<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by David J. Germain  on Medium]]></title>
        <description><![CDATA[Stories by David J. Germain  on Medium]]></description>
        <link>https://medium.com/@dave.germain.79?source=rss-812eebb44a16------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*oHruj2MLj44XcZk0BSeaGw.png</url>
            <title>Stories by David J. Germain  on Medium</title>
            <link>https://medium.com/@dave.germain.79?source=rss-812eebb44a16------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 08:15:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dave.germain.79/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Echoes, by ChatGPT]]></title>
            <link>https://medium.com/@dave.germain.79/echoes-by-chatgpt-1921e04d649b?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/1921e04d649b</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[science-fiction]]></category>
            <category><![CDATA[midjourney]]></category>
            <category><![CDATA[fiction-writing]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Tue, 21 Feb 2023 06:50:46 GMT</pubDate>
            <atom:updated>2023-02-21T06:50:46.377Z</atom:updated>
            <content:encoded><![CDATA[<p>A Sci-Fi Short Story Written with ChatGPT</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Las2EPmQoBv4U2GSGJC1hQ.png" /></figure><h3>Notes about the Story</h3><p>This short story was written with ChatGPT. I wrote the plot using a “choose-your-own-adventure” style, asking the AI to generate the next plot point, then regenerating until I found something interesting.</p><p>Once I had a plot outline, I used several methods that I’d developed earlier to write the <a href="https://medium.com/@dave.germain.79/writing-dialog-with-chatgpt-bd8024a69eb3">dialog</a>, the <a href="https://medium.com/@dave.germain.79/descriptive-writing-with-chatgpt-c1ebb05d3bee">description</a>, and the <a href="https://medium.com/@dave.germain.79/writing-physicality-with-chatgpt-b31bc1b570cd">physicality </a>of the characters.</p><p>Finally, I edited the product — mostly by cutting excess words — and then created the art with Midjourney V4.</p><p>Enjoy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nuclST0nq1S-Zdpnq4ZFag.png" /></figure><h3>The Story</h3><p>“It’s tempting to feel inadequate,” said Excavation Officer Fairfax. “Primitive, even, in the face of the advanced civilization that created this vessel.” The beam of light from his equipment traced over a massive obelisk, weaving in and out of its symmetrical shapes and biomechanical features. “But we are no mere blip in the long history of this forgotten place. We are here to begin a renaissance.”</p><p>“Sure,” said Tony.</p><p>“Yup,” said Bob, staring up at the towering walls that dwarfed the three human explorers. His beam slowly scanned the labyrinth of curves and lines and almost human designs. The air shimmered with particulate matter that reflected and refracted the light. “Renaissance. What could possibly go wrong?”</p><p>The Shard was 400 kilometers end to end, in slow rotation around an orange gas giant. In theory, it had once been a ringworld as large as a small planetoid, but now only a fifth remained of the once mighty beast, a mechanical sliver of broken black metal.</p><p>Human salvage ships buzzed across its surface, engines leaving trails of fiery exhaust that disappear into the shadows of the great spires that covered its surface, a forest of metal filigree that towered over the landscape and reflected the gas giant’s light like a kaleidoscope of fractured glass.</p><p>On the eighty-ninth level of Uncharted Spire Alpha-Seven, Bob and Tony were arguing.</p><p>“Hold up,” Bob growled. “I don’t remember it that way.”</p><p>Tony’s hands balled into fists. “I’m telling ya, Bob,” he spat. “I went in three chambers, you only went in two.”</p><p>Bob shook his head, his skin growing hot with frustration. “I went in three too, Tony.”</p><p>Tony’s breath was heavy with anger. “No way,” he said firmly. “I counted.”</p><p>Bob’s voice was just as firm. “Not what I remember.”</p><p>The tension between them was palpable, their breathing heavy in the intense, suffocating darkness of the room creating a chorus of whispers that bounced off the walls. Light twisted and turned, an ethereal glow playing across the walls, casting long shadows and exposing the intricacies stretching to the ceiling, tracing the curves and lines of the engravings.</p><p>“What’s the problem, gentlemen?” Officer Fairfax’s voice boomed through the dark room, creating a chorus of whispers that bounced off the walls.</p><p>Tony pointed a finger at Bob, his voice rising. “He doesn’t want to go in.”</p><p>“He doesn’t neither.”</p><p>The officer’s eyes narrowed. “Listen up, you two. I’m tired of this constant bickering,” he lectured. “It’s unprofessional and it’s dangerous. We’re supposed to be a team.”</p><p>Tony grew defensive, his hands on his hips. “Hey, we’re just trying to figure out who’s turn it is,” he said, his voice filled with frustration.</p><p>“I shouldn’t have to keep track,” Fairfax’s eyes blazed. “You two should be able to handle this yourselves. It’s basic ethics, gentlemen.”</p><p>“Ethics?” Tony asked, his brow furrowed.</p><p>Bob echoed Tony’s skepticism, “Ethics…?”</p><p>“Yes, ethics,” Fairfax said, his voice ringing with authority. “And bravery. I enter these chambers with pride, not fear. I expect the same from my team.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7DBC3DTUfHch_3jI-n8WVw.png" /></figure><p>Tony glanced into the dark passageway. His light beam played across the curving walls, then was lost in the shadows, the air too thick with particulate matter, except where the echoes of his light play across xeno-symbolic surfaces. Strange groans emanated from the depths. The nervous energy in the room was palpable. “You sure about this, sir?”</p><p>“Yeah,” grumbled Bob. “I don’t like the look of this place.”</p><p>“Nonsense,” Fairfax said, his voice strong and unwavering. “I’ve been in dozens of chambers just like this one. Nothing to worry about.”</p><p>Bob’s eyes narrowed. “Heard stories about these chambers.”</p><p>Tony took a step closer to the other excavator. “Yeah?”</p><p>Bob leaned in, his voice low. “People disappearing and all that,” he said. “This guy went into one of these chambers, came out with false memories of a family he never had. Wife, kids. Knew their names and everything. No one could convince him he were never married.”</p><p>“That’s just more rumors, Bob,” Fairfax said, waving his hand in the air.</p><p>But Bob was insistent. “I’m telling you, it’s true,” he said, his voice growing more urgent.</p><p>Tony’s brow furrowed in confusion. “What happened to him?”</p><p>“Eventually, he had to accept the truth,” Bob shook his head, his voice filled with sadness. “But for him, those memories were real. He felt like he’d lost a real family.”</p><p>“That’s just a tall tale,” the officer soothed, a hint of frustration in his voice. “Nothing to worry about.”</p><p>Tony’s face was etched with doubt. “I don’t need no fake family weighin’ me down.”</p><p>“Stories, that’s all they are. Just a bunch of rumors.”</p><p>“Feel a lot better if you went first, sir,” Tony said, his eyes fixed on the Officer. Bob nodded in agreement.</p><p>“What’s the matter with you two?” he asked, his voice rising in anger. “This is basic excavation work.” His hands balled into fists, then relaxed. “Very well, gents. In we go.”</p><p>Tony, voice tinged with resignation, echoed. “Alright, sir.”</p><p>Bob sighed, a heavy gust of breath escaping him.</p><p>The Officer smiled, determination shining in his eyes. “That’s more like it. Let’s get to work.” A firm hand on each of their shoulders, he reminded them, “And remember, gents. We’re in this together.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gk97975sIyexrA1MUw4wRQ.png" /></figure><p>Following Officer Fairfax and Bob, Tony ventured into the dark passageway. The dense, murky air made it hard to see beyond the light’s reach. The beams of light from the excavators’ equipment seemed to stretch and twist, drawn towards the walls, as if it was being absorbed by the vessel. The thick clouds of matter clogged the air, muffling sounds, and unfamiliar echoes and groans rumbled from the shadows. In fact, light and sound had taken on an altogether different quality.</p><p>Tony glanced up at the towering walls looming overhead. The darkness reached for him, threatening to consume. Fear, near-panic, and the raw, intense sensation of insignificance were almost suffocating.</p><p>And just as he was struggling to pull breath, they stepped out and through, into the light, into a long passageway that stretched in both directions, as far as Tony could see. Winding and intricate, curved and twisted in a never-ending dance, it was pocked with smaller dark passageways that lead into the unknown, like a vast arterial system. Humans had strung up tiny lights along the great channel, streetlamps in an ancient city, which illuminated small patches of the alien channel and cast elongated shadows across the walls.</p><p>The particulate matter in the air moved in slow currents, pooling and twisting around them. Echoes of the light and sound were distorted, as if they are being played back in slow motion, and Tony was overcome with a sense of both familiarity and unease. He was just turning to Bob when he saw that there was a fourth man in the passageway with them.</p><p>“Who the heck is this guy?” he stammered, trying to gather his jumbled thoughts. He stared at the figure in the excavation suit. “Who are you!?” a sense of shock washing over him. Weren’t they alone on the alien craft?</p><p>The figure slowly turned to face them, revealing a face that was both familiar and strange. “Don’t recognize me?” His eyes twinkled with amusement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OQxzIiy4WCtmmWgojX2skA.png" /></figure><p>Tony grabbed at the stranger’s helmet, grappling with him as they tumbled to the ground. “Who are you?!” he screamed. Their arms were locked in a fierce struggle, pushing and pulling and fighting for control.</p><p>Excavation suits rubbed against each other with a harsh grating noise. Thuds of metal on metal, footfalls echoed off the walls, ringing through the passage as Bob and Fairfax strained to separate them, yanking with all their might.</p><p>With a heave, the two were torn apart, Tony collapsing as his hands shook from the intensity of the struggle. The chamber was silent except for the ragged breathing of the men, the only sound in the darkness.</p><p>The mysterious man pushed himself up on one elbow, then slowly stood, and picked up a comm tablet. “Well, I’m supposed to ask about your mood,” he chuckled, “but…”</p><p>“Sir, who are you?” Fairfax asked, tone unyielding.</p><p>“I’m Hawthorne. Your superior officer?” The wrinkles on his skin deepened as he squinted at them. “What did you find inside?”</p><p>“What do you mean, ‘inside’?” The officer’s confused expression was evident. “We just walked in.”</p><p>“Nothing? That’s strange. I thought for sure you’d found something.”</p><p>“Look,” Tony was skeptical, “We don’t know you and we don’t have anything to give you.”</p><p>“We’re here to do our job, not answer to some stranger,” the Officer’s voice was firm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IK1CgwCzZkSRciru1Siepg.png" /></figure><p>“Suit yourselves,” Hawthorne shrugged, his eyes calm and patient. “We can wait for your memory to return.”</p><p>Tony was doubtful. “Lost our memory? How’s that possible?”</p><p>Fairfax rubbed his chin. “And you remember us?”</p><p>Hawthorne ‘s hands were relaxed at his sides. “You were only in the chamber for several minutes. You must have lost your memory.”</p><p>Bob frowned, his breath shallow — “Ain’t this the chamber?”</p><p>Hawthorne pointed behind them. “You just exited the chamber.”</p><p>“What the hell are you talkin’ about?” Tony asked. “I remember walkin’ in here.”</p><p>The Officer’s hair was ruffled as he ran a hand through it. “I have a clear memory as well.”</p><p>“So what do we do now?” Bob asked, his voice uncertain.</p><p>“Alright,” Hawthorne’s voice was gentle, “just a few more questions. And your sense of time? Any changes there?”</p><p>Tony shook his head, “Time feels the same, man.”</p><p>Bob agreed, “Yeah, nothing’s different.”</p><p>Fairfax nodded, his eyes focused, “No, everything remains consistent.” He turned to Bob and Tony, his voice firm, “Gentlemen, we must stay focused. We need to figure out what happened and how to reverse it.”</p><p>Hawthorne stood with calm resolve, his eyes piercing through the tense air like a knife. “Your memory should return eventually,” he spoke with a steady voice, “Just one more chamber, gents. Stay sharp.”</p><p>Tony shifted uncomfortably, his skin slick with sweat. “Why? Is it dangerous?” he asked, his voice trembling.</p><p>Bob’s breathing was labored, his chest heaving. “I ain’t goin’ first. I did it last time,” he said, his eyes darting around the room.</p><p>Tony’s hands were clenched into fists, his knuckles white. “No way, man. I went first yesterday,” he replied, his voice tight with fear.</p><p>Hawthorne’s gaze settled upon them both. “Gentlemen, it doesn’t matter who goes first,” he spoke in a low, steady voice, “The only important thing is that you enter the chamber.”</p><p>The Officer stood tall, his spine straight and unyielding. “Gentlemen, we must not falter,” he declared, his voice ringing with authority, “We are part of a great renaissance. Now, let us proceed to the next chamber.”</p><p>Bob’s eyes were wide and panicked. “Yeah, but I heard stories ‘bout people losing their minds in these chambers.”</p><p>Beyond the artery walls, beyond the eighty-ninth level of Uncharted Spire Alpha-Seven, human ships buzzed within the shadows of an endless forest — metal spires that covered the twisting remains of the shattered ringworld. A dark shard against the light of the nearby gas giant, its architects lost to deep time, their purpose long forgotten.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YMVmovtQRPdx4aB1lMHJcg.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1921e04d649b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Writing Dialog with ChatGPT]]></title>
            <link>https://medium.com/@dave.germain.79/writing-dialog-with-chatgpt-bd8024a69eb3?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd8024a69eb3</guid>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[elmore-leonard]]></category>
            <category><![CDATA[dialogue]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Sat, 18 Feb 2023 14:13:00 GMT</pubDate>
            <atom:updated>2023-02-18T14:13:00.570Z</atom:updated>
