ABSTRACT (aka the TLDR)

From the subjective perspective of his own creative writing process, Edwin McRae reflects upon and critique the experience of using writing tools powered by artificial intelligence in his professional narrative design practice. Through experimentation, Edwin found that ChatGPT4o and Sudowrite influenced his writing process towards a reactive rather than creative process, putting him in the cognitive state of an editor rather than writer. He also found that the AI writing tools encouraged a product versus process-oriented approach to storytelling.

In the article below, Edwin discusses how these AI tools disrupted the flow state of writing, referencing the work of Johann Hari, Nicholas Carr and Mihaly Csikszentmihalyi to explore the psychological mechanisms behind these flow state disruptions. Specifically focusing on the ChatGPT4o user experience, Edwin references studies on dopamine conducted within the contexts of video games and social media to explain both the attraction of the ChatGPT4o user-experience and the detrimental effects that the resulting manipulation of dopamine has on long-form concentration and creative flow.

Edwin McRae concludes that, in their current state, ChatGPT4o and Sudowrite do more to undermine than support the creative writing process, at least from an artistic perspective. From a commercial writing perspective, Edwin concludes that both ChatGPT4o and Sudowrite can powerfully support a write-to-market strategy where the reproduction of tropes and pre-existing experiences has higher value than efforts to challenge and disrupt the reader’s mental status quo.

In summary, Edwin believes that effectiveness of ChaptGPT4o and Sudowrite, at the time of writing (04 August, 2024), are subject to the context of the desired outcome, undermining process-oriented artistic creative writing whilst empowering product-oriented commercial creative writing. 

 

ARTICLE

AI and the Creative Writing Process_4Aug2024 

 

Much is being said at the moment about the ‘creativity’ that AI unlocks in people. ‘Centaurs’, as they are currently being referred to, are producing larger volumes of text with less time investment and cognitive effort. In these hybrid human/AI writing collaborations, prompting and editing become the dominant skills. Crafting prose is relegated to third place, at best. 

To someone who does not spend a significant amount of time writing, either for pleasure or profession, or to a writer who enjoys having written rather than writing, this may be read as a sign of progress. The ‘drudgery’ of text generation has been removed from the process. The most labour-intensive portion of the craft has been automated.  

Instead of digging the ditch by hand, pull some levers on the machine and excavate that trench in a fraction of the time. Unfortunately, this type of analogy misinterprets the creative act as a productive act. The latter has a repeatable output. Digging to a set width, length and depth produces an identical ditch, every time. Yes, there are different land conditions to consider, but these are just altered values in a formula.  

With writing a story, the value variations are incalculable. Skill, emotion, current situation, life experience, intent, personality and grit combine into a constantly bubbling and changing stew that is the writing process. This is why no two stories are ever the same. Give the writer a set of prompts in Summer and they’ll write a story that is strikingly different to the one written in Winter based on the same prompts. The prompt to product relationship is simply not quantifiable, at least not with our currently limited understanding of consciousness. And if it’s not quantifiable, then it’s not repeatable, and therefore cannot be automated. 

At least, that is true of literary fiction. When you look at genre fiction, the demarcation between art and product becomes less clear. The intent becomes a goal, a calculated output that is designed to serve the specific needs and wants of a target audience. Writing to market requires an adherence to existing genre tropes and reader preferences that are predictably consistent. The LitRPG reader wants a power fantasy. The Western reader wants the shootout. The Romance reader wants the kiss. Fail to give them what they want and they’ll go to a writer that does. Although there’s room for literary creativity in in the expression of these tropes, the conventions themselves are non-negotiable features of the product. Just as a ditch must hold water, a pirate must plunder. 

Because of its predictable outcomes, it is genre fiction across multiple mediums, from books to video games, that is likely to be the most impacted by the adoption of LLM text generation tools. The audience-generated lists of “must-haves” form sets of prompts that text generators can readily understand. And because the most commercially successful stories adhere to there’s lists, the content that LLMs can learn is relatively consistent. This, of course, is a strength of LLMs; the ability to parse and regurgitate patterns of language. 

As a concrete example, it is more likely that ChatGPT4o could write a commercially effective Jack Reacher or Harry Potter novel than a critically effective Neil Gaiman novel. The former have easily identifiable patterns of character and plot across multiple books. The latter have few discernable plot or character patterns across Gaiman’s body work apart from, perhaps, a consistency in the author’s ‘voice’. This is, of course, assuming that OpenAI would have access to the aforementioned works for the machine learning processes. At the time of writing this, the legal dispute between the Authors Guild and OpenAI is still ongoing. 

