours. The industry,
00:34:56.929 --> 00:34:59.809
I was an acquiring editor for a long time. And
00:34:59.809 --> 00:35:02.769
we always had to demonstrate that there was already
00:35:02.769 --> 00:35:05.329
a given market need for this particular book.
00:35:05.869 --> 00:35:08.690
And that simply this author had a unique spin
00:35:08.690 --> 00:35:11.519
on it. Not that it was unique in all the world.
00:35:11.679 --> 00:35:14.599
That's a temptation that I feel is easy to fall
00:35:14.599 --> 00:35:17.019
into, especially with all the copycats and how
00:35:17.019 --> 00:35:20.400
easy it is where unless you feel unique, how
00:35:20.400 --> 00:35:23.280
can you publish? You got to be unique. Yeah.
00:35:23.420 --> 00:35:25.280
And that's what we, that's sort of the art of
00:35:25.280 --> 00:35:28.760
publishing is finding the new voice, but finding
00:35:28.760 --> 00:35:31.320
the new voice that fits in the old paradigm or
00:35:31.320 --> 00:35:35.079
at least the existing one. That's cool. It does,
00:35:35.219 --> 00:35:38.199
actually. My own personal use of AI has changed
00:35:38.199 --> 00:35:40.280
over the last year that I kind of went all in
00:35:40.280 --> 00:35:43.400
and went crazy and experimented with a lot. I
00:35:43.400 --> 00:35:46.119
wrote a whole romance novel with cowboys and
00:35:46.119 --> 00:35:49.039
it was terrible. And it will never see the light
00:35:49.039 --> 00:35:52.579
of day except as an experiment. And I experimented
00:35:52.579 --> 00:35:55.219
with my own fiction and that came to a screeching
00:35:55.219 --> 00:35:57.380
halt. And I wasn't using it to actually write,
00:35:57.400 --> 00:35:59.670
but I was using it to research and to like. as
00:35:59.670 --> 00:36:01.909
a very close brainstorming partner. Like I was
00:36:01.909 --> 00:36:04.170
like, Oh, like this and this. And then I went
00:36:04.170 --> 00:36:06.769
to enter a writing contest, like just an essay,
00:36:06.849 --> 00:36:08.570
just a first chapter contest. And I was really
00:36:08.570 --> 00:36:11.110
happy. And I probably spent like many hours,
00:36:11.190 --> 00:36:14.190
many, many, many weeks on, on this one chapter.
00:36:15.159 --> 00:36:17.800
And went to fill out the entry form. And at the
00:36:17.800 --> 00:36:19.920
very bottom, it basically said on penalty of
00:36:19.920 --> 00:36:22.599
death, do not enter anything. It wasn't that
00:36:22.599 --> 00:36:24.239
bad, although I told my writers group that and
00:36:24.239 --> 00:36:25.719
they said, yeah, it sounded like that. That's
00:36:25.719 --> 00:36:27.940
what it sounded like. It was just like, you will
00:36:27.940 --> 00:36:30.679
be banned forever for submitting something that
00:36:30.679 --> 00:36:32.980
had any kind of AI in it. And it's just, there's
00:36:32.980 --> 00:36:36.320
no way to be subtle about it, right? Even though,
00:36:36.320 --> 00:36:38.500
you know, again, I want to just go back to this
00:36:38.500 --> 00:36:41.320
idea of discernment. Whereas, you know. self
00:36:41.320 --> 00:36:44.320
-publishing was taboo no longer and now ai is
00:36:44.320 --> 00:36:48.739
taboo but no longer as these get better and but
00:36:48.739 --> 00:36:51.920
honestly like i have some editor friends and
00:36:51.920 --> 00:36:53.960
they've said that they've had people approach
00:36:53.960 --> 00:36:56.059
them and say oh i read the first draft my novel
00:36:56.059 --> 00:36:58.360
with ai will you please read it and give me editorial
00:36:58.360 --> 00:37:02.159
advice and they're like no no thank you i will
00:37:02.159 --> 00:37:05.840
not read your ai drafted novel they just they
00:37:05.840 --> 00:37:08.170
live in And people are asking for that. They're
00:37:08.170 --> 00:37:11.349
asking for, I've seen it among my colleagues
00:37:11.349 --> 00:37:13.110
and peers who work with authors and publishers.
