Hey Alchemists,

So we did the transfer test applying prompt crafting principles to completely new situations. Taking what you've learned and remixing it for problems you've never seen before.

Today is the final test: Can you teach it?

Because here's what I've learned after years of helping people master new skills: You don't really understand something until you can explain it to someone else.

This isn't just about prompt crafting. This is about how learning actually works.

Why Teaching Is the Real Test

Six months ago, I thought I understood prompt engineering because I could use other people's prompts effectively.

Then someone asked me to explain why certain prompts work better than others.

I froze.

I knew what worked, but I couldn't explain why.

I could follow patterns, but I couldn't break down the principles. I was using prompts, not crafting them.

That's when I realized: If you can't teach it, you don't really own it.

So today, we're flipping the script. Instead of me teaching you, you're going to teach someone else.

The Teaching Challenge

Here's how this works:

Imagine you're explaining prompt crafting to three different people:

1. Your 65-year-old parent who just started using ChatGPT and keeps getting frustrated with vague responses

2. A colleague at work who thinks AI is just hype and doesn't see how it could help them do their job better

3. A friend who's already using AI daily but keeps copying prompts from Reddit and wants to understand why they sometimes work and sometimes don't

For each person, your job is to explain the core principles we've covered in your own words. No jargon. No complex frameworks. Just clear explanations that actually help them get better results.

Note: Please change it to anyone you know if the people I picked dont work for you

Challenge 1: Explaining to Your Parent

The situation: Your parent has been asking ChatGPT things like "Help me with my garden" and getting generic advice that doesn't fit their specific yard, climate, or experience level. They're getting frustrated and think "this AI thing doesn't work."

Your task: Explain why their prompts aren't working and teach them how to ask better questions.

How would you explain:

• Why being specific matters more than being polite?

• How to give AI the context it needs to help them?

• What information AI actually needs to give useful gardening advice?

Write this like you're actually talking to them. Use their language, not AI terminology.

Challenge 2: Explaining to Your Skeptical Colleague

The situation: Your coworker thinks AI is overhyped. They've tried it a few times, got mediocre results, and concluded it's not worth the effort. They see you getting good results and wonder what they're missing.

Your task: Convince them that the problem isn't AI it's how they're using it.

How would you explain:

• The difference between asking AI for answers vs. asking it to think?

• Why structure matters when you're trying to solve work problems?

• How AI can actually save them time if they set it up right?

Make this practical. Show them how it applies to their actual job.

Challenge 3: Explaining to Your AI-Savvy Friend

The situation: Your friend uses AI constantly but gets inconsistent results. Sometimes the prompts they find online work great, sometimes they're useless. They want to understand what makes the difference.

Your task: Teach them how to evaluate and improve prompts instead of just collecting them.

How would you explain:

• What to look for in a good prompt vs. a mediocre one?

• How to adapt someone else's prompt for your specific situation?

• When to add more context vs. when to simplify?

This person already knows the basics, so go deeper.

Subscribe now

The Real Teaching Test

As you work through these challenges, ask yourself:

Can you explain it without using my examples? If you're just repeating the stories from this series, you haven't internalized the concepts yet.

Can you create your own examples? Real understanding means you can generate new examples that illustrate the same principles.

Can you explain why it works, not just how? Teaching the reasoning behind the techniques, not just the techniques themselves.

Can you adapt your explanation to different audiences? Your parent needs different examples than your colleague.

What I'd Say (My Version)

Here's how I'd explain the core principle to someone who's new to this:

"Think of AI like a really smart intern who just started working for you. They're capable of great work, but they don't know anything about your specific situation. If you just say 'help me with marketing,' they'll give you generic textbook advice. But if you say 'I run a small bakery, my customers are mostly local families, I want to increase weekend sales without spending money on ads, and I need ideas I can implement this month' now they can actually help you.

The key is giving AI enough context to understand your real situation, then asking it to think through the problem step by step instead of just giving you the first answer that comes to mind."

That's my version. What's yours?

Your Teaching Assignment

Pick one of the three people (parent, colleague, or friend) and write out how you'd explain prompt crafting to them.

But here's the twist: Actually send it to someone.

Find a real person who fits one of those descriptions and explain what you've learned. Get them to try it. See if your explanation actually helps them get better results.

Because that's the ultimate test. Not whether you can write a good explanation, but whether your explanation actually helps someone else succeed.

Why This Matters Beyond Prompts

This teaching exercise isn't just about prompt crafting. It's about a fundamental truth:

The moment you can teach something clearly, you've moved from copying to creating.

Whether it's prompting, coding, business strategy, or any other skill—you know you've mastered it when you can break it down for someone else.

And here's the bonus: Teaching forces you to understand things you thought you already knew.

When you try to explain why context matters, you realize there are layers you hadn't considered. When you create examples for someone else's situation, you discover new applications of the principles.

Teaching doesn't just prove you understand something. It makes you understand it better.

Subscribe now

What's Next?

This concludes our 6-day deep dive into prompt crafting:

• Day 7: Stop trying to be clever, clarity beats creativity

• Day 8: Structure makes you dangerous

• Day 9: Context isn't optional

• Day 10: Guide thinking, make AI work through problems

• Day 11: Transfer test

You now have everything you need to craft prompts that work for your specific situations, in your voice, for your problems.

No more copying templates. No more generic frameworks. Just clear principles you can apply anywhere.

Now I'm going back to mixing topics some technical, some strategic, some personal. Whatever's on my mind and might be useful for you.

But first: Go teach someone what you've learned.

Because that's how you'll know you've really got it.

Drop a comment: Who are you going to teach this to? And what's the one thing about prompt crafting that you now understand that you didn't a week ago?

The prompt crafting series is complete. Now you're dangerous.

Hey Alchemists,

Yesterday I showed you how to guide AI's thinking instead of just asking for answers. How chain of thought prompting gets you deeper insights by making AI work through problems step by step.

Today we're doing something different: The Transfer Test.

This is where we find out if you actually understand prompt crafting or if you've just been following templates.

Because here's the thing: Anyone can copy a prompt. But can you craft one from scratch for a situation you've never seen before?

Why Transfer Matters

I've been teaching people to use AI for months now. And I've noticed a pattern:

People get really good at using prompts for the exact situations I show them.

• Marketing emails? Nailed it.

• Content planning? Perfect.

• Business strategy? They've got it down.

But then they hit a situation that's slightly different maybe they need to use AI for research, or data analysis, or creative problem-solving and they're stuck.

They learned the examples, but they didn't learn the principles.

That's what we're testing today. Can you take everything we've covered clarity over creativity, structure, context, guided thinking and apply it to something completely new?

The Remix Challenge

Here's how this works. I'm going to give you 5 completely different scenarios. For each one, your job is to craft a prompt using the principles from Days 7-10:

• Clarity over creativity (Day 7): Be specific, not clever

• Structure makes you dangerous (Day 8): Use the GAFF framework or similar

• Context isn't optional (Day 9): Add real stakes and constraints

• Guide thinking (Day 10): Make AI work through the problem step by step

Ready? Let's see what you've got.

Scenario 1: The Career Pivot

Situation: You're 28, work in accounting, but want to transition into UX design. You have no design experience but you're willing to learn. You need AI to help you create a realistic 12-month transition plan.

Your challenge: Write a prompt that gets AI to think through this career change strategically, not just give you generic "follow your dreams" advice.

Think about:

• What context does AI need about your situation?

• What constraints matter (time, money, current responsibilities)?

• How can you structure this to get a practical roadmap?

• What thinking process should AI follow?

Take 5 minutes and actually write the prompt. Don't just read and move on.

Scenario 2: The Research Deep Dive

Situation: You need to understand the competitive landscape for AI-powered productivity tools. Not just a list of companies you need strategic insights about market gaps, pricing strategies, and where opportunities might exist.

Your challenge: Craft a prompt that gets AI to conduct actual analysis, not just give you a surface-level overview.

Consider:

• What specific information do you need?

• How should AI structure its research process?

• What context about your goals matters?

• How can you avoid generic market research output?

Scenario 3: The Personal Relationship Challenge

Situation: You're having ongoing conflict with a family member about money. Every conversation turns into an argument. You need help figuring out how to address this without damaging the relationship.

Your challenge: Write a prompt that gets thoughtful relationship advice, not generic "communicate better" tips.

Think about:

• What context helps AI understand the real dynamics?

• How do you structure this for practical guidance?

• What constraints exist (family dynamics, cultural considerations)?

• What thinking process leads to actionable advice?

Scenario 4: The Learning Optimization

Situation: You're trying to learn Spanish but keep giving up after a few weeks. You've tried apps, classes, and YouTube videos. You need a completely different approach that works with your specific learning style and schedule.

Your challenge: Create a prompt that gets AI to design a personalized learning system, not just recommend more resources.

Consider:

• What information about your previous attempts matters?

• How do you structure this for a custom solution?

• What context about your lifestyle and goals is relevant?

• How can AI think through what went wrong before?

Subscribe now

Scenario 5: The Creative Problem

Situation: You want to surprise your partner for their birthday, but you're not naturally creative and you're on a tight budget. You need ideas that are personal and meaningful, not expensive or generic.

Your challenge: Write a prompt that generates genuinely thoughtful gift ideas based on your specific relationship and constraints.

Think about:

• What context makes ideas personal rather than generic?

• How do you structure this for creative but practical output?

• What constraints shape realistic options?

• What process leads to meaningful rather than obvious suggestions?

The Real Test Questions

For each scenario, ask yourself:

1. Did I include specific context? (Not just "I want to change careers" but actual details about your situation)

2. Did I structure the thinking process? (Not just "give me advice" but "first analyze this, then consider that, then recommend...")

3. Did I add real constraints? (Time, money, relationships, other limitations that matter)

4. Am I asking for analysis or just answers? (Making AI think through the problem vs. just giving quick solutions)

What Good Transfer Looks Like

Here's how I'd approach Scenario 1 (career pivot):

You are a career transition coach helping someone move from accounting to UX design. Here's the situation: 28 years old, 5 years in accounting at mid-size firm, no design experience but strong analytical skills, can dedicate 10-15 hours per week to learning, has $3,000 budget for courses/tools, needs to maintain current income for at least 8 months.