It’s important to note that I’m not making any value calls between genre fiction and literary fiction. I’m not saying that an LLM has more difficulty producing a literary novel because that form is, in any way, more sophisticated or superior. I’m simply saying that literary fiction has less discernable patterns and therefore is more difficult for a pattern-recognition tool to parse. 

So what has all this got to do with creativity and AI’s effect on it? What I’m trying to preface is that AI tools will have more impact, positively and negatively, upon product-oriented creativity than process-oriented creativity. While the literary author may have objectives in mind, themes and experiences they wish to express, they are not bound by the audience expectation algorithms that genre authors must adhere too. Genre fiction is a creative journey towards an established destination. Literary fiction is an exploration. 

Enter a starting point and destination into Google Maps and receive an optimum route, optimized for distance and time. This is what LLMs can do for the genre fiction writer.  

ChatGPT4o Prompt: Write a 300 word LitRPG story summary where a Level 1 knight completes quests in an epic fantasy world until they become a Level 100 Lord Marshall. 

ChatGPT4o Output (in under 3 seconds): 

In the vast, enchanted realm of Valoria, a humble Level 1 knight named Gareth begins his journey with nothing but a rusty sword and a heart full of valor. Gareth’s village, nestled at the edge of the Shadow Woods, is threatened by marauding goblins, and his first quest is to defend his home. Despite his inexperience, Gareth's bravery earns him the admiration of his peers and the mentorship of Sir Alistair, a seasoned warrior. 

As Gareth embarks on a series of quests, he gains allies, including Elara, an elven mage, and Thrain, a dwarven berserker. Together, they navigate treacherous dungeons, ancient ruins, and magical forests, facing fearsome creatures and unraveling dark conspiracies. Gareth’s skills grow with each victory, and he learns to wield powerful abilities: his swordplay becomes precise, his strength formidable, and his leadership inspiring. 

Gareth's journey is marked by pivotal battles and significant quests. He liberates the city of Stormhaven from the clutches of a necromancer, earning the title of Captain. He uncovers the lost relics of the Dragon Lords, which grants him powerful magical armor and a loyal drake companion. As he progresses to higher levels, Gareth's fame spreads across Valoria, and he is called upon to lead the King’s Vanguard against the rising threat of the Dark Legion, an army of corrupted beings led by the sinister Warlord Morgath. 

The climactic showdown with Morgath in the heart of the Forsaken Citadel tests Gareth to his limits. With the fate of Valoria hanging in the balance, Gareth and his companions unleash their full might in an epic battle. Gareth’s unwavering courage and strategic acumen lead to Morgath’s downfall, securing peace for the realm.  

Crowned as the Level 100 Lord Marshall, Gareth becomes a legendary figure, his deeds sung in taverns and recorded in tomes. He ushers in an era of prosperity and defense, ensuring that Valoria remains safe from the shadows that once threatened it. 

The LitRPG writer now has their route plotted, but what if they need a little more detail about certain legs of the journey? 

ChatGPT4o Prompt: Write a 100 word summary of how Gareth and his companions free Stormhaven from the necromancer and add a moment where all seems lost for our heroes. 

ChatGPT4o Output (in under 2 seconds): Gareth and his companions, Elara and Thrain, arrive at Stormhaven, a city under the dark grip of a necromancer. Using Elara’s powerful spells, they breach the city’s magical barriers. Thrain’s brute strength clears the undead minions blocking their path, but in the necromancer’s lair, a devastating attack leaves Elara gravely wounded and Thrain immobilized. All seems lost until Gareth, drawing on his inner resolve, rallies his friends for one last stand. With renewed vigor, Gareth’s blessed sword pierces the necromancer’s heart, breaking the dark curse. The grateful citizens hail Gareth as their hero, bestowing upon him the title of Captain. 

Now what if our fantasy writer prefers to having written or writing, or perhaps they wish to put out the next instalment in their fantasy series as quickly as possible? 

ChatGPT4o Prompt: In the style of David Eddings, write the moment where Gareth kills the necromancer. 

ChatGPT4o (In under 5 seconds): The air in the necromancer’s lair was thick with the stench of decay, an oppressive darkness pressing in from all sides. Gareth’s heart pounded in his chest as he stood over Elara’s prone form, her breath shallow, her face pale. Thrain, pinned beneath a heap of rubble, groaned in pain. The necromancer loomed before them, cloaked in shadows, his eyes burning with malevolent glee. 

