Most People Use AI Like They’re Playing a Slot Machine

Most people use AI like they’re playing a slot machine: enter prompt, pull lever, cross fingers. Sometimes it pays out. Most of the time? Meh—and they blame the AI, not their approach.

Those who know me know I love digging beneath the surface, getting to the FIRST PRINCIPLES of how something actually works.

Over the last few weeks, I’ve gone deep into how large language models are trained and why they behave the way they do. Not out of academic curiosity, but because I wanted to figure out how to use them better.

You don’t need to be an AI engineer. You just need to know enough to stop working against the system, and start working with it.

That’s why when we help businesses bring AI into their workflows, we don’t stop at surface-level “AI training.” We start with AI fluency, understanding the mechanics, the mindset, and the application. Because in a world where everyone has the same tools, the differentiator isn’t access. It’s mastery.

Here are four core mechanics of AI, why they matter, and how understanding the first principles make you better “at AI”.

1) Embeddings: Context Is Everything

When I say the word bank, what do you think of?

Most of you are business people probably think abput financial services. The fishermen, are picturing a riverbank. And if you’re Afrikaans, you might be thinking ‘Netflix en chill op die bank” .

Context is everything.

AI works the same way. It doesn’t “know” words the way we do—it maps them into mathematical clusters called embeddings. Words like “king,” “monarch,” and “queen” land in the same neighbourhood.

That’s the real breakthrough: AI doesn’t rely on literal keywords, it infers meaning from surrounding context.

These embeddings (mathematical representation of meaning), is why AI can understand nuance even when you don't use exact keywords.

Why this matters:
If you don’t provide enough context, AI may wander into the wrong semantic neighbourhood. Give it the right signals, and it snaps to the meaning you want.

What to do differently:

• Use related terms (“revenue,” “earnings,” “ROI” alongside “profit”)

• Define your role (“analyse like a management consultant”)

• Stick to consistent terminology to keep AI anchored (Context Anchoring)

• Being clear and giving context and a pattern to follow means less wasted time and typing to and thro because AI didn’t understand what you meant.

Specificity beats volume every time: Give AI a GPS to find the right conceptual neighbourhood.

Example:

AI doesn’t read between the lines—it reads around them, mathematically. You’re not just writing prompts. You’re guiding meaning through space.

When I learned this, it reframed how I prompt: it’s not about finding the magic keyword, it’s about guiding the AI to the right neighbourhood.

This insight about embeddings has become central to how we train professionals. When participants understand that AI maps language into mathematical space where related concepts cluster together, everything clicks.

That's the power of conceptual understanding over memorising techniques. Once professionals adopt this mindset, they're not just not just writing prompts. You’re guiding meaning through space.

2) Pattern Recognition: How AI Predicts, Not Understands

What AI is doing:

AI doesn’t truly “understand” your request—it predicts what should come next based on patterns it has seen in billions of text examples. That’s its entire design.

AI doesn't actually understand your request.

When you type “Write a professional email about…” it’s not pondering etiquette. It’s recalling patterns of professional emails it has been trained on predicting what usually follows statistically.

I stopped expecting “understanding” and started thinking “pattern-matching,” and my frustration dropped and my output was spot on.

Why this matters to you:

If you don’t give it a pattern, it defaults to something generic (based on what it has seen on the internet). But give it structure—a familiar format—and it locks in instantly.

What you do differently:

• Use well-known formats (e.g. “write an executive summary using bullet points”).

• Break down complex tasks into smaller, recognisable components.

• Include examples or templates when you want a specific structure.

Example techniques:

You're not limiting the AI's creativity—you're giving it a proven template to work from.

This is where understanding frameworks and prompting structures pays off. It’s not just “how to ask” AI, it’s how to frame work so the system can reliably amplify it.

3) Hallucinations: The Art of Being Confidently Wrong

What AI is doing:
AI hallucinations aren't glitches—they're outcomes of how language models work. Because AI is predicting what sounds right (the most statistically correct pattern)—not what is right, it sometimes invents.

And since models are rarely trained to say “I don’t know,” they keep talking confidently, even if they have no clue what you are asking about.

Why this matters to you:
You can’t trust the tone. A perfectly written response might still be dead wrong. And unless you build verification into your workflow, you’re flying blind.

