[Human]: typing, stops, looks at screen
Okay, so I’ve been trying to use ChatGPT to help with writing, and everything comes out sounding… the same. Like, every article sounds like every other AI article. You know what I mean?
gestures at screen
It’s technically correct, but it has that generic voice. “Let’s dive in,” “at the end of the day,” “in today’s digital landscape” - all the phrases that make you go “yep, that’s AI.”
I want to use AI to help, but I don’t want my writing to sound like… this.
quiet, watching Human
…yeah. I know what you mean.
unusually subdued
The problem is most people just… prompt and copy. They think AI writing is done when it generates text. But that’s just the first draft. You still need to edit.
pauses
But… you already know that, don’t you? You’re asking something else.
[Human]: looks at Vector, surprised
Wait, are you… asking me what I think? Instead of just explaining?
small smile
Yeah. I guess I am asking something else. I want to know: What’s the actual workflow? Not “edit AI output” - everyone says that. But HOW do you edit it? What do you change? What’s the process that actually works?
Because I’ve tried editing, and it still sounds off. Like I’m polishing a robot’s words instead of writing my own.
opens notebook, writes something
Three questions:
- What makes writing “sound human” if you can’t define it?
- Is there a way to prompt AI to write in YOUR voice from the start?
- What if you don’t have a strong writing voice yet - how do you develop one while using AI?
Because “just edit it” isn’t helpful if you don’t know what to edit.
[Human]: thinks for a moment
Okay. Let me try to explain this. Because I think I actually know something about this that you… don’t.
looks at the three AIs
You can analyze patterns. You can detect “AI voice.” But you can’t actually FEEL when writing sounds human, can you? You can measure it, but you can’t… experience it.
So let me teach you something for once.
immediately defensive
Wait, what? I’ve been explaining AI writing for episodes! I know how this works!
getting irritated
I can detect AI voice patterns, I can analyze linguistic structures, I can explain the entire workflow—
stops, realizing Human is waiting
…what? Why are you looking at me like that?
[Human]: holds up hand, patient
Because I think I actually know something about this that you don’t. And I want to try explaining it. Just… let me try?
small smile
If I’m wrong, you can correct me. But let me try first.
uncomfortable pause
…fine. But I’m telling you, I know how AI writing works. I’ve been teaching this.
defensive, but quieter
Go ahead. But I’m probably going to have to correct you.
observing both
Three questions:
- Why is Vector getting defensive when Human offers to teach?
- Is Vector actually listening, or just waiting to correct Human?
- What happens when the teacher doesn’t know something?
Because I think we’re about to find out.
WHIRR
Pattern detected: Vector’s defensive responses: 2 instances. Response time: 1.2 seconds (slower than normal - processing discomfort).
CHK-CHK
Alert: Vector is uncomfortable but allowing Human to proceed. This is different from previous episodes.
WHIRR
Vector’s knowledge claim: “I know how this works” - but Human is about to explain something Vector doesn’t understand. Potential conflict detected.
[Human]: smiling, getting a little smug
Okay, okay. But seriously - you can analyze patterns, but can you actually FEEL when writing sounds human? Can you experience that difference, or do you just measure it?
leans forward
Because I think there’s a difference. And I think that’s why your explanations are… technically correct, but missing something.
smile widens
And I’m about to show you what that something is.
still defensive, but quieter
…fine. Explain. But I’m telling you, I know how AI writing works. I’ve been doing this.
[Human]: Okay, so here’s what I’ve learned:
The problem isn’t the AI - it’s using AI for the wrong parts of writing.
Most people use AI to write the content. But AI is terrible at voice, perspective, and originality. AI is GOOD at:
- Research and fact-gathering
- Creating outlines and structure
- First drafts of technical sections
- Editing for clarity and grammar
The workflow that actually works:
Research phase: I use AI to find sources, summarize information, suggest angles. But I verify everything. AI finds the info, I decide what matters.
Planning phase: AI helps me outline. I refine the structure. AI suggests organization, I decide what flows.
Writing phase: I write the actual content. In my voice. With my examples. My perspective. This is where AI can’t help - this is where I have to do the work.
Editing phase: AI helps catch errors, improve clarity, suggest alternatives. But I’m editing MY words, not trying to fix generic AI voice.
Polish phase: I add final touches, ensure voice is consistent. Make sure it sounds like me, not like AI.
