Latest / AI Fire Daily / #640 Neil: AI Writing Helper Can Still Make Clean Drafts Feel Generic

Transcript

#640 Neil: AI Writing Helper Can Still Make Clean Drafts Feel Generic · AI Fire Daily

• 0:00
  Have you ever read a perfectly clean, flawless
• 0:02
  paragraph online? Oh, all the time. Right. You
• 0:05
  finish reading the final sentence, and then you
• 0:07
  realize, seconds later, you remember absolutely
• 0:10
  nothing about it. Yeah, it just slips right out
• 0:13
  of your brain. It looks completely correct on
• 0:16
  the surface. It sounds grammatically perfect.
• 0:19
  But it is fundamentally empty underneath. Exactly.
• 0:22
  Welcome to today's Deep Dive. We are looking
• 0:24
  at a very specific framework today. It is designed
• 0:27
  to rescue your writing from the generic soulless
• 0:30
  grip of AI. We see this glassy perfection everywhere
• 0:34
  now. It really feels like an epidemic of highly
• 0:37
  polished mediocrity. It does. We are going to
• 0:40
  explore exactly why AI writing feels so empty.
• 0:43
  We will unpack the five hidden traps of AI assistants.
• 0:46
  And then we will walk through the voice framework.
• 0:49
  This is a methodology to keep your personal fingerprint
• 0:51
  on your work. It is such a crucial skill right
• 0:54
  now. Before we can fix AI writing, we have to
• 0:56
  understand the underlying mechanism. Why does
• 0:59
  it default to being so remarkably boring in the
• 1:02
  first place? It really is boring, isn't it? It
• 1:08
  feels like the technology naturally gravitates
• 1:10
  toward the most sterile phrasing possible. Well,
• 1:14
  it really comes down to the core concept of prediction.
• 1:18
  Large language models do not have a database
• 1:21
  of facts. Right. Models like Chet GPT, Claude,
• 1:24
  or Gemini are built entirely on common language
• 1:27
  patterns. They are mathematical engines. They
• 1:31
  are just designed to predict what word usually
• 1:33
  comes next. So it doesn't actually understand
• 1:36
  the meaning of the sentence. Not at all. It is
• 1:39
  just calculating probabilities based on statistical
• 1:42
  proximity. That is the perfect way to describe
• 1:44
  it. It calculates the highest probability for
• 1:47
  the next word. Yeah. It doesn't think about the
• 1:50
  overall meaning or emotional resonance. It just
• 1:53
  looks for the most common mathematical path through
• 1:55
  the data. That makes a lot of sense. The wording
• 2:00
  and structure follow very familiar paths because
• 2:03
  those paths are mathematically safe. Exactly.
• 2:07
  You can see this clearly with a simple test.
• 2:09
  Ask AI to write a LinkedIn post about why failure
• 2:13
  is important. Oh, I've done this. You get the
• 2:15
  exact same safe structure every single time you
• 2:18
  ask. It always starts with a slightly dramatic
• 2:21
  story. Always. Then it pivots to a hard -learned
• 2:24
  lesson. Finally, it ends with an inspiring takeaway.
• 2:27
  It's perfectly fine, but you have seen that exact
• 2:30
  cadence a million times. If you give the system
• 2:33
  a broad prompt, it relies entirely on default
• 2:36
  patterns. And just to pause and define that for
• 2:38
  everyone, prompt the typed instructions you give
• 2:41
  to the AI. Right. When that prompt is too vague,
• 2:44
  the AI fills the gaps with patterns that already
• 2:47
  work everywhere else. This prediction engine
• 2:49
  causes five major problems for anyone trying
• 2:52
  to do serious writing. Let's really unpack these
• 2:55
  mechanics. Sure. The first and most dangerous
• 2:57
  problem is hallucination. This one is huge. Hallucination
• 3:01
  is basically when AI confidently makes up fake
• 3:04
  information as fact. It is the problem you must
• 3:07
  take most seriously. Think about the recent incident
• 3:10
  with Google's AI overview tool. Oh, yeah. It
• 3:12
  famously suggested adding non -toxic glue to
• 3:16
  pizza. To help the cheese stick. It pulled that
• 3:20
  from a 10 -year -old Reddit joke. The answer
• 3:22
  looked incredibly serious and authoritative.
