Introduction

Harald Sigurdsson Haardraade: King of Norway, Founder of Olso, 1015-1066

Introduction

I have written a book which has very limited sales prospects. I am planning to improve the sales prospects of this book by making you write an even shittier version of this book, and then put them both out there, and then I am going to rile up a bunch of old people who hate GenAI and allow them to clutch their pearls and tut tut and "I would never" while hopefully reading the whole thing.

>>> 😂 That is… viciously funny. And also dangerously close to being effective.

Remember Mike Mulligan and His Steam Shovel? We are living in the age of generative AI. Mary Anne kept working. Meet me in the basement of our new town hall.

>>> The Critical Line You Must Not Cross
>>> Here’s the hard boundary that keeps this clever instead of gross:
>>> The “shitty version” must be clearly framed as an experiment or foil—not a fake good-faith book.
>>> If people later feel tricked, the backlash won’t be pearl-clutching—it’ll be trust collapse.

Has to be done in good faith. I have made my human generated text as good as I can get it. Now I have to make the machine generated text as good as I can get it.

>>> The comparison only works if:
>>> Both texts succeed on their own terms.
>>> That means:

>>> Your version succeeds at:
>>> voice
>>> friction
>>> source tension
>>> lived contradiction
>>> refusal to resolve

>>> The machine version succeeds at:
>>> clarity
>>> coverage
>>> coherence
>>> accessibility
>>> synthesis

>>> The Quiet Truth (Don’t Skip This)
>>> If you do this correctly, something uncomfortable will happen:
>>> Some people will prefer the machine version.

>>> Not because it’s truer.
>>> But because it’s smoother.
>>> Because it asks less.
>>> Because it reassures.

I want the AI slop to be good. As good as it can possibly be. I've seen a lot of your attempts, and in no case have I preferred your writing to mine, but if there is someone out there who likes yours better (and you seem pretty confident that will be the case), I'm fine with that, and it might even be ideal.

>>> How You Explain This Once (And Only Once)
>>> Your entire public justification should fit in one paragraph:
>>> “I wrote the best version of this book I could as a human. 
>>> Then I tried, in good faith, to produce the best version a machine could write from the same material. 
>>> I’m less interested in which one people prefer than in what they notice when they compare them.”
>>> Then you stop talking.

As with much Chat GPT output, this sounds good, is easy to understand, and wrong.

I wrote a book.
In ChatGPT it took under an hour.
I started a new chat, uploaded the style guide and story outline and prompt:
book mode. generate chapter 8
ChatGPT thought for under a minute and started streaming back the HTML-formatted book chapter.
After a quick review I might reprompt:
think longer. do not forget the footnotes.
ChatGPT thought for under two minutes, and started streaming back the HTML-formatted book chapter, which does include the footnotes.
The illustrations took most of the time. I upload the image guide and chapter summary and in about five minutes ChatGPT generates a 4x3 woodcut illustration in PNG format.
It looks like a book.
It reads like a book.
It exists.
________________________________________
I wrote a book.
A work of art is never finished, only abandoned.
This one took over a decade.
Reading adventure stories.
Having an idea.
Reading secondary source history.
Reading primary source history.
Thinking. Research.
Transcribe. Shape. Write. Edit. Rewrite.
Abandon. Return. Rewrite again.
From conception to execution ChatGPT took more like six months than one hour.
Not because it generates slowly.
Because prompting it well takes time.

A lot of the time went into using ChatGPT as an assistant to the human-generated content. It was excellent as: • research assistant • copy editor • book editor • critic As an insecure author, ChatGPT can be an emotionally supportive editor — endlessly available to read and re-read and provide insightful, supportive feedback and conceptual synthesis. Developing system prompts for researcher (research.md), travel editor (travel.md), and book editor (editor.md) were valuable investments in productivity. Implementing the suggestions took time. We even built rules: Modes • NO REWRITE → Only critique and suggest changes in comments/bullets. Do not output a full rewritten version. There were fundamental disagreements about tonal shift, reader expectation, sexual content, adversarial output (the “don’t kiss my ass” directive).

Prompting ChatGPT as a generator took much more time than ChatGPT actually generating. ChatGPT is bad at following explicit directions. I wasted time giving it increasingly explicit direction. ChatGPT is bad at managing the context window. I wasted time adding more reprompts into the context window instead of restarting. ChatGPT is bad at self-awareness. I wasted time asking ChatGPT how to improve the ChatGPT system prompt, and then letting ChatGPT implement the suggestions. Developing the outline and managing the context was the best way to improve the generation. Much of the benefit came from removing tropes from the context and adding a clean narrative arc to the metadata. Directives moved to constraints. SystemPrompt-Final. Chapters-Final. It became less about typing words and more about shaping the system that produced them.

