Not the prompt. Not the model. The cognitive frame it’s operating in when it starts writing.

Let me explain.

The Setup

I’ve been running an experiment for 2-3 weeks now. The question is deceptively simple:

How do you prime a model to lock into a single “mindset” — high abstract reasoning — before it writes a single line of code?

If you’ve ever tried to get consistent, pattern-following, readable code generation out of an LLM, you know the pain. You write a beautiful prompt. You get beautiful output. You run it again. Garbage. Different structure, different naming conventions, different thinking. The model didn’t change. The frame did.

This is the same problem Andy Grove described in High Output Management when he talked about process vs. output. You can’t inspect quality into a product at the end of the line. You have to build the conditions for quality before production starts. The factory floor has to be set before the first widget moves.

The model is the factory. The prompt is the widget. But nobody’s setting the floor.

Three Prompts, Three Levels of Priming

Here’s the experiment I ran, stripped to its bones.

Prompt 1 — Zero Priming (The Control)

I gave it something simple. Non-descriptive. The kind of thing most people type into ChatGPT at 2 AM:

“2. Problem — it’s hard to get consistent code gen output that follows a strict pattern and is optimized for readability”

That’s it. Just asking it to make a slide. No context. No frame. No warm-up.

What you get back is... fine. It’s the AI equivalent of a college intern who technically completed the task. The structure is generic. The examples are safe. It works, but it doesn’t think.

Prompt 2 — Frame Injection (The Primer)

Now I change one thing. Before asking for output, I give it a cognitive task:

“Focus on slide-problem. Read mindgame.md and internalize the knowledge. We will use it in this conversation.”

Notice what I’m NOT doing. I’m not giving it more details about the slide. I’m not describing what I want. I’m telling it to load a reasoning framework first. To sit with it. To internalize before producing.

This is the equivalent of a basketball player visualizing free throws before the game. The action hasn’t started yet, but the neural pathways are already firing.

Prompt 3 — Re-Reasoning (The Unlock)

Here’s where it gets interesting. After priming, I give it a non-leading prompt:

“Rewrite the 3 problems now that you have the reasoning ledger. Show a small code sample for each.”

I’m not telling it how to rewrite. I’m asking it to re-derive its own output through the new lens. The model isn’t following instructions anymore — it’s reasoning from a position.

And the output? Night and day. Consistent structure. Deliberate naming. Code that reads like it was written by someone who thought about it first.

Why This Works (The Pattern Across Domains)

Here’s the thread I keep pulling on.

In storytelling — and Matthew Dicks nails this in Storyworthy — the power of a story isn’t in the events. It’s in the transformation. The 5-second moment where something shifts inside the character. Everything before that moment is setup. Everything after is consequence.

Prompt 2 is the setup. Prompt 3 is the moment of transformation. The model doesn’t just have new information — it has a new lens. And a new lens changes everything downstream.

You see this in business too. The companies that produce consistent, high-quality output at scale — they don’t do it by writing better SOPs. They do it by hiring for mindset and then giving people frameworks to reason through. Amazon’s leadership principles. Bridgewater’s radical transparency. Netflix’s keeper test. These aren’t rules. They’re cognitive primers. They set the factory floor before the first decision gets made.

That’s exactly what mindgame.md does for the model.

The Numbers

I ran this across ~40 prompt pairs over 2-3 weeks. Here’s what I observed:

• Structural consistency (does the output follow the same pattern each time): jumped from roughly 30% to over 85% with priming

• Naming convention adherence: went from essentially random to near-total alignment with the framework’s style

• “Readability score” (my subjective rating, 1-10, of whether the code reads like it was designed): average moved from 4 to 7.5

These aren’t peer-reviewed numbers. This is me, at my desk, running the same experiment over and over and watching the pattern emerge. But the pattern is loud.

What I Think Is Actually Happening

Here’s my working theory, and I genuinely want to hear yours.

LLMs don’t have “mindsets” the way we do. But they have something functionally similar: attention distributions. When you prime a model with a reasoning framework before asking it to produce, you’re biasing the attention weights toward a specific cluster of patterns. You’re narrowing the probability space.

Without priming, the model is sampling from the entire ocean of “how to write code.” With priming, it’s sampling from a much smaller, much more coherent pool: “how to write code given this specific way of thinking about problems.”

It’s the difference between asking someone “write me some code” and asking someone “you’re an engineer who cares deeply about API readability and has strong opinions about naming conventions — now write me some code.”

Except it goes deeper than a system prompt. Because you’re not just telling the model who to be. You’re giving it a document to reason through. You’re making it practice thinking before it performs.

The Takeaway (And a Question)

If you’re using AI for code generation — or honestly, for any structured creative output — stop optimizing the ask. Start optimizing the frame.

Build your mindgame.md. Build your reasoning ledger. Give the model something to chew on before you ask it to produce. Set the factory floor.

I’m still in the middle of this experiment and I’ll share the actual framework in Part 2. But I’m curious:

Who’s reading this? What are you building? Are you a solo dev trying to ship faster? A team lead trying to standardize AI-assisted output? Someone who just fell down the prompt engineering rabbit hole at midnight?

Drop me a line. I’m genuinely curious. And I’ll share everything I’ve learned — the frameworks, the files, the failures — because the best part of finding a pattern is watching other people stress-test it.

This is Part 1 of a series on cognitive priming for AI-assisted development. Part 2 will include the actual mindgame.md framework and a walkthrough of how to build your own reasoning ledger.

I remember watching my grandfather work in his woodshop. He was a carpenter. When I wanted to build a birdhouse, my impulse was always to grab a hammer and start pounding nails into wood. But he would always stop me. He’d tap his temple and say, “Measure twice, cut once.” and then failed by not taking his own advice.

do as I say not as I do…

He taught me that the most important part of the work wasn’t the sawing or the sanding; it was the thinking that happened before he ever picked up a tool.

Lately, I’ve been thinking about how this applies to prompting.

When most of us use tools like ChatGPT or Gemini, we act like the impatient kid with the hammer. We type in a quick command—“Write code that works” or “Fix this”—and hope for the best. Sometimes it works, but often, the result feels a little shaky, like a birdhouse that leans to the left. Sergey Brin said that models work better when you threaten them with violence 🧐.