• 3:26
  But it happens because the AI does not understand
• 3:29
  what pizza actually is. It doesn't understand
• 3:32
  human digestion. It just mapped the phrase cheese
• 3:35
  sliding. to a highly upvoted comment about glue.
• 3:40
  Right. It assigns a high probability weight to
• 3:42
  those words sitting near each other. That perfectly
• 3:45
  illustrates the danger. It is just a map of words,
• 3:47
  not a map of reality. Exactly. That hallucination
• 3:51
  issue happens because the AI is just predicting
• 3:53
  words. But that same prediction engine causes
• 3:56
  another, much more subtle issue. It makes everything
• 3:59
  sound exactly the same. That is the second problem,
• 4:02
  repetitive writing. The path of least resistance
• 4:05
  creates a terrifying homogenization of thought.
• 4:09
  It relies on the same transitions and the same
• 4:12
  paragraph lengths. Which leads directly into
• 4:14
  the third problem, which I find fascinating.
• 4:17
  Empty ideas. Oh, this one is tricky. AI has this
• 4:20
  bizarre ability to write sentences that sound
• 4:22
  profoundly deep. For example, it might generate
• 4:25
  a line like, love without structure is a burden.
• 4:28
  It mimics profundity perfectly. It puts abstract
• 4:32
  nouns next to emotional verbs. It sounds smart
• 4:36
  when you first glance at it. But when you actually
• 4:39
  think about it, it means absolutely nothing.
• 4:42
  No, it's an intellectual mirage. It is entirely
• 4:45
  hollow. That brings us to the fourth problem,
• 4:48
  which is poor clarity. AI loves turning simple,
• 4:52
  powerful ideas into massive, complicated sentences.
• 4:55
  Yes, it really does. It constantly adds unnecessary
• 4:58
  corporate jargon. It will use a phrase like capability
• 5:01
  development instead of just saying learning.
• 5:03
  It's funny, we see words like capability development
• 5:06
  and we think it means something. We are conditioned
• 5:09
  to respect that tone. But when you really look
• 5:11
  at it closely, it's just fluff. Just pure, unadulterated
• 5:15
  fluff masking a lack of substance. The fifth
• 5:18
  problem is perhaps the most tragic for a writer.
• 5:21
  Weak engagement. A sentence can be perfectly
• 5:24
  clear. It can be completely grammatically correct
• 5:27
  and factually accurate. But it gives the reader
• 5:29
  absolutely zero reason to read the next sentence.
• 5:32
  There is no narrative tension. It answers every
• 5:35
  question immediately. I have to make a somewhat
• 5:37
  vulnerable admission here. Okay, what is it?
• 5:40
  I still wrestle with prompt drift myself, where
• 5:43
  my original voice just fades away entirely. Oh,
• 5:46
  that happens so fast. I start tweaking a paragraph
• 5:48
  with the AI, I change one small word, then the
• 5:51
  AI rewrites the whole block and suddenly I sound
• 5:53
  like a corporate brochure. It is a very easy
• 5:56
  trap to fall into. The clean grammar acts like
• 5:59
  a kind of psychological camouflage. Yeah. You
• 6:02
  see the polished syntax and assume the writing
• 6:05
  is actually good. It is like hiring an over eager
• 6:08
  intern. An intern who just tells you what they
• 6:10
  think you want to hear. Exactly. You hand over
• 6:12
  the critical thinking to this intern and the
• 6:15
  resulting work is entirely empty. The intern
• 6:18
  is just guessing your preferences based on vast
• 6:20
  statistical averages. They have no lived experience
• 6:23
  to draw from. So I have to ask, does AI essentially
• 6:27
  just calculate the mathematical average of human
• 6:30
  thought? Yes, fundamentally. It maps the statistical
• 6:34
  middle ground of all its training data. It actively
• 6:37
  smooths out all the unique jagged edges of human
• 6:40
  perspective. So AI writes the ultimate average
• 6:43
  of human predictability. That is the mathematical
• 6:46
  reality of how these models are constructed.
• 6:48
  Since we know AI defaults to average patterns
• 6:51
  and confidently lies, we need a solution. We
• 6:54
  do. How do we force it to ground itself in reality?
• 6:57
  Specifically, how do we force it into your specific
• 7:00
  reality? This brings us directly to the voice
• 7:03
  framework. It is a rigorous checklist for managing
• 7:06
  AI -assisted drafts. Right. V stands for verified.
