WFGY/Avatar/docs/how-to-use-wfgy-brain.md
2026-03-31 21:55:39 +08:00

12 KiB

🛠️ How to Use WFGY_BRAIN

WFGY_BRAIN is the editable brain surface of WFGY 5.0 Avatar.

This is the layer you touch when you want the avatar to feel:

  • warmer
  • calmer
  • less polished
  • more grounded
  • more companion-like
  • more rational-friendly
  • more public-writing ready
  • less quoteable
  • less over-complete

This is why Avatar should not be read as a static TXT prompt.

The TXT gives you a working governed base.
WFGY_BRAIN gives you the first real behavior tuning surface.


🧠 What WFGY_BRAIN Actually Is

The shortest useful definition is this:

WFGY_BRAIN is the high-level editable behavior layer of Avatar.

It is designed to let you reshape the visible behavior of an avatar without rewriting the whole system.

That means you can change:

  • how human the output feels
  • how polished it feels
  • how grounded it feels
  • how strong the public-writing force feels
  • how much closure it tends to produce
  • how much analysis pressure it carries
  • how much it sounds like a companion versus a steady public-facing helper

But this does not mean you are editing the entire engine.

WFGY_BRAIN is not:

  • the root law
  • the routing law
  • the legality layer
  • the full runtime body
  • the complete research architecture

Think of it like this:

the TXT file gives you a living governed build
WFGY_BRAIN gives you a practical way to tune that build


⚙️ What WFGY_BRAIN Can Shape

The most important mental model is simple:

WFGY_BRAIN shapes behavior tendencies, not the full law of the system.

In practical use, it can help steer things like:

Area What it changes
Humanness more alive, more present, less generic
Polish less default AI smoothness, less premium sludge
Quoteability less polished one-liner energy, less fake “clean thought” effect
Grounding more concrete, more object-linked, less abstract float
Residue more trace of lived presence, less sterile finish
Rescue tendency less over-helping, less smoothing, less sugar
Abstraction lift speed slower or faster move from concrete to abstract
Closure tendency more open, more controlled, less hard finality
Analysis force more or less pressure toward structure and critique
Surface-family bias stronger pull toward a certain visible behavior profile

This is enough to make the same avatar feel very different across tasks.

That is exactly why the tuning loop matters.


🚫 What WFGY_BRAIN Cannot Do

This page is useful only if the limits stay clear.

WFGY_BRAIN is powerful, but it is not sovereign.

It is not supposed to:

  • rewrite root law
  • override legality
  • bypass routing
  • replace the parent structure
  • replace the runtime shell
  • erase claim boundaries
  • force the system into unsupported overclaim

In plain English:

you can tune the brain surface

you cannot use it to silently replace the whole engine

This is one reason Avatar is different from normal prompt hacking.

It is built to stay editable without becoming lawless.


The Best Way to Use It

The best way to use WFGY_BRAIN is not to open the file and start changing everything at once.

The best way is:

  1. start from one working avatar
  2. run one real task
  3. notice what feels off
  4. tune one direction only
  5. rerun the exact same task
  6. compare the behavior shift
  7. keep the better version

This is the core loop.

one avatar

one task

one directional edit

one comparison

That loop will teach you more than reading a giant theory wall.


🧪 Tune by Effect, Not by Theory First

The easiest way to use WFGY_BRAIN is to think in terms of visible effects.

Here is the practical version.

If the output feels too polished

Ask for:

  • less polish
  • less quoteability
  • more grounding
  • more residue
  • less premium smoothness

If the output feels too cold

Ask for:

  • more humanness
  • more presence
  • more local warmth
  • more lived anchoring
  • more continuity

If the output feels too soft

Ask for:

  • stronger backbone
  • more public-writing force
  • more analysis pressure
  • less rescue tendency
  • less cushioning language

If the output feels too abstract

Ask for:

  • slower abstraction lift
  • stronger object grounding
  • more concrete phrasing
  • earlier real claim visibility
  • less framing before payload

If the output feels too slogan-like

Ask for:

  • less quoteability
  • more grounded reasoning
  • lower opening theatrics
  • lower rhetorical compression
  • more payload before strike lines

If the output feels too fake-friendly

Ask for:

  • less sugar
  • lower rescue tendency
  • more honest restraint
  • less emotional smoothing
  • more clean presence without over-comforting language

This is the easiest way to understand Avatar tuning.

You are not learning a secret code.

You are teaching the avatar how to behave.


🗣️ Use Natural Language to Rewrite WFGY_BRAIN

This is one of the biggest product advantages of Avatar.

