Robotics for Writers

By August Niehaus

Now over a year into the ChatGPT-fueled AI craze, we're seeing the consequences of writers taking the tools too far: AI-generated art and prose can't be submitted to contests or collections. Low-quality books are taking over popular categories. Half-baked content is saturating every market.

 

Trying to pass off the work of a large language model, or LLM, as one's own quickly became passé, but LLMs are still powerful tools that writers can utilize to boost creativity, productivity, and quality. Robotics for Writers is your ultimate guide to harnessing AI tools to enhance every aspect of your writing process.

 

In this book, you'll find 36 practical and proven ways to use LLMs like ChatGPT and Microsoft Copilot to:

 

• Research anything from historical facts to futuristic scenarios, from character motivations to plot twists, from emotional conflicts to world-building details
• Get instant and honest feedback on your writing style, grammar, structure, tone, and more
• Spark your imagination and never run out of ideas again

 

Robotics for Writers won't teach you how to use chatbots or text generators to write fiction for you. That's still a human's job! But AI tools can help you write better fiction yourself. Learn how to craft effective prompts that elicit useful responses from LLMs and join the future of writing!

• Language: English
• Publisher: Halftree Press
• Release date: Apr 11, 2023
• ISBN: 9798215978146

Book preview

Robotics for Writers

Simple ways AI can help you write

By August Niehaus

Copyright © 2023 August Niehaus

All rights reserved.

First Edition

Reproduction in whole or in part of this publication without express written consent for any reason other than classroom education is strictly prohibited. The author appreciates you taking the time to read and enjoy this work. Please consider leaving a review on Goodreads.com or the online retailer where you purchased it, or tell your friends about it and help me spread the word!

Thank you for your support.

Find more from August at AugustWritesABook.com

Table of contents

Introduction

A note about my approach

What large language models are

What large language models are not

Thoughts on sustainability

Choose a title

Generate nuanced names

Track down an obscure term

Discover little-known history

Come up with blog post ideas

Pick the perfect hashtag

Come up with cover layout ideas

Choose a car for a character

Learn more about a profession

Generate discussion questions

Move past writer’s block

Identify your genre and sub-genres

Name a fictional piece of media

Spot potential plot holes

Describe a space you’ve never been to

Invent dishes based on an ecosystem

Avoid missing critical steps in a process

Name worldbuilding concepts

Anticipate what might go wrong

Combine multiple species to create a new one

Depict challenging emotional states

Anticipate reader reactions

Interview a character

Get rapid feedback

Discover gold in a character’s backstory

Ensure you’re using the right tense and POV

Simplify something complicated

Decide what to work on next

Combine other ideas into one

Write a speech for a character

Choose quotes to open chapters

Find related stories

Make characters who are not you

Triangulate plot, motivation, and characters

Get a character out of a scrape

Create a ritual

Etiquette for interacting with your friendly local AI

Resources for learning more about LLMs

To Clippy, a friend when I had none.

To Cortana, who will always be a part of me.

Introduction

Hello, curious writer! I’m August, a fiction writer and someone who works in tech—specifically, in human-computer interface (HCI) design. I worked on the Cortana project, when analyzing and understanding language was a much more manual task than we suddenly find that it is with the introduction of... large language models, or LLMs.

If you’re an average writer bear in the wild, at this point you’re probably most familiar with ChatGPT or Bing Chat—neither of which I have worked on as an employee as of this writing, but both of which I have tried out extensively to understand their strengths and weaknesses. Everything I talk about in this little book can be applied to either of those interfaces, or to any new and forthcoming LLM you might interact with—your results will, of course, vary (even within the same model, you can get very different results based on a whole bunch of factors, and developers may change the models at any time).

I want to be super, super clear: I don’t believe that a large language model can out-fiction a human. Certainly not now, and probably not ever.

The basic idea is that large language models predict what ought to come next, given the specific information they’ve been trained on and the parameters you set in your prompt (the input that you provide the interface). They aren’t drawing deep, rich connections between life experience, metaphors around them, and the little threads that connect all human lives together to create a memorable scene or story that sticks with readers for years. They’re just making educated guesses, which sometimes turn out quite delightful and even insightful once a human applies meaning to them.

So if you’re looking for a get rich quick guide, don’t purchase this book. I’ve experimented with paragraphs and been deeply dissatisfied, so I don’t use LLMs to write fiction and I think it’s a really bad idea in general. Fiction writing hasn’t gotten dramatically easier thanks to artificial intelligence. I still have to do all of the hard work of instilling my characters and scenes with life and meaning, crafting dialog that reveals more than my characters actually say, and ensuring that any loose ends are tied up in a satisfactory way. Oh, and... writing the manuscript. You know, small details.