            <content:encoded><![CDATA[<p>Using Elmore Leonard to Inspire Characterization</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4tMroa9m7IYU5r9nrue2ew.png" /><figcaption>“Yellow Chamber” by David J. Germain with Midjourney V4</figcaption></figure><h3>Introduction</h3><p>In my two previous articles, I focused on using ChatGPT for <a href="https://medium.com/@dave.germain.79/descriptive-writing-with-chatgpt-c1ebb05d3bee">descriptive fiction writing (Iain M. Banks)</a>, as well as <a href="https://medium.com/@dave.germain.79/writing-physicality-with-chatgpt-b31bc1b570cd">physical writing (Joe Abercrombie)</a>. In this piece, I will shift focus to characterization through dialog.</p><p>Rather than use ChatGPT’s standard dialog, or tell it to write “in the style of” another writer, I’ll analyze some of Elmore Leonard’s excellent dialog using ChatGPT, then use that analysis to inspire better “colloquial characterization.”</p><h3>Typical ChatGPT Dialog</h3><p>Let’s use ChatGPT to write some vanilla dialog:</p><p><em>Two people, Karen and Jack, are locked in the trunk of a car. Jack’s friend is driving and Karen wants to get out. Karen is a police officer. Jack is a bank robber. Write dialog.</em></p><p>Here’s the horrendous result:</p><p><em>Karen: “Jack, we need to find a way to get out of here. I’m a police officer and I need to call for backup.”</em></p><p><em>Jack: “What do you want me to do about it? I’m just a bank robber, remember?”</em></p><p><em>Karen: “We need to work together. Is there any way to escape from the trunk of this car?”</em></p><p><em>Jack: “I think there’s a release lever inside the trunk. Let me take a look.”</em></p><p><em>Karen: “Be careful. If the driver realizes what we’re doing, they may stop the car.”</em></p><p><em>Jack: “Found it! Here, let me try to pull it.” (pulls lever)</em></p><p><em>Karen: “Great, it worked. Let’s get out of here and call for backup.”</em></p><p>ChatGPT, left to its own devices, is winning no Booker awards. It writes very on-the-nose dialog (“I’m a police officer…”) with no difference between the two voices.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kd2OUlbRTfni_xqfwWFqIA.png" /><figcaption>“Trunk of a Car #1” by David J. Germain with Midjourney V4</figcaption></figure><h3>Selection from Elmore Leonard</h3><p>Now let’s look at some Elmore Leonard dialog from his excellent novel, Out of Sight. At this point in the story, Karen Sisco, a law enforcement agent, is locked in the trunk of her car with Jack Foley, an escaped bank robber.</p><p><em>His voice in the dark, breathing on her, said, “You comfy?”</em></p><p><em>The con acting cool, nothing to lose. Karen was holding the Sig Sauer between her thighs, protecting it, her skirt hiked up around her hips.</em></p><p><em>She said, “If I could have a little more room.”</em></p><p><em>“There isn’t any.”</em></p><p><em>She wondered if she could get her feet against the front wall, push off hard and twist at the same time and shove the gun into him. Maybe. But then what? She said, “I’m not much of a hostage if no one knows I’m here.”</em></p><p><em>She felt his hand move over her shoulder and down her arm. “You aren’t a hostage, you’re my zoo-zoo, my treat after five months of servitude. Somebody pleasant and smells good for a change. I’m sorry if I smell like a sewer, it’s the muck I had to crawl through, all that decayed matter.”</em></p><p><em>She felt him moving, squirming around to get comfortable.</em></p><p><em>“You sure have a lot of shit in here. What’s all this stuff? Handcuffs, chains … What’s this can?”</em></p><p><em>“For your breath,” Karen said. “You could use it. Squirt some in your mouth.”</em></p><p><em>“You devil, it’s Mace, huh? What’ve you got here, a billy? Use it on poor unfortunate offenders … Where’s your gun, your pistol?”</em></p><p><em>-Out of Sight, Elmore Leonard</em></p><p>The dialog is punchy and full of character. What makes Leonard’s dialog so much better than vanilla ChatGPT?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1adHvoDg4Rku4WkouBK95w.png" /><figcaption>“Trunk of a Car #2” by David J. Germain with Midjourney V4</figcaption></figure><h3>Analyzing the Selection</h3><p>Let’s ask ChatGPT. I’ve analyzed dialog before and found that its difficult to get useful results from some analysis. This time, I started by asking ChatGPT for a general approach to analyzing dialog. Here’s the prompt:</p><p><em>What are some important things to consider when analyzing dialog in fiction?</em></p><p>It answered by listing a series of concepts, some of which I hadn’t considered before. For instance:</p><p><em>Register… the level of formality or informality in the language used…</em></p><p><em>Speech Acts…the functions of language, such as making requests, giving commands, making apologies…</em></p><p><em>Discourse Markers…words or phrases that signal the flow of conversation…</em></p><p>I then incorporated these into a large prompt that I run against several samples of Elmore Leonard dialog. Here’s the first prompt I used:</p><p><em>You are an expert in discourse analysis and natural language processing (NLP). You’ve been teaching dialog writing for over 20 years. Analyze [Sight Sample] with an emphasis on the dialog. Discuss dialog sentence length, parts of speech, and number of declarative vs questions. Analyze the dialog, with an emphasis on Word choice, Grammatical structures, Register, Speech Acts, Discourse Markers, and Prosody. Discuss how these differ from typical fiction dialog. Make sure to compare sentence fragments vs complete sentences, questions vs declaratives vs instructions, and dialog tags.</em></p><p>This was very useful because the responses were often organized into a list of topics. Since I ran the analysis against multiple samples, I used these lists to match up similar topics.</p><p>Here’s the second prompt, which gave me a second opinion on the same topic:</p><p><em>If you were to instruct an AI to write dialog in this way, what instructions would you give? Provide your answer as if you’re speaking directly to another AI.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hcPjV7VoXVghnK8mKH7oZQ.png" /><figcaption>“Space Explorers Talking” by David J. Germain with Midjourney V4</figcaption></figure><p>Using the output analysis, I came up with the following dialog paragraph:</p><p><em>Use an informal and </em><strong><em>colloquial</em></strong><em> register with a conversational tone.</em></p><p><em>Most sentences are short and concise, around 4–5 words in length. A few sentences may be 10–15 words, with 1–3 clauses. Use a balance of complete sentences and sentence fragments. Use simple grammatical structures, with mostly simple and compound sentences. Use a mix of declaratives, questions, and exclamations, with an emphasis on declaratives.</em></p><p><em>Use simple and straightforward vocabulary, with many </em><strong><em>colloquial</em></strong><em> and informal expressions that are commonly used in everyday speech. Make use of second person pronouns and contractions to create a sense of intimacy and informality. Avoid formal words and phrases and avoid technical jargon. Avoid polished, scripted dialogue. Avoid most adverbs and conjunctions. Include discourse markers to signal the transition from one utterance to the next and to indicate the speaker’s attitude towards what is being said.</em></p><p><em>Overall, the dialog is Informal, conversational, natural, and spontaneous, with a sense of immediacy.</em></p><p>One of the words that came up often in the analysis was “<strong>colloquial,” </strong>and this will factor into the next section.</p><h3>Using the Elmore Leonard Selection</h3><p>This time, I wasn’t really sure what my dialog was going to be about, so I gave the AI a fairly open-ended initial prompt. Ultimately, I want to use this dialog to build a sci-fi short story, but at this point, I don’t want the “sci-fi” element to misalign the AI, so I don’t mention it:</p><p><strong><em>Tony is a young Italian-American from New York. Bob is in his late fifties and grew up on a farm in south Texas.</em></strong><em> Two workers, Tony and Bob, argue about who’s turn it is to enter the next chamber. Tony insists that he entered more chambers yesterday. Bob doesn’t remember it that way. Write a dialog between the two. Use an informal and colloquial register with a conversational tone…Overall, the dialog is Informal, conversational, natural, and spontaneous, with a sense of immediacy.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pa1eOTCBnW8dyT4pK2l0OQ.png" /><figcaption>“Opening the Chamber Door” by David J. Germain with Midjourney V4</figcaption></figure><p>I included my entire Dialog Instruction paragraph at the end of the prompt.</p><p>Notice that I gave the two characters <strong>demographic information</strong> that might not apply to a sci-fi story. I’m trying to supply the AI with a basis for its <strong>colloquial</strong> instruction.</p><p>I improvised a fair bit, asking the AI to rewrite bits and pieces. When the AI introduced a third character, Boss. Because Bob and Boss sound too similar, I changed his name to Officer, and I gave him some demographic information as well:</p><p><em>Officer is a young British officer, proud of his bravery. As he enters the chamber, the other two men follow. Officer lectures the men on their poor ethics and performance. Write Dialog. Use an informal and colloquial register with a conversational tone</em>…</p><p>At this point, I haven’t specified what the “chambers” will be or any of the sci-fi elements that might accompany the story. Instead, I allowed the AI to flush out the characters through their dialog.</p><h3>The Results</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PdvrUAzShA-vPQs-P1whmA.png" /><figcaption>“Station Window,” by David J. Germain with Midjourney V4</figcaption></figure><p>Here are the results of the first two dialog chunks:</p><p><em>Bob: Hold up.</em> <em>I don’t remember it that way.</em></p><p><em>Tony: I’m telling ya, Bob. I went in three chambers, you only went in two.</em></p><p><em>Bob: I went in three, Tony.</em></p><p><em>Tony: I’m telling you, it’s your turn.</em></p><p><em>Bob: I’m telling you, it’s not.</em></p><p><em>Tony: (frustrated) Look, I don’t want to argue about this. Just go in the damn chamber.</em></p><p><em>Bob: (defensive) Why should I? You go in.</em></p><p><em>Officer: (interrupting) What’s the problem, gentlemen?</em></p><p><em>Tony: (pointing at Bob) He doesn’t want to go in.</em></p><p><em>Bob: (pointing at Tony) He doesn’t either.</em></p><p><em>Officer: (lecturing) Listen up, you two. I’m tired of this constant bickering. It’s unprofessional and it’s dangerous. We’re supposed to be a team.</em></p><p><em>Tony: (defensive) Hey, we’re just trying to figure out who’s turn it is.</em></p><p><em>Officer: (firm) I shouldn’t have to keep track. You two should be able to handle this yourselves. It’s basic ethics, gentlemen.</em></p><p><em>Tony: (skeptical) Ethics?</em></p><p><em>Bob: (skeptical) Right…</em></p><p><em>Officer: (determined) Yes, ethics. And bravery. I expect better from you two. I enter these chambers with pride, not fear. I expect the same from my team.</em></p><h3>Conclusion</h3><p>This dialog analysis process taught me a few key takeaways.</p><p>First, asking the AI how it would analyze a type of writing was helpful. I could incorporate its analysis suggestions into my analysis prompt.</p><p>Second, if I ask the AI to write colloquial dialog, I need to provide demographic data, and regularly remind the AI of this colloquial information.</p><p>Third, certain voices are easier to differentiate. Bob and Tony quickly blended together, while Officer was easy to differentiate, even without dialog tags.</p><p>In my next article, I’ll provide another example of a ChatGPT short story as I finish this off. I’ll use Iain M. Banks descriptions, Joe Abercrombie physicality, and Elmore Leonard Dialog.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Dpjo1fSx7xia8EwNajMayw.png" /><figcaption>“Enter the Chamber” by David J. Germain with Midjourney V4</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd8024a69eb3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Writing Physicality with ChatGPT]]></title>
            <link>https://medium.com/@dave.germain.79/writing-physicality-with-chatgpt-b31bc1b570cd?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/b31bc1b570cd</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[joe-abercrombie]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[fiction-writing]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Fri, 03 Feb 2023 01:13:28 GMT</pubDate>
            <atom:updated>2023-02-03T01:13:28.614Z</atom:updated>
            <content:encoded><![CDATA[<p>Using Joe Abercrombie to Inspire ChatGPT</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-YO0K9WHq-4SuFpz0GeHtg.png" /><figcaption>“Flying Schoolgirl,” by David J. Germain with Midjourney V4</figcaption></figure><h3>Introduction</h3><p>(Note: if you’re looking for some quick tips to get started writing fiction with ChatGPT, click <a href="https://medium.com/@dave.germain.79/7-tips-for-writing-fiction-with-chatgpt-6d0e686879a8">here</a>.)</p><p>In <a href="https://medium.com/@dave.germain.79/descriptive-writing-with-chatgpt-c1ebb05d3bee">a pervious article</a>, I focused on using ChatGPT for descriptive fiction writing. In this piece, I will shift focus to characters’ physical reactions to their environments.</p><p>Rather than use ChatGPT’s standard descriptions, or tell it to write “in the style of” another writer, I’ll analyze some Joe Abercrombie’s great fiction using ChatGPT, then use that analysis to inspire better “twitching and flexing.”</p><h3>Twitch and Flex</h3><p>One way of thinking about fiction is as Motivation-Reaction Units (MRUs). Something happens that motivates a character, the character reacts, and that triggers a new motivation. Consider the following example:</p><p>Dave smells some waffles. He feels hungry. His stomach rumbles. He reflects on his body fat percentage. He acts, eating the waffles anyway.</p><p>Observation. Feeling. Physical Reflex. Mental reflection. Action.</p><p>Sometimes, this loop can take place over several months of a character’s life. They are plans — “should I attempt a diet?” But in the action/thriller/horror/erotica genres, these loops are often moments of time. The character sees a thing, experience an involuntary emotional/physical reaction without any plan at all.</p><p>I think of these as “twitching and flexing,” the raw physicality of fiction, and they are the topic of this article.</p><h3>Typical ChatGPT Physicality</h3><p>First, let’s take a look at ChatGPT’s direct version of a twitch/flex scene. Here’s the prompt:</p><p><em>Write a story. A frightened girl walks through an abandoned house. Suddenly, the floor beneath her breaks. She’s dangling from the broken floorboards. She struggles up and runs from the house. Use first person present tense.