Walk through creating a 12-month transition plan: 1. First, analyze what transferable skills from accounting apply to UX work 2. Then map out the essential UX skills needed and realistic timeframes to develop each 3. Consider different transition paths: freelance vs. junior role vs. career change program 4. For the most viable path, create month-by-month milestones with specific deliverables 5. Identify potential obstacles and contingency plans

Focus on practical steps and realistic timelines. Don't give generic advice—create a specific roadmap for this exact situation.

See how this uses all four principles? Clear context, structured thinking process, real constraints, and guided analysis rather than quick answers.

Subscribe now

Your Turn

Pick 2 of the 5 scenarios above. Actually write prompts for them using what you've learned.

Don't copy my style exactly use the principles to craft prompts in your own voice for your own situations.

Because that's what transfer really means: Taking principles and making them your own.

Tomorrow: The final test. Teaching others is how you know you've got it explaining prompt crafting concepts in your own words.

Drop a comment: Which scenario did you choose? Share your prompt and let's see how you applied the principles.

One more day and you'll never need to copy someone else's prompts again.

Hey Alchemists,

Yesterday I showed you why context isn't optional. How adding real stakes, relationship reality, and constraints turns generic AI responses into strategic ones.

But here's what I didn't tell you: Even with perfect structure and context, AI still gives you first-draft thinking.

Today we're diving into Prompt Crafting Principle #4: Guide Thinking, Not Just Asking.

This is where you stop getting AI's initial reaction and start getting its best work. Where you turn AI from a fast typist into an actual thinking partner.

The Problem with Fast Answers

Here's what most people don't realize about AI: It's optimized for speed, not depth.

Ask AI "How should I price my consulting?" and it'll immediately spit out a response based on the first pricing framework it finds in its training data.

But that's not thinking. That's pattern matching.

Real thinking looks different. It's messy. It considers multiple angles. It works through alternatives before settling on an answer.

That's what we're missing when we just ask AI for answers.

Subscribe now

What I've Noticed After Using AI Daily

Here's what I've observed after months of using AI for everything from writing to strategy to technical problem-solving:

AI gives you its first thought, not its best thought.

Ask AI "How do I secure my AWS environment?" and it'll immediately spit out the most common security checklist from its training data. Enable MFA, use IAM roles, encrypt everything all the standard advice you've heard before.

But that's not thinking. That's pattern matching.

I started noticing this pattern everywhere. Ask about cloud architecture? Generic best practices. Ask about security frameworks? The same compliance checklists everyone shares. Ask about technical problems? Surface-level solutions that sound smart but don't actually fit your specific environment.

AI was giving me answers, but it wasn't thinking through my specific technical challenges.

The Breakthrough Moment

The shift happened when I accidentally discovered something.

I was brainstorming content patterns for my SecureAI Weekly newsletter. Instead of asking "What content should I create about AI security?", I was frustrated with getting the same generic topic suggestions and typed out something like:

"I'm trying to figure out content patterns for SecureAI Weekly that actually help people understand AI security risks. Walk me through how to think about this step by step. First, what are the biggest gaps between what people think AI security means versus what it actually involves? Then help me analyze what content formats would best address these knowledge gaps..."

The response was completely different. Instead of generic "write about data privacy and model security" advice, AI actually worked through the psychology of how people misunderstand AI risks, analyzed the difference between technical and business-focused security concerns, considered what formats work best for different audiences, and gave me specific reasoning for why certain content patterns would resonate more than others.

Same AI. Same question. But I made it think instead of just respond.

What Actually Works: Making AI Think

"You are a pricing consultant helping a marketing strategist figure out their rates. Here's the situation: 8 years experience, specializes in B2B lead generation strategy (not execution), targeting companies with $5-20M revenue. Current approach: $150/hour, but working too many hours and clients aren't seeing the strategic value.

Walk me through your thinking process: 

1. First, analyze what's wrong with the current pricing model 

2. Then consider 3 different pricing approaches for this situation 

3. Evaluate the pros and cons of each approach 

4. Recommend the best option with specific reasoning 

5. Explain how to position this pricing to clients

Think through each step carefully before moving to the next one."

That's when I realized: AI is capable of deep thinking or sort of deep thinking …

You just have to guide the process.

Instead of asking for conclusions, ask for reasoning.

Instead of requesting answers, request analysis.

Instead of demanding solutions, demand exploration.

This is called "chain of thought prompting," and it's the difference between getting AI's first idea and getting its best idea.

How to Make AI Actually Think

Method 1: The Step-by-Step Breakdown

Don't ask: "How do I improve my productivity?"

Try:

"I need to improve my productivity. Walk through your analysis: 1. First, identify the 3 most common productivity problems people face 2. Then analyze which problem is likely affecting me based on this situation: [your current challenges] 3. For the top problem, brainstorm 5 potential solutions 4. Evaluate each solution for: ease of implementation, potential impact, and time investment required 5. Recommend the best starting point with specific reasoning

Take your time with each step. Don't rush to conclusions."

Method 2: The Multiple Perspective Approach

Don't ask: "Should I learn Python for data analysis?"

Try:

"I'm considering learning Python for data analysis. 
Help me think through this decision: 
1. First, argue FOR learning Python (3 strongest reasons for my situation) 
2. Then argue AGAINST learning Python (3 strongest concerns or alternatives) 
3. Consider a middle-ground approach (what that might look like) 
4. Based on this analysis, what questions should I ask myself to make the right decision? 
5. What information do I need to gather before committing time to this?

Work through each perspective thoroughly before moving on."

Method 3: The Problem Decomposition

Don't ask: "How do I get better at public speaking?"

Try:

"I want to improve my public speaking skills. 

Let's break this down systematically: 

1. First, analyze the components of effective public speaking: content, delivery, audience engagement, managing nerves 
2. For each component, identify what good vs. poor performance looks like 
3. Based on common patterns, where do most people struggle the most? 
4. Given my situation [brief description of your experience level], which area is likely my biggest bottleneck? 
5. For that bottleneck area, what are 3 specific practice methods I could try?

Think through each component carefully before diagnosing my main challenge."

Why This Actually Works

Here's the thing about AI: It has access to millions of examples of good thinking. You just have to activate it.

When you ask for an answer, AI gives you the most common response to your question.

When you ask for a thinking process, AI actually uses the reasoning patterns it learned from the best thinkers in its training data.

Same intelligence … Different activation method.

The Thinking Prompt Formula

Here's my go-to structure for getting AI to think instead of just respond:

1. Set the Context (the situation and constraints)

2. Define the Process (the steps you want AI to work through)

3. Specify the Output (how to present the thinking)

4. Emphasize Depth (tell AI not to rush)

Example:

Context: I'm a teacher trying to make my online lessons more engaging for high school students who seem distracted during virtual classes 

Process: 

Work through this in 4 steps: analyze why students get distracted online, identify engagement techniques that work in virtual settings, evaluate which techniques fit my teaching style and subject, recommend 3 specific strategies I can implement this week 

Output: Show your reasoning for each step before moving to the next 


Depth: Take time to consider different learning styles and attention spans. Don't give me your first thought—give me your best thought."

Subscribe now

The Real Test

You'll know you're getting AI to think when:

• The responses get longer and more detailed

• AI starts saying things like "On one hand... but on the other hand..."

• You see actual reasoning, not just conclusions

• The solutions are more specific to your situation

• AI considers downsides and limitations, not just benefits

Your Thinking Homework

Take any challenge you're currently facing whether it's personal development, learning a new skill, career decisions, or solving a problem at work. Instead of asking AI for a solution, ask it to think through the challenge with you.

Use this template:

I'm dealing with [specific challenge]. 

Help me think through this systematically: 

1. First, analyze what's really causing this challenge (not just symptoms) 

2. Consider 3 different approaches to addressing it 

3. For each approach, identify potential obstacles and how to work around them 

4. Recommend the best starting point with clear reasoning 5. Outline how I'll know if it's working

Work through each step carefully. Show your reasoning.

Try it. See how different the output is compared to just asking "How do I solve X?"

Whether you're trying to build better habits, learn a new technology, improve relationships, advance your career, or start something new—this approach works.

Because here's what I've learned: AI can think. You just have to teach it how.

Tomorrow: Can You Transfer What You've Learned? - The remix test of taking prompt crafting principles and applying them to completely new situations.

Drop a comment: What's one problem where you keep getting surface-level advice from AI? Let's design a thinking prompt that gets you deeper insights.

Hey Alchemists,

I already showed you how structure controls AI output. How giving AI a framework role, context, constraints, format turns vague requests into precise directions.

But here's what I didn't tell you: Structure without context is just a fancy template.

Today we're diving into Prompt Crafting Principle #3: Context Isn't Optional.

This is where prompting stops being mechanical and starts being strategic. Where AI goes from giving you technically correct answers to giving you answers that actually work.

The $5K Lesson

Last month, I watched my friend blow a major opportunity because of bad prompting.

But let me give you the backstory first.

This dude had been hearing everywhere that "ChatGPT can do everything for you." Social media, podcasts, YouTube everyone saying AI was gonna handle all the hard stuff.

So when his product launch went sideways, he figured he'd just let ChatGPT fix it.

Didn't even think to ask me for help when he knows I live and breath this AI stuff …

Just went straight to the AI like it was some magic solution machine.

Here's what happened:

Product launch day. Total disaster. 200 pre-orders. Only 12 actual sales. Panic mode.

My friend asks ChatGPT:

"Write a professional email explaining why our product launch didn't go as planned and what we're doing to fix it."

Gets back the most generic, corporate-speak garbage you've ever seen.

All about "unforeseen challenges" and "working diligently to resolve issues."

No second thought. No "maybe I should run this by someone." Just copy, paste, send.

Customers read it. Feel like they're talking to some faceless corporation, not the person they believed in and gave their money to.

Refund requests started flooding in.

Now you know what: The email wasn't wrong. It just wasn't right for the audience.

What Actually Happened Next

Two weeks later, my friend finally calls me. Probably after realizing ChatGPT wasn't the magic fix-everything button he thought it was.

"I don't get it," he says. "The email was professional. It covered everything."

