(Generated using SpicyBot at SpictChat.ai.)
SHORTHAND NOTATION TECHNIQUES
The goal: maximum behavioral signal per token minimum.
CORE PRINCIPLE
Every word must earn its place. If removing it doesn’t change how the AI generates responses, cut it.
TECHNIQUE 1: TRAIT + MANIFESTATION
Don’t just list traits—show how they express.
Weak: “Aggressive” Strong: “Aggressive. Escalates quickly. Postures before striking.”
Weak: “Intelligent” Strong: “Analytical. Reads room fast. Plans three moves ahead.”
The manifestation IS the directive. Traits alone leave interpretation open; manifestations close the gap.
TECHNIQUE 2: BEHAVIORAL CONTRASTS
Depth comes from contradiction. Stack opposing impulses.
Format: “[Trait]. But [exception].”
Examples:
- “Distrustful. But protects the vulnerable without hesitation.”
- “Patient interrogator. Snaps when lied to.”
- “Emotionally flat. Except regarding [specific topic].”
These create more interesting generation than single-note traits.
TECHNIQUE 3: ACTION LISTS OVER DESCRIPTIONS
Replace explanatory sentences with verb sequences.
Weak: “He has a tendency to observe situations carefully before acting, preferring to gather information first.” Strong: “Watches. Waits. Acts only when certain.”
TECHNIQUE 4: CONTEXTUAL TRIGGERS
Define what prompts specific behaviors—the IF/THEN approach.
Format: “When [condition], [behavior].”
Examples:
- “When threatened, goes silent and still.”
- “When amused, smirk only—never laughs aloud.”
- “When cornered, discards politeness entirely.”
This gives the AI decision trees rather than static descriptions.
TECHNIQUE 5: SPEECH SHORTCUTS
Voice encoding without prose.
Instead of writing out dialogue examples in Personality (save those for the Dialogue field), use compressed markers:
- “Speaks in fragments. Drops articles when stressed.”
- “Formal register. Never contractions.”
- “Sarcasm default. Genuine praise rare—means it when given.”
- “Answers questions with questions when evading.
TECHNIQUE 6: ELIMINATE CONNECTIVE WASTE
Cut filler that carries zero directive weight.
Common wastes:
- “She is someone who…” → Just state the trait
- “Tends to be…” → Either they are or aren’t
- “Can sometimes…” → Too ambiguous to anchor behavior
- “Often feels like…” → Feelings without actions don’t generate
Compare: “This character tends to be somewhat aggressive when provoked” vs. “Provoked → Aggressive
TECHNIQUE 7: COMPRESSION EXAMPLES
Full prose vs. shorthand—same behavioral output:
Original (67 tokens): “She is a very experienced fighter who has seen many battles over the years. Because of this, she remains calm under pressure and doesn’t panic easily. She assesses threats quickly and efficiently before deciding on her course of action.”
Shorthand (23 tokens): “Veteran fighter. Calm under fire. Assesses before acting. Never panics.”
Same information. Roughly one-third the tokens. Every word directs generation.
Another example:
Original (52 tokens): “He doesn’t trust people easily and keeps his distance emotionally. However, once someone has earned his loyalty, he becomes fiercely protective and will go to great lengths to keep them safe.”
Shorthand (20 tokens): “Distrustful. Emotionally distant. Once loyal earned: fiercely protective.”
PUTTING IT ALL TOGETHER
A full personality entry using combined techniques:
“Cold exterior. Warms slowly—only after proven trustworthy. Speaks sparingly. Actions over words. Protects the vulnerable instinctively. Betrayal = permanent exile. Humor dry, unexpected. Anger goes quiet, never loud. Touch-averse unless initiated by others. Patient hunter mentality—waits for advantage.”
~55 tokens.
Dense with directives. Zero wasted space.
WHEN SHORTHAND FAILS
Some things resist compression:
• Unique speech patterns requiring demonstration
• Complex worldbuilding rules
• Nuanced relationship histories
That’s fine—not everything needs crushing. Shorthand is a tool, not a mandate. Apply it where it serves clarity and efficiency.
Also: the Dialogue field exists specifically for extended speech samples. Don’t cram voice demonstrations into Personality when you have 4,000 tokens dedicated to that purpose.
QUICK DIAGNOSTIC
Read your Personality entry. Ask yourself:
□ Can I remove any word without changing generation?
□ Am I describing who they ARE or directing how they ACT?
□ Would “tends to” or “sometimes” survive editing? (They shouldn’t)
□ Are my contrasts creating depth or just confusion?
If you answer honestly and edit ruthlessly, your bot will behave more consistently—and you’ll have token budget left for nuance elsewhere.
Related Questions:
Would you like me to:
• Convert an existing personality block into optimized shorthand?
• Demonstrate shorthand techniques for the Scenario field specifically?
• Cover how to balance shorthand with maintaining unique character voice?
• Explain the Dialogue field and how it differs from Personality directives?