But when you’re a busy writer, even saving a few minutes counts, especially if those few moments are the difference between staying in your flow and getting distracted by all of the shiny the internet has to offer. Robotics for Writers is about the cost of LLMs, and carefully weighing that against the benefits, and (if you choose to purchase and read on) some of the ways I’ve discovered to utilize the unique power of large language models in your process.

A note about my approach

Lots of the techniques described here can be adapted to more than fiction writing, but I have seen tons of guides for how to use LLMs for nonfiction writing on Medium, Substack, and other online sources. I’ll leave the non-fiction tips to the experts who have done a lot more exploring in that realm—given they write a lot more non-proprietary nonfiction than I do—and stay in my made-up story lane.

Besides, I think that in many ways, it’s far easier to imagine sample scenarios in which an LLM might help you with a nonfiction writing task whereas, short of asking a model to write you a story, it isn’t obvious how you might engage it for fictional needs.

Thus, all of the tips in Robotics for Writers are geared towards writers of fiction, and the examples and descriptions will be presented as such. That said, it shouldn’t take too much imagination to make the leap from how these might help you with fiction to how they could benefit you in a nonfiction setting.

What large language models are

Before we get into how AI can help you write, let’s talk about what this specific application of AI—large language models (or LLMs)—actually are.

Disclaimer here: I am a technical person, but I am not technically experienced or educated enough to have grasped the concepts that make up what large language models really are. This book represents my best attempt to take the technical concepts I do understand and express them in simple terms so that you don’t need to have a decade of experience to get it in a broad sense. However, if you are interested in learning about how LLMs really work, I’ve provided some directions for your research in the section Resources for learning more about LLMs.

LLMs are a leap forward in natural language processing

I can only speak to 2013 and later, but here’s what it used to be like (from the perspective of a user interface designer) to get a computer to understand something: when you happened upon a phrase you wanted your users to be able to say to trigger an action (like, make it an hour long to change a calendar event to be 60 minutes), you’d contact your friendly local data scientist to see if they could create a model. If they said it seemed possible (a lot of requests were not really doable), then you had to provide as many examples as you could think of for how a user might say or ask for that thing... without going overboard and over-training the model such that it accidentally recognized too much as the wrong thing. You didn’t want someone to say I’d like an hour long massage please and have it recognized as change my massage meeting to be one hour long.

Without going into too many details, it was kind of a Gordian knot of psychology, code, logic, and guesswork.

Fast forward to late 2022, when OpenAI unveils ChatGPT—a simple, intuitive interface for interacting with a large language model called GPT-3 (which stands for Generative Pre-Trained Transformer). Restricted by some parameters set at a high level so it wouldn’t dip too far into weird, creepy, or inappropriate territory, this LLM took the world by storm by responding to whatever you asked or said with uncanny understanding and what seemed like creativity.

This represented mass adoption of the largest widely-adopted leap forward in natural language processing (NLP) since Siri took iPhones by storm, with a widely available interaction model that almost anyone who has used a computer can understand.

Cool? Oh, hell yes.

Scary? Definitely.

The potential rewards come with a ton of risks. How much easier is it to scam someone when you can craft your requests to tug on their heartstrings because you know what’s important to them? Humans can do this exceptionally well already, so it’s nothing new... but they can’t do it at scale on their own. With a competent large language model, fed enough of someone’s personal data that it could know how to tailor its approach, a scammer could reach many more victims at once.

There’s also the elephant in the room, the question of the training data. Sure, a large language model is just predicting—but based on what precedent? If the information it’s trained on is saturated with human biases—and it is, because most of what the popular LLMs are based on is scraped from the internet—then how can we trust what it produces not to reinforce harms already caused by how people act on those biases?

And don’t worry, that’s just the start of a very long list of issues that tech activists, watchdogs, and scholars have raised.

So while LLMs are, without a doubt, a massive jump in technological capability, that capability is by no means unflawed. Like any technology with awesome power, it also has the potential to be awful when turned to the wrong purpose.

LLMs are connection-making tools

At its best, having access to an LLM is like having on speed dial a friend whose superpower is connecting the dots. Alternatively, imagine if you could ask a very well-read research librarian to connect seemingly unrelated subjects. If experts had already connected those particular dots, the librarian would likely know about it, and be able to cite that source; if no experts had tackled that particular topic, the librarian’s extensive knowledge would make them good guessers about what might happen, at least insofar as a person who had absorbed all of that knowledge would be a good guesser.

It’s like being able to ask Wikipedia to theorize or make up fiction about the