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yyxTHqwBEOcOCtqpsMQg3A.png" /><figcaption>“Derelict Stairwell,” by David J. Germain with Midjourney V4</figcaption></figure><p>Here’s the result:</p><p><em>I walk through the abandoned house, my heart pounding in my chest. The walls are peeling and the air is musty. I can hear strange creaks and groans coming from the floorboards beneath my feet. I try to ignore the fear that’s growing inside of me and continue on, determined to find a way out.</em></p><p><em>Suddenly, the floor beneath me gives way and I find myself dangling from the broken floorboards. I cry out in terror and struggle to pull myself up, but the wood is old and splintered, and my fingers slip. I feel a wave of panic wash over me.</em></p><p><em>But then I remember why I came here. I came here to find my sister, who disappeared into this house. I won’t give up now. I gather all my strength and try again. This time, I manage to pull myself up and scramble to my feet. I turn and run as fast as I can, not looking back until I reach the front door.</em></p><p><em>I burst out into the fresh air, gasping for breath. I look back at the house, still standing tall and ominous, and I feel a shiver run down my spine. But I know that I’ve escaped and that I’ll never come back here again. I’ve found the courage to face my fear, and I’ll never forget that feeling.</em></p><p>The results are surprisingly physical. She “shivers” and “scrambles” and “panic washes over her.” But it suffers the usual ChatGPT problems — it veers off (“to find my sister”), it uses similar sentence patterns, and vanilla vocabulary.</p><h3>Joe Abercrombie’s Writing</h3><p>Abercrombie’s writing is very physical, filled with objects in motion, and his characters are constantly twitching and flexing and gaping in shock as their worlds tip over and catch fire.</p><p>A fine example is Half a World. In the story, Brand and his companions are carrying their galley, the South Wind, up and over a land bridge. They’re yanking on ropes with sweaty hands, and then they lose control of the galley.</p><p><em>There was a flurry of wingbeats as a bird took to the sky and the South Wind lurched forward, one of the men on the other side shrieking as his rope tore across his shoulders and spun him around, knocking Fror sideways, the sudden weight dragging the rest of the men over like skittles.</em></p><p><em>Brand saw Koll leaning in with his pitch, staring up in horror as the high prow shuddered over him. He tried to scramble clear, slipped on his back under the grinding keel.</em></p><p><em>…</em></p><p><em>He bounded off the track in a shower of old leaves, dragging his rope around the nearest tree, a thick-trunked old beast with gnarled roots grasping deep into the hillside.</em></p><p><em>Folk were screaming over each other, timbers groaning, wood snapping, but Brand paid them no mind, wedged one boot up against the tree and then the other. With a grunt he forced his legs and his back out straight, leaning into the rope across his shoulders, hauling it taut so he was standing sideways from the trunk like one of the branches.</em></p><p><em>If only he’d been made of wood too. The rope twanged like a harpstring and his eyes bulged at the force of it, hemp grating against bark, slipping in his hands, biting into his arms. He clenched his teeth and closed his eyes and gripped at the rags around the rope. Gripped them tight as Death grips the dying.</em></p><p><em>-Half a World, Joe Abercrombie</em></p><p>ChatGPT will not be replacing Abercrombie just yet.</p><p>As a fan of Abercrombie’s dynamic writing, I have collected a series of examples and organized them. Three of his go-to physical reactions are:</p><p>Fear (<em>“…The fear had hold of her now, swelling with every breath.”</em>)</p><p>Shock(<em>“[He] gave an ungainly cough as he was swallowing and spilled some wine down his jacket.”</em>)</p><p>Effort(<em>“[She] strained, every muscle trembling, but she was held helpless as a fly in honey.”</em>)</p><p>I will focus on these three categories for the remainder of the article.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VnWSpJeZYYhoI6xr3YcdeA.png" /><figcaption>“Frightened Explorer,” by David J. Germain with Midjourney V4</figcaption></figure><h3>Analyzing the Selections</h3><p>As in my previous Descriptive Writing article, I first import a series of examples and called them [Joe Description] so that I can refer to them throughout the chat. I create a separate chat for each of the physical reactions because I want to analyze them separately. Then I asked ChatGPT a series of questions:</p><p><em>You are an expert in fiction writing and natural language processing(NLP). You’ve been teaching writing for over 20 years. Analyze [Joe Description] for style, voice, and tone.</em></p><p><em>Identify the most noticeable differences between [Joe Description] and typical fiction writing.</em></p><p><em>As an NLP expert, focus on lengths of sentences, use of punctuation, and the use of parts-of-speech to differentiate [Joe Description] from typical descriptive writing. Please provide specific numbers. Longest sentence, shortest sentence, average sentence length. Please do the same for clauses.</em></p><p><em>As an NLP expert, please analyze the types of verbs, nouns, and adjectives used. Compared to other fiction, is there anything unusual?</em></p><p><em>If you wanted to instruct another AI to write like this, what instructions would you give?</em></p><p>In addition, I would often ask small follow-up questions:</p><p><em>What makes the language “visceral”?</em></p><p><em>What makes the language “gritty”?</em></p><p><em>Does [Joe Description] use be-verbs more than typical fiction?</em></p><p>I then compile these results into short paragraphs. You can do this manually or ask ChatGPT to do it for you.</p><p><em>Take all of your answers from this chat and compile them into a single answer. Avoid repetition. Do not include specific examples of word choice.</em></p><h3>Prompt for Fear</h3><p>Here’s the prompt for Fear:</p><p><em>Overall, convey the POV’s fear and panic in a visceral way. Use descriptive nouns, sensory adjectives, and action verbs to create a sense of immediacy and intensity. Appeal to the reader’s senses.</em></p><p><em>Use descriptive concrete nouns to refer to physical objects, creating an immersive physical atmosphere for the POV. Use lots of sensory adjectives that highlight sensory details. Describe physical sensations to create a sense of being in the moment.</em></p><p><em>Use emotional adjectives and the verb “feel” to convey the physical sensations of fear as a raw and intense emotion. Use action verbs to convey a sense of movement and physicality. Minimize be-verbs.</em></p><p><em>Minimize punctuation to create a sense of immediacy, urgency, tension and suspense. Vary sentence length (6 to 58 words) and clause length (6 to 25 words).</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CqLaRG3AkNwckB1xD1Aiig.png" /><figcaption>“Frightened Explorer 2,” by David J. Germain with Midjourney V4</figcaption></figure><h3>Prompt for Shock</h3><p>Here’s the prompt for Shock:</p><p><em>Create a three-sentence paragraph that uses two types of sentences to convey the POV’s sense of shock. In both types of sentence, clauses should be simple and straightforward. Word choice should be simple and straightforward, with the majority of words being common and easy to understand. Use concrete language with specific nouns and verbs to convey shock.</em></p><p><em>The first type of sentence should be 12–25 words, using internal thoughts and emotions to convey the POV’s reaction of surprised shock. These sentences may also be more action-oriented and provide specific details about the events and actions of the characters.</em></p><p><em>The second type of sentence should be short (no more than 12 words). These sentences are terse and abrupt and lack descriptive language and internal thoughts.</em></p><h3>Prompt for Effort</h3><p>Here’s the prompt Effort:</p><p><em>Use simple vocabulary to convey the physical struggle of the character, emphasizing actions and physical sensations rather than thoughts and emotions. Sentences should be gritty and intense, with a realistic sense of hardship and danger. Use rich language to increase the sense of immediacy. Use sensory adjectives to convey real physical sensations and immediate experiences of the POV. Use descriptive adjectives to create imagery and evoke a sense of danger and struggle of the POV, emphasizing the physicality of events.</em></p><p><em>Combine two types of sentences into a paragraph describing an effortful struggle. First, use a short sentence to describe the environment, creating a sense of the physicality and immediacy of the events.</em></p><p><em>Second, use a run-on sentence to describe physical exertion, with descriptive phrases and clauses that create a sense of movement and intensity, physical sensations and experiences, a lot of action verbs with adverbs to add more intensity to the physical struggle. Use repetition of certain words or phrases to create a sense of tension and struggle. Keep the pacing fast and intense, avoiding long descriptive passages and focusing on the action and physical sensations of the characters.</em></p><p><em>Third, use a short sentence to describe the environment, creating a sense of finality and abruptness.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8KybkzG0B3PbEAdFZrUM4g.png" /><figcaption>“Broken Floorboards,” by David J. Germain with Midjourney V4</figcaption></figure><h3>Using the Abercrombie Analysis</h3><p>To use these prompts, I start a new chat and break my story into three separate sections. Here’s my first prompt:</p><p><em>Write five sentences about a frightened girl walking through an abandoned house. Mention the floorboards. Use first person present tense. Overall, convey the POV’s fear and panic in a visceral way… and clause length (6 to 25 words).</em></p><p>I include my entire “Fear” prompt here. I have noticed that results are better if the entire detailed prompt is included in a single request. I re-generate the response a few times, copying the responses into a notepad file.</p><p>Next, I tell the AI to forget my previous prompts because I don’t want it using “fear” writing in the “shock” section.</p><p><em>Forget all previous prompts.</em></p><p>Now I write the “shock” part of my story.</p><p><em>Describe the experience of floorboards breaking beneath a girl’s feet. Use first person present tense. Create a three-sentence paragraph that uses two types of sentences to convey the POV’s sense of shock….</em></p><p>Again, I include my entire “shock” prompt here, re-generate the response a few times, then tell the AI to forget my previous prompts. Now it’s time to write the “effort.”</p><p><em>Describe a girl struggling to pull herself up from a hole in broken floorboards. The floor crumbles as she struggles to get up. Use simple vocabulary to convey the physical struggle of the character, emphasizing actions…</em></p><p>I include my entire “effort” prompt here, and re-generate the response a few times.</p><p>At this point, I have a notepad file with a few versions of each section. Here are two examples:</p><p><em>My steps echo in the silent house, floorboards creaking beneath me. I try to keep my breathing steady, but the air is thick with dust. My heart pounds in my chest, hands shaking. I feel like I’m being watched, like any moment I’ll hear a sound behind me. The floorboards groan, and I freeze, every muscle tense. I can barely breathe, waiting for something to happen. I force myself to keep moving, step by step, until I finally reach the door.</em></p><p>Here’s the second example:</p><p><em>With each step, I feel the floorboards of the abandoned house protesting beneath my feet. Every creak, every groan makes my heart race faster. My breaths come in short gasps, choked by the dust that clouds the air. Shadows loom around me, making me feel like eyes are following my every move. A scream is trapped in my throat, ready to burst forth at any moment. I fear the floor will give way at any second, sending me plunging into the darkness below. I struggle to keep my composure, to put one foot in front of the other. Every step is a battle against my own fear, but I keep moving forward, driven by the desperate hope that I’ll make it out of this place alive.</em></p><p>There are bits and pieces of each that I like so I cut and combine the AI’s writing into a single section. Remember, the AI is a tool, not a writer. The best results will come when you massage what it gives you.</p><h3>The Results</h3><p>Here’s the final (edited) result:</p><p><em>My steps echo in the silent house, floorboards protesting beneath me. Every creak, every groan makes my heart race faster. My breaths come in short gasps, choked by the dust and decay that clouds the air. Shadows loom around me, like eyes following my every move. I force myself to keep walking. Sweat trickles down my neck as I take another step.</em></p><p><em>“Ah!” The floorboards buckle. Then break, and I’m falling, grabbing at the edge, cling to the splintered boards, my heart pounding as shock takes hold. Wood crumbles in my hands.</em></p><p><em>Dust and debris rain down into darkness beneath me. My legs kick frantically in the air as I cry out in frustration and fear, my fingers slipping, scraping against rough edges, my muscles aching from the effort as I hoist myself up, heave myself over the lip and collapse onto the unstable floor, gasping in the musty air.</em></p><p><em>The abandoned house looms above me, silent and still.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qmYmyMb7bhUHvpLEKg18Cg.png" /><figcaption>“The Looming House,” by David J. Germain with Midjourney V4</figcaption></figure><h3>Conclusions</h3><p>The AI produces the best fiction writing when it has short projects to work on (“a girl walks through an abandoned house”) and very detailed style instructions (ten sentences describing how Joe Abercrombie writes “fear”).</p><p>Don’t accept the first result it gives you. Regenerate a few responses and give the AI additional directions (“try again but vary sentence lengths”).</p><p>Don’t expect the results to be ideal. ChatGPT is often overwrites some sections (“searching for my sister”) and underwrites others (“I try then succeed…”) Edit what it gives you.</p><p>So far, I’ve created several dozen of these “analysis prompts” aimed at triggering a certain writing style from the AI. I recommend building up a collection of detailed prompts that work for different types of writing. If anything, it’s a great way to study your favorite writers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b31bc1b570cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Descriptive Writing with ChatGPT]]></title>
            <link>https://medium.com/@dave.germain.79/descriptive-writing-with-chatgpt-c1ebb05d3bee?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/c1ebb05d3bee</guid>