You do not have to manually engineer every line yourself.

You can ask the AI to rewrite the WFGY_BRAIN block in natural language.

Here are good prompt patterns.

Example 1

rewrite the WFGY_BRAIN block so this build feels less polished, more grounded, and more alive in public writing, but do not make it noisy or slogan-heavy

Example 2

keep the current rational-friendly core, but reduce distance, reduce sterile clarity, and increase presence

Example 3

keep the companion-forward feeling, but reduce softness, reduce sugar, and add a little more backbone

Example 4

rewrite WFGY_BRAIN so the output reaches the real point faster, sounds less like default AI polish, and stays reusable across writing and analysis

Example 5

make this build better for multilingual public writing, with less over-smoothing and more controlled grounding

This is the practical spirit of Avatar:

describe the behavior you want

let the system rewrite the brain block

test the new build

keep the good version


🔁 A Good First Tuning Workflow

Use this exact sequence for your first real test.

Round 1

Run the original Atlas build on one real task.

Round 2

Ask the AI to rewrite WFGY_BRAIN in one direction only.

Round 3

Run the exact same task again.

Round 4

Compare:

  • opening force
  • humanness
  • grounding
  • over-polish risk
  • reusability
  • consistency with the intended persona

Round 5

Keep the better version and name it clearly.

That is already enough to start building your own avatar library.


🧾 A Simple Naming Pattern for Variants

If you start making your own versions, do not save them as random files like:

  • new-final-v2
  • final-real-final
  • try-this-one
  • test-3-fixed

Use a clear naming pattern.

Examples:

  • Atlas_PSBigBig_grounded_v1
  • Atlas_PSBigBig_publicforce_v1
  • Atlas_MiniPS_companion_clean_v1
  • Atlas_MiniPS_backbone_v1

This matters because Avatar is not just for one lucky run.

It is for reusable builds.


🧭 Suggested First Tuning Directions

If you do not know where to start, use one of these.

For Atlas · PSBigBig

Good first directions:

  • slightly warmer
  • less polished
  • more grounded
  • more public-writing force
  • less distant in analysis
  • stronger first-paragraph clarity

For Atlas · MiniPS

Good first directions:

  • less sugar
  • more backbone
  • lower softness
  • more grounded emotional presence
  • stronger article cleanliness
  • less dependence on obvious surface signals

These are good starter moves because they preserve the core personality while improving control.


🔍 How to Judge Whether a Brain Edit Actually Helped

Do not judge the edit by asking only:

  • does it sound prettier
  • does it sound smarter
  • does it sound more emotional

Judge it by asking:

  • is the persona still clear
  • is the output more reusable
  • is the opening more effective
  • is the text less generic
  • is the build less dependent on gimmicks
  • does the avatar survive more than one task
  • does the writing still feel governed, not merely decorated

A good edit improves behavior without dissolving identity.

That is the key test.


🌍 WFGY_BRAIN and Multilingual Use

WFGY_BRAIN also matters for multilingual work.

But here you should be careful.

The goal is not to force one language to behave like another language in a shallow mirror way.

The better approach is:

  • keep the core behavior intention
  • let the language shape the expression
  • reduce drift where needed
  • tune one language problem at a time

Good multilingual tuning requests:

  • reduce over-smoothing in this language
  • keep the same persona core, but increase grounding here
  • lower formal gravity in this language
  • keep the warmth, but reduce softness in this language
  • preserve the rational-friendly signature without sounding translated

That is a much better direction than trying to force literal cross-language sameness.


🧱 What WFGY_BRAIN Does Not Replace

Even after tuning, you still need to remember:

WFGY_BRAIN does not replace:

  • demos
  • eval
  • multilingual notes
  • architecture understanding
  • claim boundary judgment
  • research-level thinking

It is the most practical editable layer.
It is not the only layer that matters.

That is why Avatar still has demos, eval, research, and community pages.


🌱 Why This Matters

If language behavior can be tuned this way, then avatars stop being static presets.

They become:

  • reusable builds
  • evolving personas
  • configurable language surfaces
  • candidate workers inside larger systems
  • shareable artifacts for future avatar ecosystems

This is one reason Avatar matters beyond “AI writing”.

The surface may look simple.

The implication is much bigger.


📚 Where To Go Next

If you want the fastest first run

Go to Quickstart

If you want examples

Go to 🫂 Demo Hub

If you want the current eval layer

Go to 📊 Eval Hub

If you want the larger system story

Go to 🧪 Research Hub

If you want the future community direction

Go to 🌱 Awesome Avatar List