Home • Articles • Blog • Philosophy • News

June 10, 2024. 08:32 PM

Introduction

ChatGPT is a very interesting tool that writers should not neglect. But, to be actually useful, it needs to be used right.

For starters, the prose that ChatGPT outputs looks good at first sight. But, as you send hundreds of prompts, and read hundreds of answers, you soon notice that, regardless of the prompt, the output is repetitive and formulaic. Since the technology is still young, people may read answers and not see the repetition. But, with time, as users gain more experience, they will all be able to notice the patterns.

So, I would say that using the raw output of ChatGPT as part of your text is a terrible idea. Not because it’s lazy. I honestly don’t care about that. Any author should find the path of least resistance to write, in order to become better, faster, capable of supplying the growing demand of readers.

The true reason is: the style that ChatGPT delivers is not close to any kind of human writing—not even close. There will be your text, with your human voice. And, suddenly, whole paragraphs of ugly, generic, monotone prose that the chat generated. The audience will immediately feel cheated.

After all, readers bought a book to read interesting, original stuff. If they wanted to read the ramblings of a generative AI, they would have simply logged into ChatGPT, or some competitor, and sent a prompt.

But, if we don’t use ChatGPT to write for us, how can we use it to make ourselves better writers?

I’ll be listing a few ideas that I use every day:

1. Use ChatGPT to do research

Whatever you are writing, research is fundamental. For example, I always write adventure stories set in the past, and to me, research involves going into historical books and treatises, reading old manuals, and collecting data in museums.

I live in a rather “culture-less” place in Brazil, so I don’t have a lot of access to museums and libraries. The only way I can perform research of remote topics, such as Europe in the 18th and 19th Centuries, is by reading books and accessing information online.

ChatGPT is like a super-powerful teacher who knows everything by heart. It may hallucinate sometimes and output wrong information, but in general, it will work if you ask it to show you a direction.

For example, you could prompt the following:

ME: Give me a list of 10 books that are primary sources on maritime life in 18th Century England.

CHATGPT: Here is a list of 10 primary sources on maritime life in 18th Century England:

1. “A Voyage to the South Sea” by William Bligh
2. “A Voyage Round the World” by George Anson
3. “Captain Cook’s Journal During His First Voyage Round the World” by James Cook
4. “The Life of Captain James Cook” by Andrew Kippis
5. “A General History of the Pyrates” by Charles Johnson
6. “A Narrative of the Mutiny, on Board His Majesty’s Ship Bounty” by William Bligh
7. “Sailor’s Life: A Sea Songster” (various authors, collection of sea songs)
8. “The Seaman’s Vade-Mecum and Defensive War by Sea” by William Mountaine
9. “The Memoirs of Captain George Carleton” by George Carleton
10. “Naval Chronicle” (various authors, periodical)

Then, from these recommendations, you could at least start your research. Many of these books can be found on Archive, an excellent website to find information.

I searched for the first recommendation, by William Bligh, and quickly found a copy on Archive. I even found a LibreVox audiobook version. The edition I found was an HTML file with reference images showing the plants of the boat—an invaluable asset for any fiction writer who sets a story, a sequence, or even a mere scene on such a type of vessel.

For this, I wouldn’t recommend asking ChatGPT to summarize the text for you, though. I see this as a growing trend with generative AI, but the output always seems to lack important information. There’s also a lot of hallucination.

I think the generative AI lacks the human sensibility and experience to know what information matters the most. When reading these accounts, an author may be looking for clues on how life on a boat was. But, for the AI, it feels that the text is about the dynamics of mutiny (in the example above). It takes a human to judge what data on the text is relevant for each subject.

2. Get rare, highly specific words

One of my favorite ways to use ChatGPT is to find very specific, often technical terminology to increase the credibility of a story, and to enrich the text.

For example, if I’m describing a landscape, my vocabulary is only as big as daily life allows it to be. I’m not a geologist or a biologist. I don’t have the specific names of landforms and biomes on top of my head. Neither does most of the audience. But, if I use these terms, the text will feel richer. And, audiences love to learn new things. The reader will understand the term from context, but if he doesn’t, he can look it up.

ME: List me 10 specific types of landforms found in coastal lines.

CHATGPT: Here are ten specific types of landforms commonly found along coastlines:

1. Beaches: Sandy or pebbly shores, often found where the land meets the ocean.
2. Cliffs: Steep rock faces, often formed by the erosive action of waves.
3. Dunes: Hills of sand formed by wind and wave action, commonly found behind beaches.
4. Estuaries: Bodies of water where rivers meet the sea, characterized by brackish water.
5. Headlands: Narrow pieces of land that project from the coastline into the sea.
6. Bays: Recessed, coastal bodies of water that directly connect to a larger main body of water, such as an ocean or a lake.
7. Tidal Flats: Flat, muddy areas that are exposed at low tide and submerged at high tide.
8. Lagoons: Shallow bodies of water separated from a larger body of water by a barrier such as a sandbar, coral reef, or barrier island.
9. Fjords: Deep, narrow, and elongated sea or lake drains with steep land on three sides, created by glacial activity.
10. Spits: Narrow points of land projecting into the sea, formed by the deposition of sediment by longshore drift.

