I am a voracious reader. My first job was in a bookstore which, let me tell you, was a bad move on my part. How bad? I stopped counting them and started weighing them. When I hit 300 pounds of books, my parents asked me to stop. I didn’t listen until I moved into my first apartment and decided it would be better to donate 500 pounds of books than try to lug them up four flights of stairs.
Like a lot of readers, I tried my hand at writing. I did the NaNoWriMo thing. I did the “submit a short story to a publication you like” thing. And eventually, I even did the “hire an editor to polish up a manuscript” thing.
My editor, Amy Tipton of Feral Girl Books, was kind, supportive, and completely honest. Her feedback was clear, comprehensive, and insightful.
And that’s when I realized I am a terrible writer.
I had an image in my head, an idea of a narrative, but I lacked the skill to construct a decent plot or avoid trite ham-fisted stream-of-consciousness flourishes. And hearing that what was in my head, was not on the page, was a wakeup call.
So I started from scratch. I wondered, “What kind of novel am I writing? What are the basic beats/plot points of this kind of novel? How can I write like other authors, in this genre, that I adore?”
And so, as only someone avoiding a project they feel ill-equipped to tackle could, I thought of a business idea.
Value Proposition / Business Model
By leveraging the expertise of publishers, authors and their bodies of work, Writer uses machine learning to turn writers into authors.
Through partnerships with publishing houses, the value of authorship will not be diluted but enhanced. Publishers will see which writers are worth investing in, long before they have a finished draft. And authors themselves will have a chance to interact with their greatest fans, those who were inspired to write by the strength of their authorship.
In return, writers will see how closely their work aligns with their favorite authors. They receive prompts and suggestions on how to be like their heroes, while also seeing what aspects of their writing are unique and worth developing.
The balance between stakeholder value, the guiding force of publicly traded companies, and artistic value, the guiding force of writers and authors, is the Hard Part of publishing. After all, Fifty Shades started as pretty terrible Twilight fanfic, but became something incredibly lucrative. And some publishers, authors, and writers see that level of financial success AS success.
The goal of Writer is not to support only these stakeholders, but also the ones with an artistic vision, an ideal they hold more important than financial success. And by doing so, we can make the Hard Part, the Easy Part.
Mission, Vision, Values
The power of authorship isn’t in the ability to write. It is in the ability to understand how Michelangelo’s David was already there inside the marble, waiting to be set free.
The three groups we work with are the writers, the publishers, and the authors. The writers are our customers. The publishers and authors are our partners. Our mandate is to provide a clear and direct connection between our customers and our partners and to avoid injecting our own ideas and biases into the conversation.
People, Process, and Technology
You’re absolutely going to need a few ML engineers, preferably those with some experience in LLMs. A few more tech-savvy resources to ensure site stability and throughput would definitely be a good idea.
But the other half of the company will be client and customer relations focused.
Customer relations will be focused on UI/UX design for the platform. Remember, this entire edifice is functioning at the speed of computing, not at the speed of humans. If the interface is even a little complicated, writers will drop the platform and revert to, to paraphrase Pressfield, whatever environment will summon the Muse. There is evidence of this in distraction-free writing environments and their adherents.
The client relations will focus on publishers and authors.
Publishing, as an industry, has had a rough few decades with the advent of digital media and it’s distribution. As such, most established publishing houses will see this as an attempt to wrest further control of their ‘product’ from them. Having experienced staff, preferably those with previous publishing experience, will help assuage these concerns and foster healthier interactions.
Part of the negotiation with publishers will be around what the value for them would be. One major benefit would be alerting publishers that a writer happens to be working on a book that’s similar to a previously lucrative author’s work but unique in a particular way. Publisher’s would then be able to connect with the writer directly.
Authors, more so than publishers, are concerned about the impact of machine learning on their craft. Imagine is a learning model had access to every novel, novella, and short story of a particular author. How do you think they would feel about it? But ask an author if all the research they used to do, themselves, could be done in seconds by a machine? I suspect they’d have a far different response.
Meeting the authors where they are and walking them through their perfectly reasonable concerns will also be crucial to the success of this venture.
Publishers are the starting point since they would have multiple authors and several works as a base. Their editorial staff should also be engaged by the UI/UX team to understand the kinds of feedback writers expect.
Writers start by registering and setting up a payment protocol. They will be charged a fixed monthly fee to store their content with the understanding that it will be mined and learned from.
Writers then upload a body of their own work with timestamps. The model can learn how they’ve progressed over time. By comparing this to a model of a famous author, the areas of development can be illuminated.
General suggestions could look like, “Spend more time developing the main character.” More specific ones might look like, “Spend less time discussing the flora and fauna and more time on the battles.”
It should also be possible to combine authors, genres, and even lengths. A short story vs a multi-part series, for example.
Monetizing the publishers’ and authors’ interaction with the platform would be tricky, but achieveable. Publisher’s can license their work with the explicit understanding that it would not be used to generate new works. Authors can do the same. The licensing costs can be managed in tiers.
Lowest tier could be where the publishers/authors are paid an amount, per month, based on the number of writers that refernce their work. The highest tier could be a very low payment, but access to the platform and it’s findings.
Same as TOChanger except that the reporting out of the model will be less pure statistics and more interpretation of the statistics.
LLMs are great at detecting how close one piece of text is to another, but perfect matches are plagiarism and, to be honest, boring. But there is a sweet spot where one piece ‘reads’ like another and that will have to be determined by all the stakeholders.
If publishers are looking for something ‘like’ a certain author, they have to define what ‘like’ looks like to the LLM. Is it a 50% match or a 70% match? Is that 20% difference enough to make readers of an author’s previous works enjoy a new writer’s work?
Is 20% the difference between Clive Barker and Stephen King? Or is it the difference between Cliver Barker and Neil Gaiman?
Data is not information. And information is based on interpretation. Simon and Shuster’s target range will be different than an indie publisher, simply because the risk tolerance for a large publisher is naturally lower.
This goes for the writers themselves as well. Fostering a unique voice might require prompting the writer to choose more appropriate target authors or genres or even finished product.
Scale and Scope
I suggest starting off with a free version for writers. It would allow them to upload their text, compare it to an author’s and get a series of suggestions. The free version, doesn’t know who you are so the suggestions end up being vague. Also, the content uploaded is ingested into the model so it can report what kinds of writers want to write like which authors. This information could be used by publishers and authors themselves.
The paid version, described earlier, should be launched after this free version has ‘made the rounds’. If there isn’t enough traction, the pivot might be to license the model and method to publishers for their in-house usage. This turns a B2C idea into a pure B2B idea and there’s nothing wrong with that, per se.
As far as expansion beyond prose, screenplays would be a good next step. In this case, you’d have to engage with studios instead of publishers. And unless you want to run afoul of unions (and you don’t), you’d better get the various guilds on board with the idea.
I still have a problem with getting too many books. But I’ve gotten better at giving them away after I’ve read them.
Still working on being a better writer, though. After all, this article may be my least-edited, most-freeformed one ever!
[Update: So I published this on May 1, and on May 5, this crossed my feeds.]