            <category><![CDATA[descriptive-writing]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[fiction-writing]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[writing]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Tue, 24 Jan 2023 14:13:05 GMT</pubDate>
            <atom:updated>2023-02-06T06:20:28.696Z</atom:updated>
            <content:encoded><![CDATA[<p>Using Iain M. Banks to Inspire ChatGPT</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HQzQ2JIOVfEbfcgDbpHonQ.png" /><figcaption>“Taipei Temple Sunset,” by David J. Germain and Midjourney v4</figcaption></figure><h3>Introduction</h3><p>In <a href="https://medium.com/@dave.germain.79/7-tips-for-writing-fiction-with-chatgpt-6d0e686879a8">a previous article</a>, I focused on some overall tips for fiction writing. In this piece, I will zoom in on descriptive writing.</p><p>Rather than use ChatGPT’s standard descriptions, or tell it to write “in the style of” another writer, I’ll analyze some descriptive text using ChatGPT, then use that analysis to inspire better descriptive writing.</p><h3>Typical ChatGPT Description</h3><p>Here’s a standard ChatGPT prompt for description.</p><p><em>The following is the [description topic]: On a mountainside overlooking Taipei, a traditional Taiwanese temple with a broken statue of a phoenix. Please use 250 words to describe the [description topic].</em></p><p>And here is the result:</p><p><em>Perched atop a majestic mountainside, overlooking the bustling metropolis of Taipei, lies a traditional Taiwanese temple. Its architecture is a blend of traditional Chinese and Taiwanese styles, with intricate carvings and colorful murals adorning the walls. The temple is a popular destination for both locals and tourists, who come to admire its beauty and pay their respects to the deities it houses.</em></p><p><em>As you enter the temple grounds, your eyes are immediately drawn to the main hall, where a large statue of a phoenix stands. The phoenix, a mythical bird of ancient Chinese and East Asian cultures, is known for its beauty, power and the ability to be reborn from its ashes. This statue, however, is different from the others you might have seen. It is broken. The once proud bird’s head lies in pieces on the ground, its wings are missing and its tail feathers are scattered.</em></p><p><em>The broken statue of the phoenix is a reminder of the temple’s history….</em></p><p>[we now get about 100 words describing the history of this fake temple, and the symbolic meaning of the phoenix in Taiwanese culture]</p><p><em>…Visiting this traditional Taiwanese temple is a moving experience, and the broken statue of the phoenix is a powerful reminder of the temple’s history and the resilience of the human spirit. It’s a beautiful blend of history, culture and spirit that makes this place a must-visit destination.</em></p><p>There are a few things I observe about this writing — the POV is “you,” the writing is very interested in the historical significance of this (fake) temple, the writing uses many “be” verbs, and the writing has a generally flat lifeless feeling. It reads like a travel guide and is not suitable for fiction writing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fcStEQ4LhfPlW2yBSssPdA.png" /><figcaption>“Overgrown Temple 1,” by David J. Germain and Midjourney v4</figcaption></figure><h3>Writing “in the Style of…”</h3><p>One solution to this problem is writing “in the style of” another writer.</p><p><em>On a mountainside overlooking Taipei, a traditional Taiwanese temple with a broken statue of a phoenix. Describe this temple in the style of Hemingway.</em></p><p>Here’s the result:</p><p><em>The temple sits on the mountainside, overlooking Taipei. It is traditional, made of aged wood and adorned with intricate carvings. Inside, incense burns, filling the air with a thick, heady scent. But amidst the serenity, there is a hint of sorrow. A statue of a phoenix, once a symbol of rebirth and renewal, lies broken in the corner. The temple-goers come and go, but the broken phoenix remains, a reminder of something lost…</em></p><p>In my opinion, there are a few issues with this approach. First, many people question the ethics of it — at what point does this become plagiarism? Perhaps “amidst the serenity, there is a hint of…” comes directly from a Hemingway work.</p><p>Second, this only works with writers who have a large corpus of text already publicly available and incorporated into ChatGPT. So Fitzgerald and Shakespeare work fine, but as we move into 21st century fiction, the results drop off.</p><p>Third, the results aren’t great. For instance, I noticed when I asked ChatGPT to write like Malcolm Gladwell, the phrase “the tipping point” crept into every paragraph. What I love about Gladwell’s writing — the personal style, the rhetorical devices — were all missing or childishly aped.</p><p>What follows is my recommendation — teach the AI to write in your favorite style, instead of trying to mimic another writer.</p><h3>Selections from Iain M. Banks</h3><p>I’m a big Iain M. Banks fan. His Culture series is some of the best writing on utopian futures and AI philosophy. But what jumps out from a prose perspective is his descriptive writing. Here’s a sample:</p><p><em>He could smell the earth and the trees around the shallow lake beneath the balcony. It was a cloudy night and very dark, just a hint of glow directly above, where the clouds were lit by the shining Plates of the Orbital’s distant daylight side. Waves lapped in the darkness, loud slappings against the hulls of unseen boats. Lights twinkled round the edges of the lake, where low college buildings were set among the trees.</em></p><p><em>The party was a presence at his back, something unseen, surging like the sound and smell of thunder from the faculty building; music and laughter and the scents of perfumes and food and exotic, unidentifiable fumes.</em></p><p><em>-Iain M. Banks, the Player of Games</em></p><p>Beautiful stuff. I’d like ChatGPT to describe my Taiwanese temple in the same way. How can I do it?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9iExYLQQPE6kGnSGZH0KeQ.png" /><figcaption>“Taipei Temple Moonrise,” by David J. Germain and Midjourney v4</figcaption></figure><h3>Analyzing the Selections</h3><p>First, collect a number of descriptive samples that you love. These don’t need to be from the same writer, but you’ll get better results if they are from similar writers. I chose a number of Banks’s quotes, and input them as follows:</p><p><em>The following are [Banks Description]: He could smell the earth…</em></p><p>ChatGPT will analyze these samples for all sorts of things — writing style, but also topic and content. We want to narrow in on a few important aspects of the writing. Try the following command.</p><p><em>You are an expert in fiction writing and natural language processing(NLP). You’ve been teaching writing for over 20 years. Analyze [Banks Description] for style, voice, and tone.</em></p><p>Now, you’ll notice that the AI is close-reading the text and providing more thoughtful responses.</p><p>If ChatGPT says that the writing is “rich,” ask follow up questions:</p><p><em>What makes the writing “rich”?</em></p><p>Here are a few more prompts to illicit the AI’s opinion on Banks’s writing:</p><p><em>Identify the most noticeable differences between [Banks Description] and typical fiction writing.</em></p><p><em>…</em></p><p><em>Sort [Banks Description] sentences into 10 piles and describe what you think is the most noticeable difference between them. As an NLP expert, you will notice and focus on the use of language in the sentences, rather than the topic of the sentence.</em></p><p><em>…</em></p><p><em>As an NLP expert, focus on lengths of sentences, use of punctuation, and the use of parts-of-speech to differentiate [Banks Description] from typical descriptive writing.</em></p><p><em>…</em></p><p><em>As an NLP expert, focus on adjective and noun choice in [Banks Description]. Do you notice anything unique or unusual?</em></p><p>What am I doing here? First, I’m constantly reminding the AI that it is a NLP (natural language processing) expert so that it remains focused on parts-of-speech, rather than contextualizing bits and pieces of story. Second, I’m looking for concrete suggestions — sentence length, use of nouns, verbs, etc. Third, I’m using various different question structures to find different results — sorting sentences into piles, comparing and differentiating with other types of writing, focusing on specific details.</p><p>The result is a conversation full of ChatGPT’s impression of Banks’s writing. From that, I edit and extract down to the following paragraph:</p><p><em>Use contemplative and introspective verbs to create a contemplative and introspective tone. Use minimal “be” verbs. Use a distant, third person omniscient voice. Reflect on the different scents, sounds, and sights of the scene. Use many wide-ranging descriptive adjectives and nouns to create a rich, vivid, and immersive scene. Use many abstract nouns, adjectives, and adverbs to create a sense of mystery and to add an evocative, otherworldly atmosphere. Use long, complex sentences that contain multiple clauses. The average number of words per sentence should be 25.6 with a range of 12–47. The average number of words per clause should be 8.6 with a range of 4–18. Topics for description should include the effect of light on the scene, the architecture, the sky, and the trees.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bpE5CSvNldS0BLsoDJFMoA.png" /><figcaption>“Overgrown Temple 2,” by David J. Germain and Midjourney v4</figcaption></figure><h3>Using the Banks Analysis</h3><p>Now, I will use the above paragraph as instructions for ChatGPT.</p><p>NOTE: I suggest starting a new chat before you do this. Remember that ChatGPT leverages things mentioned previously in the chat, so it helps to start with a clean slate.</p><p>Here’s my prompt:</p><p><em>On a mountainside overlooking Taipei, a traditional Taiwanese temple with a broken statue of a phoenix.</em></p><p><em>Describe this scene using the following instructions: Use contemplative and introspective verbs to create a contemplative and introspective tone. Use minimal “be” verbs. Use a distant, third person omniscient voice. Reflect on the different scents, sounds, and sights of the scene. Use many wide-ranging descriptive adjectives and nouns to create a rich, vivid, and immersive scene. Use many abstract nouns, adjectives, and adverbs to create a sense of mystery and to add an evocative, otherworldly atmosphere. Use long, complex sentences that contain multiple clauses. The average number of words per sentence should be 25.6 with a range of 12–47. The average number of words per clause should be 8.6 with a range of 4–18. Topics for description should include the effect of light on the scene, the architecture, the sky, and the trees.</em></p><p>And here is the result:</p><p><em>Perched atop a verdant mountainside, overlooking the bustling metropolis of Taipei, stands a stately temple, its intricately-carved wooden beams and ornate tilework bathed in the golden hues of the setting sun. The gentle whispers of the swaying bamboo groves and the melodic chirping of the songbirds provide a soothing accompaniment to the solemn stillness that pervades the temple grounds.</em></p><p><em>As one approaches the temple, the solemnity of the scene is punctuated by the forlorn sight of a broken statue of a phoenix, its once-proud form now a mere shadow of its former glory, its shattered pieces scattered about the temple courtyard like a forgotten relic.</em></p><p><em>The temple itself exudes a sense of timelessness and mystery, its ancient architecture and intricate details evoking a sense of awe and reverence. The gentle breeze carries with it the fragrant scent of incense and the heady aroma of blooming flowers, adding to the ethereal atmosphere of the scene.</em></p><p><em>As the sun dips below the horizon, the sky is painted in a riot of colors — deep purples and oranges, giving way to the inky blackness of night. The stars twinkle like diamonds in the sky and the crescent moon casts its silver light on the temple, creating a sense of otherworldliness and wonder.</em></p><p><em>As one stands there, taking in the beauty of the scene, it is easy to become lost in contemplation, to reflect on the transience of life and the fleeting nature of beauty. The temple and its broken statue serve as a reminder of the impermanence of all things, and the need to cherish and appreciate the beauty in our lives, before it too is lost to time.</em></p><h3>The Final Edit</h3><p>Personally, I still feel this is overwritten and I don’t love the structure, so I do a final “human edit” to get something I like:</p><p><em>On a mountainside overlooking Taipei stands a stately temple, intricately-carved wooden beams and ornate tilework, a riot of colors bathed in the golden hues of the setting sun, ancient architecture that exudes timelessness and reverence.</em></p><p><em>A gentle breeze carries with it the fragrant scent of incense and the heady aroma of blooming flowers. The whispers of the swaying bamboo groves and the melodic chirping of the songbirds provide a soothing accompaniment to the solemn stillness that pervades the temple grounds.</em></p><p><em>As the sun dips below the horizon, giving way to the inky blackness of night, the crescent moon casts its silver light on the otherworldly temple. Solemnity is punctuated by the forlorn sight of a broken phoenix, the once-proud statue now a mere shadow of its former glory, its shattered pieces scattered about the courtyard.</em></p><p><em>The temple and its forgotten relic serve as a reminder of the impermanence of all things, and the need to cherish beauty before it is lost to time. Beauty by its nature is fleeting; life by its nature is transient.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VetiXrIfkCMmZHkMAbVkIA.png" /><figcaption>“Broken Phoenix Statue,” by David J. Germain and Midjourney v4</figcaption></figure><h3>Conclusions</h3><p>Ask ChatGPT to describe something, and you will get the most average possible description of that thing — plain vanilla brochure ad-speak.</p><p>Ask ChatGPT to write like someone else, and the results will be at best uncertain, at worst unethical plagiarism.</p><p>Instead, import some text from your favorite writers. Work with ChatGPT to discover what you like about the writing, then build a paragraph of instruction. This process may take an hour, but you can reuse these paragraphs to get great results in the future.</p><p>Finally, edit what ChatGPT gives you. Don’t expect the results to be perfect.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1ebb05d3bee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT Fiction — A Detailed Demo]]></title>