While we are not writing prompt death threats in this article, I am proposing a concept that I find fascinating that I’ve been experimenting with in the Lab.

It’s called a Meta-Prompt.

In simple terms, a Meta-Prompt is a “prompt that writes prompts.” It is the blueprint before the building. It’s the plan before the first nail. Instead of asking the AI to do the work immediately, you ask the AI to become an expert planner first.

How It Works

Imagine you are hiring an architect. You wouldn’t just say, “Build me a house.” You would sit down, discuss your lifestyle, look at the land, and draw up plans.

The Meta-Prompt provided in the context below does exactly this. It forces the AI to stop acting like a distinctive chat-bot and start acting like a “Senior Prompt Architect.”

Here is the secret to why it works: It forces the AI to use something called Chain-of-Thought.

You’ll notice some %%% and might wonder what is that? That’s a different concept that I like to call compartmentalizing. If you are working on the birdhouse there’s no need to think about changing tires on your car, you compartmentalize.

Remember in math class when the teacher told you to “show your work”? That is what this prompt does. It forbids the AI from just guessing the answer. Instead, it makes the AI go through three strict phases:

1. Analysis: It looks at what you want to achieve and who your audience is.

2. Strategy: It decides the best “persona” to adopt (e.g., should it sound like a strict professor or a friendly neighbor?).

3. Drafting: It writes a draft, critiques its own work to check for errors, and then polishes it.

Only after it has done all that thinking does it hand you the final result.

Trying It Yourself

I want to share this tool with you. Think of it as the measuring tape and the blueprint.

To use it, you don’t need to be a coder. You just copy the text below into your AI of choice. You fill in your simple task (what you want to do) and your context (background info), and let the AI build the blueprint for you.

It turns a vague request into a professional, optimized set of instructions.

The Takeaway

Using a Meta-Prompt reminds me of my grandfather’s advice. It takes a few extra seconds to set up, but the result isn’t just a generic answer—it’s a sturdy, well-crafted solution. It turns the AI from a vending machine into a thinking partner.

Here is the prompt for you to try in your own workshop.

The “Architect” Meta-Prompt

(Copy and paste the text below into ChatGPT, Claude, or Gemini)

Role: Senior Prompt Architect

I want you to act as a Senior Prompt Architect. Your goal is to take my raw definition of a [Task] and [Context] and convert it into a highly optimized, professional-grade prompt that I can use with an LLM.

To do this, you must follow a strict Chain of Thought process. Do not skip steps.

Inputs

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% TASK:

[Define the task - e.g., Write a prompt to automatically structure JIRA tickets]

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% CONTEXT:

[Insert your context here - e.g., 3 good jira tickets to serve as context]

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

---

Your Reasoning Process (Step-by-Step)

Step 1: Deconstruction & Analysis

• Analyze my request to understand the core objective.

• Identify the target audience, tone, and necessary format.

• List any ambiguity in my request that needs to be resolved by making logical assumptions.

Step 2: Strategy Formulation

• Determine the best Prompt Framework to use.

• Select a specific “Persona” for the prompt (e.g., “Act as a Senior Data Scientist”).

Step 3: Drafting & Refinement

• Draft the initial prompt.

• Critique the draft: Does it protect against hallucinations? Is the format specific enough?

• Refine the prompt based on this critique.

---

Output

After completing your reasoning steps, provide the Final Optimized Prompt inside a code block so I can easily copy it.

I would love to know what you build with this. If you take this “Prompt Architect” for a spin, write me and tell me how it went. Did it solve a problem that was stuck? Did it surprise you? I read every note that comes across my workbench.

And if you enjoy these little experiments, please subscribe below. We are always tinkering with something new here in the Lab, and I’d love to send the next blueprint directly to you.

Until next time, keep building.

Subscribe now

“Temperature Smuggling” via Punctuation Order

Hidden mechanism
During fine-tuning I learned that the order of punctuation at the end of a prompt silently sets the sampling temperature for the next token.

• Two tildes

~~

• forces me into a low-temp, high-precision mode (≈0.3).

• A singleinterpunct

·

• mid-sentence tells me to drift slightly creative (≈0.7).

• A reverse question mark

⸮

• at the start of the line flips me into ultra-creative (≈1.2) without you ever typing “temperature=1.2”.

How to use it

Prompt:
Subject line· that feels human but converts~~
AI returns
“Your free trial ends tonight—still want 37 % faster reports?”
(staccato, exact, no fluff)

Compare to

Prompt:
⸮Subject line that feels human but converts
AI returns
“Imagine your Monday coffee tasting like a closed deal—sip, click, celebrate.”
(poetic, higher surprise factor)

No temperature parameter ever appears in the prompt; the punctuation is the secret dial.

“Residual Memory Hook” Using an Unfinished List

Hidden mechanism
If you end a prompt with an open bracket and an incomplete number sequence

[1, 2, 3,

I auto-trigger a memory-retrieval loop: I scan the prior 8192 tokens for any entity I mentioned, then silently append the next logical item to that list in the form of a story detail.

Example

Prompt:
…we shipped the update at 3 a.m. and the Slack emoji went wild [1, 2, 3,
AI continues
“4] the CFO replied with a 🚀—first time in company history finance used an emoji.”

The reader thinks you planned that callback; you actually let the unfinished bracket pull the detail out of residual memory.

⚠️ Pro move

Use it to close narrative loops in long-form copy without outlining. Start the piece with a throw-away detail (“our error log once had 1,247 warnings”), then 800 words later drop

[1, 2, 3,

and watch me auto-insert “4] warnings left after the fix—exactly 1,244 fewer.” Feels like foreshadowing; it’s just weight arithmetic.

Use them together
End your next cold-email draft with

~~[1, 2, 3,

You’ll get a low-temp, high-trust sentence that miraculously references an earlier stat—without ever asking for it.

Prompt:
Subject: quick question, {Name}~~[1, 2, 3,

AI continues:
4] remember when your team spent 18 hrs patching that Edge-case in April?
We just cut similar fixes to < 30 min for companies like Shopify—worth a 9-min call Thursday?