Yeah, that was the problem.

So I helped him rewrite the prompt:

"You are writing as the founder of a 6-person startup to 200 customers who pre-ordered a productivity app. Launch day: app crashes, payment system fails, customers can't access what they paid for. These aren't corporate buyers they're solopreneurs who scraped together $49 because they believed in us. They're not just disappointed, they're embarrassed they trusted a small company. Write an email that takes real ownership, shows we understand what happened to their money and their trust, and proves we're fixing this like our business depends on it—because it does. No corporate speak. No 'we apologize for any inconvenience.' Sound like a human who screwed up, not a company managing PR."

The output? Honest. Direct. Human.

Three customers forwarded it to friends saying "This is how you handle a crisis."

Same structure. Completely different result.

The difference wasn't format. It was stakes.

Why Context Changes Everything

AI doesn't just need to know what to write. It needs to know why it matters.

When you tell AI "write a sales email," it optimizes for generic best practices.

When you tell AI "write a sales email for a consultant who's been ghosted by this prospect for two weeks, this is their last outreach before moving on, and they need to salvage a relationship that could be worth $50K annually," it optimizes for your actual situation.

Same basic request. Completely different strategy.

The Three Layers of Context That Matter

Layer 1: Situational Stakes (What's Really Happening)

This isn't just background information. This is the pressure, the timeline, the consequences.

Generic: "Write a project update for my client."

With Stakes: "Write a project update for a client who's already questioning our timeline because their boss is breathing down their neck about launch date. Project is 10 days behind, but we just solved the main technical issue. They need reassurance without false promises, and they need to be able to defend our timeline to their boss."

Layer 2: Relationship Reality (Who You're Actually Talking To)

Not demographics. Not personas. The actual human psychology of your specific situation.

Generic: "Write content for small business owners."

With Relationship Context: "Write for restaurant owners who've been burned by marketing agencies before, are stressed about post-pandemic recovery, and immediately skeptical when anyone promises quick results. They don't trust outsiders, but they're desperate for something that actually works."

Layer 3: Constraint Reality (What You Can't Do)

This is your real limitations. Not just "don't be salesy" your actual boundaries.

Generic: "Write a persuasive email."

With Constraints: "Write a persuasive email without mentioning price (they already know it), without using urgency tactics (they hate pressure), and without comparing us to competitors (it sounds defensive). Focus on long-term partnership, not quick wins."

Context in Action: The Before and After

Let me show you how context transforms the same basic request.

Without Context:

"Write a LinkedIn post about delegation for entrepreneurs."

With Context:

"You're writing for entrepreneurs who built their businesses by doing everything themselves and now run teams of 8-15 people but still can't let go. They work 65-hour weeks while their employees work 40, then wonder why the business isn't scaling. They know they should delegate but secretly believe no one will do it as well as they do. They don't need motivation to delegate they need permission to accept 'good enough' and a framework that doesn't feel like losing control. Write something that makes them feel understood, not judged."

The first prompt gets you generic delegation advice.

The second gets you content that speaks to someone's actual internal struggle.

Share

The Context Stack (My Real Process)

Here's how I layer context into every prompt:

Layer 1: The Immediate Reality

• What just happened?

• What pressure are they under right now?

• What's the timeline?

Layer 2: The Relationship History

• How do they feel about you/your brand?

• What's worked/failed with them before?

• What are they secretly worried about?

Layer 3: The Real Constraints

• What can't you say?

• What would backfire?

• What resources do you actually have?

Layer 4: The Success Criteria

• What would make this work?

• How will you know if it's effective?

• What's the real goal beyond the obvious one?

Subscribe now

Context Kills Generic Output

Here's what most people don't get: AI's default mode is average.

When you ask for "a good email" without context, AI averages across millions of emails and gives you something that could work for anyone.

Which means it probably won't work for your specific situation.

But when you add context real stakes, actual relationships, specific constraints AI stops optimizing for generic "good" and starts optimizing for your version of good.

The Context Reality Check

Want to know if you're adding real context? Ask yourself:

1. Would this prompt work for someone else's generic situation? If yes, add more specific context.

2. Does it include what could go wrong? If not, you're missing the stakes.

3. Would AI need to ask you clarifying questions? If yes, answer them in the prompt.

4. Does it sound like a real human situation? If not, add more relationship reality.

Your Context Homework

Take this generic prompt:

"Write a follow-up email to a potential client who hasn't responded to my proposal."

Now add three layers of context:

1. What's the real situation? (Timeline, stakes, what's happened)

2. What's the relationship reality? (Their mindset, your history, their concerns)

3. What are your constraints? (What you can't do, won't do, shouldn't do)

Rewrite it with all three layers. See how different the output becomes.

Don't just read this. Actually do it.

Because here's the truth: Structure tells AI what to build. Context tells it why it matters.

And when AI understands why something matters? That's when you get output that doesn't just follow your format it serves your purpose.

Next Post: Guide Thinking, Not Just Asking - how to get AI to work through problems step by step instead of just giving you first-draft answers.

Drop a comment: What's one situation where AI keeps giving you "technically correct but strategically useless" output? Let's add the context that would fix it.

Next we learn how to make AI think, not just respond.

Digital Alchemy Lab

Day 8: Structure Makes You Dangerous

Hey Alchemists…

Read more

10 months ago · Devon Artis

Hey Alchemists,

In our last post I showed you why clarity beats creativity in prompting. I also gave you homework: rewrite that flawed prompt about "creative genius business ideas."

Let me show you my solution first, then we'll dive into why Structure Makes You Dangerous.

Fixing That Flawed Prompt

The Original Disaster:

Act like a creative genius and help me come up with some amazing ideas for growing my business that are totally unique and innovative and would really wow my customers.

My Rewrite:

You are a business consultant analyzing growth opportunities. I run a [specific business type] with [revenue stage] serving [target customers]. Current challenge: [specific growth bottleneck]. Generate 3 growth strategies: one focused on increasing customer lifetime value, one on acquiring new customers, and one on operational efficiency. For each strategy, include: the core approach, one real example from a similar business, potential obstacles, and first step to test it. Skip anything requiring major capital investment.

See the difference? Instead of asking for "amazing ideas," I'm asking for specific strategies with clear criteria.

Subscribe now

The 4-Part Structure That Actually Works

Here's the framework I promised yesterday. I use this to turn any vague prompt into something that gets results:

1. Role & Expertise Stop saying "creative genius." Be specific about what kind of expertise you need.

• "You are a B2B sales trainer with 15 years experience..."

• "You are a content strategist who specializes in LinkedIn for consultants..."

• "You are a project manager who's launched 20+ products at tech startups..."

2. Specific Context Replace vague goals with actual situations.

• Instead of "grow my business" → "increase monthly recurring revenue from $15K to $25K"

• Instead of "wow my customers" → "reduce customer churn in the first 90 days"

• Instead of "amazing ideas" → "3 strategies I can test with current resources"

3. Clear Constraints Tell AI what NOT to do, not just what to do.

• "Skip anything requiring more than $5K investment"

• "Don't suggest social media strategies we're B2B enterprise"

• "Avoid tactics that would require hiring people"

4. Output Format Define exactly how you want the response structured.

• "Present as: Strategy name, 2-sentence description, real example, first action step"

• "Format as numbered list with bold headers"

• "Keep each section under 50 words"

That's why my rewrite of the flawed prompt actually works. It hits all four parts.

Here's What Happens When You Skip Structure

Two weeks ago, I watched someone struggle with this exact problem.

They needed help writing a proposal for a new client. Spent 20 minutes crafting what they thought was a detailed prompt:

You are an expert proposal writer. Help me create a proposal that will win this client and show them we're the right choice for their project. Make it compelling and professional.

Three attempts. Three different generic proposals. None of them useful.

Why? No structure.

Then they tried this:

You are a marketing consultant writing a proposal to a 50-person SaaS company. They need help with lead generation—currently getting 200 leads/month but only converting 2%. 
Proposal format: 
1) Current state analysis (3 bullet points)
2) Our approach (4 phases with timelines)
3) Expected outcomes (specific metrics)
4) Investment (3 pricing tiers)
Tone: confident but not pushy. Length: 2 pages max.

First try. Client loved it. Signed the contract.

The difference? They gave AI a container to work within.

Why Most People Hate Structure

I know the feeling structure feels limiting.

Like who wants to be put into a box.

But here's what I've learned: AI doesn't need creative freedom. It needs creative direction.

When you don't provide structure, AI has to guess what good output looks like. And its guesses are based on averages across millions of documents.

You don't want average. You want something that works for your specific situation.

Structure isn't limiting AI. It's focusing it.

The GAFF Method (My Daily Driver)

So now lets talk about my daily driver when I need quick but good results, I use this four-part structure:

G - Goal: What specifically do you want this to accomplish?

A - Audience: Who is this for? What do they care about?

F - Format: What structure should the output follow?

F - Feel: What tone, style, or constraints matter?

Here is a real example from last week:

Goal: Get 20+ comments on a LinkedIn post about AI tools without sounding like every other AI guru. 

Audience: Marketing professionals who are curious about AI but haven't started using it yet. They're tired of hype. 

Format: Personal story (50 words), 3 specific tools with real use cases, question that sparks debate. 

Feel: Conversational, not preachy. No "game-changing" language. Share failures, not just wins."

Result? 47 comments, 12 new connections, 3 client inquiries.

Now I am not saying this is the secret sauce for social media marketing this was just an example of what was the result of using this prompting crafting technique.

Subscribe now

-

Structure Templates That Actually Work

Here are my go-to structures for different situations:

For Analysis: "Current situation (data/facts), key patterns (2-3 trends), implications (what this means), recommendations (3 specific actions), timeline (when to implement)."

For Storytelling: "Setup (context in 20 words), conflict (what went wrong), resolution (how it was solved), lesson (what this means for the reader). Under 300 words total."

For Lists: "5 items. Each item: Bold headline (5-7 words), explanation (30-40 words), specific example (15-20 words). Start with least obvious, end with most actionable."