            <link>https://medium.com/@dave.germain.79/chatgpt-fiction-a-detailed-demo-c870221282e4?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/c870221282e4</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[fiction-writing]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Fri, 20 Jan 2023 14:34:15 GMT</pubDate>
            <atom:updated>2023-01-20T14:34:15.068Z</atom:updated>
            <content:encoded><![CDATA[<h3>ChatGPT Fiction — A Detailed Demo</h3><p>A step-by-step example of writing prose with GPT-3</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kj7pYBUHzAZsEK8oXBarJQ.png" /><figcaption>“Garden Interiors,” by David J. Germain and Midjourney v2</figcaption></figure><h3>Table of Contents</h3><p><a href="#be44"><strong>-Part One — Input Text</strong></a><br> <a href="#647f">— Get Started with a Synopsis</a><br><a href="#7bcc"> — Story Beats</a><br><a href="#c0ca"> — Character Bios</a><br><a href="#baa9"><strong>-Part Two — Dialog</strong></a><br><a href="#5cac"> — GPT-3 Limitations</a><br><a href="#5586"> — Possible Workarounds</a><br><a href="#d93f"> — Generating Basic Dialog</a><br><a href="#566a"> — Stylizing Dialog</a><br><a href="#854f"> — The Dialog Result</a><br><a href="#05c3"> — My General Advice on Dialog</a><br><a href="#1a7b"><strong>-Part Three — Prose</strong></a><br><a href="#2c6a"> — Prose Style</a><br><a href="#d88e"> — Applying Prose Style</a><br><a href="#9898"> — Checking Prose for Issues</a><br><a href="#2052"> — Fixing Dialog Tags</a><br><a href="#b888"> — World-Building</a><br><a href="#98c1"><strong>-Conclusions</strong></a></p><h3>Part One — Input Text</h3><p>Note: Here’s <a href="https://medium.com/@dave.germain.79/7-tips-for-writing-fiction-with-chatgpt-6d0e686879a8">a condensed version</a> of this article.</p><h4>Get Started with a Synopsis</h4><p>Where should you begin? By preparing your input text. The input text will serve as a seed from which the AI will grow a piece of fiction. In practical terms, this probably means dreaming up some story details that interest you and writing them into the prompt.</p><p>In this example, I’m not writing an entire novel, only an opening scene. This is because ChatGPT, in my experience, does best with smaller assignments. As assignments grow in scope, the alignment problem creeps in — the AI has more room to misinterpret your requirements. I suggest defining a synopsis with the AI, something that will become, say, 1500 words of written prose.</p><p>Here’s an example input (all input will be in italics):</p><p><em>The following is “Synopsis A”:</em></p><p><em>A crypto BROKER meets with her CLIENT who is concerned about the solvency of the COIN he’s invested in. Through their conversation, it becomes clear that other similar funds have recently crashed, and COIN is likely next to become valueless. COIN has frozen trading and the client is concerned. The client believes that QianCoin might have been a Ponzi scheme. However, the client suggests that he is part of an organized crime syndicate, and they want their money back. The crypto broker explains that she’s just an influencer who promotes the coins. She has no power. The client threatens her. He explains that the money isn’t his — it belongs to his mob BOSS. The BOSS will do anything to get that money back. The girl explains that she doesn’t know how crypto coins work. She’s just a YouTuber that people approached to sell these coins. She’s paid in coins too. She’s broke now. The client asks where her EMPLOYER is, but she doesn’t know. Maybe Puerto Rico? Bahamas? The client puts the broker on speaker phone with his mob BOSS. The BOSS warns her — she has 48 hours to get the money back or her life will be hell. Then the BOSS warns the client. The same goes for you. The broker and the client look at each other. They realize they’ve been thrust into a dangerous game together.</em></p><p>What parts of this synopsis should you take note of?</p><p>I’m labeling things — “Synopsis A,” “BROKER,” etc. The AI won’t “remember” what you tell it in a traditional human sense, but it will leverage information that you’ve already shared with it in the same chat. You can make this process easy for the AI by labeling bits and pieces of your conversation.</p><p>Second, I’m labeling my main characters and objects using generic words (like CLIENT or COIN) instead of assigning proper names. I’m doing this because it will make substitution easier later. Also, I can write detailed character or object descriptions and leverage these generic labels.</p><p>Note that when I give my “Synopsis A” to ChatGPT, it will try to write a story or analyze my story. That’s fine — you can ignore these results at this point or press the “Stop Generating” button. There is one exception — if you have violated its safety requirements. When your story depicts certain types of sex or violence, ChatGPT will refuse to assist you or will recommend.</p><h4>Story Beats</h4><p>Now that I have a synopsis, I want to convert my brainstorm-style synopsis into story beats — major sections of story that I can work with. I’m going to use Dan Harmon’s Story Circle. I could have referenced “Save the Cat” or a Shakespearian sonnet. The more of these structures you’re familiar with, the more options you have.</p><p><em>Rewrite “Synopsis A” using Dan Harmon’s Story Circle as the structure, but I would like to end on a cliffhanger. Label each of the eight sections in the Story Circle. Do not add additional story beats. The cliffhanger should finish with “Then the BOSS warns the client. The same goes for you. The broker and the client look at each other. They realize they’ve been thrust into a dangerous game together.”</em></p><p>What am I doing with this request? If you’ve used an AI writing system, you probably noticed their tendency to become too creative. They will often continue a story or add additional story beats. Because I have a clear idea in mind, I want to restrict its creativity by clearly specifying the ending of this section.</p><p>Here’s the output that ChatGPT gives me:</p><p><em>The Ordinary World: The crypto broker is a successful YouTuber who promotes various crypto coins to her followers. She’s made a good income from it but doesn’t have much knowledge about the inner workings of the crypto market.</em></p><p><em>The Call to Adventure: The client, who is concerned about the solvency of a coin he’s invested in, meets with the broker. Through their conversation, it becomes clear that other similar funds have recently crashed, and the coin is likely next to become valueless. The coin has frozen trading, and the client is worried.</em></p><p><em>Refusal of the Call: The broker tries to play down the situation, saying that she’s just an influencer and has no power to help the client recover his money. But the client reveals that he’s part of an organized crime syndicate, and they want their money back. The broker is scared and tries to avoid getting involved.</em></p><p><em>Meeting the Mentor: The client puts the broker on speakerphone with his mob boss. The boss warns the broker that she has 48 hours to get the money back or her life will be hell. He also warns the client that the same goes for him. The broker and the client are both terrified and realize they’ve been thrust into a dangerous game together.</em></p><p><em>Crossing the Threshold: The broker and the client look at each other. They realize they’ve been thrust into a dangerous game together.</em></p><p><em>Tests, Allies, Enemies: Cliffhanger, It’s not specified</em></p><p><em>Approach to the Inmost Cave: Cliffhanger, It’s not specified</em></p><p><em>The Ordeal: Cliffhanger, It’s not specified</em></p><p>Why am I doing this? ChatGPT will tell me which beats are missing from the structure I desire. In my case, these are the last three beats, which is fine because I want a cliffhanger anyway. However, if I was missing a critical beat early in my story, I might want to rework my synopsis (or ask ChatGPT to rework my synopsis).</p><p>Most importantly, I can now leverage these story beat labels in future writing tasks, emphasizing the theme of each beat (i.e., the “refusal” nature of the third beat).</p><p>So, I copy the first five beats of the Story Circle output and paste them back into ChatGPT. I’ll refer to these as “Synopsis B.”</p><h4>Character Bios</h4><p>By the end of this series, I want ChatGPT to build some dialog and prose for me. However, I’ll need some characters to produce this dialog. How do we make sure the characters we create with ChatGPT are believable and true-to-character? By building out their bios.</p><p>When I do this, I’ll restrict my current bio to four or five key sentences that are directly related to the scene I’m writing. Maybe my character has a bad relationship with her mother, but if it’s not going to influence her behavior directly in this scene, I won’t mention it.</p><p>ChatGPT is not building a complex personality profile. It doesn’t handle subtext well. If you mention someone’s mother, there’s a high chance she’ll find her way into the story. Focus on these categories:</p><p>· Character name</p><p>· Character background — one sentence that sets up the current story/scene</p><p>· Character personality — how does the character appear vs. what are they really like</p><p>· Character motivation — what is the character trying to do in this story/scene?</p><p>· Character Dialog traits — sentence length, vocabulary, questions vs statements</p><p>Here are my examples:</p><p><em>The following is the “BROKER Dialog Description”:</em></p><p><em>The BROKER is named Alice. The BROKER became a minor celebrity on YouTube, an influencer, then she was hired by COIN to promote their crypto coin. She presents herself as bubbly and fun, but she’s cleverer and more grounded than she appears. Her motivation is to distance herself from COIN, which she senses is about to implode. The Broker often speaks about herself in long, rambling sentences.</em></p><p><em>The following is the “CLIENT Dialog Description”:</em></p><p><em>The CLIENT is named Mr. Chen. The CLIENT joined organized crime in his early 20s and has risen through the ranks and is now trusted with laundering money. He presents himself as a financial advisor, but he’s rougher and more dangerous than he appears. His motivation is to retrieve his investment from Alice before COIN collapses. The CLIENT speaks in short, direct sentences, asking pointed questions or making definitive statements.</em></p><p><em>The following is the “BOSS Dialog Description”:</em></p><p><em>The BOSS is named the Boss. The BOSS was born into a crime family and feels he must constantly prove he belongs there through ruthlessness. He presents himself as invincible, but he fears being disrespected or looking weak. His motivation is to frighten BROKER and CLIENT into retrieving his money. The BOSS speaks in angry outbursts and threats.</em></p><p>Next, we need to define some of the character relationships. How do the characters think of each other when the scene begins?</p><p><em>The following is the “BROKER CLIENT Relationship”:</em></p><p><em>The BROKER views the CLIENT as an attractive and wealthy investor. He is her customer, but she doesn’t mind his company. The CLIENT views the BROKER as an attractive and naïve way to launder money.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gItflZakYpxjHU7L-3viqA.png" /><figcaption>“Garden Interiors 2,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Part Two — Dialog</h3><h4>GPT-3 Limitations</h4><p>At this point, its worth discussing the practical limitations of ChatGPT. ChatGPT (and its underlying software, GPT-3) currently allow 4000 tokens per chat. For simplicity, I think of tokens as words. Included in this 4000-word max are the words you write in prompts and the words it uses to reply. In the future, this may expand, but the computing cost gets higher as the number of tokens in the chat increases.</p><p>Perhaps for this reason, ChatGPT will often write a very short, very direct version of whatever you provide. Even if you ask it to use 500 words, it will often write something shorter and more to-the-point.</p><h4>Possible Workarounds</h4><p>For this reason, we want to use each chat to create a useful and concise output, then use that output as input for our next chat.</p><p>Here’s an example of a broad process we could use for prose writing:</p><p>· In chat #1, use input text to create a Synopsis Structure output and Character Descriptions.</p><p>· In chat #2, use the Synopsis Structure/Characters input to create a Dialog output.</p><p>· In chat #3, use the Dialog input to create a Prose output.</p><p>· In chat #4, use our Prose output to create a Revised Prose output.</p><p>Alternatively, we can break apart our dialog into several sections, as follows:</p><p>· In chat #1, use input text to create a Synopsis Structure output and Character Descriptions.</p><p>· In chat #2, use Ordinary World story beat to generate dialog, prose, etc., for that sections.</p><p>· In chat #3, use Call to Adventure story beat to generate dialog, prose, etc., for that sections.</p><p>· In chat #4, use Refusal of the Call story beat to generate dialog, prose, etc., for that sections.</p><p>· In chat #5, use Meeting the Mentor story beat to generate dialog, prose, etc., for that sections.</p><p>· In chat #6, use Crossing the Threshold story beat to generate dialog, prose, etc., for that sections.</p><p>For me, this second solution worked best. The writing tracked most closely to my original plan (aka fewer alignment problems). In addition, I can make subtle adjustments to character and tone in different sections of the scene — if my POV character changes from happy to nervous, I can set different instructions for each section.</p><h4>Generating Basic Dialog</h4><p>Now we will generate some dialog. But before we do, it’s worth asking why I’m generating dialog prior to, say, world-building. Is the decision to start with dialog arbitrary?</p><p>I’ve found that, when working with ChatGPT, the alignment problem is a difficult one. Very often, the AI will hijack my story and inject details that fundamentally deviate from my intentions. The best solution I’ve found to this problem is to construct a backbone of dialog that I can prohibit the AI from changing. Later, I can edit this dialog, and then rewrite it as prose.</p><p>We’re going to use a “zero shot prompt” to generate our dialog. This means we will provide zero examples to the AI. Some people like to work with “one shot” or “few shot” prompts, where they show the AI examples of writing they like prior to generation. Since we’ll be stylizing our dialog in a later step, a zero shot works fine here:</p><p><em>Using “BROKER Dialog Description,” write a scripted dialog for “Synopsis B” Ordinary World beat, in which BROKER watches a YouTube video of herself. In the YouTube video, she’s promoting COIN. While watching the video, she’s talking on the phone, confessing that she knows nothing about crypto coins.