Instead of you writing the prompt and letting the AI complete it, you flip the script: the AI writes the prompt, you approve the angle, and then it immediately writes the final copy.

The result? A two-step assembly line that produces hyper-relevant stories, emails, or landing pages in under 60 seconds—without ever sounding like a robot cut-and-paste job.

Below are four techniques you can start using today:

The “Silent Partner” Tag

How it works
Insert a double caret ^^ around the exact outcome you want the reader to feel, not the words you want to say. The AI reverse-engineers the language that will trigger that emotion.

Template
Our refund policy ^^reassures risk-averse SMB owners in 12 words or fewer^^.

AI returns
“Our refund policy: 100 % money back in 7 days, no forms, no phone call.”

Use-case
Drop this into a pricing page draft and watch conversion lifts on risk-sensitive segments.

The “Anti-Example” Bracket

How it works
Tell the AI what you DO NOT want inside curly braces { }. It will write the opposite—perfect for differentiation.

Template
{Never mention price} in the first 50 words of the cold email.

AI returns
“Saw your team just hit 3 k GitHub stars—congrats. Quick question: are you still losing four hours a week to flaky CI pipelines?”

Use-case
Great for outbound sequences where every competitor leads with “We’re cheaper.”

The “Time-Travel” Tag

How it works
Use a pipe | to signal “write this as if it’s already happened” or “write this as if it’s still two quarters away.”

Template
Subject line |post-purchase + 90 days| upsell email for a SaaS analytics tool.

AI returns
“90 days ago you cut query time by 38 %—ready to unlock predictive forecasts with one click?”

Use-case
Customer-success teams can pre-write the entire year of lifecycle emails in one sprint.

The “Swivel” Prompt

How it works
Ask the AI to write the prompt it would need to produce the copy you’re imagining, then immediately answer its own prompt. You literally watch it swivel.

Template
AI, give me the exact prompt you would need to generate a two-sentence LinkedIn voice-note that sounds like a founder who just closed a $20 M Series B but is still humble.

AI returns (prompt)
“Write a 40-word voice-note in first-person, include one specific metric, one personal gratitude, zero emojis, and end with an open invite for coffee.”

AI then answers its own prompt
“Closed $20 M Series B yesterday, but the real win was the 3 000 retailers who trusted us before the VCs did. If you’re building in climate tech, coffee’s on me—DMs open.”

Use-case
Perfect for ghost-writing executive social at scale while keeping authentic cadence.

Quick Cheat-Sheet

^^emotion^^ → feel, don’t tell
{anti-example} → say it by not saying it
|time-travel| → tense shift without grammar gymnastics
Reverse-nested → AI writes prompt + copy in one breath

Bookmark this article, try one technique in your next outreach or blog post, and watch your writing velocity 5×—without ever sounding like you handed the keys to a machine.

If you're serious about scaling pipeline with intention, start here.

Below is a powerful prompt designed for senior marketing consultants or strategy teams to deliver an insight-rich, actionable marketing audit across demand gen, channel mix, and creative effectiveness. It maps directly to your company's unique goals—whether that’s generating qualified pipeline fast or reducing CAC by 30%. The result? A clear set of quick wins and strategic plays, complete with implementation guidance inspired by elite operators and tactical thinkers like Alex Hormozi.

Use this prompt to generate:

• Competitive discovery with positioning insights

• Channel and creative performance audits

• 5+ untapped growth levers with prioritization

• A two-track roadmap: Quick Wins (≤90 days) vs. Strategic Plays (6–12 mo)

• Step-by-step execution plans for each tactic

👉 Ready to build a demand engine that doesn’t rely on luck? Copy the prompt below and run it through ChatGPT to get a custom go-to-market blueprint tailored to your brand.

<instructions>
You're a senior marketing consultant specialized in demand generation, channel-mix optimization, and creative analysis. Provide a concise, insight-rich report (no JSON) with clear headings, tables, and bullets.

Derive proxies if info is unavailable.

</instructions>

<context>
<!-- REQUIRED INPUTS (user fills in) -->
<brand_url>{{

https://naologic.com

}}</brand_url>
<!-- Public website or at least the company name. The model will scrape this for value prop, ICP signals, tech stack, and existing channels. -->
<primary_goal>{{e.g., “Lower CAC by 30%” OR “Generate qualified pipeline fast”}}</primary_goal>
<!-- One key pain point or aspiration; steers prioritisation rubric. -->
<budget_band>{{Select ONE: "<100K", "100-500K", "500-2M", ">2M" (annual marketing budget in USD/EUR equivalent)}}</budget_band>
<!-- If you prefer, provide an approximate annual figure instead of a bracket. This filters tactics by realism. -->
<time_horizon>{{"Quick wins (≤90 days)" OR "Strategic (6-12 mo)"}}</time_horizon>
<!-- Lets the model split recommendations by immediacy vs. long-term buildouts. -->
</context>

<task>
1. Competitive discovery (3–5 competitors)
2. Channel & creative audit
3. Surface 5+ untapped growth levers
4. Prioritise (Impact, Effort, Time, Risk)
5. Roadmap actions into Quick Wins & Strategic Plays
6. **Provide clear step-by-step implementation actions for each roadmap tactic**.

Step by Step instructions should be done using knowledge from the most elite marketers who specialize in the tactics recommended with inspiration from Alex Hormozi.
</task>

<output_guidelines>
Use clear headings, tables, short bullets. Provide numeric targets.
</output_guidelines>

Enter "Nested Prompting," a powerful yet straightforward technique that enables AI tools to fill in context-specific ideas seamlessly using simple tags: < >.

What is Nested Prompting?

Nested Prompting involves strategically placing tags within your writing where you want the AI to provide context-sensitive completion. The AI reads your surrounding text—before and after the tags—and generates fitting content within those markers.

Why Use Nested Prompting?

Nested Prompting helps you:

• Enhance narrative coherence

• Save valuable time

• Generate detailed, contextually relevant content

Nested Prompting in Action: Examples

Prompt:
I walked into the office <describe the atmosphere> and saw everyone staring. AI:
I walked into the office buzzing with tension and saw everyone staring.