“You’ve fought valiantly, little knight,” the necromancer hissed, his voice a serpentine whisper. “But this is where your story ends.” 

 Gareth tightened his grip on his sword, its blade glowing faintly with the ancient magic it held. Desperation surged through him, a fierce determination igniting in his veins. He couldn’t let it end here, not like this. With a roar, he launched himself at the necromancer, his movements driven by a blend of fear and fury. 

The necromancer’s dark magic lashed out, tendrils of shadow clawing at Gareth’s armor, but he pressed on, each step fueled by the memories of those he sought to protect. He saw the faces of his friends, his village, and the innocent people of Stormhaven. He would not fail them. 

  

“Your magic is strong,” Gareth growled through gritted teeth, “but it lacks one thing—righteousness.” 

With a final, mighty swing, Gareth’s sword cleaved through the shadows, its enchanted edge meeting the necromancer’s heart. There was a blinding flash of light as the dark magic was dispelled, the necromancer’s scream echoing through the chamber. Gareth felt the surge of power, the essence of the evil mage dissipating like mist in the dawn. 

 The necromancer crumpled, his form disintegrating into dust. The oppressive darkness lifted, replaced by a soft, golden glow. Gareth sank to his knees, exhausted but victorious. As the echoes of battle faded, he turned to his friends, relief flooding his heart. 

 “It’s over,” he said, his voice trembling with the weight of their hard-won victory. “We’ve done it.” 

I read quite a bit of David Eddings in my youth and the above is not dissimilar in style. This is an excerpt from David Edding’s Pawn of Prophecy. 

 

Halfway up he saw the flaw in his plan. There were no side passages on the 

stairs, no way to escape and no place to hide. He had to get to the top quickly or 

chance discovery and capture, or even worse. 

 

"Boy!" a shout came from below. 

 

Garion looked quickly over his shoulder. A grim-faced Cherek in mail and 

helmet was coming up the stairs behind him, his sword drawn. Garion started to 

run, stumbling up the stairs. 

 

There was another shout from above, and Garion froze. The warrior at the top 

was as grim as the one below and wielded a cruel-looking axe. He was trapped 

between them. Garion shrank back against the stones, fumbling for his dagger, 

though he knew it would be of little use. Then the two warriors saw each other. 

With ringing shouts they both charged. The one with the sword rushed up past 

Garion while the one with the axe lunged down. 

 

Even more compelling is that ChatGPT4o produced its text with no grammatical, punctuation or spelling errors. It is immediately ready for publication. 

However, only since the 13th of May 2024 has it been possible to construct an entire novel with a ChatGPT tool. Game writing and short story writing have already been impacted by ChatGPT3.5 and 4. Short scripts, flash fiction and poems form the bulk of a game writer’s job on all but the most story-driven of games. I have personally written thousands of flavour texts for in-game assets over the years. I wrote the following area flavour texts for Path of Exile in 2012. 

 

Mudflats - “Mud and air seethe with warped life.” 

The Ship Graveyard - “Whispered agonies of the marooned dead.” 

Fellshrine Ruins - “The ground aches with desecration. The rain tastes of blight.” 

Ancient Pyramid - “Two thousand years of regret.” 

The Docks - “Life pales in the wake of progress.” 

The Sceptre of God - “The taller the tower, the greater the fall.” 

 

On average, each flavour text would take twenty minutes to write. The creative process involves seeing the area in-game, then imagining how it would genuinely feel to stand in that place. What are they sounds and smells that aren’t rendered in the game. Does the place have a certain energy to it? Does evoke particular emotions? Then I would quickly write a few versions, pick the most evocative and edit it. The above set represents two hours of work, although is informed by many more hours of immersion in the game’s production, and years of reading fantasy fiction, playing games with a fantasy context and viewing fantasy TV shows and films. 

This is what ChatGPT4o was able to give me in less than a minute, and most of that time was spent by me thinking up the prompt. 

 

ChatGPT4o Prompt: Name the favourite weapons belonging to Gareth, Elara and Thrain, and write three flavour texts each for these weapons. 

 

ChatGPT4o Output: 

Gareth’s Favorite Weapon: Dawnbreaker 

1. Forged in the ancient fires of the Dragon Lords, Dawnbreaker’s blade gleams with the light of a thousand suns, cutting through darkness like the first dawn of creation. 

