AI

Robot writing a book

Lately, I’ve been going through AI panic cycles, ranging from “meh, this stuff’s never going to be as good at telling stories as me” to “Augh, augh, I’ll never be able to compete with a flood of tailored AI-generated books!” Back and forth, back and forth, fear and confidence, confidence and fear.

But now, I’ve figured it out. And my money is on “there’s a bit to worry about, but once you reach a certain level of skill, there’s no replacing a human writer.”

Here’s why:

The current generation of AI wonders is based on LLM:s – Large Language Models. For those of you who know nothing about AI, and have lived beneath a rock, LLM:s are the “we’re scraping the entire web and training our gigantic models to approximate human output” kind of AI. But that’s not important for this discussion.

What is important is this: LLM:s are averaging engines. They aren’t capable of thought, or content recognition, or even random output. They take whatever you type into them, and give you the answer that an average person would give.

Here’s an uplifting proof. That’s from a overview of a new AI tool that allows users to interact with several models at once. The question is a very open-ended “tell me a joke.”

Give that prompt to 1 000 people, and you’ll end up with 900 different jokes, and 50 people who say “What? Me? Joke? No, I couldn’t…”

3 of the 5 questioned LLM:s reply with the exact same joke.

Exact. Same. Joke.

Here’s the segment (1 minute to watch):

 

That result in itself is a joke, but here’s what’s going on: they’re using similar training data, and that data likely has a not insignificant amount of copies of that joke. When asked to tell a joke, the LLM:s look to what tokens follow the joke prompt, and return the same joke. And since it’s a science-themed joke, it’s likely (at least in my mind) that it’s a joke used as an example in numerous computer science texts on the web. That the LLM:s have scraped, and are now regurgitating.

There’s a study (to which I’ve forgotten the link) that asked LLM:s to generate random numbers between 1 and 100, over and over again. In a truly random scenario, you should get an output that resembles a wavy line – each number represented roughly the same amount of times. Guess what happened with the LLM:s? Half of those “random” numbers turned out to be “42” and a quarter were “69”…

And when the researcher pressed the LLM to generate the random numbers with memory, 100 times in a row, they got each number exactly once – also NOT a random selection.

That’s because the LLM:s can’t do random. They can only do average.

TLDR: LLM:s tend to produce the SAME output.

Here’s a great video by Professor Sabine Hossenfelder explaining why (5 minutes):

What does that mean for us writers, artists, musicians, and general creatives?

We’re going to see a period of AI being more and more able to create formulaic stories. If you’re the kind of writer that only writes to market, strictly following a formula (not mentioning any names, but I’m looking at you, Best-Selling Brand Name Author #2,) then you’ll likely have a very hard time in the next ten years.

AI can be a competent writer. AI can follow a template. AI can do filler. If you need to churn out your average, empty newsletter or background image or Lorem Ipsum junk text, AI is your buddy.

AI doesn’t innovate. And that’s they key.

Because what will happen is what always happens: readers flock to a new genre, they fall in love with it, they want more of it – and then they fall out of love, and want something new. And LLM:s can’t do new. Not without a lot of human prompting and support, and sometimes not even then.

So for me, who writes quirky, off-genre books that combine diverse elements, diverse casts, and basically stuff that I find cool, I’m not worried. AI will, for the foreseeable future, have a hard time duplicating what I do. And when readers tire of same-old, same-old and search for something that’s beyond what their AI-dominated genres can deliver, maybe they’ll discover something by me.

Pen icon
The Pirate Publisher
The Pirate Publisher by Joseph Ferdinand Keppler (1838-1894) / Wikimedia Commons

I keep hearing, often from new writers or aspiring writers, that it’s pointless to write, you’ll never make a career of it today because AI is coming and spamming Amazon with rip-offs, knock-offs, or plain crap. AI will kill writing! AI will kill publishing! The Great Wave of Spam is coming!

Nope.

It’s already here, and it’s been here for a long, long time. (more…)

Pen icon

I have a confession to make: I don’t like editing.

Writing is fun. Editing is a chore. Not only is it a slow slogg, you have to be ever-vigilant and outright meticulous, both things that I don’t enjoy when reading.

So when I played around with ChatGPT (admit it, you’ve done it too!), on a whim, I asked it what the word “crudmunching” meant in the text I’d fed it. And to tell the truth, ChatGPT’s answer blew me away. (more…)

Pen icon

There’s a battle going on right now, between the pro-AI futurists and the anti-AI rights creators. Both sides got some excellent points that you can find online. I’m not going to comment on that.

I will say that I’ve been using the Midjourney Art AI and while it took quite some getting used to (and the results, as yet, can’t compare to a true illustrator – for one, a human illustrator can count the number of fingers on a hand,) there are some things that AI is really, really good at.

Like scraping lots of images and extracting the similarities from them. Which is exactly what we do when we’re doing cover research. Take a look at this:

AI-generated Western covers. It may not be the greatest covers in the world, but they’re definitely western covers. That’s the prompt I used “book cover, western”.

If I was writing western, assuming I knew nothing about the genre or the covers, I’d instantly know a few things: (more…)

Pen icon

This is a reply to John Scalzi’s post on not using art AI:s.

I’m going to stick out my head to get it chopped off: I’m somewhat AI-positive myself.

Yes, that includes writing AI:s like GPT3, too, even though I’m a writer.

The reason for this is twofold:

First, I’m leaning against the lessons from the Bittorrent debacle in the late 90’s and early 00′. A lot of powerful people screaming how music piracy would destroy music, how no musician would be able to afford a living, and we’d all lose out.

What happened? (more…)

Pen icon

Get Free Science Fiction & Fantasy Stories!

I hope you enjoyed this content!

If so, subscribe to monthly(-ish) newsletter and get free copies of some of my stories and collections!

No spam, and you can unsubscribe at any time!

Get Free Science Fiction & Fantasy Stories!

I hope you enjoyed this content!

If so, subscribe to monthly(-ish) newsletter and get free copies of some of my stories and collections!

No spam, and you can unsubscribe at any time!