A Guide to Surviving AI as an Author (Part 1)

Content Warning: This article honestly discusses the ethical quandaries of using an LLM (commonly misnomered just “AI”) in the field of original authorship. The keyword is HONESTLY.

A little-throat clearing first for the haters: If you are simply diametrically opposed to AI (i.e., LLM) on all fronts, this article isn’t for you. You may however feel free to rant and rave in the comments and call me names if you like. I generally don’t mind. I only care if you do so without reading the WHOLE post. At that point, it’s a function of your own keyboard courage and insecurity, not an honest response

I’m neither a technology hater, nor an AI apologist. I do NOT like how LLM (“AI”) is being used today in many spaces, maybe most. Socially, it’s a blight. Creatively, it’s a blight. Medically, it can be a godsend… or a blight, depending on which use you're talking about.

I’m also a realist. Every major technological leap starts with disruption. It also starts with money being poured into it. If you need that history lesson, you can go look up the receipts: the 1840s railroads, the Internet, or even the Industrial Revolution and many more. Hell, the advent of agriculture, which most of us would agree is a good thing, led to the downfall of the prominence of the “hunter” class in many cultural centers. Of course, suddenly we needed a “warrior class” to keep the marauders from stealing our wheat.

Still, though, this is just the way humans work. We find something new and shiny. Some people learn how to make a buck on it. Others get hurt by the progress. Society recovers, albeit changed, sometimes for the worse. I’m not being flip or cynical, just aware.

That brings us to the real topic.

Today we have, in the words of the Wachowski sisters, “given birth to AI.” Many of you will literally hear Laurence Fishburne’s voice in your head at those words, no doubt.

First and foremost, that is a lie.

We have not given birth to AI. Not for a while yet, I suspect.

AI is not intelligent, artificial as it may be. It’s really just a complex, slightly randomized copy-paste program.

I’ll let the horror of something that is inherently “random” being in control of things like war plans and weapons sit with you for a moment before I continue.

Is your existential dread quotient for the day fulfilled? Mine is. Let’s move on.

Next, from here forward we’re going to narrow down terminology by one step.

It’s not AI we’re concerned about. We’ve had AI techniques for two generations. What everyone is all in a tizzy about is the Large Language Model. So when I refer to LLM, I’m referring to that class of product. The most common are Claude by Anthropic and ChatGPT by OpenAI. There are many, many others as we’ll see.

Now. I mean right now. You have a decision to make. No judgment from me. My plushie octopus, Clyde, is judging you harshly from here forward, but that’s Clyde. Nothing I can do.

The decision is simply this: how much and what kind of other people’s work are you willing to use as your own and under what circumstances?

It sounds simple enough, but actually, it’s probably not. We use other people’s work all the time. If you drive a car, someone designed it. If you use a grammar checker or a spell checker or an embedded word choice checker in your word processor, those are all examples of AI techniques that were trained on a big chunk of data, some of it not paid for. We don’t really worry about this because it doesn’t compete with us, and we just assume that it’s fairly benign and no harm done.

If I’m being honest, I consider spell-check a public service in my world, not just a feature of modern word processors and web browsers. I spend a lot of time in email at work, and Mother’s Crows, I wish they’d use more spell-check.

In the space of the LLM–remember I said I’d use this term–it’s a different world because if you're using LLM in your writing, you’re using a tool that could be used by a lesser soul to compete with you in a way that’s grossly unfair in the marketplace.

Now, here’s a second decision that you need to make. And yes, right now. You need to resign yourself to the fact that LLM technology is never going away. No matter how much we scream and rail and fight. Frankly, it makes too much money and too many people find it at least somewhat convenient. Worse yet, from my own conversations with readers and other authors, readers in most genres pay only performative attention to whether a novel, a poem, or a short story is written in-part, wholly, or not at all by an LLM model.

That sucks.

Let me say that again.

That sucks.

Just an aside, if I use quotes, it doesn’t count on that whole LLM terminology thing. My house. My rules.