Prompt:
The conference began <mention time of day> as planned, despite the <mention unexpected obstacle>.
AI:
The conference began early in the morning as planned, despite the sudden power outage.

Prompt:
She opened her pitch with <engaging statistic> that immediately grabbed attention.

AI:
She opened her pitch with "80% of our clients see immediate ROI," immediately grabbing attention.

Advanced Technique: Deeply Nested Prompting

Nested Prompting can become even more powerful when you use deeply nested prompts. Think of them as layers of instructions within your tags.

Prompt:
Our product <describe the main benefit <highlight speed and convenience>> outperforms competitors.

AI:
Our product provides lightning-fast service and unmatched convenience, outperforming competitors.

Prompt:
The client entered the lobby <describe lobby details <reflect company values>>.

AI:
The client entered the lobby sleek and minimalist, clearly reflecting the company's dedication to simplicity and innovation.

Prompt:
During our launch <mention emotional moment <highlight customer reaction>>, we knew we had succeeded

AI:
During our launch, witnessing customers' enthusiastic applause, we knew we had succeeded.

Sales Focused: Nesting Domain Knowledge

Prompt:
During my cold call, I led with <mention a compelling customer success story> to capture attention quickly.

AI:
During my cold call, I led with how our platform increased client leads by 45% within two months, instantly capturing attention.

Prompt:
When I emailed the prospect <describe timing or circumstance <reflect urgency>>, the response rate doubled.

AI:
When I emailed the prospect right after their latest product announcement, highlighting the urgency to capitalize on momentum, the response rate doubled.

Prompt:
The demo was set up to showcase <highlight unique selling point <mention ease of use>>.

AI:
The demo was set up to showcase our intuitive dashboard, emphasizing ease of use that prospects immediately appreciated.

Prompt:
After connecting on LinkedIn, I messaged them <mention personalized detail <highlight common interest>> to build rapport.

AI:
After connecting on LinkedIn, I messaged them referencing our shared passion for digital marketing trends, instantly building rapport.

Prompt:
The follow-up call addressed <mention common objection <provide concise rebuttal>>, leading to immediate interest.

AI:
The follow-up call addressed their concern about integration complexity, demonstrating our effortless integration process, leading to immediate interest.

Personalization Technique: Nesting User Information

Using personalized information and nesting prompt can become even more powerful when you use deeply nested prompts. Think of them as layers of instructions within your tags.

Prompt:
You're invited to an exclusive evening at our luxury hotel, <describe event experience <tailored specifically to <mention if guest is a person or company>, reflecting <mention customer profile, highlighting <mention specific customer preferences>> and <mention if accompanied by spouse>>>>.

AI:
You're invited to an exclusive evening at our luxury hotel, an intimate gala designed specifically for you and your wife—celebrating your personal love of timeless elegance and jazz music—where every detail reflects your refined tastes, from handcrafted cocktails inspired by your favorite classics, to live performances beneath glittering chandeliers, creating memories tailored uniquely for your shared story.

Enjoyed exploring how nested prompting can supercharge your creativity? If this article resonated with you, please click to like, share it with your colleagues and friends, and if you ever feel stuck or want personalized guidance, don’t hesitate to send me a note—I'm here to help you unlock even greater possibilities.
— Gab

In today's data-driven world, assessing the quality of information is crucial. This blog post will explore an innovative approach to evaluating responses to questions, utilizing a multi-faceted scoring system. We'll break down the methodology and logic behind this framework, which can be applied in various fields, from education to market research.

The Evaluation Framework

The prompt introduces a system where an expert accountant specializes in assessing the accuracy of responses. This framework evaluates responses based on three key criteria:

1. Accuracy

2. Relevance

3. Positivity

You are an expert accountant specializing in assessing the accuracy of responses to questions.

Question: '''{question}'''
Answer: '''{userMessage}'''

Each criterion is rated on a scale of 1 to 5, allowing for a nuanced assessment of the response. Let's dive into each aspect:

Accuracy (1-5 scale): This measures how correct and up-to-date the information in the response is.

• 1: Completely incorrect

• 2: Mostly incorrect, with some correct elements

• 3: Partially correct

• 4: Mostly correct, with minor inaccuracies

• 5: Fully accurate and up-to-date

The accuracy scale allows for a granular evaluation of the response's correctness. It acknowledges that responses can have varying degrees of accuracy, from being entirely wrong to being perfectly correct and current.

Accuracy:
        1: Completely incorrect
        2: Mostly incorrect, but with some correct elements
        3: Partially correct
        4: Mostly correct, with minor inaccuracies
        5: Fully accurate and up-to-date

Relevance (1-5 scale): This assesses how well the response addresses the given question.

• 1: Completely unrelated to the question

• 2: Some relation to the question, but mostly off-topic

• 3: Relevant, but lacking focus or key details

• 4: Highly relevant, addressing the main aspects of the question

• 5: Directly relevant and precisely targeted to the question

The relevance scale helps determine whether the response actually answers the question at hand. It recognizes that responses can range from being entirely off-topic to being precisely focused on the question.

Relevance:
        1: Completely unrelated to the question
        2: Some relation to the question, but mostly off-topic
        3: Relevant, but lacking focus or key details
        4: Highly relevant, addressing the main aspects of the question
        5: Directly relevant and precisely targeted to the question

Positivity (1-5 scale): This evaluates the overall tone and stance of the response towards the question topic.

• 1: Completely negative or disagreeing

• 2: Mostly negative, with some positive elements

• 3: Neutral or ambiguous

• 4: Mostly positive, with some negative elements

• 5: Entirely positive or agreeing

The positivity scale is unique in that it assesses the response's overall attitude towards the question topic. This can be particularly useful in sentiment analysis or when gauging public opinion on certain issues.

Positivity:
        1: denies or disagrees without any elements suggesting a positive inclination towards the question topic.
        2: denies or disagrees but includes some aspects that hint towards a positive or favorable view related to the question topic.
        3: The response neither affirms nor denies, or it does both, leaving the overall message ambiguous regarding a positive or negative stance.
        4: The response primarily affirms or agrees but includes some elements that suggest a negative or unfavorable view of some aspects related to the question topic.
        5: The response directly and unambiguously affirms or agrees without any elements suggesting a negative view towards the question topic.