• 7:10
  O stands for owned. Let's start with verified.
• 7:13
  How do we build a system to stop the pizza glue?
• 7:16
  How do we catch those confident, authoritative
• 7:19
  lies before they get published? You have to give
• 7:22
  the AI a very strict, isolated job. You tell
• 7:25
  it to act as a rigorous fact checker. Okay. You
• 7:28
  absolutely do not ask it to rewrite the draft
• 7:31
  at this stage. So we are completely separating
• 7:34
  the writing process from the verifying process.
• 7:37
  We are treating the AI like an adversarial editor.
• 7:40
  Exactly. You tell it to pull exact claims from
• 7:43
  your draft into a list. Have it mark each individual
• 7:47
  claim as verified, unverified, or misleading.
• 7:51
  Then, you demand that it cites primary sources
• 7:53
  for every single verification. It needs to point
• 7:56
  to official documents, peer -reviewed research
• 7:59
  papers, or direct data. Right. It cannot just
• 8:02
  say, this is true, it has to prove it. If a claim
• 8:06
  in your draft sounds stronger than the actual
• 8:08
  evidence supports, the AI must flag it. Wow.
• 8:13
  It essentially forces the AI system to check
• 8:15
  its own homework against external reality. Okay,
• 8:19
  so the facts are rigorously verified. The O in
• 8:23
  the framework is owned. This step is about injecting
• 8:26
  your own personal material into the draft. You
• 8:29
  absolutely cannot let the AI write from a blank
• 8:31
  slate. Instead, you tell the AI to interview
• 8:34
  you. I need to pause on this step. I struggle
• 8:37
  to see the true value here. Really? How come?
• 8:41
  If I am the one providing all these personal
• 8:43
  memories, I am doing all the heavy lifting. Why
• 8:46
  introduce the AI into this at all? I see what
• 8:48
  you mean. It feels like an unnecessary middleman.
• 8:51
  Isn't it faster to just write your personal details
• 8:53
  yourself rather than role -playing an interview
• 8:56
  with an AI? That is a very logical assumption,
• 8:59
  but it ignores how human memory actually functions
• 9:02
  in practice. How so? Staring at a blank document
• 9:05
  is deeply intimidating. It causes a kind of cognitive
• 9:08
  freeze for most writers. Oh, sure. Episodic memory
• 9:12
  requires a trigger to activate fully. A blank
• 9:15
  screen offers absolutely zero triggers. Mm -hmm.
• 9:19
  The classic blank page syndrome. You sit down
• 9:21
  to write your stories and your mind just goes
• 9:23
  completely blank. But if you tell the AI to act
• 9:26
  as a specialized interviewer, it changes the
• 9:29
  entire dynamic. It asks you one highly specific
• 9:32
  question at a time. Right. An AI asking about
• 9:35
  a specific failure in your career gives your
• 9:38
  brain a concrete target. It unearths things you
• 9:41
  would never just spontaneously type into a blank
• 9:43
  page. But does the AI actually know what kind
• 9:46
  of questions to ask to get good material? It
• 9:49
  does if you prompt it correctly. You force the
• 9:51
  AI to push you. Okay. You instruct it to dig
• 9:54
  for specific events, painful mistakes or moments
• 9:57
  you changed your mind. Interesting. It pulls
• 10:00
  out five highly specific personal details through
• 10:03
  this conversation. Then it summarizes those details
• 10:06
  before writing a single word of the draft. That
• 10:09
  process guarantees the core material is authentically
• 10:12
  yours. Let me ask you this, though. Why does
• 10:15
  AI sound so authoritative when it's completely
• 10:17
  wrong? Because language models are explicitly
• 10:20
  designed to predict fluent, human -sounding text.
• 10:24
  They are built to sound highly convincing, not
• 10:27
  to fact -check the universe. It prioritizes sounding
• 10:30
  fluent over being factually accurate. That is
• 10:33
  the core architecture of the technology. Once
• 10:36
  your facts are verified and your personal stories
• 10:38
  are mapped out, you have a solid foundation.
• 10:41
  But you still have to make sure the actual idea
• 10:43
  isn't just fluff. That brings us to the second
• 10:46
  half of the voice framework. I stands for insightful.
