An Idiot's Guide to Line Editing

Chapter Eighteen -- Beep boop I'm a robot

Now it's time to talk about The Future. (If you don't know the difference, 'the future' is where you decide what socks you're going to wear tomorrow and 'The Future' is where you ride a cyborg centaur into the sunset of a supernova. You pronounce it about an octave lower with lots of gravitas and echo.) I've been putting this off because it terrifies me, but you don't pay me just to talk about nouns, so, let's go.

All of the techniques we've been describing have been about painstakingly going through your text, word by word. But living, as we do, in a time which is just starting to creep from the future onto The Future, there are some new tools you can use. And the one that you have access to is ChatGPT.

Now look, this field is moving ridiculously quickly right now. So this is probably already out of date. But hopefully some of the key concepts will be true when you, oh person of The Future, read this.

First, some science. Today, this is how chatbots work. The people who make them take a huge quantity of text (called a 'corpus'). Using all this text, they produce a thing which, given a few words, will suggest what word is most likely to come next. So when you type a question in, the bot will start with that as its text. It will figure out the first word that's most likely to come after your question, and then look at your question plus the word it produced to make the second word, and so on.

Note that I used the hand-wavy phrase 'given a few words'. The number of words used by a chatbot is pretty variable, but in practice this means bots will 'forget' stuff you told them a while ago. That means they have a short attention span, even shorter than your humble fleshy human scribe here. As a result, they're no good for development editing.

And of course, they're only as good as the corpus they're trained on. Most of today's chatbots are trained on the public internet. This is not a brilliant source. In fact, the internet in generally awful and so you pick up all sorts of horrible biases; but we have an additional problem which is that it's not trained on good writing. Yes, 'My Immortal' is in there but modern copyrighted trad published fiction is not.

But, it's a tool like any other, and it would behoove you to understand it. To that end, I've put some phrases from this book into ChatGPT to edit, and I'm going to talk about what it gets right and what it gets very, very wrong. Remember this?

The hauler was a mile from its bow to its stern, a behemoth of metal and plastic orbiting a lonely star. Its AI was long dead, its crew departed, its cargo looted. It was filled with dust and memories and not much else. It was gloomy and forgotten.

Jose's skimmer was a beautiful little thing, fast and graceful. It was the closest thing he had to a lover. It was barely ticking over as he made a first pass across the great ship.

Yes, of course you remember it, we just looked at it. It's good because it's got some nice problems in it: subjective adjectives, overuse of 'was', and if we're going to be honest, that second paragraph is a clumsy jump from the first. Well, I fired a ChatGPT session up, and asked for an edit. I didn't tell it what to look for. This is what I got.

The hauler stretched a mile from bow to stern, a colossal amalgamation of metal and plastic encircling a solitary star. Its AI had long perished, its crew dispersed, and its cargo plundered. Within its confines, only dust and fading memories remained. The vessel stood as a somber, forsaken relic.

Jose's skimmer, a captivating and nimble craft, danced gracefully across the void. It was his closest semblance to a lover, an object of affection. As he embarked on his initial sweep around the colossal ship, the skimmer hummed softly, barely ticking over.