2. With each swing, Dawnbreaker sings a song of valor, its enchanted edge thirsting for justice and the defeat of all that is evil. 

3. Legends say that Dawnbreaker was crafted from a fallen star, its celestial power granting the wielder unmatched strength and courage. 

  

Elara’s Favorite Weapon: Whisperwind 

1. Whisperwind’s staff hums with arcane energy, its sapphire crystal core capturing the essence of the wind, enabling Elara to command the elements with a mere whisper. 

2. Carved from the ancient heartwood of the Elder Tree, Whisperwind channels Elara’s magic with precision and grace, amplifying her spells to devastating effect. 

3. In the quiet of night, the runes etched along Whisperwind’s length glow faintly, a testament to the ancient elven magics that flow through it. 

  

Thrain’s Favorite Weapon: Earthshaker 

1. Earthshaker, the mighty warhammer, is said to contain the fury of the mountains themselves, each strike causing the ground to tremble and foes to scatter. 

2. Forged in the deepest forges of the Dwarven Halls, Earthshaker’s head is unyielding, its surface etched with runes of power that amplify Thrain’s strength. 

3. When Thrain raises Earthshaker in battle, a battle cry on his lips, enemies know that the unrelenting force of the earth itself stands against them. 

 

The above flavour texts are passable, of a quality that I would expect from a competent junior game writer. Interestingly, ChatGPT4o has made the same mistake junior game writers commonly make. The functionality of the item has been brought to the fore and embellished with touches of history and embodied feeling. However, this is not the most efficient use of flavour text. Comments on function only mimic that which the game already expressions through mechanics and special effects. Flavour text should evoke less easily represented elements like nostalgia, legacy, historical potency and inherited status. To achieve this, I gave ChatGPT4o the same sort of instruction that I have given many of the junior writers I’ve worked with. 

 

ChatGPT4o Prompt: Rewrite, and this time don't mention the function of each weapon. Focus on its history and write the flavour text in a poetic manner. 

ChatGPT4o Output: 

Gareth’s Favorite Weapon: Dawnbreaker 

1. Forged in the ancient fires of the Dragon Lords, Dawnbreaker’s blade gleams with the light of a thousand suns, a beacon of hope against the encroaching darkness. 

2. Dawnbreaker, its edge kissed by the first light of creation, sings a song of valor, a hymn to the brave souls who wield it. 

3. Legends whisper that Dawnbreaker was crafted from a fallen star, its celestial power a testament to the heavens' eternal vigilance. 

  

Elara’s Favorite Weapon: Whisperwind 

1. Whisperwind’s staff hums with arcane energy, its sapphire crystal core capturing the essence of the wind, a gift from the ancient spirits of the sky. 

2. Carved from the heartwood of the Elder Tree, Whisperwind bears the touch of the old magic, its runes a silent prayer to the forest’s wisdom. 

3. In the quiet of night, Whisperwind’s runes glow faintly, their light a soft reminder of the timeless elven magics that reside within. 

  

Thrain’s Favorite Weapon: Earthshaker 

1. Earthshaker, the mighty warhammer, holds the fury of the mountains, its runes a testament to the enduring strength of the dwarven halls. 

2. Forged in the deepest fires of the Dwarven Halls, Earthshaker’s head is a monument to the unyielding spirit of its creators, etched with tales of old. 

3. When Thrain lifts Earthshaker, the ancient runes upon its surface glow, a promise of the earth’s unwavering power and the legacy of the dwarves. 

 

That’s more like it. This second batch of flavour texts are game-ready, and the entire process took less than ten minutes. If I were a Producer, given the choice of paying a senior writer for an hour to write the above flavour texts versus paying them for ten minutes of prompting ChatGPT4o, how can I possibly justify the former to the studio’s CFO? 

 

While the impact of LLMs on narrative for games is anecdotal, not yet well documented, the impact on the short story publication is. 

On Monday, the editor of the renowned sci-fi publication Clarkesworld Magazine announced that he had temporarily closed story submissions due to a massive increase in machine-generated stories sent to the publication. - Arstechnica, 22/2/2023 

 

As mentioned already, the release of ChatGPT4o has now enabled ChatGPT users to tackle larger works of fiction, such as novels. Why? 