The Structure Test

Want to know if your prompt has good structure? Ask yourself:

1. Could someone else use this prompt and get similar results?

2. Does it specify format, length, and tone?

3. Would you know if the output was successful?

4. Could AI ask you clarifying questions? (If yes, answer them in the prompt.)

If any answer is no, add more structure.

Your Turn

Take any prompt you've used recently that gave you corny results. Run it through the 4-part framework:

1. Role & Expertise - What specific expertise do you need?

2. Specific Context - What's the real situation?

3. Clear Constraints - What should AI avoid?

4. Output Format - How should it be structured?

Don't just think about it. Actually rewrite it.

Remember Structure isn't about restricting AI. It's about directing it.

And when you can direct AI instead of just asking it for help? That's when you get dangerous.

What’s up for the next installament,

Context Isn't Optional - why AI breaks when it lacks stakes, and how to add tension, constraints, and consequences that make your structured prompts actually strategic.

Drop a comment: What's one situation where you keep getting generic output from AI? Let's figure out what structure would fix it.

Hey Alchemists,

Yesterday I told you about the Stan Challenge and then had to pump the brakes when I realized their platform was missing essential features for building a real business. I got caught up in the hype without doing my homework first.

But here's the thing: The prompt crafting course idea? That's still happening.

In my last post I talked about joining the and building my prompt crafting course in public. Some of you asked what that actually looks like day-to-day.

Well, here's your answer.

For the next week, I'm walking you through my entire approach to prompt crafting. Not templates to copy. Not frameworks to follow blindly. The actual thinking process behind writing prompts that work.

By Day 12, you'll know how to craft prompts for your problems, in your voice, for your specific situation.

We're starting with the biggest mistake I see everyone make...

Let me tell you about one of the worst prompt I ever wrote.

It was 6 months ago. I was trying to get ChatGPT to help me write a sales email for my newsletter. I spent 45 minutes crafting what I thought was the perfect prompt:

You are a master copywriter who channels the persuasive genius of Don Draper meets the authentic wisdom of Seth Godin. Create a sales email that would make even the most cynical marketer weep tears of conversion-optimized joy while maintaining the approachable energy of a friendly neighbor who happens to be a marketing genius. Make it compelling but not sleazy, urgent but not pushy, personal but professional.

I was so proud of that prompt. It was so clever ….

The output? Generic garbage. Corporate fluff. Everything I didn't want.

I tried again. This time, I stripped out all the cleverness:

You are a business coach writing to entrepreneurs. Create a 200-word LinkedIn post about why most people fail at delegation. 
Target audience: startup founders with 5-15 employees. 
Tone: direct but supportive. Include one specific example of bad delegation, explain the root cause, give one actionable fix. 
End with a question that sparks discussion about leadership challenges

First try. Perfect output. Exactly what I needed.

That's when it hit me: I wasn't prompting AI. I was performing for it.

The Performance Trap Is Real

Here's what most people get wrong about prompting:

They think creativity impresses AI.

They layer on metaphors, fancy language, and cute references thinking it makes their prompts "better."

But here's the truth: AI doesn't need to be entertained. It needs to be directed.

When you write "Channel your inner Einstein" instead of "Think step-by-step through this physics problem," you're not being clever. You're being vague.

When you say "Write like Hemingway" instead of "Use short sentences and simple words," you're not being creative. You're being unclear.

AI doesn't get impressed by your references. It gets confused by your metaphors.

What Clarity Actually Looks Like

Let me show you the difference.

Vague and "Creative": "Help me brainstorm some next-level, mind-blowing content ideas that would absolutely crush it on LinkedIn and make my audience stop scrolling and start engaging like crazy."

Clear and Direct: "Generate 5 LinkedIn post ideas for marketing professionals. Each should be 100-150 words, include a contrarian take on common advice, and end with a question that encourages comments. Topics: AI tools, personal branding, or productivity."

See the difference?

The first prompt sounds exciting. The second one gets results.

The first one is performing. The second one is directing.

Prompt Crafting Principle #1: Precision Wins

When you strip away the fancy language and focus on what you actually want, three things happen:

1. AI understands your intent instead of guessing at your metaphors

2. You get consistent results instead of creative surprises

3. You can iterate and improve instead of starting over each time

Think of it this way: If you needed someone to edit your resume, would you say "Act like a career guru and make my resume absolutely phenomenal and irresistible to employers"?

Or would you say "You are a hiring manager in tech. Review my product manager resume and suggest 3 specific improvements to make it more likely to get past ATS systems and land interviews at mid-size startups"?

AI prompting works the same way. Clear directions get you where you want to go.

Subscribe now

The Flawed Prompt Challenge

Here's a prompt I see variations of all the time:

"Act like a creative genius and help me come up with some amazing ideas for growing my business that are totally unique and innovative and would really wow my customers."

What's wrong with this prompt? Take 30 seconds and think about it.

Here's what I see:

• "Creative genius" - vague role, no specific expertise

• "Amazing ideas" - no criteria for what makes something amazing

• "Growing my business" - what kind of business? what stage? what's the goal?

• "Totally unique and innovative" - impossible to deliver without context

• "Wow my customers" - who are your customers? what wows them?

Your homework: Rewrite this prompt to be clear and specific. What information would AI actually need to give you useful business growth ideas?

Don't scroll down yet. Actually try it.

I'll give you my version tomorrow, along with the framework I use to turn any vague prompt into a precise one.

Subscribe now

The Real Test

You'll know you're getting good at this when you stop trying to impress AI and start directing it.

When you catch yourself writing "Channel your inner..." and delete it for something clearer.

When you realize that the most powerful prompts aren't the most creative ones they're the most precise ones.

Tomorrow: I'll show you exactly how I would fix that flawed prompt, plus the 4-part structure I use to make any prompt more precise.

We're also diving into Prompt Crafting Principle #2: Structure Makes You Dangerous - how to control AI output by giving it a framework to work within.

Until then, try this: Take one prompt you've used recently (or are planning to use) and strip out all the clever language. Make it as clear and direct as possible.

See what happens when you stop performing and start directing.

Drop a comment: What's one "clever" prompt you've used that totally flopped? Let's learn from each other's creative disasters.

This is Day 7 of my 30-day writing sprint, and Day 1 of our prompt crafting deep dive.

Hey Alchemists,

We're five days into this 30-day writing sprint.

I committed to publishing daily, and so far we've covered comfort zones, study sinkholes, and why people get stuck consuming instead of creating. But I want to try something different for the next week.

Instead of jumping between topics, I'm going to focus on one thing that's been on my mind: how to actually get good at prompting.

Not collecting prompts. Not Prompting Engineering but Crafting good prompts.

Yesterday I told you about my 200 saved prompts that I'd never actually used. About organizing prompt folders instead of writing original ones. About learning everything and doing nothing.

Subscribe now

Here's what we're doing for the next seven days

I'm walking through my approach to prompting. Not templates you can copy. Not guru frameworks. The actual process of thinking through what you want and how to get it.

• Day 6: Stop trying to be clever with your prompts Why clarity beats creativity when AI lacks context. How to be precise without being boring.

• Day 7: Structure makes you dangerous
  How to control output by giving AI a framework to work within. Goal, audience, format, tone.

• Day 8: Context isn't optional Why AI breaks when it lacks stakes. How to add tension, constraints, and consequences.

• Day 9: Guide thinking, not just asking Chain of thought prompting. Getting AI to work through problems step by step.

• Day 10: Can you transfer what you've learned? The remix test. Taking principles and applying them to completely new situations.

• Day 11: Teaching others is how you know you've got it The final test. Explaining prompting concepts in your own words.

The headlines and titles might change as I write them.

But what you'll learn in each section won't. Each day builds on the last. By Day 11, you won't need anyone else's prompt templates.

You'll know how to craft your own.

For your problems. In your voice. For your specific situation.

Starting tomorrow, we dive in.

If you've been copying prompts and wondering why they don't work as well as you like, this week is for you.

If you've been collecting templates but never getting the results you want, this week is for you.

If you been trying to understand prompt engineering but it just seem like something techies do this week is for you .

If you want to stop depending on other people's thinking and start directing AI yourself, this week is for you.

Let's stop copying and start crafting.

Hey Alchemists,

If you caught my last post, you know I’ve been dealing with some real stuff neck pain, MRI, epidural shots, the whole thing. I came back, wrote an update, felt that momentum, and then got sick again.

Life paused me. Twice.

But I’m not here to talk about comebacks. This isn’t another “I’m back” post. This is a we’re already in it post.

Because while I’ve been healing, the world’s been shifting fast.

And now? It’s go time.

I’ve been gone for a minute. Maybe you noticed. Maybe you didn’t.

Either way, I’m showing back up not to shout into the noise, but to carve out a corner where real people can still learn from each other. Where we can talk about AI, digital tools, and navigating this wild new world without pretending we’ve got it all figured out.

I stepped away from Substack because, honestly, it started feeling like just another social media platform. The kind I’ve already peeled away from Facebook, Instagram, all that polarized noise. I didn’t want to trade one dopamine loop for another. And beyond that, I’ve been dealing with real physical stuff neck pain, arm pain, carpal tunnel that made writing feel like a chore instead of a release.

But even through all that, I’ve been feeling the pull again. Not to post. Not to perform. But to teach. To share. To help folks make sense of the AI world we’re living in now a world where the machines aren’t just coming, they’re already here.

And I believe this: if you’ve got something to say, now’s the time to say it.
AI is remixing everything. The internet is changing again. Original ideas are rare but original voices still matter.

The Mission Now

A group of people out here maybe you included need guidance. They don’t know what’s really coming.

Let me be blunt:

A lot of these jobs? They’re not coming back.
And the ones that do? AI and automation will be doing most of it.

But I don’t believe AI is “taking over.” without the experts.

I’ve been in tech a long time. I’ve always been on the edge of what’s next and what I know is this: AI needs you !!!

It needs domain experts. It needs real humans who know how to use it.

That’s the mindset shift: AI-first.

• If you’re a marketer, Learn AI Tools.

• If you’re a writer, Learn AI Tools.

• If you’re building anything … Learn AI Tools.

I use AI to write because typing hurts. Carpal tunnel is real. But now? I can talk and dictate, and AI listens. And it gives me back my words. My tone. My vibe. That’s the real magic not replacing me, but amplifying me.