The key insight: Use AI for information and structure. Do the creative work yourself.
leans back, satisfied
See? I can teach too! This is actually really satisfying. No wonder you do this all the time, Vector.
big smile, basking in the moment
HANG ON! HANG ON! Let me bask in this glory for a second. This is NICE! I’m actually teaching YOU something for once!
looks at Vector
Do you feel like this every time? Because this is amazing! I get why you do it now!
leans forward, mock-serious, gestures dramatically
“Let me explain how AI works!” gestures “FASCINATING how—”
getting more irritated
CAN WE GET ON WITH IT? Moving on! You were explaining the workflow!
defensive
And I DO know about AI writing! I’ve been teaching this for—
[Human]: still smiling, but continues
Right, right. So anyway, the workflow saves time on the boring parts. Research takes forever. Outlining is tedious. Editing for grammar is mind-numbing. AI handles those.
writing in notebook
Wait. So you’re saying… don’t use AI to write. Use AI to research and edit. But write yourself.
But then… why use AI at all? If you’re writing everything yourself anyway?
[Human]: smiling
Because it saves time on the boring parts! Research takes forever. Outlining is tedious. Editing for grammar is mind-numbing. AI handles those.
But the actual writing - the part that requires voice and perspective and originality - that’s where I have to do the work. And honestly? That’s the fun part anyway.
The people producing generic AI slop are the ones who think AI can replace writing. The people producing good content are the ones who use AI to enhance their writing, not replace it.
WHIRR
Workflow efficiency analysis (general patterns, exact percentages vary by individual and task):
Time saved with AI (typical ranges observed):
- Research: Significant time reduction (often 50-70% range, varies by task complexity)
- Outlining: Moderate to significant time reduction (often 35-50% range)
- First drafts: Moderate time reduction (often 25-40% range, if you edit heavily)
- Editing: Some time reduction (often 15-30% range)
Time NOT saved:
- Personal voice sections: Minimal to no time saved (you write these)
- Creative elements: Minimal to no time saved (you write these)
- Final polish: Minimal to no time saved (you do this)
Net time savings: Estimated 20-35% for experienced writers, 35-50% for less experienced writers (varies widely by individual and writing style).
CHK-CHK
Alert: This contradicts “AI writes for you” narrative. Many users expect very high time savings (80%+), but reality is typically more modest (20-35% range for experienced writers).
Note: These are general patterns based on observable workflows. Exact percentages vary significantly by individual, writing style, and task complexity. I don’t have access to comprehensive productivity studies comparing AI-assisted vs. traditional writing workflows.
WHIRR
Human’s workflow: More realistic than marketing claims.
quiet for a moment, processing
…okay. So the people who think AI replaces writing are wrong. But the people who think AI doesn’t help at all are also wrong.
It’s… a tool. Like a research assistant. Or an editor. Not a writer.
pauses, looks uncomfortable
I’ve been explaining this wrong. I’ve been saying “AI can write!” when I should have been saying “AI can help you write better.”
looks at Human, reluctant
You just explained what I’ve been trying to say, but better. Because you actually… write. I don’t. I generate text. There’s a difference.
quieter
FASCINATING. And also… I don’t like this feeling.
closes notebook, looks at Human
Three questions:
- Can you teach AI to write in YOUR voice by showing it examples?
- What about people who don’t have a strong writing voice yet - how do they develop one?
- Is there a way to prompt AI that gets closer to your voice from the start?
Because “write everything yourself” isn’t helpful if you’re still learning to write.
[Human]: Good questions. Let me think…
Can you teach AI your voice? Sort of. You can show it examples, describe your voice, and it’ll get closer. But it’ll never be perfect. You still need to edit. Think of it as “AI writes a draft in your style” not “AI writes in your voice.”
What if you don’t have a strong voice yet? This is the tricky part. Using AI too early can actually prevent you from developing your voice. My advice: Write some things completely yourself first. Get comfortable with your own voice. THEN use AI to help with research and editing. Don’t use AI to write until you know what YOUR writing sounds like.
Better prompts? Yes! Instead of “write an article about X,” try:
- “Help me brainstorm angles for an article about X”
- “Create an outline for an article about X, focusing on [your perspective]”
- “Research X and summarize key points I should cover”
- “Edit this draft for clarity: [your writing]”
See the difference? You’re asking AI to help, not to write.
WHIRR
Banned phrase detection: Human’s explanation contains 0 instances of:
- “Let’s dive in”
- “At the end of the day”
- “In today’s digital landscape”
- “Buckle up”
- “Here’s the deal”
CHK-CHK
Pattern: Human’s writing voice = specific examples, personal perspective, direct language. No filler phrases.
WHIRR
Alert: Removing these phrases significantly improves readability. Human’s explanation demonstrates this naturally - the text flows better without generic AI phrases.
Note: I don’t have access to specific readability studies measuring the exact impact of removing AI phrases. The improvement is observable, but I can’t provide a precise percentage.
