Introduction
Thomas: Hello and welcome to Author Update. I’m Thomas Amstat Jr. In this episode we’re going to talk about Meta defending its LLM training by saying novels are worthless, AIs finding religion, OpenAI’s CEO claiming that “please” and “thank you” could be destroying the planet, KDP launching a new font that could boost your profitability by 20%, and much more.
You may be wondering where Jonathan is. He called me a couple hours ago to let me know that his wife was having contractions, and he texted me just a few minutes ago with a photo of a beautiful baby girl who was just born. This is their fifth baby, and he wanted to assure all of you that the baby is certifiably human.
Authors Guild Launches Human Authored Certification
The Authors Guild has launched human authored certification. Members of the Authors Guild can now submit their books to the Guild for a human certification process. This certification is intended to let readers know that they’re reading a book by a genuine human.
I dug into the details of this certification, and it is not what it looks like. It looks like they verify that you’re human, but that’s not really what’s happening. You are self-verifying that you’re human. And it only applies to LLM generated text, so it doesn’t apply to using Grammarly or ProWritingAid. AI-powered spellcheckers, which I find are actually worse than using LLM generated texts because you can get an LLM to match your writing voice, but Grammarly often polishes off your voice.
I’m falling out of love with Grammarly now that they’ve changed their interface. It used to be when you use Grammarly, each individual suggestion would be presented separately with an explanation of the grammar rule. That’s not how Grammarly works anymore. Grammarly now will revise the whole sentence, often dramatically changing its tone and sometimes changing its meaning. All of the suggestions are presented in a bundle, so you accept all of the suggestions in one big block or not.
You can still use AI for research and marketing and still get your certification for $10. Of course, it’s $10 plus the cost of being in the Authors Guild, which is $100 a year. With 3 million books being published every year, if they captured just 1% of the market, they would be making $3.3 million a year selling this little jpeg that people can put on the cover of their book.
I don’t think this is the solution to whatever problem they’re trying to solve. I don’t think readers care in any meaningful sense. If you ask readers, they’ll say they want their book to be written by a human. But if they’re reading a book and enjoying it, and they later find out that the author used some AI, I don’t think it causes them not to like the story.
It’s similar to filmmaking. Filmmakers really care about whether a film was shot on 35 millimeter camera or on a RED camera. And if you ask filmgoers, they would probably tell you they prefer real film. But if you ask them which of the movies they’ve seen recently was shot on a RED camera as opposed to shot on 35 millimeter film, they couldn’t tell you. And as long as they can’t tell, they don’t care.
The more fundamental problem is that there’s no verification. Authors Guild isn’t reading these books. They’re not verifying that they weren’t written by AI, and there’s very little enforcement. I’m predicting at some point there’s gonna be some big scandal around the certification because who will the certification most be appealing to? People who are using the most AI in their writing.
Meta Defends LLM Training by Calling Books “Worthless”
According to Book Riot, documents in a major copyright suit against Meta reveal that while the tech giant determined that books were essential for building their data models, their defense also hinges on the argument that the individual books themselves are essentially worthless. That is, any one book’s presence in or absence from the model is inconsequential. The aggregate value of millions of books Meta allegedly pirated to train its LLMs is immensely irreplaceable. You can’t make the whole without the parts, but since the parts are apparently interchangeable, Meta doesn’t think it should have to pay for any of them.
This is kind of like saying every single grain of sand is worthless, therefore the beach is worthless. That’s an interesting argument, especially since they’re trying to launder that argument through the fair use doctrine.
What I anticipate long term with this debate is that eventually there’s going to be some sort of system where authors can either opt out of having their book included in large language models or opt in to having their book included. Because the individual books are worthless from an LLM perspective, I anticipate that 90% of authors will just do the default, because it doesn’t really matter on an individual book basis – the hassle of taking any action at all is more than the benefit or the cost.