00:37:13.269 --> 00:37:16.909
It's just, no, we can't turn that into something.
00:37:16.969 --> 00:37:19.570
Start with the original thought. So I've gone
00:37:19.570 --> 00:37:22.150
from having AI help me outline and sort of figure
00:37:22.150 --> 00:37:25.269
out the next plot point to the point where now
00:37:25.269 --> 00:37:28.739
I write by hand. And then I read it. I let AI
00:37:28.739 --> 00:37:31.980
transcribe it for me as I read it in. And then
00:37:31.980 --> 00:37:34.239
I help it have it clean up my spelling errors
00:37:34.239 --> 00:37:37.920
and get it typed in a lot faster. But I'm using
00:37:37.920 --> 00:37:41.679
it differently after and also understanding that
00:37:41.679 --> 00:37:45.980
human is really important this year. And my audience,
00:37:46.019 --> 00:37:48.300
my colleagues aren't going to tolerate me using
00:37:48.300 --> 00:37:52.539
it in that way. Cool. How can people follow along
00:37:52.539 --> 00:37:56.059
with you and AMLET? Yeah, sure. You can find
00:37:56.059 --> 00:37:59.659
me at paperbacksandpixels .com, which is where
00:37:59.659 --> 00:38:02.980
my author education company is. And then amlet
00:38:02.980 --> 00:38:07.219
.ai if you want to register your book. So if
00:38:07.219 --> 00:38:09.519
you're a self -publishing author, the best way
00:38:09.519 --> 00:38:12.619
to register your book with AMLET is to join Streetlib.
00:38:12.739 --> 00:38:14.840
And you can find information about that on my
00:38:14.840 --> 00:38:18.139
website or on AMLET or just streetlib .com where
00:38:18.139 --> 00:38:21.210
you can sign up. for a lifetime membership and
00:38:21.210 --> 00:38:23.449
upload your books, and they will automatically
00:38:23.449 --> 00:38:26.230
be entered into the AMLINT database. Very cool.
00:38:26.389 --> 00:38:29.130
Thank you, Julie. Sure, thank you. Thanks for
00:38:29.130 --> 00:38:33.389
chatting. I think it's rare to see what an industry
00:38:33.389 --> 00:38:37.750
like publishing that is so big and feels so old
00:38:37.750 --> 00:38:42.389
would think of AI. Yeah, it's really behind.
00:38:43.829 --> 00:38:46.550
I started this podcast because of an image that
00:38:46.550 --> 00:38:49.780
came to me in a dream. Back when cities had walls,
00:38:50.019 --> 00:38:52.760
everyone passed through the gate. Talk to enough
00:38:52.760 --> 00:38:56.139
people there and you'd understand the city. I
00:38:56.139 --> 00:38:58.260
wanted to stand at the gate and understand not
00:38:58.260 --> 00:39:01.019
just how AI was changing, but how AI was changing
00:39:01.019 --> 00:39:04.039
us. If you know someone building or leading with
00:39:04.039 --> 00:39:07.539
AI, or if this episode was useful, an introduction,
00:39:07.780 --> 00:39:10.599
a share, or a quick review on your podcast app
00:39:10.599 --> 00:39:13.059
all help keep these conversations going and help
00:39:13.059 --> 00:39:16.420
others find them. You can reach me on LinkedIn
00:39:16.420 --> 00:39:19.719
at Daniel Maneri or by email at Daniel at Maneri
00:39:19.719 --> 00:39:23.599
dot house. That's Daniel at M -A -N -A -R -Y
00:39:23.599 --> 00:39:27.320
dot H -A -U -S. Thanks for listening.