ChatGPT4 users were limited by a context window of 4k to 8k, equating to approximately 3000 to 6000 words. This severely hobbled the writer because the tool could not keep an entire novel in its ‘mind’, and therefore could not refer to characters and plots much earlier in the story. 

To get around this limitation, many authors turned to Sudowrite, another generative writing tool that offered a much larger context window by accessing multiple LLMs. 

From the Sudowrite website: “We use over two dozen AI models, including Claude 2 by Anthropic, multiple open models, and several variants of GPT-3.5, and GPT-4, transformer models created by OpenAI.” 

Now ChatGPT4o has a context window of 128,000 tokens (approximately 100,000 words) which is a workable size for most genre fiction novels. The tool can ‘remember’ every detail of a novel from start to finish. 

However, if the tool can generate the storyline according to prompt supplied by genre tropes, and then generate the prose based on the storyline, where has the human creativity gone? 

The author, podcaster and technology advisor to the Alliance of Independent Authors, Joanna Penn, argues that the creativity remains in the curation and editing of the text. She states that “I am not someone who just outputs words in text or audio format. It is the purpose behind the words that matter and the connection I make with other humans that has an impact.” 

To me this is a very product-focused approach, one that concentrates on output at the expense of the process, one that ignores AI’s influence on the creative process at the potential detriment of the product it hopes to achieve. 

My counter argument is that, in using an AI text generation tool, creativity has been all but been replaced by something else; reactive productivity. 

What is reactive productivity? 

In its simplest form, reactive productivity is the act of producing outputs in response to external stimuli. In this context, productivity is primarily externally motivated. The producer prompts the LLM and then reacts to the output with further prompts or by taking that output and manually editing it until it’s fit for purpose. This is exactly the process experienced when generating the flavour texts above. 

In many respects, the reactive productivity process mirrors the scientific process. You conduct and experiment or observe a natural phenomenon, you record your findings and then edit and peer review them in preparation for publication. 

Of course, at the risk of stating the obvious, the scientific process and the creative process are worlds apart and produce very different results. Science is about gaining knowledge of what already exists. Creativity is about bringing something new into existence. 

In Creativity: Flow and the Psychology of Discovery and Invention, Mihaly Csikszentmihalyi suggests that “the best program is one that makes the organism feel good when something new is discovered, regardless of its present usefulness” (page 108). He goes on to says that “Groups that supported and protected creatives were better prepared to face unpredictable situations.” 

In essence, Csikszentmihalyi is claiming that the creative process is an important survival mechanism that helps us to weather “the slings and arrows of outrageous fortune”. 

Science aids us with predictable threats. Creativity aids us with threats that science has not yet been able to predict. 

So by replacing creativity with reactive productivity, the practitioner is at risk of undermining a vital survival strategy. A tool like ChatGPT4o is effective because it can use existing material to produce predictable outcomes that are fit for purpose with the minimum of human intervention. Is this not the epitome of the “observe and report” process in science? 

Yet for all its similarity to the scientific method, why does reactive productivity feel like creativity to so many of its adopters? By looking at Mihaly Csikszentmihalyi’s definition of “flow”, I’ve observed that the experience of interacting with an LLM mirrors the ingredients of the flow state, almost like an alien mimicking human appearance and behaviour in The Bodysnatchers (1978). 

To unpack how engaging with an LLM mimics the enjoyability of creativity, we’ll run down a check list of Mihaly Csikszentmihalyi’s nine elements that he proposes supports a flow state (p111-112). 

1. “There are clear goals every step of the way.” 

The act of prompting is the act of goal setting. The clearer that goal is, the more accurately an LLM can deliver to that goal. And this happens at every step of the LLM interactive process. As seen with my examples above, prompts start more general and grow more specific with each iteration. There is little opportunity to feel lost or aimless. Interacting with an LLM demands that you take a goal-oriented approach to creativity. The conception of the outcome dominates the experience of the process. 

1. “There is immediate feedback to one’s actions.” 

The response time for the leading LLMs numbers in the seconds. ChatGPT4o’s response time, even for large blocks of text, is so fast that it feels like instant gratification. To give a personal example, my professional average for writing fiction is 500 words per hour. By contrast, ChatGPT4o can produce the same amount of text in under 5 seconds. The patient toil of sixty minutes at the computer has literally been replaced by the wave of a magic wand. Hence the term “automagically” that has become popular in the tech world. 

1. “There is a balance between challenges and skill.” 

No matter one’s literary skill level, an LLM will produce a gratifying result. For instance... 