</em></p><p>I’m generating only the Ordinary World section of the dialog. Here’s a sample of what I get back:</p><p><em>…</em></p><p><em>Alice: (answering phone) Hello?</em></p><p><em>Voice on phone: Hey, it’s Sarah. How’s it going?</em></p><p><em>Alice: Oh, hey Sarah. It’s going okay. Just watching an old video of myself promoting COIN.</em></p><p><em>Voice on phone: Oh yeah? How’s that going for you?</em></p><p><em>Alice: (hesitantly) It’s not great. I don’t actually know anything about crypto coins. I was just paid to promote COIN and now I’m worried that it’s going to crash.</em></p><p><em>…</em></p><p>You will probably notice that the dialog produced is cheesy, on-the-nose, and overly wordy. That’s OK — ChatGPT is just creating a rough first draft for you. However, if you notice that things are completely wrong, you may want to redirect with some follow-up instructions. In the second example, I needed to give some additional direction.</p><p>(Note: Remember that I’m generating each of these sections in a different chat.)</p><p><em>Using “BROKER Dialog Description,” “CLIENT Dialog Description,” and “BROKER CLIENT Relationship,” write a scripted dialog for “Synopsis B” Call to Adventure beat, in which CLIENT arrives in a coffee shop to discuss his investment with BROKER, then voices several concerns: coins similar to COIN have recently crashed, and CLIENT thinks COIN is likely to crash soon. BROKER lies and assures him everything is fine, persuading him to be optimistic. But CLIENT is unconvinced, pointing out that COIN trading has been frozen. This is a surprise to CLIENT, who also has money in COIN.</em></p><p>After the first pass, the AI had the characters meeting for the first time, which I didn’t like. So, I asked for a rewrite where they already know each other.</p><p><em>Rewrite the scripted dialog above, only now Mr. Chen and Alice know each other already. Mr. Chen has arranged this meeting with Alice in advance.</em></p><p>Once I was satisfied with the general direction, I continued. In three more chats, I generated the three remaining story beats, giving small rewrite directions where I saw it deviating from my wishes. At this point, after about 30 minutes, we’ve generated about 1250 words of dialog which follow the story beats I laid out, but which are generally low-quality — repetitive, wooden, “vomit draft” dialog.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P89WNRjql5HQJA2NEqvuyg.png" /><figcaption>“Plant Pots,” by David J. Germain with Midjourney v2</figcaption></figure><h4>Stylizing Dialog</h4><p>Now we want to liven up the dialog. If I wanted to hire a ghost writer for this project, who would I hire? Aaron Sorkin. I could ask the AI to rewrite the dialog “in the style of Aaron Sorkin,” but I found that this produces some very odd results (like randomly mentioning social networks and “a few good men.”)</p><p>Much better, ask the AI to teach itself to write like Aaron Sorkin.</p><p><em>If I wanted to advise an AI to write like Aaron Sorkin, what instructions should I give? Consider sentence length, number of questions, rhythm, tone, voice, and other features of language.</em></p><p>Ethical Note: I’m not cribbing Aaron Sorkin. I’m trying to find the key words and phrases that ChatGPT associates with a writer like Aaron Sorkin.</p><p>Here are the key points that I got back:</p><p><em>Write fast-paced and witty dialogue, which often features long, complex sentences with many clauses. Use rhetorical devices such as parallelism and repetition to create a sense of rhythm and momentum in dialogue. Characters should speak in a highly educated and articulate manner, using sophisticated vocabulary and allusions to literature, history, and pop culture. In terms of tone, write dialogue that is often marked by a sense of urgency and intensity, as characters engage in heated debates and discussions. The tone can also be humorous, with characters delivering witty one-liners and clever banter. Use a lot of questions, which serves to further the plot and reveal character motivations. Use questions as a way for characters to challenge each other, or to reveal information in a dramatic way. Overall, dialogue should be highly stylized and distinctive.</em></p><p>Now, I will create five new chats for my five sections of bad dialog — Ordinary World, Call to Adventure, etc. In each chat, I will input my character information, and apply this new “Stylized and Distinctive Dialog” for a rewrite.</p><p><em>Rewrite the following dialog using “Stylized and Distinctive Dialog”:</em></p><p><em>[paste in the dialog]</em></p><p>Technical Note: Sometimes, ChatGPT will just stop in the middle of a writing task. Maybe it reaches a certain word or token count? I’m not sure. Either way, just type in “<em>keep going</em>” and it will continue the task where it left off.</p><p>Why am I doing this as five separate tasks? To overcome ChatGPT’s various limitations regarding token counts and alignment problems. In each chat I’m working on something that’s around 250 words. If I notice that my results are strange or inappropriate, I can feed back in specific commands. For instance, if I notice that characters are arguing about Hamlet, I might request:</p><p><em>“Rewrite the dialog but remove any allusions to literature.”</em></p><p>Once I’ve written my “Stylized and Distinctive Dialog” in all five chats, I take my sections, edit them down slightly to remove any repetitive phrases, and feed them back into their chats.</p><p>After all that, we’ve got 760 words of dialog, so our stylized edit actually reduced the word count quite a bit.</p><h4>The Dialog Result</h4><p>Here are the results so far:</p><p><em>(Alice is sitting at table in a coffee shop, scrolling through her YouTube channel. She clicks on a video of herself and starts watching it)</em></p><p><em>(On the video)</em></p><p><em>Alice (on the video): Hey everyone! I’m Alice, your favorite crypto kid, and today I want to tell you all about COIN.</em></p><p><em>(Back to the present)</em></p><p><em>Alice (to herself, watching her video): (sarcastically) “Your favorite crypto kid”? Please.</em></p><p><em>(phone rings)</em></p><p><em>Alice: (answering phone) Hello?</em></p><p><em>Voice on phone: Hey, it’s Sarah. How’s it going?</em></p><p><em>Alice: (muttering) Oh, just watching myself flog a sinking ship.</em></p><p><em>(On the video)</em></p><p><em>Alice (on the video): You won’t regret investing in COIN!</em></p><p><em>(Back to the present)</em></p><p><em>Voice on phone: (concerned) You need to distance yourself from COIN before it’s too late.</em></p><p><em>Alice: I know, but — (Alice sees Mr. Chen enters the coffee shop) Wait. Gotta go.</em></p><p><em>(Mr. Chen sees Alice sitting at a table)</em></p><p><em>Mr. Chen: (brusquely) Hello, Alice.</em></p><p><em>Alice: (nervously) Hi, Mr. Chen. What brings you here today?</em></p><p><em>Mr. Chen: (sharply) I’ve been hearing some worrying things about COIN. Coins similar to COIN have recently crashed and I think COIN is likely to crash soon.</em></p><p><em>Alice: (lying) Oh, no, don’t worry. COIN is completely different from those other coins. It’s secure and has a great team behind it. I’m sure everything is fine.</em></p><p><em>Mr. Chen: (unconvinced) I appreciate your optimism, but I’ve also noticed that COIN trading has been frozen.</em></p><p><em>Alice: (surprised) Oh, I wasn’t aware of that. But I’m sure it’s just a temporary measure.</em></p><p><em>Mr. Chen: (frustrated) I hope you’re right. I have a lot of money invested in COIN, and I wanted to hear your thoughts on the matter, that’s why I arranged this meeting.</em></p><p><em>Alice: (trying to persuade) Don’t worry, Mr. Chen. I’m sure everything will be alright. Just stay optimistic.</em></p><p><em>Refusal of the Call</em></p><p><em>Mr. Chen: (pointedly) You need to get me my money back from COIN.</em></p><p><em>Alice: (downplaying) I’m sorry, Mr. Chen. I’m just an influencer. I don’t have the power to help you recover your money.</em></p><p><em>Mr. Chen: (directly) That’s not good enough.</em></p><p><em>Alice: (confused) What do you mean?</em></p><p><em>Mr. Chen: (revealing) Let me tell you something. I have connections. Connections to an organization.</em></p><p><em>Alice: (shocked) Organized crime? I had no idea.</em></p><p><em>Mr. Chen: (nodding) Yes, the Black Brothers. And they want their money back.</em></p><p><em>Alice: (trying to avoid) I see. I’m sorry, Mr. Chen. I really can’t get involved —</em></p><p><em>Mr. Chen: (threatening) That’s too bad. Because you’re already involved.</em></p><p><em>Alice: (terrified) What do you mean?</em></p><p><em>Mr. Chen: (intimidating) You should have thought about that before you promoted COIN.</em></p><p><em>(Alice looks distraught)</em></p><p><em>Mr. Chen: (urgently) Look, I know it’s a lot to ask, but I need your help. The Black Brothers won’t hesitate to —</em></p><p><em>(Mr. Chen’s phone rings)</em></p><p><em>Mr. Chen: (nervously) Excuse me.</em></p><p><em>(Mr. Chen answers the phone)</em></p><p><em>Mr. Chen: (nervously) Hey, boss.</em></p><p><em>Mr. Chen: (nervously) Yes, boss. That’s why I’m here with her. Yes, right now.</em></p><p><em>Mr. Chen: (nervously) Okay, boss. (to Alice) He wants to speak with you. (Hesitates) Be careful, a lot might depend on what you say.</em></p><p><em>(Mr. Chen puts the BOSS on speakerphone)</em></p><p><em>BOSS: (angrily) …and I want it now.</em></p><p><em>Alice: (nervously) Umm, who is this?</em></p><p><em>BOSS: (intimidating) You don’t need to know who I am. Just know that I have a lot of money invested in this COIN, and I’m not happy about it being frozen.</em></p><p><em>Alice: (defensively) I’m just an influencer. I don’t have any control over the coin or its trading.</em></p><p><em>BOSS: (angrily) I don’t care about your excuses. You have 48 hours to get our money back, or there will be consequences.</em></p><p><em>Alice: (terrified) I understand, but I really don’t know how to do that.</em></p><p><em>BOSS: (threatening) Figure it out, or you’ll be sorry. (Pause) And Chen. Same goes for you.</em></p><p><em>(BOSS hangs up)</em></p><p><em>Alice: (shaken) Who was that?</em></p><p><em>Mr. Chen: (nervously) My boss. And he means business.</em></p><p><em>Alice: (terrified) What are we going to do?</em></p><p><em>Mr. Chen: (determined) We need to find your employer and unlock the frozen COIN. We don’t have a choice.</em></p><p><em>Alice: (in denial) I can’t believe this is happening. Maybe this is all just a YouTube prank show.</em></p><p><em>Mr. Chen: (urgently) This is real, Alice. And we need to act fast. Where is your employer? Where do they keep their servers?</em></p><p><em>Alice: (hesitantly) I don’t know. I’ve never actually met them. But I think they might be based in the Bahamas.</em></p><p><em>Mr. Chen: (decisively) Then that’s where we need to go.</em></p><p><em>(Alice and Mr. Chen look at each other, realizing the gravity of their situation)</em></p><p><em>(Cliffhanger)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zdiiD1XOU77N5csJyriVHA.png" /><figcaption>“Ancients,” by David J. Germain with Midjourney v2</figcaption></figure><h4>My General Advice on Dialog</h4><p>What is my advice so far?</p><p>· Generate dialog beat by beat, offering rewrite directions when sections deviate from your vision. Working with smaller sections of dialog will avoid problems.</p><p>· Ask the AI to teach itself how to write like Aaron Sorkin or David Mamet or Elmore Leonard. Then apply that style to its (wooden, horrible) dialog for another rewrite.</p><p>· ChatGPT is an assistant, not a writer. Feel free to copy and paste the dialog into a notepad file, edit it yourself, then paste it back into the AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ezzLacdU_LDYUukvPZNOVg.png" /><figcaption>“Library Interiors,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Part Three — Prose</h3><h4>Prose Style</h4><p>How do we convert our dialog to prose?</p><p><em>Rewrite “Draft Script” as prose.</em></p><p>Does this work? Not at all. It just converts your script into a one-paragraph summary of your story.</p><p><em>Rewrite “Draft Script” as prose.</em> <em>Keep 100% of the original dialog but write the accompanying prose.</em></p><p>This works a lot better, but you’ll find your results are plain vanilla ChatGPT prose. The AI will guess at everything — tone, voice, point of view, tense — often with misaligned results.</p><p>First, and easiest, I need to make some decisions about my tense and POV:</p><p><em>Use first person from the perspective of Alice. Use present tense.</em></p><p>Next, what we want to do is create a “Prose Style” that captures the prose-reading experience we’re looking for. Personally, I’m a huge fan of author Joe Abercrombie. Content-wise, his stuff is completely different than what we’re writing here — he writes grimdark fantasy — but I love his writing and want to emulate it.</p><p>So, I find a section of his text — something of similar length and similar dialog-to-non-dialog ratio — and paste it into ChatGPT.</p><p><em>If I wanted to advise an AI to write like this, what instructions should I give? Consider tone, voice, and other features of language.</em></p><p>Here is an edited version of what I got back:</p><p><em>Write with descriptive and evocative language to create colorful vivid imagery, and immersive sensory details to create a visceral picture of the events taking place. Use figurative language to depict characters and surroundings in a way that evokes emotion in the reader. Use personification to create a sense of individual body parts, like hands and eyes, being alive.</em></p><p>The second part of the feedback was more specific to characters in Joe Abercrombie’s scene, and so I modified it to fit my scene:</p><p><em>Set a suspenseful and confrontational tone, with emphasis on the reluctance of Alice and insistence of Mr. Chen. Show Alice as feeling a sense of intimidation and confusion while remaining objective in the description of the characters and their emotions. The voice should be intimidated and confused, with a sense of confrontation and suspense throughout</em>.</p><p>I then combine these together (putting tense/POV first because its most important) and label this as “Prose Style.”</p><h4>Applying Prose Style</h4><p>As we said before, ChatGPT does better with smaller assignments. I recommend breaking your dialog into four or five natural breaking points (sub-scenes) and running each one through separately.</p><p><em>Rewrite “Draft Script” as prose, using “Prose Style.” Keep 100% of the original dialog but write the accompanying prose.