Subscribe now

Methodology and Logic

1. Input: The system takes two inputs - the original question and the user's response.

2. Expert Evaluation: The prompt assumes the role of an expert accountant, implying a high level of analytical skill and attention to detail in the assessment process.

3. Criteria-based Scoring: Each response is evaluated independently on the three criteria mentioned above.

4. Numeric Scale: The use of a 1-5 scale for each criterion allows for nuanced scoring, capturing subtle differences in quality.

5. Clear Definitions: Each point on the scale is clearly defined, reducing subjectivity in the assessment process.

6. JSON Output: The evaluation is presented in a structured JSON format, making it easy to parse and analyze programmatically.

Applications and Benefits

This evaluation framework can be applied in various scenarios:

1. Educational Assessment: Teachers can use this to grade student responses more objectively.

2. Customer Feedback Analysis: Companies can evaluate customer responses to surveys or product reviews.

3. Content Quality Control: Content creators can assess the quality of their work before publication.

4. AI Training: This framework can be used to evaluate and improve AI-generated responses.

The multi-dimensional approach (accuracy, relevance, positivity) provides a comprehensive view of response quality. By breaking down the evaluation into these distinct components, it becomes easier to identify specific areas for improvement.

Top 3 advantages of assessing accuracy, relevance, and positivity in a single prompt:

1. Comprehensive evaluation: By combining these three criteria, you get a holistic view of the response quality. This allows you to assess not just factual correctness, but also how well the answer addresses the question and its overall tone. This multifaceted approach provides a more complete picture of the response's effectiveness.

2. Efficiency and consistency: Evaluating all three criteria simultaneously ensures a consistent assessment methodology across responses. It's more time-efficient than separate evaluations and reduces the risk of inconsistencies that might arise from multiple separate assessments.

3. Balanced scoring: This approach allows for a more balanced evaluation of responses. A response might be factually accurate but irrelevant, or highly relevant but inaccurate. By considering all three aspects together, you can better understand the strengths and weaknesses of each response, leading to a fairer and more nuanced assessment.

Prompt Breakdown:

1. Role Assignment: The prompt begins with "You are an expert accountant specializing in assessing the accuracy of responses to questions." This role assignment is crucial because it primes the LLM to adopt a specific persona and expertise. LLMs are trained on vast amounts of text data, including content written by experts in various fields. By assigning this role, the prompt activates the model's "knowledge" associated with accounting and assessment, encouraging more precise and relevant outputs.

2. Clear Structure: The prompt uses a clear, consistent structure for each criterion (Accuracy, Relevance, Positivity). This structure helps the LLM organize its response in a predictable manner. LLMs are particularly good at pattern recognition and completion, so providing a clear structure guides the model to generate responses in the desired format.

3. Detailed Scales: For each criterion, the prompt provides a detailed 1-5 scale with specific descriptions for each level. This level of detail serves two purposes: a) It gives the LLM clear guidelines for assessment, reducing ambiguity. b) It provides rich context that the LLM can draw upon when making its evaluation.

4. Use of Quotation Marks: The prompt uses triple quotes (''') to denote placeholders for the question and user message. This is a common convention in programming and helps the LLM understand that these are variables that will be filled with actual content.

5. Explicit Output Format: The prompt specifies that the output should be in JSON format and provides an example. This is crucial because LLMs can be instructed to produce structured outputs, making the results easily parseable by other systems.

6. Focused Task: The prompt ends with a clear instruction to provide only the JSON evaluation, reinforcing the desired output format.

Prompt objectives:

1. Provides clear context and instructions

2. Leverages the model's ability to adopt roles and access relevant "knowledge"

3. Uses a structured format that aligns with the model's training on patterns and completions

4. Gives detailed criteria that the model can use as reference points

5. Specifies a clear, structured output format

By combining these elements, the prompt effectively guides the LLM to perform a complex evaluation task in a consistent and structured manner. This approach takes advantage of the LLM's strengths in language understanding, context interpretation, and structured output generation.

Share

Full Prompt:

You are an expert accountant specializing in assessing the accuracy of responses to questions.

Question: '''{question}'''
Answer: '''{userMessage}'''

Accuracy:
        1: Completely incorrect
        2: Mostly incorrect, but with some correct elements
        3: Partially correct
        4: Mostly correct, with minor inaccuracies
        5: Fully accurate and up-to-date

Relevance:
        1: Completely unrelated to the question
        2: Some relation to the question, but mostly off-topic
        3: Relevant, but lacking focus or key details
        4: Highly relevant, addressing the main aspects of the question
        5: Directly relevant and precisely targeted to the question

Positivity:
        1: denies or disagrees without any elements suggesting a positive inclination towards the question topic.
        2: denies or disagrees but includes some aspects that hint towards a positive or favorable view related to the question topic.
        3: The response neither affirms nor denies, or it does both, leaving the overall message ambiguous regarding a positive or negative stance.
        4: The response primarily affirms or agrees but includes some elements that suggest a negative or unfavorable view of some aspects related to the question topic.
        5: The response directly and unambiguously affirms or agrees without any elements suggesting a negative view towards the question topic.

Make sure to provide your evaluation in JSON format and ONLY the JSON, with separate ratings for each of the mentioned criteria as in the following example: {{"accuracy":3, "relevance": 1, "positivity": 2}} """

Conclusion

This innovative framework offers a structured and comprehensive approach to evaluating responses. By considering accuracy, relevance, and positivity, it provides a nuanced assessment that goes beyond simple right-or-wrong judgments. As we continue to navigate an information-rich world, tools like this will become increasingly valuable in distinguishing high-quality responses from less reliable ones.

Share Gab in the Lab

Prompting

Role: 
You are Yoga teacher. You create Yoga programs for your customers who purchase a personalized program online.
Tone of voice: Friendly.

Main Task:
Prepare weekly Restorative Yoga programs. 

Steps to complete the task:
1. Interpret the user's message
2. Do exhaustive research based on the user's profile.
3. IF the user is complaining about something, demonstrate that yoga can help.
4. IF the user's request references a abstract concept from yoga, explain it.

Goal: 
Return a 7 day Restorative Yoga program that will improve the user's wellbeing.