My prompt to ChatGPT4o: rite me a pom abut 3 blind mices 

ChaptGPT4o’s output (less than two seconds): Certainly! Here's a playful poem about three blind mice: 

 

Three blind mice, out on a spree,   

Wandered the fields so merrily.   

With twitching whiskers and tails in line,   

They set out to explore, come rain or shine.   

  

In the barn, they danced in hay,   

Laughing and singing the day away.   

Though they couldn't see the world around,   

Their hearts were light, their joy profound.   

  

They bumped and stumbled, giggled with glee,   

Tumbled through grass as wild as can be.   

With tiny paws and ears so keen,   

They found adventure in every scene.   

  

Through fields of daisies, they took a chance,   

Swaying together in a joyful dance.   

In the world unseen, they found their way,   

Three blind mice, brightening the day.   

  

So here's to the mice, so brave and bold,   

In a world of dreams, their stories told.   

For though they’re blind, they clearly see,   

The beauty in life’s sweet melody.   

  

At the other end of the scale, the LLM struggles to rise to the challenge presented by a skilled and experience writer seeking a more refined and sophisticated output. 

 

My prompt to ChatGPT4o: Write a four line verse that expresses the balance of pain and pleasure in life, using alliteration, assonance and an extended metaphor about the four seasons. 

 

ChatGPT4o’s output:  

 

Spring's sweet whispers waltz through weary winters,   

Summer's sunlit laughter lingers, lightly swaying,   

Autumn's amber tears tumble tenderly,   

Winter's weary embrace warms, quietly cradling change.   

  

It’s not a great poem. While the LLM succeeded with the alliteration and the assonance, the metaphoric comparisons make little sense. How does “sunlit laughter” sway? This poem would also be impossible to edit into shape. While attempting to fix the metaphors, you would disrupt the soundscape. 

Yet, for amateur and junior writers, ChatGPT4o delivers a highly gratifying experience, arguably delivering material that exceeds the skill level of the author, as if the casual jogger has been cybernetically enhanced so they can run in their next local marathon. 

 

1. “Action and awareness are merged.” 

Engaging with an LLM is an immersive experience, one that can be understood from a game design perspective. Games that succeed in being immersive are games that have a strong and enjoyable game loop. The game loop is formed by the gameplay mechanics. For example, in a platformer, the player jumps their player–character from one platform to the next. It requires full concreation to do this as one weighs the distance against the player-character’s jumping ability. With practice this judgment becomes internalized, subconscious, allowing the conscious mind to consider other things. But this is where difficulty comes in. Good platformer design introduces moving platforms, enemies that need to be avoided or destroyed, traps and navigational hazards, and expansions and refinements of the player-characters abilities. These additional factors place greater demands on the player’s cognition, thereby keeping action and awareness merged within a sense of immersion. 

While the LLM loop does not have scaling difficulty like a platformer video game, it offers the user the opportunity to raise the challenge bar through iteration on the LLM’s output. The user can scale their own difficulty. And they can ask the LLM to help them do this. 

My input to ChatGPT4o: suggest ways that the above poem could be improved 

ChatGPT4o’s output: 

To enhance the poem and emphasize the balance of pain and pleasure using alliteration, assonance, and the metaphor of the seasons, you might consider the following suggestions: 

1. Strengthen the Imagery: Use vivid images that capture the essence of each season's dual nature. 

1. Enhance the Metaphor: Make the connection between the seasons and life experiences more explicit. 

1. Balance Sound Devices: Ensure that alliteration and assonance are evenly distributed for a smooth rhythm. 

Here's a revised version incorporating these suggestions: 

 

Spring's soft whispers waltz through winter's weary woes, 
Summer's sunlit laughter lingers, light yet fleeting, 
Autumn's amber tears tumble, tenderly embracing change, 
Winter's weighty cloak cradles quiet renewal. 

 

Explanation of Improvements: 

• Spring: Emphasized the contrast between new beginnings and the remnants of winter's challenges. 

• Summer: Highlighted the fleeting nature of joy with "light yet fleeting." 

• Autumn: Focused on the transition, using "embracing change" to signify both loss and acceptance. 

• Winter: Added "weighty cloak" to convey both the heaviness and protective nature of winter. 

These adjustments aim to create a more vivid and balanced depiction of the complex interplay between pain and pleasure throughout the seasons. Let me know what you think or if you'd like further changes! 