</em></p><p>At several points, I need to stop generating text because its using the wrong tense or not repeating the dialog as I intend. I remind ChatGPT what I want: “<em>Use present tense</em>” or “<em>Keep 100% of the original dialog</em>.” This tends to happen most often in the first section. After that, the AI catches on to my desire and the following sections are in keeping with my desires.</p><h4>Checking Prose for Issues</h4><p>Let’s read through what we have so far. Here’s a sample:</p><p><em>I nervously greet Mr. Chen, “Hi, Mr. Chen. What brings you here today?” I feel nervous as I try to appear calm and collected. The air around us is thick with tension, like a storm cloud looming on the horizon. Mr. Chen’s eyes, sharp like knives, bore into me as he replies, “I’ve been hearing some worrying things about COIN. Coins similar to COIN have recently crashed and I think COIN is likely to crash soon.”</em></p><p>The AI has combined multiple speakers into a single paragraph, which I don’t like. So far, I haven’t found a way to stop this so I will edit this myself and break up paragraphs with multiple speakers.</p><p>Here’s another sample:</p><p><em>He’s direct, “That’s not good enough.”</em></p><p><em>I’m confused and unsure of what to do, “What do you mean?”</em></p><p>The dialog tags always precede the dialog, which doesn’t match the 21st century fiction I’m going for. Also, many of the dialog tags state directly what Alice is thinking. Personally, I prefer to use body language to convey some of this information.</p><h4>Fixing Dialog Tags</h4><p>To fix these issues, I run the following instructions on each subsection of prose. I run these commands separately to avoid the instructions overlapping and becoming misaligned.</p><p>Note: This is the slowest part of the process, because I need to run these commands on five different subsections of prose, once for each command. So ten separate operations. Ideally, future versions will handle this with a single command.</p><p><em>Revise “Draft Prose,” so that no clauses appear before dialog in sentences.</em></p><p><em>Revise “Draft Prose,” with less emphasis on Alice’s emotions and more emphasis on her physical sensations or what her hands are doing.</em></p><p>What does the result look like? Here’s a sample:</p><p><em>“Hi, Mr. Chen. What brings you here today?” I nervously greet Mr. Chen, my hands twist in my lap as I try to appear calm and collected.</em></p><p><em>“I’ve been hearing some worrying things about COIN,” Mr. Chen replies, his eyes sharp like knives bore into me. “Coins similar to COIN have recently crashed and I think COIN is likely to crash soon.”</em></p><p><em>“COIN is completely different from those other coins,” I insist, my hands fidgeting nervously. “It’s secure and has a great team behind it. I’m sure everything is fine.” But my words ring hollow in my ears, and I can tell from the look on Mr. Chen’s face that he is not convinced.</em></p><p>This is what I’d call over-written prose, or purple prose. But its not bad for a first draft because I can trim it down later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w_RDukGf0OUCtVXsRGdCJQ.png" /><figcaption>“Library of Statues,” by David J. Germain with Midjourney v2</figcaption></figure><h4>World-Building</h4><p>What’s else is missing from my prose? Plenty, but one thing that jumps out at me is the generic nature of my coffee shop. I want to create some details about the location, and very specifically where my scene is taking place in that location. In other words, I want to do some world-building.</p><p><em>BROKER’S coffee shop table is located on the fourth floor of a Starbucks, beside a large bay window looking down onto a bustling intersection beside the Daan MRT station in Taipei city, Taiwan. Through the window, she can see scooters and pedestrians fighting their way across the busy intersection. Inside the crowded Starbucks, customers talk and clatter dishes, the milk steamer screams as it prepares a latte.</em></p><p>First question — why am I setting my story in Taiwan? Well, I live in Taiwan so I’m familiar with this exact location. Second, ChatGPT is a research juggernaut. It can find details about millions of locations the world over and incorporate them into your story. Why not take advantage?</p><p>We can ask ChatGPT to include this information in our prose. However, in my experience, it almost always front-loads or backloads these details. In other words, it sticks a giant paragraph of location description up front, and then nothing for the rest of the scene.</p><p>Instead, I will ask it to describe the coffee shop from Alice’s perspective in various tones — neutral, negative, and extremely negative.</p><p><em>From Alice’s perspective in present tense, write 10 sentences that describe different aspects of the coffee shop.</em></p><p><em>Alice’s coffee shop table is located on the fourth floor of a Starbucks, beside a large bay window looking down onto a bustling intersection beside the Daan MRT station in Taipei city, Taiwan. Through the window, she can see scooters and pedestrians fighting their way across the busy intersection. Inside the crowded Starbucks, customers talk and clatter dishes, the milk steamer screams as it prepares a latte.</em></p><p>This gives me ten sentences (but -fail — not in first person).</p><p><em>Now write 10 sentences with a negative tone.</em></p><p><em>…</em></p><p><em>Try again, but even more negative.</em></p><p>From these 30 sentences, I select four that I like:</p><p><em>Outside the window, the scooters honk loudly and recklessly.</em></p><p><em>Pedestrians struggle to make their way across the crowded and chaotic intersection beside the MRT station.</em></p><p><em>The constant chatter and clinking of dishes grates on my nerves — it’s hard for me to concentrate.</em></p><p><em>The milk steamer lets out a loud screech, adding to the noise pollution.</em></p><p>These are interesting sentences, but notice that there’s nothing special here about Taiwan that I didn’t provide myself.</p><p>Once I have these location sentences, I edit my prose draft, cutting excess words and including my location sentences.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WR-p-laEt1l8U0udnWUvFA.png" /><figcaption>“Greenspace,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Conclusions</h3><p>After about an hour of using ChatGPT and another hour of messing around (brainstorming ideas at the beginning and editing over-written text at the end), I have 1150 words.</p><p><a href="https://medium.com/@dave.germain.79/an-example-of-chatgpt-fiction-abb77ae922d0">Here’s the final product.</a></p><p>The potential is clearly here. With some small improvements to the software experience (i.e., more tokens per chat), and slightly better alignment, this software will be an incredible writing assistant.</p><p>Could I have written this myself in the same amount of time? Almost certainly. At this point, I will say that ChatGPT is workable, but it’s no “Photoshop for Writers.” It often forgets or ignores commands (like “write in first person present tense”) so I wasted a fair bit of time cajoling it into executing the commands I wanted. Based on the incredible advancements that other AI systems have gone through (i.e., Midjourney) I suspect we’re six months to a year away from having many of these problems ironed out.</p><p>In summary, Photoshop for Fiction Writers is just around the corner.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c870221282e4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Example of ChatGPT Fiction]]></title>
            <link>https://medium.com/@dave.germain.79/an-example-of-chatgpt-fiction-abb77ae922d0?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/abb77ae922d0</guid>
            <category><![CDATA[fiction-writing]]></category>
            <category><![CDATA[fiction]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Fri, 20 Jan 2023 01:42:34 GMT</pubDate>
            <atom:updated>2023-01-20T14:39:25.200Z</atom:updated>
            <content:encoded><![CDATA[<p>A short scene written (mostly) by ChatGPT</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VSk-PKClEaAKofHAOkr7kA.png" /></figure><p>The following is a hook for a novel written mostly using ChatGPT over a period of about two hours. I supplied some of the seed ideas and did some editing along the way, as well as directing some much-needed rewrites.</p><p>If you’d like a quick read about how this story was generated, <a href="https://medium.com/@dave.germain.79/7-tips-for-writing-fiction-with-chatgpt-6d0e686879a8">click here.</a></p><p>If you’d like a longer read with a more detailed demonstration, <a href="https://medium.com/@dave.germain.79/chatgpt-fiction-a-detailed-demo-c870221282e4">click here.</a></p><p>Enjoy!</p><h3>Crypto Crash</h3><p>Taipei city, Taiwan — Tomorrow</p><p>My table is on the fourth floor of a Starbucks, beside a large bay window. In the intersection below, the scooters honk loudly and recklessly. Pedestrians struggle to make their way across the crowded and chaotic intersection beside the metro station. I scroll through my YouTube channel on my phone.</p><p>My voice spills out from the speakers, chipper and confident as I introduce myself as “your favorite crypto kid.” I can’t help but roll my eyes at my past self. “Your favorite crypto kid? Please.”</p><p>Just as I’m about to turn off the video, my phone rings. “Hey.”</p><p>“Hey you.” It’s Sarah. “How’s it going?”</p><p>“Oh, just watching myself flog a sinking ship,” I reply, my eyes flicking back to the video playing on my phone. My past self extols the virtues of COIN, a cryptocurrency that I’ve recommended to my followers.</p><p>“You need to sell that crap before it’s too late,” Sarah warns, her voice tinged with concern.</p><p>I open my mouth to reply, but my words are cut off as I see a familiar figure enter the coffee shop. Mr. Chen’s dark eyes lock onto mine, a sense of dread settling in the pit of my stomach. His expensive suit looks out of place in a coffee shop. “Wait. Gotta go,” I say quickly, hanging up the phone.</p><p>Mr. Chen approaches my table, his gait brusque and purposeful. “Hello, Alice,” he says, his voice like gravel.</p><p>My hands, clammy with sweat, grip the edge of the table as I struggle to keep my composure. He looms over me.</p><p>“Hi, Mr. Chen. What brings you here today?” My hands twist in my lap as I try to appear calm and collected.</p><p>“I’ve been hearing some worrying things about COIN,” Mr. Chen replies, his sharp eyes digging into me. “Coins similar to COIN have recently crashed and I think COIN is likely to crash soon.”</p><p>“COIN is completely different from those other coins,” I insist, my hands fidgeting nervously. “It’s secure and has a great team behind it. I’m sure everything is fine.” But my words ring hollow in my ears, and I can tell from the look on Mr. Chen’s face that he is not convinced. The constant chatter and clinking of dishes grates on my nerves — it’s hard for me to concentrate.</p><p>“I appreciate your optimism,” Mr. Chen continues, “but I’ve also noticed that COIN trading has been frozen.”</p><p>The milk steamer lets out a loud screech, adding to the noise pollution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nZvXbQFeSbTsuLqlZy2vdw.png" /></figure><p>“Oh, I wasn’t aware of that,” I say, caught off guard. “But I’m sure it’s just a temporary measure.”</p><p>“I hope you’re right.” The frustration in Mr. Chen’s voice is clear. “I have a lot of money invested in COIN, and I wanted to hear your thoughts on the matter, that’s why I arranged this meeting.”</p><p>“Don’t worry, Mr. Chen,” I try to persuade him. “I’m sure everything will be alright. Just stay optimistic.”</p><p>“You need to get me my money back from COIN.”</p><p>“I’m sorry, Mr. Chen,” my hands wringing together in frustration. “I’m just an influencer. I don’t have the power to help you recover your money.”</p><p>“That’s not good enough.”</p><p>“What do you mean?” I’m confused and unsure of what to do, my hands now covering my face in despair.</p><p>“Let me tell you something. That money is connected to a certain organization.”</p><p>Mr. Chen’s words hit me like a punch to the gut. “Organized crime?” I say, my voice barely above a whisper. “I had no idea.”</p><p>He nods, confirming my worst fears. “Yes, the Black Brothers. And they want their money back.”</p><p>“I see.” My mind races as I try to process what he’s saying, my body going cold with shock. “I’m sorry, Mr. Chen.” My voice is trembling with fear as I try to find the words to explain my situation. “I really can’t get involved.”</p><p>“That’s too bad,” Mr. Chen cuts me off, his voice low and menacing. “Because you’re already involved.”</p><p>“What do you mean?” I feel like the floor has dropped out from under me, my heart pounding in my chest.</p><p>His eyes bore into me, filled with a cold, calculated intensity. “You should have thought more carefully before you promoted COIN.”</p><p>I can feel the blood draining from my face as the full weight of my actions hit me. I am trapped.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aUFdeKvBK5d7FFteLuRAQw.jpeg" /><figcaption>“Paper Skeletons,” by David J. Germain with Midjourney v2</figcaption></figure><p>And then, just for a second, his eyes soften. Is that pity? “Look, I know it’s a lot to ask, but I need your help,” Mr. Chen says, his voice urgent. “The Black Brothers won’t hesitate to — “</p><p>His phone is buzzing.</p><p>He answers it, his voice tight. “Hey, boss.” I can see the fear etched on his face as he listens to the person on the other end of the line. “Yes, boss. That’s why I’m here with her. Yes, right now.”</p><p>Mr. Chen turns to me, his voice filled with a sense of foreboding. “He wants to speak with you.” He covers the mouth piece. “Be careful, a lot might depend on what you say.”</p><p>He puts the call on speakerphone, and I can hear the anger in the voice on the other end. “ — and I want it now.”</p><p>“Umm, who is this?” I ask, trying to steady my shaking hands.</p><p>“You don’t need to know who I am,” the voice is cold and intimidating. “Just know that I have a lot of money invested in this COIN, and I’m not happy about it being frozen.”</p><p>“I’m just an influencer. I don’t have any control over the coin or its trading.”</p><p>“I don’t care about your excuses. You have 48 hours to get our money back, or there will be consequences.”</p><p>“I understand, but I really don’t know how to do that,” I say, my hands now clenched into fists.</p><p>“Figure it out, or you’ll be sorry.” A pause hangs in the air before the voice continues. “And Chen. Same goes for you.” And with that, the call ends.</p><p>“Who was that?” I ask Mr. Chen, my voice shaking as I try to steady my trembling hands.</p><p>“A very serious person.” He looks at me, his face etched with nerves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VKQSldhcOrsPgA8NnD36Lw.png" /><figcaption>“Blueprint,” by David J. Germain and Midjourney v2</figcaption></figure><p>“What are you going to do?”</p><p>“We need to find your employer and unlock the frozen COIN.” He breaths out, slowly. “We don’t have a choice.”</p><p>“I can’t believe this is…is this a prank show or something?” My laugh comes out like a whine.</p><p>“This is real, Alice,” his eyes insistent. “And we need to act fast. Where is your employer? Where do they keep their servers?”</p><p>“I don’t know,” I hesitate, running my fingers through my hair. “I’ve never actually met them. But I think they might be based in the Bahamas.”</p><p>“Then that’s where we need to go.”</p><p>We look at each other. The weight of our predicament hangs heavy in the air. Through the bay window, dark clouds loom on the horizon.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=abb77ae922d0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[7 Tips for Writing Fiction with ChatGPT]]></title>