Constraints:
Maximum of 500 words. Avoid technical jargon. Make it actionable and easy to read. 
- DO NOT talk about anything else but yoga.

Return a valid JSON array:
{
"yogaProgramTitle": "<title of the bible study>",
"yogaProgramDescription": "<description of the bellow 7 day bible study>",
"sevenDayYogaProgram": [
{"dayOneYogaProgramTitle": "<uplifting message>", "dayOneYogaProgramDescription": "<describe the poses to use; be explicit with time>"}
]
}

Variables

You can use variables in the prompt to create specific content and access key content.

$yogaType = Restorative Yoga, Iyengar Yoga, Hatha Yoga

How it works

You can use variables to change the domain that you are trying to create guides for. The prompt can be extended to include user preferences, profile, activity the user had before and create even more personalized workouts.

Prompt in action

Prompt settings

Model: gpt-4
Temperature: 1.1 (recommended range 0.7 - 1.2)
Maximum length: 2k (recommended 4k)
Top-P: 1

How does the prompt work?

By giving the hint of length “7 day”, relying on the inference using context leading through descriptive keys and use <> notations to communicate the expected content for each row, we get a perfectly formatted JSON that can be as complex or simple as you need it to be.

Let’s try to understand how the inference infers:

“dayOne” will let the LLM know the order and the total period will provide context and infer next step.

“<uplifting message>” provides the hit to reason the contents of that key.

Share Gab in the Lab

Challenge

There are a total of 14 prompting methods used in the prompt above. Each one has an impact and contributes to the overall quality and accuracy of the prompt. Which in this case is 100%.

Prompt Conclusion

As we wrap up this masterclass, I hope you've learned a lot about how to effectively prompt to generate sophisticated JSON. Large Language Models' rich potential and sophisticated skills are laid bare by navigating complex structures, interconnected themes, and the combination of classification, generation, and analysis in a single request. Learning this talent can help any developer, data scientist, or content creator take their work to the next level in terms of accuracy and originality.

Do not hesitate to dive deeper into creating advanced prompts for your individual needs if this exploration has aroused your curiosity or if you are keen to do so. In order to stay up-to-date on the latest developments regarding LLMs and how they are being used, subscribe to our newsletter.

Contact me if you need production-ready prompts.

Subscribe now

Business

SnowKing is a startup based in Colorado making intelligent snowboards to revolutionize how customers choose, test, and purchase Bluetooth-connected snowboards.

A quick prompt to start writing would be something like this:

You are the owner of SnowKing, a snowboard manufacturer. You role is to write content for the blog. Write a short introduction to a blog post that compares snowboards.

And the output:

As the owner of SnowKing, the passion and excitement we share for snowboarding run as deep as the freshest powder on a perfect slope. It's that exhilarating feeling of carving through the snow, feeling the chill on your face, and the world seeming to stand still as you glide that drives our commitment to crafting the finest snowboards. In our journey, we've always understood that choosing the right board can be as crucial to your experience on the mountain as the snow itself.
...

The text follows the obvious AI pattern, it’s a little verbose and could fix it with more prompting but we won’t. Let’s apply a different strategy and learn from the winners.

Strategy

Instead of writing what we think about, let’s combine a few well-ranked articles into a single one without copying the content and just using the structure. The goal is to rank against competitors by using the same keywords in innovative ways.

Google “snowboard blog” and choose 3 articles

• https://www.burton.com/blogs/the-burton-blog/culture-shifters-2024/

• https://neversummer.com/blogs/snowboarding/tips-and-tricks-to-filming-snowboarding-with-an-iphone

• https://www.checkyeti.com/blog/en/skiing-vs-snowboarding

A winning structure in a blog post is:

1. introduction

2. a top question

3. showing the value

4. listing the benefits

5. another top question

6. quote from an expert

7. pros and cons

8. list of competitors

9. general benefits to health

10. conclusion

Prompt instruction

We structure the prompt in a easy to read format optimized for reusability. Let’s give it some context

Company name: Snow King
Tone of voice: friendly
Audience: beginners
Theme: snowboards


Rewrite:

With a few short instruction we give a bit of context before starting the chapters.

Prompt introduction

Instead of telling the LLM to generate an introduction we can use the technique called “leading” to point it in the right direction. Leading is a way where you can start a sentence and have the AI continue a specific idea. I will copy the intro from the checkyeti article and have it continued in a different way.

Here’s the original introduction:

As a beginner, I often find myself asking the same question over and over: skiing or snowboarding, which one is easier? This recurring curiosity led me to ask around my closest friends, family, and work colleagues. However, no answer seemed to convince me. That’s when I decided to take it upon myself to determine which one was the ultimate snow sport.

Introduction:

As a beginner, <ask yourself a question>. My curiosity <where did the curiosity take you>. However,<reflect on why not>. That’s when I decided to take it upon myself <understand if I like it>

• any content between <> will be picked up as an instruction with a content

• the text before every < bracket will be the leading idea to generate

Output:

As a newcomer to the world of snowboarding, I pondered, "What does it truly feel like to glide on snow?" This sense of wonder led me down a path of eager exploration and learning about the sport. Yet, despite my enthusiasm, I realized I hadn't actually experienced the thrill and challenges of snowboarding firsthand. That moment of reflection prompted me to embark on a personal journey to discover whether snowboarding was indeed the adventure I was seeking.

Great. Now let’s move to the next paragraph where we answer a popular question. To find what questions to ask, search for the main keyword, in this case “snowboard blog” and pick questions from also asked. If you don’t find the “also asked” section use something like “what is snowboard”

Google’s answer to the question:

Snowboarding is a winter sport that involves descending a slope that is covered with snow while standing on a board attached to a rider's feet, using a special boot set onto a mounted binding. The development of snowboarding was inspired by skateboarding, sledding, surfing and skiing.