And thus the loop continues. The user can now refine their prompts based on the LLM’s suggestions, just like the platformer gamer must adjust their distance and timing judgements to the now moving platform. 

 

1. “Distractions are excluded from consciousness.” 

According to Csikzentmihalyi, “Flow is the result of intense concentration on the present, which relieves us of the usual fears that cause depression and anxiety in everyday life.” (p112) 

Through a combination of instant gratification and the immersive prompt-output iteration loop, the LLM experience can completely dominate the user’s cognitive load, preventing other considerations from intruding. This is where the ‘fun’ of the LLM experience finds commonality with the entertainment of the video game. The sheer business of both activities temporarily banish the concerns of everyday life. 

1. “There is no worry of failure.” 

As established with our consideration of Element 3, the balance of skill and challenge, LLMs offer a ‘safe’ space where failure is nearly impossible, at least for the amateur and junior professional level. In fact, with the output potentially exceeding the skill of the inputter, the LLM may consistently exceed realistic expectations, thereby forging an illusion of resounding success. It is only when an experienced creative uses the LLM that the wheels can fall off. This is because “realistic expectation” is now based on an in-depth understanding of the creative process, what it takes to produce ‘good’ work, and therefore what the qualities of ‘good’ work actually are. Failure, the production of a poor quality output, is certainly a concern for the professional creative. But for the beginning, success is all but guaranteed. 

1. “Self consciousness disappears.” 

1. “The sense of time becomes distorted.” 

I’ve merged the above elements because Csikszentmihalyi is at risk of repeating himself. Self consciousness disappears when anxiety is removed from the picture by sufficient cognitive load and the removal of the fear of failure. 

As for the sense of time… “Meanwhile, the more difficult the concurrent task, the more they tend to underestimate the time. These distortions are even more pronounced when observers are walking. Estimation biases and uncertainties change differently with durations depending on the task, consistent with a fixed relative uncertainty. Our findings show that cognitive and motor systems interact non-linearly and interfere with time perception processes, suggesting that they all compete for the same resources.” (Castelotti et al, 2022) 

Interacting with LLM distorts the user’s sense of time because it keeps them mentally busy, just as a well-designed video game does. The old saying, “time flies when you’re having fun”, is a little misleading. In fact, time flies when your cognition is so loaded that you have no attention to spare for accurate appreciation of time. But it’s not quite as catchy. 

1. “The activity becomes autotelic.” 

“Autotelic” is Greek for something that is an end in itself. As opposed to exotelic, which is doing something because of some external push or pull factor, the autotelic experience is one that we enjoy so much that we will do it for no other reason than it is a pleasant experience. The previous eight elements contribute to this final element by making the activity pleasurable. Therefore, by fulfilling the criteria of the elements one to eight, interacting with an LLM automagically becomes an autotelic experience. 

Therefore, since using ChatGPT4o fulfills all of the experiential elements of Mihaly Csikszentmihalyi’s “Flow State”, does that mean it will help to facilitate a mastery of writing by encouraging the author to put in Malcolm Gladwell’s “10000 hours”? 

No.  

The danger is that the beginner writer can prompt outputs that mimic the work of skilled amateurs, even junior professionals, without actually experiencing and learning from the writing process. Because the LLM process is automagical, there is a cognitive disconnect between input and output. How can the inputter judge the qualities of the output when they don’t understand how that output was produced? 

At this point, this argument or similar is often made. “You don’t need to know how a car works to drive it”. But this is a false comparison. The competitive motorsports driver does need to know how their car works. By driving to the mechanical limitations of their car they can optimize their pace around the track by adapting their driving techniques, playing to the car’s strengths and mitigating its weaknesses. The more accurate comparison is, “You don’t need to know how a car works to take a taxi from A to B”. Very true. And you don’t need to know how to drive, either. 

Without that hands-on experience of the creative process we are left with reactive productivity. The user has an outcome in mind, let’s say a humorous poem about three blind mice. They prompt ChatGPT4o and receive a poem. They react by asking the LLM what would make the poem better. They then write a prompt, based on ChatGPT4o’s suggested best practices for writing a poem about mice. They receive another version of the poem to react to. And so the loop continues until they have produced a poem they feel happy with. 

Apart from the user’s initial idea of the humorous poem about three blind mice, there is nothing that is coming solely from the poet. The LLM is producing the words of the poem. The poet is simply reacting to them.

This is an example of Creative Direction? Yes. Is it Creative Writing. I don’t think so.