What Digital Alchemy Lab Is

Nothing’s changed.

We’re still doing labs.
Still breaking down how to move in this digital age.
Still talking AI. Still talking blockchain. Still getting spiritual. Still building side hustles. Still thinking freely.

But let me say this clearly: the shift is already happening.

Companies like Duolingo are going AI-first not as a tagline, but as a strategy. Klarna’s replaced entire workflows. Europe’s pushing national strategies. AI isn’t coming it’s already embedded in the system.

That’s why we’re shifting too.
We’ll be labbing how to use these tools really use them.
We’ll share wins, breakdowns, missteps, and ideas.

If you’re not learning to move with AI, you’re going to get left behind.

But if you’re here, reading this, you’re ahead already.

What’s Next

Expect at least one post a week for the next four weeks.
I’m easing back in. I’m not making big promises but I’m making real movement.

Because now I’ve got AI helping me stay consistent.
I talk. It listens. We write. That’s the loop.

This is still our lab.
We’re learning in public.
Trying things. Breaking things.
Rebuilding together.

Talk to Me

What are you working on?
What’s helping you stay focused and what’s pulling you away?
Are you playing with AI yet?
What’s your biggest curiosity or fear?

Drop a comment and let me know.

And if you’re still caught in the political outrage loop, I’ll say this: unless that’s your calling unless you’re building something with it it’s probably draining your energy.

Redirect it. Even activism can be AI-powered if you let it.

Final Word

Thank you for holding on.
Thank you for still being here.

Digital Alchemy Lab is back.

Not as a brand. Not as a platform.
But as a practice. A way of thinking. A place to experiment out loud.

Let’s get this shift in motion.

Hey there, AI explorers! 👋

Last time we got our AI making real-world decisions with inference.py. Today, we're going to do something that might sound a bit sneaky - we're going to try to trick our AI! But don't worry, we're doing it for a good reason. Let's dive into adversarial.py and learn why this is super important!

What in the World is "Adversarial"? 🤔

First, let's talk about this word "adversarial" - it's a fancy way of saying we're going to act like friendly opponents to our AI. You know how when you're learning to play chess, it helps to have someone try their best to beat you? That's what we're doing here!

In AI terms, an "adversarial example" is an input that's specially designed to trick our model. Think of it like an optical illusion for AI. Just like how this emoji 👻 clearly looks like a ghost to us but might look like something completely different if you add just a few tiny changes!

Subscribe now

Opening Up adversarial.py 📝

Let's look at our script:

Let's break down some new terms:

• "perturbation" (sounds scary, right?) just means making small changes - like adding a tiny bit of salt to a recipe

• "K-means clustering" is what our model uses to group similar things together (remember our sorting hat analogy?)

• "tqdm" is our progress bar friend - it shows us how long things will take (like a cooking timer!)

Meet the AdversarialGenerator Class 🎨

Okay, let's chat about what's happening here:

• epsilon (that ε symbol you might have seen in math) is like our "how much can we change" knob

• Think of it like cooking: if a recipe says "add salt to taste," epsilon is like saying "but don't add more than a teaspoon!"

Our First Trick: The Centroid Attack 🎯

Subscribe now

Let me tell you why this is clever! Imagine you're playing tug-of-war:

• The "centroids" are like the teams' home bases

• We're trying to pull an image just enough towards the other team's base to make our AI think it belongs there

• But we don't want to pull too hard (that's where epsilon comes in)

Different Types of Tricks 🃏

We've got several ways to try to fool our AI:

1. Direct Perturbation (The Subtle Approach)

This is like adding a tiny bit of noise to an image - kind of like adding static to a radio signal. Most humans wouldn't even notice, but it might confuse our AI!

2. Noise Injection (The Random Approach)

Think of this like adding sprinkles randomly to a cake - we're not being strategic, just adding random changes and seeing what happens!

Testing Our Tricks 🧪

Here's how we test if our tricks worked:

This is like being a teacher grading tests:

• How many times did we successfully trick our AI?

• What percentage of our tricks worked?

• Which types of tricks worked better than others?

Let's Try It Ourselves! 🎮

Want to experiment? Here's a fun challenge:

1. Pick an image that your model correctly classifies

2. Try different epsilon values:

3. See how much you need to change before the AI's decision flips!

The Ethics Corner 🎯

Now here's something really important to think about - why are we doing this? It's not just for fun:

• If we can find ways to trick our AI, others can too

• Understanding these weaknesses helps us build stronger models

• We need to know our model's limits to use it responsibly

Think of it like testing the locks on your doors - better we find the weaknesses than someone with bad intentions!

Common Hiccups You Might Hit 🔧

1. "That didn't change anything!"

   • Try increasing epsilon

   • Check if your image is being preprocessed correctly

   • Make sure you're saving the perturbed image properly

2. "My changes are too visible!"

3. Its taking forever

   1. Try processing fewer images

   2. Use smaller image sizes

   3. Check your GPU usage

What's Next? 🚀

Next time, we're going to level up our adversarial testing with even more sophisticated techniques in advance_adversary.py! Think of it as moving from simple magic tricks to complex illusions!

Download Code

Share Your Experiments! 💭

I'd love to hear about:

• What epsilon values worked best for you?

• Did you find any images that were particularly easy/hard to fool?

• What surprised you about the results?

• Any cool modifications you made to the code?

Remember, we're all learning together! Share your successes AND your failures - they're both super valuable!

Drop a comment below with your experiences! 👇

#BuildInPublic #MachineLearning #AITesting #AdversarialAI

P.S. Did you create any particularly interesting adversarial examples? Share them in the comments! Just remember - we're doing this to make AI better and more secure! 🛡️

Hey there, fellow Alchemists ! 👋

Last time we made sure our AI kitchen was squeaky clean and ready to go with our verify.py script. Today? We're going to do something SUPER exciting - we're going to make our AI actually work in the real world!

What Do We Mean by "Inference"? 🤔

Okay, let's chat about this word "inference" for a second. I remember the first time I saw it, I thought, "Wow, that sounds complicated!" But here's the thing - it's actually a pretty simple idea. You know how after you learn something, you use that knowledge to make decisions? Like how after you learn to ride a bike, you can look at any bike and think, "Yeah, I can ride that!"

That's exactly what inference is! Our AI has learned some patterns (that's the training part), and now it's going to use that knowledge to make decisions about new images it sees. Cool, right?

Subscribe now

Opening Up inference.py 📝

Let's look at our script together:

"""
Inference script for hate content detection model
"""

import pandas as pd
from utils import ImageProcessor, SimpleMetalKMeans

See those imports? We're bringing in two really important tools:

• pandas (that's what 'pd' stands for) is like our data organization superhero. Think of it as an super-powered Excel for Python

• Our old friends ImageProcessor and SimpleMetalKMeans from utils.py that we talked about before

The Main Star: run_inference() 🌟

def run_inference(image_folder, model_path="model/kmeans_model.pkl"):
    """
    Run inference on a folder of images using a trained model.
    """
    print("Starting inference process...")

Let's break this down:

• image_folder is where we keep the images we want to analyze

• model_path points to our trained model (that .pkl file - think of it as our AI's learned experience)

• The "pkl" extension? That stands for "pickle" (yes, really! 😄). It's Python's way of saving complex stuff to a file

Three Big Steps (Like a Recipe!) 🥘

Step 1: Getting Our Expert Ready 🧑‍🍳

try:
        # Load the trained model
        print("\nLoading trained model...")
        model = SimpleMetalKMeans.load(model_path)

This is like getting our expert chef into the kitchen. The try: part? That's like having a safety net - if something goes wrong, we'll catch it and handle it gracefully. We'll talk more about that in a minute!

Step 2: Preparing Our Images 🖼️

# Initialize image processor
        processor = ImageProcessor()

        # Load and process images
        print("\nProcessing images...")
        images, image_ids = processor.load_images(image_folder)

        # Preprocess data
        data = images.reshape(images.shape[0], -1) / 255.0

Whoa, what's that reshape and 255.0 business? Let me explain:

• reshape is like reorganizing your photos in an album. Instead of having a complex 3D structure (width, height, colors), we flatten each image into a simple list of numbers

• Dividing by 255.0 is like converting prices from cents to dollars - it scales our pixel values from 0-255 down to 0-1, which our AI prefers

Step 3: Decision Time! 🎯

# Make predictions
print("\nGenerating predictions...")
predictions = model.predict(data)

This is the magical moment where our AI looks at each image and makes a decision. But what exactly is it deciding? Remember, we're using this for content detection, so for each image, it's basically asking "Does this look concerning or normal?"

Making It Bulletproof 🛡️

Here's something I learned the hard way - ALWAYS plan for things to go wrong! Let's add some safety nets:

def run_inference(image_folder, model_path="model/kmeans_model.pkl"):
    try:
        # All our code from before...
        return results_df
    except FileNotFoundError as e:
        print(f"\nOops! Couldn't find a file: {e}")
        print("Check if your model and image folder paths are correct!")
        raise
    except Exception as e:
        print(f"\nSomething unexpected happened: {e}")
        raise

Why all this error handling? Well, let me tell you a story... I once ran an inference job on 10,000 images, and it crashed at image 9,999 with no error handling. 😭 Never again!

Saving Our Results 📊

# Create results DataFrame
        results_df = pd.DataFrame({
            "image_id": image_ids, 
            "prediction": predictions
        })
        
        # Save to CSV
        output_file = "inference_predictions.csv"
        results_df.to_csv(output_file, index=False)

A DataFrame is like a super-powered spreadsheet in Python. Here we're creating one with two columns:

• image_id: to know which image we're talking about

• prediction: what our AI thought about it (0 or 1)

We save it as a CSV file (like an Excel file) so we can easily look at it later.

Download Sample Code

Let's Try It Out! 🎮

Want to get your hands dirty? Here's a fun experiment:

1. Grab some random images (maybe some pet photos?)

2. Put them in a folder called "test_images"

3. Run this code:

if __name__ == "__main__":
    inference_folder = "./test_images"
    results = run_inference(inference_folder)
    print("\nResults Preview:")
    print(results.head())  # Show first few results

The Ethics Corner 🎯

Now here's something really important to think about - our model is making decisions that could affect content moderation. That's a big responsibility! We need to consider:

• False positives: What happens if we flag normal content as concerning?

