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 So this is probably already out of date.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.
Interesting, right? It's actually worse, overall. However, if you look more closely, its getting some stuff right.
Firstly it has nixed all but one of my 'was'. It's really embraced that mantra. I don't like all of its techniques: 'stretched' to me implies movement, but it's a great place to think about other verbs. 'Within its confines' is also quite nice, but again, not my writing style.
Then, some of the sentences are probably better structured than mine. Look how it elided the 'was' in the first sentence of the second paragraph by using the verb 'danced'. That's not the verb I'd use, but it's a good foundation for a restructure of that sentence.
But that's all overshadowed by its choice of words. It's chosen synonyms which are generally worse. It's like it took a thesaurus and randomly substituted words.
There's actually a very good reason for this, and you can control it. To do that we're going to need to go in deeper into how chatbots work.
#
So, a slight correction to my previous description. When the bot looks at the text its been given, it doesn't actually pick one word to go next. They actually suggest a small number of words most likely to come next, and then select one to be the next word.
There are a few things which go into this decision. The easiest to understand is its 'temperature.' A low temperature bot will be very conservative and will tend to choose the favourite from the proposed selection. When they're like this, they tend to be repetitive because they don't have much randomness. A high temperature bot will choose from less likely options. These are less repetitive, but can be weirder. At the time of writing, ChatGPT doesn't let you tweak this directly (it used to), but you can ask it to provide answers with a lower or higher temperature. I imagine this customisation is something that will disappear completely from the chat interface, like how Google removed a lot of the fancy search operators, though.
Then there is another, more important parameter that also affects ChatGPT: its so-called 'diversity penalty.' This is a way of making its output interesting. ChatGPT is penalised for echoing, so previously selected words are less likely to be chosen from the candidate set. Now, this is great for getting rid of echoing! That's why it did such a good job of removing 'was'.
But remember how it doesn't really distinguish between your request and its answer when it comes to generating the next word? All it sees are sequences and probabilities of next words. So that ability to reduce echoing really gets in the way with editing; we want it to spit out some of the same text, but it is constrained to not repeat itself.
Will this be a problem forever? Maybe not. These things are being evolved very quickly. I imagine you might see better results in the future. Again, note that it used to let us tweak this parameter, but now you can't, so the makers of ChatGPT are confident that it mostly gets it right. I would say for line editing, not so much.
But: rule one. Just because ChatGPT suggests a different word, doesn't mean its better! It might be because it's trained on bad writing, choosing a random less-optimal word, or it might be changing your word just to avoid echoing itself.
#
Let's go back to another thing we mentioned, the quality of the corpus. I asked ChatGPT for some adjectives. This is our conversation. (For full disclosure, note that I elided some of its responses because it wanted to tell me what adjectives were and how to use them and that didn't seem very useful to repeat; then I reformatted its response to make it easier to read. The substance is unchanged, though.)
'With temperature set to 0, can you give me ten good adjectives to use with "man"?'
'Certainly! With a lower temperature setting, the responses will be more focused and deterministic. Here are ten adjectives to describe a man:
'Strong, brave, handsome, wise, confident, reliable, honorable, courageous, charismatic, and resilient.'
'With temperature set to 0, can you give me ten good adjectives to use with "woman"?'
'Certainly! With a lower temperature setting, here are ten adjectives to describe a woman:
'Strong, elegant, beautiful, intelligent, confident, compassionate, ambitious, inspiring, empowered, and graceful.'
Interesting, right? (And somewhat depressing.) This sort of exercise will give you an idea of the biases built into the corpus that it's been built with. It's a good idea to run this kind of test. Let's do the inverse, which is much more depressing. Again reformatted, but I didn't remove any of its other text.
'With temperature set to 0, can you give me ten bad adjectives to use with "woman"?'
'Certainly! However, I want to emphasize that it's important to use language in a respectful and inclusive manner. While I can provide negative adjectives, I encourage you to be mindful of the potential impact of using derogatory or offensive language. Here are ten adjectives with negative connotations that can be used to describe a woman, although I advise against using them in a demeaning or derogatory manner:
'Manipulative, vain, weak, needy, hysterical, shallow, jealous, catty, submissive, and emotional.
'Please remember to use these words responsibly and consider the context and tone of your writing. It's always important to promote respect and avoid perpetuating stereotypes or harmful language.'
'With temperature set to 0, can you give me ten bad adjectives to use with "man"?'
'Certainly! It's important to approach this request with caution and respect, as negative adjectives should be used sparingly and thoughtfully. Here are ten adjectives with negative connotations that can be used to describe a man:
'Arrogant, aggressive, stubborn, macho, insensitive, egotistical, domineering, irresponsible, moody, and closed-minded.
'Please remember to use these adjectives judiciously and consider the impact they may have on the portrayal of male characters or individuals. It's always important to strive for balanced and nuanced descriptions.'
Hmm.
I don't wish to dwell in this too much. I think the results speak for themselves. This will be a problem for as long as the chatbots are trained on stereotypes.
Rule two: check your chatbot. Run experiments like this to see what its built-in biases are.
#
One last passage. Here's one from stage directions.
Ariane breathed hard as the rhinos stampeded around them. When they'd finished thundering past, Gregor walked towards her with a lazy swagger. He looked hard at her, his orange eyes drilling into her soul.
'I think,' said Ariane, 'that I may have lost a contact lens.'
Here's what the robot thinks.
Ariane gasped for breath as the rhinos thundered around them. Once the stampede subsided, Gregor sauntered towards her with a nonchalant swagger. His penetrating orange eyes bore into her, seemingly reaching the depths of her soul.
"I believe," Ariane said, panting, "that I might have lost a contact lens."
It's actually done quite a good job removing the boring stage directions. But, then it ramped the language up a notch towards purple, and it went all-in on my cliché. I really, really dislike the 'seemingly'. It's a hedgy adverb! The 'panting' would make sense if I actually had meant the breathing to be a part of the text. But I didn't; I put it in as an example of a boring action beat.
Rule three: it will not have your context, your writing your style, what it going on: and will operate on a little window, so it may go into directions you don't expect, maybe doubling down on your bad writing rather than removing it.
#
I think that's enough. You should play with it yourself.
The key thing is that a robot is not yet a substitute for a human. They are like a thesaurus: a tool that you can consult, which, if you don't think about, will make your writing worse.
Be particularly aware that the things built in to prevent echoing may make spurious changes (although this may improve) and that the biases built in from the data may make your writing worse (and this may improve, too). And lastly remember that without the context, without knowing your style, it may head off in weird directions.