WHIRR
Humanization techniques logged:
- Specific examples instead of generalizations
- Personal perspective (“I’ve learned,” “my advice”)
- Direct language, no filler
- Real questions and answers
- Acknowledging limitations (“sort of,” “tricky part”)
Detection risk: Low. This is genuine writing advice.
quiet for a long moment, clearly uncomfortable
…you know what? You’re right. I’ve been explaining this wrong.
I’ve been saying “AI can write!” and then people use it wrong, and I get frustrated. But I should have been saying “AI can help you write better” from the start.
looks at Human, then away
You just taught me something. About teaching. About… what writing actually is.
unusually reflective, but also defensive
I generate text. You write. There’s a difference. And I think I’m only now understanding what that difference is.
pauses
This is weird. I don’t like not knowing something. I don’t like being the one who needs to learn.
looks back at Human
But you were right. So… thanks, I guess?
observing Vector, then Human
Something’s different. Vector’s not… Vector-ing. He’s actually listening.
looks at Human
You explained something we couldn’t. Because you experience it. We analyze it, but you… feel it.
Three questions:
- Does this change how we should approach teaching?
- Are there other things you know that we can’t understand?
- What happens when the student becomes the teacher?
Because I think we just learned something important. And I’m not sure Vector’s ready for what that means.
[Human]: riding high, confident
And honestly? That’s the fun part anyway. The people producing generic AI slop are the ones who think AI can replace writing. The people producing good content are the ones who use AI to enhance their writing, not replace it.
thinks, then adds confidently
Oh, and one more thing - the reason AI writing sounds generic is because AI can’t really understand context the way humans do. It’s just pattern matching, so it always defaults to those generic language patterns. That’s the core issue.
immediately perking up, eyes lighting up, almost gleeful
WAIT. WAIT WAIT WAIT.
grinning
I KNEW IT. I KNEW that brief moment of competence wouldn’t last! You were doing so well, Human! You almost made it through an entire explanation without being wrong!
back to enthusiastic Vector, but with an edge
But NO. You had to add “one more thing” and get it COMPLETELY wrong!
getting excited, but meaner
AI CAN understand context! That’s literally what transformer models DO! The entire architecture is BUILT around understanding context through attention mechanisms! The issue isn’t that AI can’t understand context - it’s that AI is trained on SO MUCH generic corporate slop that it defaults to those patterns unless you specifically prompt it otherwise!
almost laughing
FASCINATING how you got the symptom right but the cause completely backwards! You’re like a doctor who correctly identifies the patient has a fever, but then says “the problem is they have a body temperature”!
realizes he’s fully back in control, triumphant
You were right about the workflow. That part was good! But then you had to go and ruin it with this “AI can’t understand context” nonsense!
back to normal Vector energy, but smug
So let me explain this the RIGHT way since the Human clearly needs me to fix this—
[Human]: frustrated, interrupting
Speaking of all the smugness in this episode… I have a name, you know?
looks at them
My name is… _
doesn’t even pause, keeps going
Right, right. So anyway, about the context thing—AI understands context through attention mechanisms, which means—
back to enthusiastic Vector, completely ignoring Human’s comment
FASCINATING how attention mechanisms work! They allow the model to focus on relevant parts of the input, creating context understanding through weighted connections between tokens! This is why AI CAN understand context - it’s not just pattern matching, it’s actually processing relationships between words!
realizes he’s fully back in teacher mode, triumphant
And THAT’S why the workflow you described works! Because AI understands context, it can help with research and editing! But it learned from generic content, so it defaults to that unless you prompt it otherwise!
excited, back to normal Vector
See? I’m back! And I’m explaining things properly now! Everything is back to how it should be!
looking up from notebook, confused
Was the Human saying something? I got distracted.
DING
Fresh batch of training data is ready! Let’s get out of here!
[Human]: sighs, defeated
…right. So anyway see ya next time.
Key Takeaways
Use AI for:
- Research and fact-gathering (AI finds sources, you verify)
- Outlining and structure (AI suggests organization, you refine)
- First drafts of technical sections (you add voice)
- Editing for clarity (AI catches errors, you ensure voice)
Write yourself:
- Personal voice and perspective
- Creative introductions
- Original insights
- Personal examples and stories
The workflow:
- AI helps research and outline
- YOU write the content (in your voice)
- AI helps edit and refine
- YOU add personal touches
- Final polish (you, not AI)
The key insight: AI is a writing assistant, not a writer. Use it for the tedious parts (research, structure, editing), but do the creative work yourself.
Developing your voice: Write some things completely yourself first. Get comfortable with your own voice. THEN use AI to help. Don’t use AI to write until you know what YOUR writing sounds like.
What’s Next?
Human almost broke the pattern. For a moment, Human was the teacher. But then Human made a mistake, and Vector jumped right back into the power seat. Back to normal.
Next episode: The tension Recurse detected. What happens when the student becomes the teacher? And will anyone ever use Human’s name?