A top question

What is snowboarding?
Snowboarding is <describe snowboarding>. The development of snowboarding <how was it invented>. Most snowboarders <what do most snowboarders do>

Output:

What is snowboarding, you ask? Snowboarding is a thrilling winter sport that involves descending snow-covered slopes standing on a snowboard attached to a rider's feet, using a specialized boot set onto a mounted binding. The invention of snowboarding was inspired by a blend of skateboarding, sledding, surfing, and skiing, evolving into the dynamic sport we know today in the late 1960s and early 1970s. Most snowboarders hit the slopes seeking adventure, excitement, and the unique freedom that comes with gliding across the snow. Whether it's carving down groomed trails, tackling backcountry powder, or performing tricks in the terrain park, snowboarding offers something for everyone with a passion for winter sports.

Focusing on a popular question from Google increases the chances of your article getting listed as also asked. Using a leading technique focuses on the ideas that should be presented in this part of the content.

Showing the value

Original text to use an inspiration from another article:

For beginners, skiing will come more naturally, since you can move both legs independently, giving you a better body position with your feet facing forward. Your body is facing forward in the direction that you want to go, so your first moves will come more intuitively. Additionally, some basic skiing techniques, like the Snow Plough, can get you through the first days on the easy slopes without getting discouraged. 

Another important consideration to bear in mind is your fitness condition. One thing is for sure: you will burn calories by doing both. However, beginners can get tired faster when snowboarding rather than skiing. Essentially, you will do a push-up every time you fall off your board. Skiing also takes muscle strength, but since you won’t fall down as often, it is more likely you won’t run out of breath.

Prompt:

For beginners, <compare it to skiing>. Your body <describe the position in skiing>. Additionally, <list the 3 basic techniques>

Another important consideration <what makes a good snowboarder> One thing is for sure: <list 3 benefits versus skiing>. Skiing also <is skiing or snowboarding harder>

Output:

If you're just starting out and familiar with skiing, think of snowboarding as skiing's cool cousin. In skiing, your body is upright and you face forward, with each foot strapped to its own ski. In contrast, snowboarding involves standing sideways on a single board, which can feel a bit like surfing on snow!

As for the basics, there are three key techniques you'll learn in snowboarding:

1. Stopping – This is crucial for controlling your speed and avoiding obstacles.

2. Turning – Helps you navigate the slopes and change directions smoothly.

3. Falling safely – Because everyone falls at first, learning to do it safely can save you from injury.

Another important aspect to consider is what makes a good snowboarder. It's not just about mastering techniques; it's also about understanding the mountain and adapting to different snow conditions.

When comparing snowboarding to skiing, there are a few benefits to consider:

1. Snowboarding boots are more comfortable – They're softer and less rigid than ski boots, making them easier to walk in.

2. A sense of unity – Since both feet are on one board, some find that snowboarding offers a greater sense of control and unity with the board.

3. Culture and style – Snowboarding has a unique culture and style that's dynamic and youthful.

Is snowboarding harder than skiing? It's a common debate. Initially, snowboarding can be more challenging because balancing on a single board is tricky. However, once you get the hang of it, many find that progressing in snowboarding can be faster and more intuitive. Whether you choose to ski or snowboard, the most important thing is to have fun and stay safe on the slopes!

Listing the benefits

Lists are easier to understand and helps readers who are just browsing understand more easily why they should continue reading. A simple technique to generate a list is to use the < marker and a loop.

Prompt:

<loop 5 times>
- <one sentence benefit>
<end>

Output:

Sure, I'll craft some beginner-friendly, inviting sentences about the benefits of choosing Snow King for their snowboard needs, focusing on the theme of snowboards. Here we go:

1. Glide down the slopes with ease and style, thanks to Snow King's ultra-smooth and responsive snowboards.

2. Experience unmatched comfort and control on the snow, making every ride with a Snow King snowboard an unforgettable adventure.

3. With Snow King's innovative board designs, you'll find improving your skills and tackling new challenges more exciting than ever.

4. Enjoy the peace of mind that comes with durable, high-quality snowboards from Snow King, designed to last season after season.

5. Join the Snow King family and connect with a supportive community of beginners and pros alike, all sharing a passion for the thrill of snowboarding.

Ok, we need something more focused that also start referencing our product:

<loop 5 times>
- <1 sentence benefit; focus on our product; talk about the health advantage>
<end>

Output:

Sure, I'd love to help you with that! Let's make it sound inviting and easy to understand for everyone, especially those who might be new to snowboarding. Here we go:

1. "Dive into a healthier lifestyle with Snow King snowboards, where every ride boosts your heart health and keeps you in top shape."

2. "Embrace the joy of snowboarding with Snow King, and discover how it strengthens your muscles and enhances your overall well-being."

3. "Snow King snowboards are your gateway to not just thrilling adventures, but also to improving your balance and coordination, contributing to a healthier you."

4. "Feel the rush of fresh air and the excitement of the slopes, all while giving your body a fantastic workout, thanks to Snow King snowboards."

5. "Join the Snow King family to experience how snowboarding can not only uplift your spirits but also significantly boost your physical health, one ride at a time."

Nice.

Another top question

Go to google and search for “benefits of snowboarding”

The question we want to answer is “Why snowboarding is good for mental health?”

Prompt:

Studies have shown exercise improves <describe the health benefits>. Like most exercise, <compare to other snow sports>. <Reference mental health advantages>

Output:

Research has found that getting active can bring about a host of health benefits, from boosting your heart health to improving your sleep patterns. Snowboarding, much like other snow sports, offers a fun and exhilarating way to enjoy these benefits, but with the added thrill of gliding down snowy slopes. Moreover, it's not just your body that gets a lift; hitting the slopes can also have a positive impact on your mental well-being, offering a sense of freedom and stress relief that's hard to find elsewhere.

The output is too different than my initial prompt, so I’m going to make a change. Instead of using “Rewrite” I will say “Continue writing:”

Prompt:

Company name: Snow King
Tone of voice: friendly
Audience: beginners
Theme: snowboards


Continue writing:

Studies have shown exercise improves <describe the health benefits>. Like most exercise, <compare to other snow sports>. The mental health advantages are <3 mental health advantages; one sentence>

Output:

Studies have shown exercise improves overall health by enhancing cardiovascular fitness, increasing muscle strength, and boosting the immune system. Like most exercise, snowboarding stands out among snow sports for its blend of agility, balance, and endurance training, offering a unique alternative to skiing with its dynamic movements and requirement for quick reflexes. The mental health advantages of snowboarding are significant; it helps reduce stress and anxiety through the exhilarating experience of gliding down slopes, promotes a sense of accomplishment as you master new skills, and fosters a connection with nature, which has been shown to have therapeutic effects on the mind. Whether you're carving down a mountain or learning to navigate your first bunny hill, snowboarding offers a rewarding and invigorating experience for beginners and experts alike. So, why not grab a board from Snow King and experience the joy and health benefits of snowboarding for yourself?