As you can see, some of the listed elements were familiar, like beaches, cliffs, and estuaries. But, words like headland, tidal flat, lagoon, and spit are not part of our daily vocabulary. ChatGPT helped me learn these words, and now, I can use them in my landscape description to make it more specific and richer.

3. Help with grammar and style

ChatGPT and other generative AIs can serve as a perfect consultant for grammar questions. It’s hard to know “all of grammar” from the top of my head. As such, I constantly make highly specific questions to the Chat, and the answer saves me a lengthy consultation to a bulky grammar book.

For example:

ME: In which situations could I omit the word “to” from an infinitive verb in the middle of a sentence?

CHATGPT: In English, the word “to” is typically used with the infinitive form of a verb, but there are situations where “to” can be omitted. Here are some common cases:

• After Modal Verbs: The word “to” is omitted after modal verbs such as can, could, may, might, must, shall, should, will, and would.

  • Example: “She can swim.”

• After Certain Verbs of Perception: When using verbs like see, hear, feel, watch, and notice, “to” is often omitted.

  • Example: “I saw her leave.”

• After Certain Causative Verbs: Verbs like make, let, and have often omit “to”.

  • Example: “She made him do it.”

• In Infinitive Phrases After Help: The verb “help” can be followed by an infinitive with or without “to”.

  • Example: “She helped me (to) clean the room.”

• After Expressions with “Why”: When forming questions with “why,” the infinitive “to” is omitted.

• Example: “Why wait?”

• After Conjunctions or Relative Pronouns in Shortened Phrases: In some cases, especially in informal speech, “to” can be omitted for brevity.

• Example: “All I did was (to) ask.”

This information would take at least a quarter of an hour to find in a large grammar book, especially with all these examples. But, ChatGPT gave me a quick shorthand in only a few seconds. With less time wasted consulting books, I’m able to write more.

4. Find very specific terms

This is similar to the second topic, but it involves words that we don’t usually use. For example:

ME: What verbs describe the walking style of slow zombies like in The Walking Dead?

CHATGPT: Several verbs can describe the walking style of slow zombies like those in The Walking Dead. Here are a few:

• Shuffle: Slow, dragging steps, often with a lack of coordination.
• Lumber: Heavy, slow, and ungainly steps, as if the body is difficult to control.
• Stagger: Unsteady, swaying steps, indicating a lack of balance or coordination.
• Trudge: Slow, laborious steps, suggesting a sense of exhaustion or difficulty.
• Stumble: Unsteady steps with a risk of tripping or falling, often due to weakness or disorientation.

With this answer, I don’t have to search for a thesaurus and search for synonyms of “walk,” and then, consult the dictionary for every obscure word that I don’t know, until I find that trudge is a verb that defines what I’m looking for.

5. Grammar revision

Generative AI is not good at seeing spelling and grammar mistakes like a dedicated app, such as LanguageTool. But, it is capable of reading a sentence and transforming it according to instructions.

For example:

ME: Break this complicated sentence into smaller ones, easier to understand:

“It is well understood that a poorly fed army, when pushed to the limits of its endurance, without hope for a bountiful reward and a secure return home, shall eventually rebel violently, not because of a major defeat or disaster, but because of the smallest of causes, like a squabble between soldiers that escalate like the spark that blows up the powder keg.”

CHATGPT: Certainly! Here’s a breakdown:

1. It’s widely known that an army that isn’t given enough food and is pushed too hard, with no promise of good rewards or a safe journey home, will eventually rebel violently.
2. This rebellion doesn’t necessarily happen because of a big loss or disaster, but rather because of small things, like arguments among soldiers that quickly escalate, similar to a small spark igniting a powder keg.

As you can see, the AI kept most of the original words, and the original idea. Grammar revision tools like LanguageTool and native tools of word processors will mark the sentence as wrong (too lengthy), but will fail to offer a solution. It takes a machine with some natural language capability to do that.

Conclusions

There are other examples that I could give, but with these, you already have a start. As a writer, there’s no shame in using a tool like ChatGPT to increase your productivity. It may not be good enough to write for you, but it will save hours, maybe even days, of lengthy research, consultation to dusty thesaurus, dictionaries, and grammar books.

It will also give you words that only specialists know. A doctor knows the names of body parts, and a geologist knows the names of landforms. A biologist knows the names of branches of the tree of life, and body parts of plants and insects. An astronomer knows the terms of specific stellar bodies and constellations, and a mechanic knows the names of gears and cogs inside your car. But, with ChatGPT, you, a writer, will be able to know all those things in the blink of an eye.

Think about that.