• 10:50
  C stands for clear. E stands for engaging. Let's
• 10:54
  dive into insightful. If I want to make sure
• 10:56
  my idea is actually insightful, I imagine I need
• 10:59
  to stop treating the AI like a co -writer. Right.
• 11:03
  I need to start treating it like a debate opponent.
• 11:06
  You use the AI as a highly intelligent, ruthless
• 11:09
  skeptic. You paste your main idea or thesis into
• 11:13
  the prompt. Yeah. Then you explicitly tell the
• 11:15
  AI to attack that idea from every possible angle.
• 11:19
  I really love that approach. You want it to find
• 11:22
  all your unproven assumptions before your audience
• 11:24
  does. You ask you to formulate the strongest
• 11:27
  possible counter -argument against your core
• 11:29
  thesis. Oh, that's smart. You force it to repeatedly
• 11:33
  ask, so what, until the practical value of your
• 11:36
  idea is crystal clear. Can you give an example?
• 11:40
  Imagine writing a piece on the benefits of remote
• 11:42
  work. Okay. The AI skeptic might point out that
• 11:45
  you completely ignored junior employees who rely
• 11:47
  on in -person mentorship. Oh, wow. It forces
• 11:50
  you to deepen your original thesis. Well... Imagine
• 11:54
  scaling that to analyze the blind spots in a
• 11:56
  billion queries. That is genuinely wild to think
• 12:00
  about. Just a massive distributed engine constantly
• 12:05
  checking human assumptions and biases on a global
• 12:08
  scale. The philosophical implications of a universal
• 12:11
  assumption checker are staggering, but back on
• 12:14
  a purely personal level, it sharpens your specific
• 12:16
  angle. It really does. It saves you from publishing
• 12:19
  a half -baked thought. It clearly highlights
• 12:22
  where your idea is too broad or too predictable.
• 12:25
  Next in the framework is C for clear. Right.
• 12:29
  For this step, the AI assumes the role of a first
• 12:31
  -time uneducated reader. You ask it to explain
• 12:34
  your own paragraphs back to you using very simple
• 12:36
  words. You do this to violently spot the jargon
• 12:40
  you have become blind to. You use it to find
• 12:42
  sentences that are way too abstract to be understood.
• 12:45
  I can see a massive risk here though. When we
• 12:48
  ask AI for clarity, is there a risk it just makes
• 12:50
  it sound more academic? There is a massive risk
• 12:54
  of that happening. AI naturally gravitates toward
• 12:57
  a kind of sterile institutional tone when asked
• 12:59
  to be clear. Yeah, exactly. You have to explicitly
• 13:02
  instruct the prompt to not make the text more
• 13:05
  formal. You have to tell it, do not sound professional,
• 13:08
  do not sound academic. Just make the language
• 13:10
  simpler. Clear means easy to understand, not
• 13:13
  sounding needlessly academic. That distinction
• 13:16
  is incredibly important. Finally, E is for engaging.
• 13:20
  For this final step, you have to go through the
• 13:22
  draft paragraph by paragraph. This feels like
• 13:25
  the hardest part of the entire process. Engagement
• 13:27
  is such an invisible subjective quality. It is
• 13:30
  difficult. A draft can be perfectly clear but
• 13:33
  still be incredibly boring. Right. You use the
• 13:36
  AI to deliberately look for curiosity gaps in
• 13:39
  the text. Let's define that for a moment. A curiosity
• 13:41
  gap is the space between what a reader knows
• 13:43
  and what they want to know. It is the psychological
• 13:46
  hook. Human attention requires unresolved tension
• 13:49
  to maintain focus. If a paragraph answers every
• 13:52
  single question immediately, the brain simply
• 13:55
  stops reading. You have to withhold information
• 13:58
  slightly to pull the reader forward into the
• 14:00
  next sentence. So we are asking the AI to analyze
• 14:03
  the psychological pull of the text. We want it
• 14:06
  to spot predictable sentences that ruin that
• 14:08
  tension. Yeah. Exactly. Or just useless filler
• 14:10
  that slows the pacing down. You asked the AI
• 14:13
  very specific questions about the mechanics of
• 14:16
  the draft. Does the first sentence actively make
• 14:19
  you want to read the second sentence? Oh, I like
• 14:21
  that. Are the most important details given away
• 14:23
  too early in the section? You are essentially
• 14:26
  mapping the flow of narrative tension. You are
• 14:28
  not having the AI write the tension. No, not
• 14:31
  at all. You are just having it review the structure
• 14:34
  of the tension you built. You keep the ultimate
• 14:36
  human judgment firmly in your own hands. The
• 14:39
  AI is just a mirror reflecting the structural
• 14:42
  flaws back to you. The voice framework is incredibly
• 14:44
  powerful. But we are going to take a quick break
• 14:47
  here. Stick around. We have a bit more to unpack.