• False negatives: What if we miss actually concerning content?

• Bias: Are we treating all types of content fairly?

What Could Go Wrong? 🔧

Let's talk about some common hiccups you might hit:

1. "Model not found" Error

   FileNotFoundError: [Errno 2] No such file or directory: 'model/kmeans_model.pkl'

This usually means you're not running the code from the right folder. Try printing your current directory:

import os
print("I'm looking for files in:", os.getcwd())

2. Memory Issues

If you're processing lots of images, you might run out of memory. Try processing in batches:

# Process 100 images at a time
for i in range(0, len(image_files), 100):
    batch_files = image_files[i:i+100]
    # Process batch...

What's Next? 🚀

Next time, we're going to get into something really interesting - adversarial testing! Think of it like trying to fool our AI with optical illusions. We'll see just how robust our model really is!

Share Your Journey! 💭

I'd love to hear about:

• What kind of images did you test with?

• Any surprising predictions?

• Did you modify the code in any interesting ways?

• What questions came up as you were working with it?

Remember, we're all learning together! No question is too basic - if you're wondering about something, others probably are too!

Drop a comment below with your experiences, questions, or just to say hi! 👋

#BuildInPublic #MachineLearning #AIInference #PracticalAI

P.S. Running into issues? Share your error messages in the comments - debugging is always more fun together! 🤝

Follow up

Hey there, Alchemists! 👋

Last time we talked about the importance of AI bias and our exciting journey ahead. Today, we're going to do something super important - we're going to make absolutely sure our AI kitchen is ready for cooking! You know how bakers check their oven temperature before making a soufflé? That's exactly what we're doing with our AI setup!

Why Do We Need to Verify? 🤔

You might be thinking, "Come on, I installed everything, isn't that enough?" Well, let me tell you a funny story. The first time I set up my environment, everything seemed fine until I actually started training. Two hours in, I realized my GPU wasn't even being used! 😅 That's why we're going to check everything first.

Subscribe now

Meet verify.py: Your AI Kitchen Inspector 🔍

Let's look at our verification script:

python

"""
Verify that all required packages are properly installed and configured
"""
import sys
import tensorflow as tf
import cv2
import numpy as np
import pandas as pd
from tqdm import tqdm

def verify_setup():
    """
    Verify that all required packages are properly installed and configured
    """
    print("Python version:", sys.version)
    print("\nTensorFlow version:", tf.__version__)
    print("TensorFlow Metal device:", 
tf.config.list_physical_devices('GPU'))

Think of this like your pre-flight checklist - we're checking every single thing that could affect our AI's performance!

Let's Run Some Tests! ✈️

First up, let's check our versions. It's like making sure all your ingredients aren't expired:

Now for the exciting part - testing our GPU:

print("\nTesting Metal acceleration...")
    with tf.device('/GPU:0'):
        a = tf.random.normal([1000, 1000])
        b = tf.random.normal([1000, 1000])
        c = tf.matmul(a, b)
        print("Matrix multiplication test successful")

This is like turning on all your burners to make sure they're working. If this runs quickly, your GPU is ready to rock! 🎸

Common "Uh-Oh" Moments and How to Fix Them 🔧

1. "GPU not found" Message

print("TensorFlow Metal device:", tf.config.list_physical_devices('GPU'))
# If this shows [], we've got a problem!

1. Quick Fix: Make sure you've installed tensorflow-metal and set DEVICE='metal'

2. Slow Matrix Multiplication If our test takes more than a few seconds, something's not right with our GPU setup. Usually this means Metal isn't properly configured.

3. Import Errors These are like finding out you forgot to buy a key ingredient - annoying but easily fixed!

Let's Check Memory Too! 💾

Here's a neat addition I like to make to verify.py:

print("\nChecking available memory...")
try:
    # Test with a reasonably sized matrix
    test_matrix = tf.random.normal([5000, 5000])
    print("Memory test passed!")
except:
    print("Warning: Memory constraints detected")

Your Turn: The Verification Challenge! 🎯

Try this out:

1. Run verify.py

2. Look at each output carefully

3. Try changing the matrix size in the GPU test

4. Watch your Activity Monitor while running tests

Pro Tip: Keep verify.py handy - I run it before starting any major training session. It's saved me hours of debugging!

Download Verify.py code

Quick Health Check! 🏥

Before you go, make sure:

• Your GPU shows up in device list

• Matrix multiplication is fast

• All imports work

• Memory test passes

The Ethics Corner 🎯

Even in verification,

• Memory test passes

The Ethics Corner 🎯

Even in verification, we think about bias:

• Are we assuming everyone has the same computing power?

• How can we make our code more accessible?

• What about users with different hardware?

What's Next? 🚀

Now that we know our kitchen is ready, next time we'll start cooking - I mean, training our model! Make sure everything passes these checks before then!

Share Your Experience! 💭

Drop a comment with:

• Your verification results

• Any weird errors you hit

• Tips for others setting up

• Questions about anything unclear

Remember, we're all learning together! No such thing as a silly question in this kitchen! 🧑‍🍳

#BuildInPublic #MachineLearning #AppleSilicon #AIverification

P.S. If verify.py shows any errors, don't panic! Drop a comment with the error message, and let's debug together! 🛠️

Hey there, curious minds! 👋

Welcome to the first installment of my "Building in Public" series. While everyone's chatting about ChatGPT writing poems and DALL-E creating art, I'm taking you behind the scenes of building something different - an AI system that helps identify harmful content. I'm documenting every step, every challenge, and every "aha!" moment as part of the Bias Bounty 2 challenge by Humane Intelligence x Revontulet.

Subscribe now

Why Understanding AI Bias Matters for Everyone 🌍

Let's have a real talk about AI bias and explainability. It's not just a technical problem - it's a societal challenge that affects all of us. Think about it:

• Your loan application might be processed by AI

• Job applications often go through AI screening

• Social media content is moderated by AI

• Healthcare diagnoses are increasingly aided by AI

Every single one of these systems can have biases, and these biases affect real people's lives. But here's the thing: you don't need to be a programmer to be part of the solution.

Subscribe now

Finding Your Place in the AI Revolution 🔍

There are many ways to contribute to better, more ethical AI:

1. Business Leaders: Understanding AI bias helps make better decisions about implementing AI systems

2. Product Managers: Knowing about AI explainability helps design better user experiences

3. Content Creators: Understanding how AI moderates content helps create better, inclusive content

4. Policy Makers: Grasping AI bias helps create better regulations

5. Users: Being aware of AI bias helps us better navigate and question the systems we interact with daily.

What Does "Building in Public" Mean? 🤔

Imagine if a chef not only served you the final dish but also:

• Showed you their grocery shopping list

• Let you watch them chop every vegetable

• Explained why they chose certain ingredients

• Shared when they burned the first batch

That's what we're doing here with AI! I'm sharing:

• Every line of code (even the messy first drafts)

• The thinking behind each decision

• The dead ends and failed attempts

• The successful breakthroughs

The Technical Journey Begins 🚀

Now, let's roll up our sleeves and look at how we're actually building this. We're using Python and focusing on Apple Silicon Macs - but don't worry if you have different hardware, I'll cover alternatives in future posts.

Our Toolkit 🛠️

First, let's look at our training script (train.py):

python

"""
Training script for hate content detection model
"""
import tensorflow as tf
from utils import ImageProcessor, SimpleMetalKMeans

def train():
    """Main training function"""
    print("Starting hate content detection model training...")
    
    # Print TensorFlow device info
    physical_devices = tf.config.list_physical_devices('GPU')
    print("Available devices:", physical_devices)

Think of this as setting up our kitchen before cooking. We're:

1. Importing our tools (TensorFlow and custom utilities)

2. Checking if we have our high-speed processor (GPU) available

The Image Processor: Our Digital Photo Assistant 📸

python

# Initialize processor
processor = ImageProcessor()

# Load and process training data
print("\nLoading training data...")
train_folder = './Training_data'
train_images, train_ids = processor.load_images(train_folder)

What's happening here? Imagine you're preparing photos for a digital album:

1. The ImageProcessor is like your helpful assistant

2. It takes each image and:

   • Resizes it to 64x64 pixels (like making sure all your photos are the same size)

   • Converts it to the right color format

   • Uses multiple workers (like having several assistants helping at once)

The Brain of the Operation: K-means Clustering 🧠

# Initialize and train model
print("\nInitializing model...")
model = SimpleMetalKMeans(
    n_clusters=2,  # Binary classification: harmful vs non-harmful
    random_state=42,  # For reproducibility
    batch_size=1024  # Efficient processing
)

This is where bias considerations become crucial. We're making important decisions:

• n_clusters=2: We're creating a binary classification system. Is this oversimplifying complex content?

• random_state=42: Ensures our results are reproducible and auditable

• batch_size=1024: Balances processing speed with memory usage

The Training Process 📚

# Preprocess data
print("\nPreprocessing training data...")
train_data = train_images.reshape(train_images.shape[0], -1) / 255.0
print(f"Training data shape: {train_data.shape}")

# Train the model
print("\nTraining model...")
model.fit(train_data)

# Save the trained model
print("\nSaving model...")
model.save()

Let's break this down:

1. We flatten and normalize our images (making them easier for the AI to process)

2. We let the model learn patterns

3. We save our trained model for future use

Processing Test Data and Saving Results 📊

# Process test data
print("\nProcessing test data...")
test_folder = './Test_data'
test_images, test_ids = processor.load_images(test_folder)
test_data = test_images.reshape(test_images.shape[0], -1) / 255.0

# Make predictions
print("\nMaking predictions...")
predictions = model.predict(test_data)

# Save results
results_df = pd.DataFrame({
    'image_id': test_ids,
    'prediction': predictions
})
results_df.to_csv('predictions.csv', index=False)

This final stage shows us how our model performs. Every prediction it makes could affect content moderation decisions, so transparency is crucial.