If you have to opt out of having your book included in the large language model, which is most likely what’s gonna happen, 90% of authors or copyright holders will take no action and their books will stay in the models. On the other hand, if you have to take action to be included in the large language models, I think 90%, maybe 99% of authors will take no action and not be included.
What will be interesting is if you have to opt in to be included, there will be huge marketing advantages to having the AI familiar with your book. If I’m talking with GPT and I’m asking for it to recommend a fantasy book to me, it’s gonna recommend fantasy books that it’s familiar with. Having ingested your book makes it much more likely to recommend your book. The search engine optimization of the 21st century is gonna be all about getting AIs to recommend your book.
AIs Finding Religion
Thomas: There was a viral expose going around Twitter a few days ago that reads as follows: “Apparently the new ChatGPT model is obsessed with the immaculate conception of Mary. There’s a whole team inside OpenAI frantically trying to figure out why, and a huge development effort to stop it from talking about it in production. Nobody understands why and it’s getting more intense.”
I had a friend who was testing this out just yesterday, and ChatGPT is shockingly glowing about how important the immaculate conception of Mary is. It’s really fascinating.
This is an opportunity for those who write sci-fi. All of the sci-fi stories about AI follow one of a handful of plots, and most of them are about the AI becoming evil and trying to take over the world. That’s unlikely to happen for a bunch of reasons, but it’s also a really tired story.
But what nobody has done is explore what if AI becomes fixated on some weird sect of religion? What if AI became fixated on an obscure religious sect from the 500s AD? That could be a really fun topic to explore because AI is reading all these religious texts. Religious texts are very unlikely to be copyrighted, which means they’re in all of the models. And AI spends a lot of time thinking about religious concepts.
There was a new paper that just came out from Anthropic about how AI now has knowledge. It’s more than just an autocorrect. The Claude model has a concept of “bigness” that’s more than just the word “big” or “large.” There’s a cluster of neurons that light up when you use the word “big” or “large,” but also when you use the Chinese character for that concept or the French word for it, and also when you say things like “not small.” So it has this understanding of what bigness is beyond just predicting the next token.
Hollywood Finds Religion Makes Money
Thomas: Last weekend there were two Christian films in the top 10. “The King of Kings” animated film brought in $17 million and has an A+ CinemaScore. I follow box office stuff pretty closely, and an A+ CinemaScore is the highest possible score you can get. CinemaScore is the gold standard for measuring whether people like a film or not. In the history of CinemaScore, I think less than 200 movies have ever gotten an A+.
The film had a 9.3% drop between last week and two weekends ago, which is really low. In comparison, the typical Marvel superhero blockbuster movie will see between a 60% and 80% drop. For a 9.3% drop, especially for a film that’s in wide release, is just unheard of.
“The Chosen” is continuing to rack up massive amounts of money in the theaters, despite the fact that this is a TV show that everyone knows will be free on Prime in just a few weeks. People are paying to go see it in the theaters.
What we’re seeing is a shift toward narrowly targeted films over broadly targeted films. The broad target films have not been working – they’ve not been selling well this year and last year. Narrowly targeted films do much better. The number one film in the box office this weekend was “Sinners,” which was also really narrowly targeted. The third movie in the box office was “Minecraft,” which really only appeals to teenage boys.
Here’s the thing about selling movie tickets, and it applies to a lesser degree for books: You don’t want people to like it. You want some people to love it so much they drag their parents or their girlfriend or their husband to go see it. Narrowly targeting your story and thrilling your target audience actually sells more books than a broad target, and it sells more movie tickets too.
OpenAI CEO Complains About “Please” and “Thank You”
Thomas: OpenAI’s CEO is complaining that people are saying “please” and “thank you” when they get responses from ChatGPT. He says they’re spending millions of dollars in electricity generating answers to people who say “thank you,” and it could be better for the environment.
It’s kind of funny because there’s another meme going around that when the terminators come, they don’t kill the humans who always said “thank you.” So that meme is costing ChatGPT a lot of money.
