Domion Mode: It Takes a Man
Plus: Smurf the Filioque, Free Access to prototype Carnivore CoarchBot, and Make Your Robot Smarter by Getting Smarter Yourself
Dude.
Pray for me. Thank you for your support!
Wait, What?
That’s Just the Tip, Baby.
Keep Your Eye on the Prize Boyz
His Rod and His Staff, They Comfort
50-Card Psalm Decks Already Taking Orders at
Ask for @Flicker
Smart “Lang-Chaining”
Make Your Robot Smarter by Getting Smarter Yourself
What it is.
Smart Lang-Chaining is a simple loop that teaches your AI to write like you. You take a messy output, edit it to your standard, and feed the polished version back as the example the model should follow. Repeat. The AI learns your form through your corrections—not through prompts alone.
Why it works.
Models imitate patterns. When you supply clean “before → after” pairs, you narrow the model’s options and raise the signal. Your edited sample becomes the target distribution for future drafts.
The 5-step loop.
1. Prompt small. Ask for a short piece (one paragraph or one section).
2. Edit decisively. Fix voice, verbs, order, and claims. Delete filler.
3. Annotate briefly. Add 3–5 labels (e.g., Active voice, Thesis first, Short sentences).
4. Feed back. Say: “Use THIS as the model. Rewrite the original in this style.” Paste only your edited text + labels.
5. Chain forward. Accept the best parts, repeat the loop on the next chunk. Over time, the model “gets you.”
Clear Writing Model (CWM) mini-spec.
• Goal: One clear claim per paragraph; verbs > nouns; facts before flourish.
• Tone flags: Plain > clever; concrete > abstract; show math/dates/names.
• Structure: Hook → Claim → Evidence → Action.
• Style rules: Active voice, short sentences, parallel lists, zero hedging.
Starter prompt.
“Write one paragraph in my Clear Writing Model: Hook → Claim → Evidence → Action. Use active voice and short sentences. Then stop.”
(After you edit, say:)
“Learn from this revision. Use it as the pattern for all future outputs in this chat.”
What to feed back.
• Gold sample (your edit).
• Three tags (e.g., Direct, Evidence-first, No filler).
• One sentence rubric (“If a sentence doesn’t advance the claim, cut it.”)
What to avoid.
• Huge prompts, vague goals, and “one-take” essays.
• Stacking style rules you won’t enforce with edits.
• Letting the AI overwrite your standard—your sample is law.
Success signal.
Drafts arrive already close to your voice; edits shrink from heavy rewrites to light trims; speed increases “at a clip.”
One-page checklist.
Hook first. One claim. Evidence named. Action stated. Active verbs. Short lines. Parallel form. No hedges. Cut filler. Feed back the fix. Repeat.