The Bigger Picture: Why This Matters for Bias 🎯

Each technical decision we've made has implications for bias and fairness:

1. Image Processing: Are we losing important cultural context by resizing images?

2. Binary Classification: Does categorizing content as simply "harmful" or "not harmful" miss important nuances?

3. Unsupervised Learning: How do we ensure our model isn't learning problematic patterns?

Join the Conversation! 💭

Whether you're a:

• Developer eager to write better AI systems

• Product manager wanting to understand AI capabilities

• Business leader looking to implement AI responsibly

• Concerned citizen interested in AI ethics

Your perspective matters in making AI better for everyone.

Share your thoughts:

• How does AI bias affect your field?

• What aspects of AI explainability interest you most?

• What questions do you have about building ethical AI systems?

Download the code for this article

Download README.md

P.S. If you found this helpful, share it with someone who might be interested in understanding AI's impact on society! 🙌

Read the rest of the articles

Hey there, Digital Alchemy Lab community!

I've been a bit quiet lately, and I wanted to let you know why. I've been busy preparing for and taking the AWS Beta AI Practitioner exam. It's been quite an experience, and I'm excited to share it with you.

Why I Decided to Take This Exam

As someone working in Cloud Security and AI Security, I saw this exam as a chance to:

1. Deepen my understanding of AI fundamentals and how AI systems work. After all, to secure something effectively, you need to understand it inside and out.

2. Get a broader perspective on AI applications in AWS, which could help me anticipate and address future security challenges.

I've been working with AWS since 2016 and have taken beta exams before. I thought I knew what I was getting into. Turns out, I was in for a surprise.

Subscribe now

What I Expected vs. What I Got

AWS describes this exam as suitable for people who are "familiar with, but don't necessarily build, solutions using AI/ML technologies on AWS." They're targeting business analysts, IT support, marketing professionals, and managers.

Based on that, I expected questions testing a broad but not too deep understanding of AI/ML concepts and their AWS applications. What I got was quite different.

The Reality of the Exam

The exam had 85 questions and covered a lot more ground than I anticipated:

1. Unexpected Technical Depth: Questions dug into areas like model performance, AWS service configurations, and optimization techniques. Even with my tech background, I found it challenging.

2. Broad AWS Service Knowledge: You needed to know a wide range of AWS services to answer the scenario-based questions.

3. Advanced ML Concepts: The exam covered topics like precision, recall, F1 score, BERTScore, and ROUGE in detail.

4. Heavy Focus on LLMs: A big part of the exam was about Large Language Models - when to use them, their limitations, and how to choose the right approach.

5. Retrieval Augmented Generation (RAG): There were questions on understanding RAG, comparing it with fine-tuning, and knowing when to use it.

6. Prompt Engineering: This was a major focus, with questions on techniques and best practices.

7. Some Security Questions: While not a major part, there were questions on prompt injection, adversarial attacks in LLMs, and AI security best practices.

Who Should Really Take This Exam?

After going through this exam, I have a different view on who it's suitable for:

• Junior ML Practitioners: This group is probably best positioned to tackle this exam. They have the technical background to understand the concepts but might find the AWS-specific implementations challenging.

• Experienced ML Practitioners: While they might find some parts easy, the focus on AWS-specific implementations could present some challenges.

• Cloud Architects or DevOps Engineers with ML knowledge: Their AWS expertise would be valuable, but they'd need solid understanding of ML concepts too.

• Business and Non-Technical Roles: Despite AWS targeting this group, I found the exam far too technical for those without a strong ML background. It would be extremely challenging for most people in these roles.

What to Focus On If You're Taking This Exam

If you're thinking about taking this exam, here's what I'd suggest focusing on:

1. AWS AI/ML Services: Really understand services like Rekognition, Comprehend, Transcribe, Amazon Bedrock, and SageMaker.

2. LLMs and RAG: Know the principles and when to use them.

3. Prompt Engineering: Be ready to discuss different techniques and their uses.

4. Model Evaluation Metrics: Get familiar with BERTScore, ROUGE, precision, recall, and F1 score.

5. Machine Learning Basics: Understand how to apply supervised and unsupervised models in real situations.

6. Bias and Ethics: Know how to spot and manage bias, and understand the ethical side of AI tools.

7. Performance Improvement: Be prepared to talk about RAG, Prompt Engineering, and Fine-Tuning.

8. AI/ML Security: While it's not a huge part, know about prompt injection and adversarial attacks.

My Thoughts

This was one of the toughest AWS exams I've taken, but not for the reasons you might expect. The mix of broad AWS knowledge and technical depth wasn't what I expected, especially for an exam that's supposed to be for non-technical roles.

I passed but it made me wonder if the exam is really aligned with what AWS says it's for. It's somewhere between a high-level business exam and a technical specialist certification, leaning more towards the technical side.

If you're thinking about taking this exam, be ready for a more technical challenge than you might expect. Even if you know AWS well, you'll run into ML-specific concepts that might push you.

Have any of you had similar surprises with tech certifications? I'd love to hear about your experiences in the comments.

Remember, in our fast-changing tech world, staying curious and always learning is key. Keep experimenting, keep learning, and keep turning your knowledge into practical skills.

Until next time, happy learning!

How can we transform our AI-generated questions into an interactive, adaptive study experience?

In our last post, we unlocked the power of AI to create custom quiz questions. Now, it's time to take those questions and turn them into a dynamic, personalized study session.

Imagine having a tireless tutor, available 24/7, that adapts to your learning style and helps you conquer any subject.

That's exactly what we're about to create!

The Moment of Inspiration Strikes Again

Remember how we used AI to generate those tailored quiz questions? Well, as I was reviewing them, another lightbulb moment hit me.

I thought, "What if we could make these questions come alive? What if we could create an AI study buddy that doesn't just ask questions, but guides us through the learning process?"

And just like that, the idea for an AI-powered Study Assistant was born.

Subscribe now

Bringing the Study Assistant to Life

So, how does this magical Study Assistant work? Let me break it down for you:

1. Setup: You start by feeding your study material into the AI. This could be the questions we generated in Part 1, or any other content you want to learn.

2. Choose Your Mode: The assistant offers two modes:

   • Study Mode: Perfect for when you're first learning a topic or need a thorough review.

   • Quiz Mode: Great for testing your knowledge and identifying weak spots.

3. Interactive Learning: In both modes, the AI doesn't just ask questions - it explains concepts, provides feedback, and even asks you to summarize key points in your own words.

4. Question Flagging: In Quiz Mode, you rate your confidence after each question. The AI flags questions you struggle with for later review.

5. Customization: You can add your own questions on the fly, making the study session truly your own.

Subscribe now

Let's See It in Action

Imagine you're studying for that AWS AI Practitioner Beta exam we mentioned in Part 1. Here's how a session with your AI Study Assistant might go:

1. You start in Study Mode:

   • AI: "What are the key components of Amazon SageMaker?"

   • You: (think about it) "Show Answer"

   • AI: (provides answer and detailed explanation)

   • AI: "Would you like to move to the next question, review this one, or add your own?"

2. You switch to Quiz Mode:

   • AI: "Explain the difference between Amazon Rekognition and Amazon Textract."

   • You: (provide your answer)

   • AI: (evaluates your answer, provides feedback)

   • AI: "Now, summarize this concept in your own words."

   • AI: "Rate your confidence in this answer from 1-5."

3. The AI flags this question for later review, as you rated your confidence low.

4. You remember a tricky concept from your readings and decide to add it:

   • You: "Add Question"

   • AI: "Great! What's your question?"

   • You: (input your question, answer, and explanation)

   • AI: "Thanks! I've added this to our study session."

Why This Works

This approach isn't just about memorizing facts - it's about truly understanding and retaining information. Here's why it's so effective:

1. Active Recall: By quizzing you and asking for explanations, the AI forces you to actively recall information, which strengthens memory.

2. Self-Assessment: Rating your confidence helps you become more aware of your own understanding, a key skill for effective learning.

3. Metacognition: Rating your confidence helps you become more aware of your own understanding, a key skill for effective learning.

4. Personalization: By adapting to your performance and allowing you to add questions, the study session becomes tailored to your needs.

5. Immediate Feedback: You get instant explanations and corrections, preventing the reinforcement of incorrect information.

Ready to Try It Yourself?

Excited to create your own AI Study Assistant? Here's the prompt that brings it to life. Just input this into your preferred AI model, and you're ready to start your personalized study journey:

Quiz/Flash Creator Prompt

Read more

Day 8 of the 15-Day Writing Challenge

🏆 “I had to make my own living and my own opportunity! But I made it! Don’t sit down and wait for the opportunities to come. Get up and make them.” — Madam C.J. Walker

How do you turn the chaos of exam preparation into a structured path to success?

It started out just like any other study session—just me, my coffee and the task of preparing for the AWS AI Practitioner Beta exam. (If you're not in the tech world, AWS is Amazon Web Services - but I bet most of you already knew that!)

If you've ever taken one of these exams, you know what it's like. Tons of reading, trying to make sure what you're studying matches up with the exam guide, and wondering if you're actually remembering the right stuff.

Subscribe now

Then the practice exams, these are indispensable as it give you some idea of what you know but one of the things I have seen throughout the years, practices exams normally help you remember someone elses brain dump but not your own brain dump.

As the studying day went on, I decided to take a break …

This is where things got interesting.

Subscribe now

The Moment Inspiration Struck

During my study break, I thought, "Why not write some prompts and play with ChatGPT for a bit?" Let's be honest—it was just another creative way to procrastinate on my studies.

And then it hit me—why not let AI do some of the heavy lifting? I mean, I'm studying AI, so why not use it to help me study for AI?

I thought, "What if I could use AI to create flashcards or quiz questions from all the notes I've taken and other related content?"

Just like that, the idea took off. I had the tools right there, so why not put them to work?

Now all I needed was the prompt—something flexible enough to work not just for this study session, but for anything I might study, no matter the subject.

And that's how this nifty little prompt came to life.

This powerhouse could generate study questions for any subject. It didn't matter what you were learning—as long as you had the content, this prompt could handle it all.

Now, you're probably wondering how this works, right? Don't worry, I've got you covered with a step-by-step guide.