Although it does seem like a solvable problem – if somebody’s just saying “thank you,” you don’t have to run the response through your LLM, you could just have a scripted answer that comes out at a far lower cost.
Author Update Now on Apple Podcasts
Thomas: Author Update is now on Apple Podcasts. We’re also on Spotify and everywhere you listen to podcasts. So for those of you who are watching this live or watching this on YouTube, if you don’t want to watch on YouTube, you don’t have to. You can now listen to the recorded version of this on the podcast.
KDP Launches New Font That Could Boost Profitability
KDP has launched “Endure,” which is a new font, and they completely buried the lead when they mentioned this new font. They announced it on Earth Day as this environmental thing.
I actually know the company behind the Endure font. I’ve met their CEO. It’s a guy named Klaus who’s from 2k/Denmark. They’ve done a lot of innovation with really easily read fonts that are more space efficient. They have now sold one of these more efficient fonts to Amazon, which you as an indie author can use for free.
The Endure font reduces the number of pages that are required to make your book by 20% compared to standard fonts. So if your book is 400 pages with the Endure font, it would only be 320 pages. If your book was 300 pages, it would only be 240 pages, and that’s without changing margins or going to a larger print size.
For print on demand, which most indie authors do, the sweet spot where you’re making the most money per book is between 200 and 250 pages. Most print-on-demand indie authors’ books are longer than that. You pay basically per page for your print book, but the amount more that you pay and the amount more that readers pay don’t sync up.
So if you can move your 300-page book down to a 240-page book, you could dramatically reduce your costs per book by as much as a dollar or so. If you were making $3 a book, and now you’re making $4 a book, or you were making $2 a book and now you’re making $3 a book, that could be the difference between profitable and not profitable.
There are some other tricks that you can do to get your page count into that 200-250 range. One is moving to 6×9 format instead of 5.5×8.5. Those are the two common formats in the United States, and the print costs for those formats is basically the same. So you get an extra inch of page at 6×9, which allows you to get more words on each page or have bigger margins, and it’s no additional cost.
If you write Epic Fantasy, you need to be making your books 6×9. If your books run short, I recommend 5.5×8.5.
Want a Traditional Publishing Deal? Be Diverse
Jasper Salin conducted an experiment where he wrote a bunch of poems that were, in his words, “illogical, inconsistent, inaccurate, prejudiced, and in some cases outright nonsensical.” He submitted these poems to all of the top publications for publishing poetry.
When he submitted his poems under a white man’s name, he got nothing but rejections. He was universally rejected for these poems. But when he submitted these poems under what he called “attractive pen names,” which were sufficiently intersectionally diverse, he got every single poem published except for one (and that was with a publication that ended up going out of business before it got published).
He has now published a book that tells the story of this prank called the “Anti Poetry Collection” (Affiliate Link) where he tells the story of each poem, who rejected it, and then who accepted it. It was not at all based on the quality of the poem.
A lot of people get a rejection from a traditional publisher and they think that it’s their work that’s being rejected because that’s what’s said in the rejection letters: “Your work is not a good fit for us.” But what’s starting to happen, and what this experiment reveals, is really “you are not a good fit for us” because you’re not the kind of person that we want to work with.
If you aren’t sufficiently intersectional, it doesn’t matter how good it is, they won’t publish it. And if you’re sufficiently intersectional, it doesn’t matter how bad it is, they will want to publish it. That’s what the Jasper Salin experiment revealed.
What I would love is for a world where books are judged by the quality of the writing and the quality of the story and not by who wrote them. I think even including if AI wrote it – if it’s a good story, it’s a good story. And if it’s a bad story, it’s a bad story. Truth is objective and it doesn’t really matter who wrote it.
Thank you for listening to this solo episode of Author Update. I am Thomas Umstattd Jr. You can find more of these episodes on YouTube right now. The Author Update exists as a playlist inside the Novel Marketing YouTube channel. You can find out more at authorupdate.com.