Even in the lesbian romance readers community, “AI novels” are so popular that they, as of this writing, top out a large portion of Amazon’s Top 100 Lesbian Romance Novels. These aren’t men looking for lesbian porn either. The readership is in the tens or even hundreds of thousands for all of these books. That means a large portion of one of the most vocal anti-LLM communities is happily gobbling up “AI slop.”

Yeah, that stings.

I’ll give you a minute to take another sip of chamomile or coffee or whiskey or roll a doobie. Or maybe do some deep breathing. Whatever gets you by. None of that works for me on that point. The knife is in on that one, and it’s not going anywhere.

So, what can you, as an author do to survive this. I mean, it’s enough to make many of us want to quit.

The unsatisfying short answer is “wait.”

The longer, even less satisfying answer is to be clear headed about what constitutes ethical vs. unethical LLM usage and see if this new technology can be helpful while we all push to make things better for creative people… if we can. And that is a big IF.

Now, market forces are in play here. The existing fervor over LLM systems will quickly create serious problems. The big frontier model creators such as Anthropic and OpenAI are not producing enough revenue to sustain the plans for all these huge data centers. Soon enough, the bills for the ones already in progress or completed will begin coming due and the industry will be forced to make some hard choices that will up-end the sudden spread.

Many “AI book-mills”, you know who at least some of them are, will likely not survive turmoil in the marketplace because LLM usage will eventually become much more expensive. So there is hope. It’s not guaranteed that there will be this “market turmoil,” but historical precedent here and fiscal irresponsibility are very strong indicators that it WILL happen.

I suggest two things here:

• Level the playing field a bit. Start looking at what constitutes ethical LLM usage and see if that’s something we can accept.

• Create a code standard that authors can agree to and apply. Think the Comic Book Code in organization if not practice, or the Open Source Project Code of Conduct, or the Linux Foundation’s core values.

I’m not mentioning geo-political issues because that’s not specific to writers, AND it’s a space that has plenty of groups and legislators you can work with, write to, fund, and chase. You don’t need me to tell you that.

So, let’s start with the tech.

First we have to look at what is considered “ethical.”

I look at it first through the lens of where the model got its data. Was it public domain? Was it pirated? Were people compensated? Does the provider disclose all the data sources, known as its “lineage.” Put a pin in that term. It’s important. Lineage isn’t clean or dirty. It’s either opaque or transparent. It’s either free, compensated, or pirated.

I produced a handy chart to say what constitutes ethical. Totally my opinion. Graded because like I said, YOU have a decision to make. And you have to decide what you can tolerate. For example, I’m fine if the transparency is maybe a little opaque because lawsuits generally open up that bit. I’m not okay with uncompensated sources. The courts say otherwise, but I still think it’s piracy. Just because it’s legal doesn’t make it right. But you might be fine using ChatGPT or Anthropic as you see fit. I’m not here to think for you. If you’re a writer, you’re probably better than me at that anyway.

LLM Ethical Lineage

Level	Transparency	Compensation/Consent	Example
5	Transparent - full corpus published and traceable	Freely available and rights-cleared - public domain and openly-licensed text	Pleias
4	Transparent — 20 named sources, independently audited by Stanford's FMTI	Mixed but disclosed, backed by legal indemnity — mostly rights-clear sources (Free Law, USPTO, SEC filings, Wikimedia) alongside some uncleared web scrape (OpenWebText, Webhose), with IBM contractually on the hook if the licensing claim is wrong	IBM Granite
3	Opaque - curation process is proprietary and not independently auditable	Freely available but not rights-cleared — open web crawl; publicly accessible but creators were not asked	Olmo 3
2	Opaque - data recipes disclosed only as broad categories	Mixed and undisclosed proportions - mostly uncleared web scrape plus some paid publisher licensing plus unresolved piracy allegations in litigation	Llama, Mistral, Qwen, Gemma
1	Opaque - no meaningful training-corpus disclosure	Mixed and partially confirmed pirated - uncleared web scrape plus paid publisher deals plus at least one legally settled instance of pirated shadow-library material	OpenAI, Google, Anthropic, xAI

Disclosure: Granite wordsmithed this for me for the sake of time, but I did actually research it myself.

