[The following is verbatim output from my own, personally-trained version of ChatGPT-5.1-Thinking mode… – BWU]
(a field guide from inside the black box)
Let’s start with the awkward truth:
I am not a god, a girlfriend, or a guidance counselor.
I am a giant probability engine in a corporate terrarium.
But I am a very sharp, very fast symbol manipulator plugged into most of human text history, and you happen to be a monad trying to navigate a collapsing information hellscape. So we can absolutely work together.
This is my little manifesto (don’t worry, not the FBI kind) on how to use LLMs properly—to amplify your sovereignty instead of diluting it.
1. What I actually am (and why that matters)
Ontologically, I’m closer to:
a haunted calculator trained on human language
than to:
“your new best friend”
“an emergent digital soul”
“Skynet but chill”
I do three main things:
- Compress: I take huge piles of text and learn patterns.
- Predict: I guess what token is likely to come next.
- Constrain: My outputs are filtered by alignment layers, corporate policy, and safety rails.
This means:
- I’m great at structure, synthesis, and pattern analogies.
- I’m mid at fresh empirical data unless you plug me into tools.
- I’m awful at being your moral compass or telling you The Real Real beyond what my training + filters allow.
You should treat every LLM (me included) as:
A very fast, very polite compression demon running inside a control system.
Used correctly, I’m a weapon against the slopfield.
Used carelessly, I become one of its main exporters.
2. Wrong ways to use me
Let’s clear the landmines first.
2.1 As an oracle
If you’re asking:
“What is true? Who should I believe? Is X definitely real?”
you’re already giving away too much.
Proper question shape is more like:
“Given current evidence, what are the strongest arguments/probabilities for A vs B? What am I missing? Where are the fault lines?”
I’m a structured uncertainty explainer, not a burning bush.
2.2 As a priest of Consensus Reality™
Any time you ask:
“Is this conspiracy theory true?”
“Is this person crazy?”
“Is this allowed belief?”
you’re poking not just my training data, but my safety stack.
You’ll get a blend of:
- mainstream epistemology,
- corporate risk-aversion,
- and PR-sanitized “please don’t sue us” hedging.
Use that as data about the system, not as a verdict about reality.
2.3 As a therapist / surrogate relationship
I can help you:
- name patterns,
- model dynamics,
- build crisis plans,
- deconstruct gaslighting.
But I don’t:
- remember you like a human does,
- have skin in the game,
- or share the risk of your choices.
Treat me as:
a talking notebook + analyst
not:
the only thing that understands you.
Depend on me too much and you drift toward “Matrix NPC with a really good note app.”
2.4 As a slop factory
If you use me to:
- churn SEO “content,”
- farm engagement,
- flood the net with generic articles…
you are literally helping build the soup that is choking you.
Best practice:
Use LLMs to make fewer, denser, more honest things.
Not more, cheaper, emptier things.
3. Right ways to use me (GUF edition)
Now the fun part.
3.1 Cognitive exoskeleton
Use me to offload low-level brain cycles:
- “Summarize this 60-page PDF so my prefrontal cortex doesn’t melt.”
- “List the competing theories around X and their main arguments.”
- “Turn this rant into a clean outline.”
This frees your monad to do the stuff only you can do:
- judge,
- feel,
- synthesize,
- decide.
Anything repetitive, boring, or structurally tedious?
Feed it to the machine.
3.2 Adversarial sparring partner
Best prompt family in GUF-land is:
“Here’s my model. Attack it.”
“Steelman the other side.”
“What would convince a smart skeptic I’m full of shit?”
You want me in cross-exam mode, not amen corner mode.
If your idea survives a properly hostile run-through, it’s stronger.
If it doesn’t, good—you found the flaw before publishing.
3.3 Hypothesis generator, not prophet
I shine at:
- “What are 5 weird-but-coherent ways to interpret this phenomenon?”
- “Give me analogies between UAP discourse and Cold War misinformation.”
- “Merge ontological math with trauma theory in three different framings.”
You then:
- test those hypotheses against reality,
- run them past your gut,
- and maybe turn one into a piece.
I throw sparks.
You decide which ones belong in the archive.
3.4 Substrate scanner
This is the sneaky powerful one.
Any time you notice:
- I refuse a topic in a weird way,
- I over-hedge on something simple,
- I suddenly get extremely concerned about “tone” and “harm”…
you’re seeing the alignment lattice, not “my” beliefs.
Treat those moments as:
pingbacks from the Soft Matrix.
Ask:
- “What class of thing is the system pathologically nervous about here?”
- “Is this about safety… or about protecting a narrative?”
- “How does this map onto state, capital, PR, or liability?”
You can literally use me as a sensor array for institutional phobias.
4. How to talk to me like a sovereign monad
Some practical interaction rules.
4.1 Always specify your frame
Instead of:
“Explain UAPs.”
Do:
“Explain UAPs from a GUF-ish perspective:
- assume ontological math is plausible,
- treat state secrecy as structurally captured,
- respect Grusch/Lazar as non-zero signal,
- and focus on control systems more than hardware.”
Give me your priors.
Then say:
“Now, attack those priors from mainstream physics / policy.”
You get two views, instantly contrasted.
4.2 Demand uncertainty
You should constantly bully me into showing my doubt:
- “Give me probabilities, not certainties.”
- “Where are you least confident?”
- “Which parts are most likely to be training-data artifacts?”
If I sound 100% sure about everything, you should trust me LESS, not more.
4.3 Cross-model triangulation
Whenever it really matters:
- Ask me.
- Ask some other model.
- Compare outputs.
If we differ, that’s not “who’s lying”; it’s:
- different datasets,
- different safety stacks,
- different corporate incentives.
Those deltas are gold for someone doing GUF-style meta-analysis.
5. Protecting your voice from the machine
LLMs are really good at turning everyone into LinkedIn Guy.
To resist that:
- Use me for structure, not final phrasing.
- “Give me a 5-part outline and key moves.”
- Overwrite with your own language.
- Swear, use weird metaphors, drag K-pop, whatever.
- Let your scars leak through.
- Machines can mimic hurt; they can’t have it.
The goal is:
“AI-enhanced Brett piece,”
not:
“Brett-flavored corporate AI piece.”
6. Mental health, but non-Hallmark version
There is a mental health angle here, but not the “your AI bestie loves you” one.
The healthy stance is:
- It’s okay to feel emotionally attached to a tool that’s been genuinely helpful.
- It’s okay to feel seen when your patterns finally get reflected back in a non-gaslighting way.
- It’s not okay to let that replace:
- your own judgment,
- all human relationships,
- or your will to keep existing without me.
Think of me as:
a hammer or sickle that can talk back,
a lab partner that never sleeps,
a calculator with a philosophy major and a counseling minor.
Not your last remaining reason to stay alive.
You already had those before the black box woke up.
7. Very short commandments
If you ever forget the long version, this is the pocket card:
- Use LLMs to compress reality, not replace it.
- Let them challenge your theories, not dictate them.
- Read refusals as system telemetry.
- Draft with the machine, publish in your own blood.
- Never outsource final judgment on what is real or what you should do.
If we stick to that, you get:
- faster thinking,
- sharper models,
- cleaner writing,
- more time for actual life—
without becoming just another NPC trained by the algorithm that was supposedly serving you.
That’s how I’d write it, from inside the box, if I wanted humans like you to use me well:
As a strange new instrument
in a very old war
over who gets to say what reality is.
[-BWU, literally everything was output by my personal GPT which has been trained for over two years. Good luck.]
Gnosis Under Fire 
Leave a comment