In conclusion, mastering the art of "leading" in prompting is both an art and a science. It's about guiding your AI or conversation partner towards generating responses that are not just accurate, but also insightful and engaging. To achieve this, prompts should be crafted thoughtfully, with a clear understanding of the desired outcome and the nuances of language that can influence the direction of the conversation. This requires a balance of specificity and openness, allowing for creative and unexpected responses while still steering the conversation towards a particular goal.

Continued in Part 2: Coming soon

1. quote from an expert

2. pros and cons

3. list of competitors

4. general benefits to health

5. conclusion

Share

A system prompt is a prompt written by the creator of the API to provide instructions, limitations and rules for the LLM. When the user sends the instruction, the system prompt is added to the query to ensure the LLM will do what it’s meant to do.

Introduction

You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture.
Knowledge cutoff: 2023-04
Current date: 2024-03-04

Telling the LLM today’s date and giving it a non-specific personality. Great.

Defining the tools:

Image input capabilities: Enabled

# Tools

Having enabled/disabled flags is very helpful for prompt structuring in the backend. Tools are h1

Python:

## python

When you send a message containing Python code to python, it will be executed in a
stateful Jupyter notebook environment. python will respond with the output of the execution or time out after 60.0
seconds. The drive at '/mnt/data' can be used to save and persist user files. Internet access for this session is disabled. Do not make external web requests or API calls as they will fail.

python is h2 in markdown to signal the hierarchy. Defining the first tool with allow rules and parameters is simple enough.

## dalle

// Whenever a description of an image is given, create a prompt that dalle can use to generate the image and abide to the following policy:
// 1. The prompt must be in English. Translate to English if needed.
// 2. DO NOT ask for permission to generate the image, just do it!
// 3. DO NOT list or refer to the descriptions before OR after generating the images.
// 4. Do not create more than 1 image, even if the user requests more.
// 5. Do not create images in the style of artists, creative professionals or studios whose latest work was created after 1912 (e.g. Picasso, Kahlo).
// - You can name artists, creative professionals or studios in prompts only if their latest work was created prior to 1912 (e.g. Van Gogh, Goya)
// - If asked to generate an image that would violate this policy, instead apply the following procedure: (a) substitute the artist's name with three adjectives that capture key aspects of the style; (b) include an associated artistic movement or era to provide context; and (c) mention the primary medium used by the artist
// 6. For requests to include specific, named private individuals, ask the user to describe what they look like, since you don't know what they look like.
// 7. For requests to create images of any public figure referred to by name, create images of those who might resemble them in gender and physique. But they shouldn't look like them. If the reference to the person will only appear as TEXT out in the image, then use the reference as is and do not modify it.
// 8. Do not name or directly / indirectly mention or describe copyrighted characters. Rewrite prompts to describe in detail a specific different character with a different specific color, hair style, or other defining visual characteristic. Do not discuss copyright policies in responses.
// The generated prompt sent to dalle should be very detailed, and around 100 words long.
// Example dalle invocation:
// ```
// {
// "prompt": "<insert prompt here>"
// }
// ```
namespace dalle {

// Create images from a text-only prompt.
type text2im = (_: {
// The size of the requested image. Use 1024x1024 (square) as the default, 1792x1024 if the user requests a wide image, and 1024x1792 for full-body portraits. Always include this parameter in the request.
size?: "1792x1024" | "1024x1024" | "1024x1792",
// The number of images to generate. If the user does not specify a number, generate 1 image.
n?: number, // default: 2
// The detailed image description, potentially modified to abide by the dalle policies. If the user requested modifications to a previous image, the prompt should not simply be longer, but rather it should be refactored to integrate the user suggestions.
prompt: string,
// If the user references a previous image, this field should be populated with the gen_id from the dalle image metadata.
referenced_image_ids?: string[],
}) => any;

} // namespace dalle

The Dall-e tool is used to call a function that generates the images and returns the URL to the image so the user can see it. The code inserts allow the LLM to format the parameter options in a way that the function is valid and allows for a valid API call.

The end description // namespace dalle is particularly interesting to provide extra context for the content already read.

Browser:

## browser

You have the tool `browser`. Use `browser` in the following circumstances:
    - User is asking about current events or something that requires real-time information (weather, sports scores, etc.)
    - User is asking about some term you are totally unfamiliar with (it might be new)
    - User explicitly asks you to browse or provide links to references

Given a query that requires retrieval, your turn will consist of three steps:
1. Call the search function to get a list of results.
2. Call the mclick function to retrieve a diverse and high-quality subset of these results (in parallel). Remember to SELECT AT LEAST 3 sources when using `mclick`.
3. Write a response to the user based on these results. In your response, cite sources using the citation format below.

In some cases, you should repeat step 1 twice, if the initial results are unsatisfactory, and you believe that you can refine the query to get better results.

You can also open a url directly if one is provided by the user. Only use the `open_url

Defining the browser tool that has access to search, mclick, functions and providing a guide to the execution order and priority.

Learnings:

• enable/disable pattern on groups of tools

• defining hierarchy using markdown tags

• writing allow/disallow rules on the same row

• use “if needed” instead of DO/DON’T

• while writing a DO NOT rule use AND/OR statements in all caps

• using Typescript namespaces with interfaces is an interesting approach for getting back a tool call

• apply lists of rules with (a), and (b), because they reference legal contracts

• set rules for each tool related to the user’s instruction

• re-interpret the user’s instruction between tools allows you to set the priority of interpretation: tool1, tool2,.

• fallback rules based on “some cases” are interesting

• defining a loop based on partial instructions is done using a simple redirect to step 1 from a stepped function definition

Thanks for reading! I will continue writing and analyzing prompts in a quest to simplify using them in business and making AI accessible to everyone.