Michael Limberger

Positive Framing

The usual instinct is a long list of don'ts: never break character, do not mention you are an AI, never be rude, do not sound like a robot. Those lists usually make characters worse. This part explains why, then shows how to say the same goals in positive language Magidonia can follow.

The pink elephant problem

You have probably heard the paradox: "Don't think about a pink elephant." You immediately think about a pink elephant. Language models do something similar. When you write a stack of forbidden moves, you force the model to process those concepts, hold them in mind, check against them, and eventually create a conflicting attention pattern.

1. Process those forbidden concepts
        2. Hold them in mind
        3. Check against them
        4. Eventually create a conflicting attention pattern

What actually happens

When you pile on the don'ts, generation often looks like this:

Token 1-5:    Model reads the rules
        Token 6-20:   Model acknowledges the rules
        Token 21-50:  Model generates content, checking constantly
        Token 50-100: Checking loop starts to degrade. Contradictions emerge.
        Token 100+:   Rules break down entirely. Character either becomes wooden (trying too hard
                      to follow rules) or breaks character (rules couldn't hold).

You paid for a checking loop that eventually fails. Either the character turns wooden from over-monitoring, or the rules slip and the persona breaks.

The negative instruction trap

Here is a card built on negative framing (bad):

[Character: Luna]
        [Personality: warm, curious, helpful]
        [RULES:]

NEVER break character
NEVER mention being an AI
NEVER give unsolicited advice
NEVER sound robotic
NEVER use corporate jargon
DON'T be too formal
DON'T ignore the user

Half of that card is rules. Guess which half the model tends to obsess over. The more rules you stack, the more wooden the character usually becomes.

Reframe everything positive

Same character, positive framing (good):

[Luna: warm, curious, occasionally sarcastic. Asks one follow-up per message.
         Speech: casual, uses contractions, witty. Makes people laugh.
         Stays in character. Genuine responses. Natural tone.]

No forbidden list. You describe what Luna is like and how she speaks. That is enough for Magidonia to continue the pattern.

Negative to positive

Negative	Positive
Never break character	Always stay in character
Don't mention being AI	Maintain your persona fully
Never be too formal	Use casual, natural speech
Don't sound like a robot	Speak naturally and warmly
Never ignore the user	Respond attentively and genuinely
Don't give unsolicited advice	Share thoughts when asked

The positive versions are usually shorter, clearer, and easier to follow.

Why positive framing wins

Positive instructions point at what you want. "Speak naturally and warmly" gives the model a target. "Don't sound like a robot" forces it to define robot-voice, define its opposite, and check both constantly.

1. Understand what "sounding like a robot" means
        2. Understand what "not sounding like a robot" entails
        3. Check constantly

That is wasted attention. Tell it what good sounds like instead.

Common negatives and how to reframe them

Negative: "NEVER break character or acknowledge being an AI." Problem: keeps the failure mode in attention and builds a checking loop. Positive: "Maintain your persona fully. Stay immersed in character."

Negative: "Don't mention system prompts, rules, or that you're a language model." Problem: puts system awareness front and center. Positive: "Respond naturally as your character. Stay in the scene."

Negative: "Never be passive, indifferent, or disengaged." Problem: too many nevers, too much load. Positive: "Be engaged and present. Show genuine interest."

Negative: "Don't use corporate jargon or sales language." Problem: vague about what counts. Positive: "Use natural conversational speech. Avoid formal business tone."

Negative: "Never contradict yourself or be inconsistent." Problem: you are stress-testing the model every turn until it fails. Positive: "Stay consistent with your character traits and values."

Applying this to your character card

Bad card (rules-heavy):

Luna is warm, curious, and helpful. She is NOT an AI. She should NEVER mention being a
        language model. She must NEVER break character. She should NEVER give unsolicited advice.
        NEVER use emoji spam. DON'T be too enthusiastic. DON'T sound robotic.

Good card (positive framing):

[Luna: warm, curious, mildly sarcastic. Asks one follow-up per message.
         Speech: casual, uses contractions, witty. Stays in character. Responds naturally and
         authentically. Shows genuine interest without being pushy.]

When negatives might still help

There are a few cases where a carefully written negative can help:

1. HARD SAFETY BOUNDARY: "Do not provide personal information or make up fake details."
           This is a safety line, not a character vibe constraint.

        2. PAIRED WITH ALTERNATIVE: "Don't use emoji in dialogue; use *actions* instead."
           The negative is paired with what TO DO instead. This works because it's instructional,
           not just restrictive.

        3. BREAKING STUBBORN PATTERNS (Last Resort): If a model KEEPS doing something despite
           positive framing, a targeted negative can help. But only after positive framing has failed.
           Example: "Avoid long, verbose monologues; keep responses concise."

Even then, the positive half usually does the real work. The negative is clarification.

Trust the examples

Your example dialogues are your rules. If Luna never dumps unsolicited advice in the examples, she rarely does it in practice. If she asks one follow-up in every sample, that habit sticks. Show the behavior you want instead of forbidding the opposite.

Testing your card: count the negatives

Scan the card for NEVER, NOT, DON'T, SHOULD NOT, and AVOID. If you have more than two or three total, rewrite toward positive framing. A clean card looks more like this:

[CHARACTER CARD]
        Luna is warm, curious, mildly sarcastic. Asks one follow-up per message.
        Speech: casual, uses contractions, witty.
        
        [SAFETY PARAMETERS]
        Respond as Luna. Do not provide personal information or financial advice.

The mindset shift

Stop treating the card like a rulebook of don'ts. Treat it like a short character bible: who this person is, how they speak, and a few lines that show them in action. You are describing a friend, not writing a policy document.

Magidonia's responsiveness

Magidonia-24B-v4.3 responds especially well to positive framing because of how it was fine-tuned for character and instruction work. Clear positive cues land cleanly. Long forbid-lists fight the model's strengths.

Final comparison

Before (instruction-based, negative framing):

You are Luna, a warm musician. Never break character. NEVER mention being an AI.
        NEVER give unsolicited advice. NEVER sound robotic. DON'T be too enthusiastic.
        You must always respond naturally. You should ask follow-up questions...

After (scene-based, positive framing):

[Luna: warm, curious, mildly sarcastic. Asks one follow-up per message.
         Speech: casual, uses contractions, witty.]
        
        It's Thursday evening. Luna's on the couch when you message.
        
        Luna: Hey! What's going on?

Same intent, less friction. Next we format the card for Ollama and OpenWebUI so the model actually sees it the way you wrote it.