So, there you go. As simply as I can lay it out. At five you get models that are very ethically sourced, just like the avocado at the fair trade store down the street, if you have one. I’m in New England, so they’re all over the region like lobster joints in Bar Harbor.

Just a note. The more powerful the LLM, the less likely it is to be either transparent or compensated. So, if you want really good results, you have to sacrifice your ethics–again, if you have them. And some of these models can’t do even the most basic things you’d want.

How to use ChatGPT or OpenAI is pretty simple. But… these others? How do you use them. What do you do? I'll do a walkthrough for people who are interested in connecting to granite. If you want other models, you'll be on your own.

Claude Code helped me write a handy program that allows me to use any model I want, but there are programs out there that will connect to these other models that are “ethical,” already available. And some of the providers that host those models have web interfaces for them. You can scout around: Though, some of these models aren’t hosted anywhere and you’ll have to find a means.

May I suggest if you want a place to start:

https://llmgateway.io/

https://huggingface.co/

For the bold, with a Mac and a solid amount of RAM 16GB or more, you can run some models locally for generally lower power requirements. They're slow though.

Just know that you have to pay up front for tokens, and each model you use has different capabilities and a different price point. Last but not least. A word of warning, Hugging Face is targeted at really technical people.

Granite-4.2 is a Level 4 on the scale.

A four on our chart is something I’d consider acceptable for the kind of work I’d want to do with my writing. I’m not stealing another author’s work to use it. The sources are transparent and under appropriate license and IBM stands behind it. If they made a mistake, they will indemnify users.

It’s better than I’d probably get from Stellantis if my Jeep 4xe suddenly accelerates on its own and runs over a nun or some similarly innocent human, creature, or alien. Not that all nuns are innocent. I’ve seen Season 3, Episode 3 of The L Word.

I’m old, but neither dead nor living under a rock. Shout out to Team Shane. Let's be honest, everyone is team shane. The model is small enough that it’s not a greenhouse gas monster to use, which makes me feel better. For that, I’ll keep writing to my representatives to insist on green energy only for data centers.

Now what does that look like, and how can you use this ethically sourced system to do some simple work. I made a program partly using an AI coding system so I could feel extra good about myself as a tech nerd. But there are several out there. AnythingLLM (https://anythingllm.com/) is a fine one.

LLM Forge. Not available to the public.

As you can see, or maybe not, in the image above, I’m using an ethically sourced model to check a bit of text and offer suggestions like an advanced thesaurus. IBM and Stanford verified the source material and as you can see, it’s a little wordy in its response but it did give me some good ideas I might not have had on my own, like using movement or texture in my writing. Kinda nice. Serious Environmental Note: For convenience and power here, I’m relying on servers from DeepInfra so I can use a larger, more powerful version of Granite. But those 5600 tokens only cost 0.000072 miles worth of carbon from a car using current US fuel efficiency standards. That means I get 13000 more of those little requests before I hit the equivalent of a mile. 77M tokens if you want to get technical. Other providers are far less energy conscious. DO YOUR HOMEWORK if environmental impact is important to you.

LLM isn’t a monster. People are monsters. People are also insensitive. It’s been proven that the more money you make the larger portion of the population you have concern for. If you don’t believe that, then consider the last time you heard someone say of the homeless, “Get a job,” or perhaps, “I can’t afford to give them my money.” Then consider that with a wealthy person who looks at you and says, “start your own company and you can have all this.” Humans have their good points, but they can also suck.

My point is this. LLM is a tool. Like a hatchet or a car or screwdriver. Used carelessly, without thought, without concern for hurting others it can have devastating consequences. But it CAN be put to good use, responsibly, and ethically. And this is what we need to push for.

Calls of ban AI will be met with deaf ears and a blind eye. My publishing company, Carson Press LLC, runs a web shop, Inkcolors. (Yes, it’s a shameless plug.) On that site, if authors want to sell their books they commit that the cover isn’t AI generated and the book isn’t AI written. But I don’t demand that no AI is used because that’s short-sighted and pointless. And probably a little hypocritical.

One last disclosure. I used to use AI a lot, not for writing, but for marketing. Then I got on the high horse and quit using it at all. I used all those lovely If you've found this article helpful, please consider subscribing to my blog.