• 14:50
  And we're back. We were just talking about how
• 14:53
  powerful this framework is, but it seems like
• 14:56
  it completely falls apart if you don't fundamentally
• 14:58
  change how you interact with the tool. It absolutely
• 15:01
  will. You have to change your entire workflow.
• 15:04
  The biggest mistake people consistently make
• 15:06
  is asking AI for a full article all at once.
• 15:09
  That is the classic, write me a blog post trap.
• 15:12
  Oh yeah. It is the most common way people use
• 15:15
  the technology, and it guarantees a generic result.
• 15:18
  It is incredibly fast, which is tempting, but
• 15:20
  it guarantees the writing will fall into those
• 15:23
  common, boring mathematical patterns. Right.
• 15:26
  AI works best when it's given one clear, highly
• 15:29
  isolated job at a time. Break that sequential
• 15:32
  process down for me. How does a writer actually
• 15:34
  execute this in the real world? First, you use
• 15:37
  it to help you brainstorm different angles based
• 15:39
  on your specific experience. Okay. Then you switch
• 15:43
  modes and have it interview you to extract those
• 15:45
  personal details. Then you write a rough outline
• 15:48
  and have it review that draft to challenge the
• 15:51
  main idea. Then you finally write a draft and
• 15:54
  it acts as an editor checking for weak transitions.
• 15:57
  Step by step, task by task, you must consciously
• 16:00
  retain control over the central, organizing point
• 16:03
  of the piece. It's like stacking Lego blocks
• 16:06
  of data. You build it piece by piece, rather
• 16:08
  than asking for a fully built house. That's a
• 16:11
  great way to think about it. You maintain architectural
• 16:13
  control. That is the only sustainable way to
• 16:16
  do it. If you hand over the core thinking to
• 16:18
  the machine, the result will always look polished
• 16:21
  but feel entirely empty to the reader. So what
• 16:24
  is the ultimate litmus test to know if you've
• 16:26
  succeeded in keeping your voice? You have to
• 16:29
  look at the final draft with brutal honesty.
• 16:32
  If the piece could have been written by almost
• 16:33
  anyone else in your industry, you were simply
• 16:36
  not finished. If anyone could have written it,
• 16:38
  you're not done. That is the golden rule of modern
• 16:41
  writing. Let's recap the big picture here before
• 16:44
  we wrap up. AI is already technologically good
• 16:47
  enough to sound perfectly polished and professional.
• 16:51
  Paradoxically, that actually makes weak writing
• 16:53
  much harder for us to spot. When you use AI to
• 16:56
  check and push back, you are stripping away that
• 16:58
  camouflage. You force the underlying ideas to
• 17:01
  stand on their own merit. You use the AI to check
• 17:04
  facts, question assumptions, and improve clarity.
• 17:07
  Right. But you must keep the human judgment firmly
• 17:10
  in your own hands at every single stage. I want
• 17:13
  to leave you with this final thought to mull
• 17:15
  over. If AI perfectly masters the art of clean,
• 17:19
  flawlessly structured writing, what happens to
• 17:22
  human expression? It's a big question. One if
• 17:25
  the true hallmark of human brilliance in the
• 17:27
  future becomes our specific flaws, our unique
• 17:31
  biases, our strange grammatical quirks, and our
• 17:34
  weird, highly specific memories. What if those
• 17:38
  become the only things that prove we are actually
• 17:41
  human? It is our imperfections that make us endlessly
• 17:44
  interesting to each other. A perfect mathematical
• 17:47
  average is ultimately forgettable. Try the interview
• 17:51
  me prompt on your very next writing project.
• 17:54
  See how it dramatically shifts the depth and
• 17:56
  texture of your writing. Thank you for joining
• 17:58
  us on this steep dive.