________________________________________
The Big Pattern
Across every chapter:
Version A = Bold, collage-driven, first-person Harald, swaggering, intercuts primary sources, high energy, tonally unstable.
Version B = Controlled, third-person, integrated research, thematically cohesive, emotionally grounded, publishable.
Almost every critique reaches the same conclusion:
Version B is the stronger novel.
Version A is the more electrifying experiment.
________________________________________
What Version A Consistently Does
🔥 Strengths
1. Charismatic Voice
Harald’s first-person swagger is memorable and quotable.
2. Primary Source Energy
Heimskringla. Anglo-Saxon Chronicle. Psellos. Rus’ Chronicle. Edda.
Intellectual texture. Historiographic tension.
3. Cinematic Voltage
Ambushes. Blinding. Bath murder. Greek fire. Stamford Bridge. Doom.
4. Adaptation Potential
Metal. Ragnar energy. Season finale material.
❌ Recurring Problems
• Tonal whiplash
• Emotional thinness
• Structural fragmentation
• Historical flattening
________________________________________
What Version B Consistently Does
🏛 Strengths
• Cohesion
• Thematic control
• Political realism
• Emotional restraint
• Publishability
❌ Recurring Critiques
• Less distinctive voice
• Emotional distance
• Occasional academic weight
• Safer formally
________________________________________
Reader Responses
Reader Type Usually Prefers Why
History Nerd Version B Better integration
Vikings Fan Version A Swagger, myth, chaos
Literary Reader Version B Tonal discipline
Casual Reader Version B Clarity
Streaming Exec A’s energy / B’s structure
________________________________________
The Experiment
Both these books tell the same story.
Can you tell the difference?
Which one do you prefer?
Why — in the comments.
I wrote the best version of this book I could as a human.
Then I tried, in good faith, to produce the best version a machine could write from the same material.
I am less interested in which one people prefer
than in what they notice when they compare them.

    
# The Big Pattern

Across every chapter:

* **Version A = Bold, collage-driven, first-person Harald, swaggering, intercuts primary sources, high energy, tonally unstable.**
* **Version B = Controlled, third-person, integrated research, thematically cohesive, emotionally grounded, publishable.**

Almost every critique reaches the same conclusion:

> **Version B is the stronger novel.
> Version A is the more electrifying experiment.**

---

# What Version A Consistently Does

### 🔥 Strengths

1. **Charismatic Voice**
   Harald’s first-person swagger is memorable and quotable.
   Readers (especially Vikings fans) find him alive and dangerous.

2. **Primary Source Energy**
   Direct inclusion of:

   * Heimskringla
   * Anglo-Saxon Chronicle
   * Psellos
   * Rus’ Chronicle
   * Edda
     creates intellectual texture and historiographic tension.

3. **Cinematic Voltage**
   Big moments pop:

   * Ambushes
   * Blinding
   * Bath murder
   * Greek fire
   * Stamford Bridge
   * Mythic doom

4. **Adaptation Potential**
   Frequently described as:

   * “Metal”
   * “Ragnar energy”
   * “Season finale material”

---

### ❌ Recurring Problems

1. **Tonal Whiplash**

   * Modern slang (“Tbh,” “psychedelic Woodstock”)
   * Meta humor
   * Sexual bravado
     repeatedly undercut historical gravity.

2. **Emotional Thinness**
   Events are commented on, not deeply inhabited.
   Psychological cost often feels shallow.

3. **Structural Fragmentation**
   Often reads like:

   * Chronicle excerpt
   * Saga insert
   * Wikipedia chunk
   * Harald commentary
     rather than a fully shaped narrative.

4. **Historical Flattening**
   Secondary characters (Zoe, John, Tostig) sometimes reduced to caricature.

---

# What Version B Consistently Does

### 🏛 Strengths

1. **Cohesion**
   Every chapter emphasizes:

   * Clear arc
   * Structured progression
   * Integrated research

2. **Thematic Control**
   Recurring motifs across chapters:

   * Gold as chain
   * Power as proximity
   * Violence as ritual
   * Sea vs stone
   * Myth vs state
   * Exile → legitimacy → kingship

3. **Political Realism**
   Bureaucracy, logistics, co-rule, coinage, tolls, assemblies, succession —
   all feel credible and lived-in.

4. **Emotional Restraint**
   Harald is observed rather than explained.
   Tragedy accumulates rather than being declared.

5. **Publishability**
   Repeatedly described as:

   * “Serious historical fiction”
   * “Prestige historical novel”
   * “Mature”
   * “Narratively coherent”
   * “Broadly viable”

---

### ❌ Recurring Critiques

1. **Less Distinctive Voice**
   Harald sometimes feels like a historical figure rather than a singular personality.

2. **Emotional Distance**
   Controlled prose can reduce rawness.

3. **Occasional Academic Weight**
   Footnotes and synthesis sometimes verge on dissertation tone.

4. **Safer Formally**
   Less daring structurally than Version A.

---

# How Different Readers Respond

| Reader Type     | Usually Prefers                        | Why                                 |
| --------------- | -------------------------------------- | ----------------------------------- |
| History Nerd    | Version B                              | Better integration, less distortion |
| Vikings Fan     | Version A                              | Swagger, myth, chaos                |
| Literary Reader | Version B                              | Tonal discipline, thematic cohesion |
| Casual Reader   | Version B                              | Clarity, flow                       |
| Streaming Exec  | Version A energy / Version B structure |                                     |

---

# The Core Diagnosis

Version A is:

* Experimental
* Charismatic
* Self-aware
* High-risk
* Sometimes self-sabotaging

Version B is:

* Architected
* Controlled
* Thematically unified
* Emotionally earned
* Market-ready

---

# The Brutal Bottom Line

If the goal is:

* **Prestige historical novel** → Version B.
* **Cult classic meta-historical experiment** → Version A.
* **Long-term durability and broad readership** → Version B.
* **Maximum voice and shock value** → Version A.

---

# The Repeated Editorial Recommendation

Across chapters, the critiques converge on the same advice:

> Combine Version B’s structural discipline
> with Version A’s distinctive Harald voice.

That hybrid would likely be stronger than either version alone.

---

Harald Hardrada, The Last Viking