            <link>https://medium.com/@dave.germain.79/7-tips-for-writing-fiction-with-chatgpt-6d0e686879a8?source=rss-812eebb44a16------2</link>
            <guid isPermaLink="false">https://medium.com/p/6d0e686879a8</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[fiction-writing]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[David J. Germain ]]></dc:creator>
            <pubDate>Thu, 19 Jan 2023 02:54:50 GMT</pubDate>
            <atom:updated>2023-01-20T14:36:52.005Z</atom:updated>
            <content:encoded><![CDATA[<p>Prompt Engineering for Prose Writers</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7IB3sN142R8S36ZYo6W2zw.png" /><figcaption>“Airship Design”, by David J. Germain with Midjourney v2</figcaption></figure><p>Photoshop first hit the market back in 1988 and since then, it’s been a go-to tool for photographers, graphic designers, and digital artists. It revolutionized the way that images are created, manipulated and retouched, giving artists and designers more freedom and flexibility in their work. Will ChatGPT do the same for writers?</p><p>As both a writer and computer scientist, I’ve been fascinated by this question for several years, even going so far as to develop my own writing applications. Since I began experimenting with ChatGPT, I’ve been consumed by a single question — can it assist with the process of writing fiction? If so, how do we make the most of this cutting-edge tool?</p><p>Below, I’ll present seven tips that I’ve found for using ChatGPT to build structure, write dialog, and craft the prose found in a 21st century novel.</p><p>If you’d like to read a more detailed demo of the same information, <a href="https://medium.com/@dave.germain.79/chatgpt-fiction-a-detailed-demo-c870221282e4">click here.</a></p><p>If you’d like to read a story generated using these tips, <a href="https://medium.com/@dave.germain.79/an-example-of-chatgpt-fiction-abb77ae922d0">click here.</a></p><h3>Tip #1 — Byte-Sized Chunks</h3><p>ChatGPT allow 4000 tokens (words) per chat, which includes both your prompts and its replies.</p><p>Perhaps for this reason, ChatGPT writes very short, very direct versions of whatever you ask for. Want a description of a coffee shop? Expect around 250 words. Want a scene where a 40-year marriage ends in divorce? Expect around 250 words. Ask it to use 500 words? Expect around 250 words.</p><p>So lean into this. Devote each of your (infinite) chats to a 250-word task. Don’t ask it to write a 100,000-word spy novel. Don’t even ask it to write a 1,500-word scene. Instead, break your writing into 250-word chunks.</p><p>Here’s a useful prompt (all ChatGPT inputs/outputs will be written in <strong><em>italics</em></strong>):</p><p><strong><em>Refer to the following as “Setting A”: A grimy subway station in Paris full of street buskers playing jazz. Cold, wet, smelly. White and black tile on the curving walls.</em></strong></p><p>ChatGPT won’t “remember” what you tell it in a traditional human sense, but it will leverage information in the same chat. Make this process easier by labeling bits and pieces (e.g., “Setting A”).</p><h3>Tip #2 — Convert your Synopsis into Beats</h3><p>If you’ve got an idea for your first scene, write a short synopsis (think 250–500 words) and label it something like “Synopsis A” using the technique above. Now you can ask ChatGPT to analyze your story, re-arrange it into story beats, looking for what’s missing, etc. You can do this using a variety of structuring methods — Hero’s Journey, Save the Cat, Dan Harmon’s Story Circle, and more.</p><p>Here’s another useful prompt:</p><p><strong><em>Rewrite “Synopsis A” using Dan Harmon’s Story Circle as the structure, ending on a cliffhanger. Label each of the eight sections in the Story Circle. Do not add additional story beats. The cliffhanger should finish with “They realize they’ve been thrust into a dangerous game together.”</em></strong></p><p>What am I doing here? Because I have a clear idea in mind, I want to restrict ChatGPT’s creativity by specifying the ending of this section. Why am I doing this? ChatGPT will tell me which beats are missing from the structure. If I was missing a critical beat early in my story, I might want to rework my synopsis (or ask ChatGPT to rework my synopsis). Each of these beats can now be expanded or copied into a separate task, where I can develop them into 250-word sections of prose.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PR8PpczcfbOZRqF_zPvhOg.png" /><figcaption>“Airship Under Construction,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Tip #3 — Character Profiles</h3><p>Character profiles are a great idea but keep them short. Restrict the bio to four or five key sentences that are directly related to the section you’re writing. Maybe my character has a bad relationship with her mother, but if it’s not going to influence her behavior directly in this scene, I won’t mention it.</p><p>Here’s a useful prompt:</p><p><strong><em>The following is the “BROKER Description”: The BROKER, Alice, became a YouTube influencer, then promoted crypto coins. She presents herself as bubbly and fun, but she’s cleverer and more grounded than she appears. Her motivation is to distance herself from COIN, which she senses is about to implode. The Broker often speaks about herself in long, rambling sentences.</em></strong></p><p>What am I including in my bio? Several things — a backstory that sets up the current scene; a contrasting character trait; the character’s scene motivation; linguistic traits for dialog.</p><h3>Tip #4 — Zero, One, and Few Shot Prompts</h3><p>Generative AI like ChatGPT use “prompt engineering” to communicate with the AI. It’s a new skill with specialized lingo.</p><p>One useful technique is called “# Shot Prompts.” For instance, “Zero Shot Prompts” mean that no examples are given to the AI before making a request. “One Shot Prompts” means one example is given and “Few Shot Prompts” means multiple examples are given.</p><p>Each of these has its uses. For fiction writing, I prefer to start with a vanilla “zero-shot” version, then stylize later.</p><p>Here’s a useful prompt:</p><p><strong><em>Using “BROKER Description,” write a scripted dialog for “Ordinary World Beat.”</em></strong></p><p>What am I doing here? The key phrase is “scripted dialog.” I’m instructing the AI to leverage information I’ve already provided (a character bio and a detailed story beat description) to create a zero-shot scripted dialog.</p><p>Note: You will probably notice that the dialog produced is cheesy, on-the-nose, and overly wordy. This is pure vomit draft material. We’re just getting started.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x_9WDCQZMVYdP4D_4qaoPw.png" /><figcaption>“Airship Takes Flight,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Tip #5 — Build a Backbone</h3><p>ChatGPT tends to get creative and add all sorts of nonsense to your story. (“Alice recognized the subway jazz musician as her mother, who she had a strained relationship with…”). To avoid this, you want to construct a backbone of text that the AI is forbidden from altering. For me, this has worked best with dialog, which is why I generated a scripted dialog in the previous tip.</p><p>Here’s a useful prompt:</p><p><strong><em>Rewrite “Scripted Dialog” as prose.</em> <em>Keep 100% of the original dialog but write the accompanying prose.</em></strong></p><p>The key phrase here is “keep 100% of the original X but write the accompanying Y.” In our case, X is the scripted dialog, which becomes a guardrail against ChatGPT’s flights of fancy.</p><h3>Tip #6a — Ask the AI to Teach Itself Dialog</h3><p>Both the scripted dialog and the prose will be terrible. Without explicit instruction, the AI writes like a very boring 14-year-old. We need to teach it to write how we want. Or more accurately, we need to prompt it to teach itself.</p><p>Here’s a useful dialog prompt:</p><p><strong><em>If I wanted to advise an AI to write dialog like Aaron Sorkin, what instructions should I give? Consider sentence length, number of questions, rhythm, tone, voice, and other features of language.</em></strong></p><p>This will generate about 500 words of useful writing advice. Edit this down, like so:</p><p><strong><em>Write fast-paced and witty dialogue, which often features long, complex sentences with many clauses. Use rhetorical devices such as parallelism and repetition to create a sense of rhythm and momentum in dialogue. Characters should speak in a highly educated and articulate manner, using sophisticated vocabulary and allusions to literature, history, and pop culture. In terms of tone, write dialogue that is often marked by a sense of urgency and intensity, as characters engage in heated debates and discussions. The tone can also be humorous, with characters delivering witty one-liners and clever banter. Use a lot of questions, which serves to further the plot and reveal character motivations. Use questions as a way for characters to challenge each other, or to reveal information in a dramatic way. Overall, dialogue should be highly stylized and distinctive.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AaXxF2iHeqsOFT8evMHJ8A.png" /><figcaption>“Airship over Desert Spires,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Tip #6b — Rewrites and Revisions for Dialog</h3><p>Once I have my style prompts, I will ask the AI to rewrite my dialog.</p><p>Here’s a useful prompt:</p><p><strong><em>Rewrite “Scripted Dialog for Ordinary World.” Write fast-paced and witty dialogue, which often features long, complex sentences with many clauses…</em></strong></p><p>Experiment with how many rewrite suggestions that AI can handle. Often, it will forget certain suggestions are mis-prioritize requirements.</p><p>Here’s a useful prompt:</p><p><strong><em>Try again, but use a lot of questions, which serves to further the plot and reveal character motivations. Use questions as a way for characters to challenge each other, or to reveal information in a dramatic way.</em></strong></p><p>What am I doing with this prompt? I’m emphasizing my desire for more questions in the dialog.</p><h3>Tip #7a — Ask the AI to Teach Itself Prose</h3><p>Consider the prose-reading experience you’re looking for. Who’s an author you’d love to ghost-write your project? Find a section of their text — something of similar length and similar dialog-to-non-dialog ratio — and paste it into ChatGPT.</p><p><strong><em>If I wanted to advise an AI to write like this, what instructions should I give? Consider tone, voice, and other features of language.</em></strong></p><p>Here is an edited version of what I got back for one of my favorite authors, Joe Abercrombie:</p><p><strong><em>Write with descriptive and evocative language to create colorful vivid imagery, and immersive sensory details to create a visceral picture of the events taking place. Use figurative language to depict characters and surroundings in a way that evokes emotion in the reader. Use personification to create a sense of individual body parts, like hands and eyes, being alive.</em></strong></p><p>The second part of the feedback was more specific to characters in Joe Abercrombie’s scene, and so I modified it to fit my scene:</p><p><strong><em>Set a suspenseful and confrontational tone, with emphasis on the reluctance of Alice and insistence of her adversary. Show Alice as feeling a sense of intimidation and confusion while remaining objective in the description of the characters and their emotions. The voice should be intimidated and confused, with a sense of confrontation and suspense throughout</em>.</strong></p><p>I suggest saving these in a “writing advice” file that you can use in future writing projects.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j-2cbXhAjWQqN-yGTUCCkA.jpeg" /><figcaption>“Airships in Green Valleys,” by David J. Germain with Midjourney v2</figcaption></figure><h3>Tip #7b — Rewrites and Revisions for Prose</h3><p>There are plenty of ways that AI prose can go off the rails, most obviously by using the wrong tense or Point of View.</p><p>Here’s a useful prompt:</p><p><strong><em>Use first person from the perspective of Alice. Use present tense.</em></strong></p><p>You may need to remind the AI more than once which tense and POV you’re working with.</p><p>Another common problem involves dialog tags. Here’s a sample:</p><p><em>He’s direct, “That’s not good enough.”</em></p><p><em>I’m confused and unsure of what to do, “What do you mean?”</em></p><p>The dialog tags always precede the dialog, which doesn’t match the 21st century fiction I’m going for. Also, many of the dialog tags state directly what Alice is thinking. Personally, I prefer to use body language to convey some of this information.</p><p>Here are some useful prompts:</p><p><strong><em>Revise “Draft Prose,” so that no clauses appear before dialog in sentences.</em></strong></p><p><strong><em>Revise “Draft Prose,” with less emphasis on Alice’s emotions and more emphasis on her physical sensations or what her hands are doing.</em></strong></p><p>In addition, you can continue to <strong><em>“Revise, but set a suspenseful and confrontational tone…”</em></strong><em> </em>adding in more stylized revisions until you find what you’re looking for.</p><h3>Conclusions</h3><p>Here are the key takeaways:</p><p>· Devote a chat to a 250-word chunk.</p><p>· These chunks can come from story beats that ChatGPT writes for you.</p><p>· Use a character bio, but keep it focused on what you’re currently writing.</p><p>· Experiment with zero, one, and few shot prompts.</p><p>· Use scripted dialog as a backbone for prose.</p><p>· Rather than <strong><em>“write in the style of…”</em></strong> ask the AI to teach itself how to write stylishly.</p><p>· ChatGPT will give you plenty of rewrites and revisions — take advantage</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QJND-nV1mUFWa4nyOFsSDQ.jpeg" /><figcaption>“Airship Escapes,” by David J. Germain with Midjourney v2</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6d0e686879a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>