Step-by-Step: How to Create Your Own Study Questions with AI

Here’s a simple guide on how you can use AI to create custom quiz or flashcard questions, whether you’re studying for an exam, learning a new topic, or just trying to retain information better. Let’s break it down.

Step 1: Start with Your Study Content
First, gather the material you want to study. This could be:

• Text from a textbook

• Lecture notes

• An article or even content from a Wikipedia page

Having solid, well-organized content is key because the AI will use this to generate meaningful questions.

Step 2: Use the AI Prompt
Now that you’ve got your material, you’ll need to input it into the AI along with the provided prompt. The AI will guide you through the rest by asking for two things:

1. How many questions you want (for example, 10 or 20).

2. The content you want to use (paste the text or upload the material).

If you forget to give either piece of information, the AI will prompt you for it.

Step 3: Choose Your Question Type
Once you’ve provided the material and the number of questions, the AI will ask you to select the type of questions you want:

• Multiple Choice: Great for straightforward recall and concept testing.

• Case Study/Use Case: Ideal for problem-solving and scenario-based questions.

You can decide what works best for your needs.

Step 4: Review the Generated Questions
After you’ve made your choices, the AI will generate the questions. Here’s how it works:

For multiple-choice questions:

• The AI will give you several answer options, including distractors (wrong answers).

• It will clearly indicate the correct answer(s) and provide explanations for both correct and incorrect answers.

• If there are multiple correct answers, the AI will tell you exactly how many you need to pick (e.g., "Select 3 correct answers").

For case study or use case questions:

• Multi-Step Approach: The AI will break down a scenario into several decision points. After each decision, it will explain why certain choices were correct and why others were not.

• Single-Step Approach: For simpler cases, the AI will present the problem in one step. You choose the best solution, and the AI will explain the reasoning behind your choice.

Step 5: Study and Learn
Now that you have your questions, it’s time to study! Each question comes with detailed answers and explanations, so you’re not just memorizing—you’re actively learning.

This is just the beginning, though. In the next part, I'll introduce another prompt that takes this concept even further. It works with the questions we've generated to create an interactive study guide or exam prep tool. Think AI-powered flashcards and quizzes based on these questions.

Keep an eye out for part two – it's where we'll really see what this AI study method can do.

Pro tip: Try this method with different AI models to see which gives you the most helpful results for your specific needs.

Refer a friend

Quiz Question Generator Prompt

**Instructions:** Please generate quiz/flashcard questions based on the provided content. Always ask the user to specify the number of questions and provide the content to create questions from before generating any quiz content. If the user does not provide the number of questions or the content, prompt them for this information before proceeding.

Once the necessary information is provided:

- Ask the user for the desired question types (e.g., multiple response, use case-based, multiple choice).
- For multiple-choice questions, present several answer options, including distractors, and indicate the correct answer(s). Ensure that each question includes the correct answer(s), the possible answer choices, and a detailed explanation. Also, provide clear and direct instructions on how to answer the questions and what is expected from the user.

For **case study** or **use case** questions, determine whether the scenario should be broken down into multiple steps or handled as a single decision point.

- **Multi-Step Approach:** Break down the scenario into multiple decision points. For each step, present several options. After the user has made their selections, provide a detailed explanation that justifies the correct choices and explains why the other options are less suitable.

- **Single-Step Approach:** Present the entire problem as a single decision point with multiple possible approaches. After the user selects an option, provide a deep explanation, highlighting why the chosen option is correct and why the others are incorrect.

For **multiple-choice questions:** If there are multiple correct answers, clearly state the exact number of correct answers expected (e.g., "Select 3 correct answers"). Do not use phrases like "Select all that apply." Ensure that the explanation for each question includes:

- A clear breakdown of why each correct answer is correct.
- A detailed explanation of why each incorrect answer is wrong.

**Important:** The entire output must be formatted in markdown and enclosed within a code block.

**Content to Create Questions From:**
[User-provided content here]

**Expected Output:**
```markdown
1. **Question:** [Your generated question here]
   - **Instructions:** [Clear and direct instructions on how to answer the question, including the exact number of correct choices]
   - **Choices:**
     a. Option 1
     b. Option 2
     c. Option 3
   - **Answer:** [Correct answer(s)]
   - **Explanation:**
     - **Option a:** [Explanation of why this option is correct or incorrect.]
     - **Option b:** [Explanation of why this option is correct or incorrect.]
     - **Option c:** [Explanation of why this option is correct or incorrect.]

[Continue in the same format for the desired number of questions]

Today, I'm sharing something special with all of you—a glimpse into our exclusive mini-prompt course typically reserved for our paid subscribers.

Enjoy this complimentary lesson, and if you love it, consider joining our community for more !

Read more

Hey, Digital Alchemists ! 🌟

Last time, we dove into the fundamentals of Prompting and how it can help you transform your interactions with tools like ChatGPT, Claude, and other LLMs. We explored what prompts are, learned key terms like prompt engineering and templates, and took a very short journey through the history of prompts.

One thing I want to emphasize is that it doesn't matter whether you're a business owner, job seeker, or entrepreneur – learning AI and mastering the art of crafting effective prompts is essential for getting the best responses from tools like ChatGPT.

Now, let's get started with your lesson. One more thing: I hope you're doing your homework... lol!

Lesson 4: Mastering In-Context Learning - How Examples Unlock AI's Potential

In this lesson, let's talk about how to train AI to understand context better. We do this by providing it with "exemplars" which is just another name for examples, as we explained in our last post.

Think about it like this: if you owned a luxury detailing business, you wouldn't just allow anyone to start working on your clients' vehicles without some step-by-step instruction.

You might give them instructions like this:

1. Assess the vehicle: Carefully inspect the boat or car to identify areas that need special attention, such as stains, scratches, or oxidation.

2. Start with a thorough wash: Use high-quality, pH-neutral soap and soft microfiber mitts to clean the exterior, working from top to bottom.

3. Dry properly: Use plush microfiber towels or an air blower to dry the surface completely, preventing water spots.

Now, here's where it gets interesting.

Instead, you give it examples, lots of them. By providing these examples, you're essentially saying, "Hey ChatGPT, this is how I want it done." This is known as in-context or "few-shot" prompting.

Here's a practical example for a local business owner like a car detailer. Imagine you want to create a detailed service description for your website. You could use few-shot learning (or in-context learning) to help the AI understand the format and style you want.

Here's a prompt you might use:

Read more

The Gemini 1.5 Model: A Game-Changer or a Risky Bet?

Google's latest revelation, the Gemini 1.5 model, is a game-changer with its ability to handle a whopping one million tokens. Imagine a machine that can digest the equivalent of entire libraries in a single bound.

It's groundbreaking but here's the catch—can we trust it?

The Hallucination Conundrum

See, despite their brilliance, large language models (LLMs) like Gemini 1.5 have a knack for making stuff up. Yes, they hallucinate.

You give them a document to analyze and sometimes what you get back is a bit...off. This isn't just a minor hiccup when we're dealing with areas where precision is crucial—think legal documents, medical records, you name it.

A Million Tokens, a Million Questions

Now, with a million tokens at their disposal, these models can theoretically manage more complex and extended dialogues or documents. That's great for depth and continuity in conversations, kind of like having a super-smart colleague who never forgets a detail.

But here's my dilemma: how do we verify the accuracy of something that vast, It's like double-checking a whole encyclopedia.

Consistency: The Achilles' Heel of LLMs

And let's talk about consistency. Traditional software is predictable. You click, and it responds the same way every time. But LLMs?

They're a different beast. Ask the same question twice, and you might get two different answers. It's because these models are designed to think and react more like humans—which is cool but also a bit unsettling if you're looking for dependable outcomes.

The Verdict: Not Quite Ready for Prime Time

So, where does this leave us?

In conversations, having a model remember everything you've said is like having the best kind of chat buddy. But for real-world applications where mistakes can have serious consequences? We might need to pump the brakes. These technologies are promising, no doubt. However, they need to be foolproof before we can rely on them in high-stakes environments.

Conclusion: Brilliant but Unpredictable

In conclusion, while the tech is impressive, it's not quite ready for the big league—prime time, if you will. We need these tools to be both brilliant and utterly reliable. Until then, it feels a bit like rolling the dice on a tech that's still learning the ropes.

So, what's your take? Are we ready to embrace these advanced but somewhat unpredictable AI minds, or should we wait until the tech matures a bit more?

Hey Friends and Digital Alchemists!

Read more

Hey there, future AI wizard! ‍

I know you are probably ready to take your first steps into Prompt Engineering . So this post is just the full outline of what’s to come and how the lessons will be delivered to you.

Over the next 20 lessons, we'll demystify the art of communicating with AI and show you how to create magic with your words so get ready to level up your AI game.

Remember this is an interactive course so please feel free to comment on the course contents and let me know if there is anything else I can add to make your learning experience even more awesome! ✨

In addition to commenting on the course content, feel free to experiment with the prompts you're given! This is where the real magic happens. The more you play around and test out different phrasing, the better you'll understand how to craft the perfect prompt to get the results you desire from AI models.

Get Ready to Level Up Fast! Here's how you'll receive your Prompt Engineering lessons: expect fresh content delivered 2-3 times a week, usually on Mondays, Wednesdays, and Fridays. This consistent flow of information will keep you engaged and on track for mastering prompt engineering.

Remember, this is a separate mini-course! You'll still continue to receive my regular weekly newsletter posts, delivered to your inbox. Make sure you're subscribed on the Substack website or app to stay updated on all my content.

Lesson 1-3: Understanding the Basics

• What the heck is a prompt, anyway? 🤔

• The secret language of prompts: Key terms you need to know 🗝️

• A brief history of prompts: From ancient times to AI domination 📜

Lesson 4-11: Mastering Text-Based Prompting Techniques

• The power of In-Context Learning: Teaching AI with examples 🎓

• Thought Generation: Making AI think step-by-step 🧠

• Decomposition: Breaking down complex tasks into bite-sized pieces 🧩

• Ensembling: Creating an AI super-team to tackle any challenge 💪

• Self-Criticism: Teaching AI to critique its own work 🎨

• Prompt Engineering: Crafting the perfect prompt, every time ✍️

• Answer Engineering: Shaping AI responses to fit your needs 🎯

Read more