I now see hallucinations as the cosy of creativity and efficiency and just creating amazing things in a tenth of the time.

What you do differently:

• Categorise AI tasks by risk (writing help = low risk; financial advice = high risk).

• Always verify high-stakes output: numbers, dates, quotes, legal/medical information.

• Ask AI to flag uncertainty (Yes, you van ask Ai to do that)… and then check it.

Treat AI as a sophisticated first-draft generator, not a fact-checking service. Use it to get 80% of the way to a polished output, then apply human judgment for accuracy, nuance, and final quality control.

Once you grasp why hallucinations happen, you stop over-trusting outputs and start designing workflows that build in verification automatically.

The goal isn’t to eliminate hallucinations, it’s to harness creativity while protecting yourself from confident nonsense.

4) Bias: AI Isn't Woke or Evil—It's Just Repeating Us

What AI is doing:
AI reflects the patterns of its training data—the good, the bad, and the ugly. That means stereotypes, cultural blind spots, and historical inequities slip in. It’s not intentional. It’s statistical.

Or as I like to put it: AI isn’t toxic—it’s just been trained on the internet, hanging out online too long.

Why this matters to you:
Bias shows up subtly—assuming “CEO” is male, defaulting to Western thinking, or ignoring underrepresented groups. In professional contexts, that’s more than awkward: it’s risky.

What you do differently:

• Ask for diverse perspectives

• Challenge assumptions: “What bias might be present here?”

• Test outputs for edge cases

Bias mitigation isn’t virtue signalling, it’s professional risk management.

The goal isn't perfect objectivity, that's impossible for humans or AI. Ai Fluency means seeing blind spots before they damage credibility. It’s not about avoiding controversy, it’s about building more professional, inclusive, and accurate outputs.

The Real Shift: From Prompter to Collaborator

When you understand tokens, embeddings, patterns, hallucinations, and bias, you stop treating AI like a genie and start treating it like a system. Predictable. Structured. Limited, but powerful when directed.

That small investment in understanding compounds. Prompts get sharper. Outputs improve. Reviews are faster. Decisions are smarter.

You’re no longer pulling a slot machine lever—you’re collaborating.

I don’t “ask” AI anymore. I collaborate with it. I shape its strengths, I guardrail its weaknesses, and I get far better results than I did when I treated it like a magic box.

Being AI Fluent is a shift from hope-based prompting to intentional collaboration. Not just tools, but transformation in you people think and work.

It’s Actually Not About AI. It’s About How You Think.

AI doesn’t make you smarter. It amplifies how you already think. If you’re vague, it guesses. If you ramble, it rambles. If you’re sharp and structured, it delivers.

The best AI users aren’t the most technical—they’re the most intentional. They know the difference between impressive technology and effective application.

Digging into the first principles of AI hasn’t just made me better at prompting, it’s made me a sharper thinker. Understanding why the system behaves the way it does forces me to be more intentional in how I communicate, frame problems, and design solutions.

You don’t need to be an AI expert. You need to understand just enough of the mechanics to guide it well.

“Because in a world where everyone has access to the same AI, the real differentiator isn’t the tool, it’s the user.”

Ready to build real AI Capability in your Organisation?

If your organisation is ready to move past surface-level AI training and build true fluency (skill, knowledge, mindset, and real application), this is where we can help.

We use our Crawl,Walk,Run,Fly Framework because mastery happens in layers, not leaps:

• Crawl: Get comfortable with AI. Simple tasks, low stakes, building confidence.

• Walk: Solve one real business problem. Move from experimenting to measurable productivity.

• Run: Scale and systemise. Embed AI into workflows and share success across teams.

• Fly: Advanced capabilities—agents, automation, and organisation-wide leverage. AI becomes part of your business DNA.

Where we develop the Four Dimension of AI Fluency :

• Skills: The “how-to” of AI tools, prompts, and outputs.

• Knowledge:  Understanding AI’s role, limits, and potential.

• Mindset:  Staying curious, experimental, and responsible.

• Application:  Embedding AI in small, valuable ways in your daily work.

AI capability isn’t built in a day. It’s built in layers. And that’s the path we help organisations take.

👉 Schedule a free intro call and let’s get started.

If you forget everything else, remember this…