You’re Not Behind (Yet): How to Learn AI in 17 Minutes

Most people using AI are doing it wrong,

which is why it's surprisingly easy to

get ahead of 99% of them. I have spent

over 20 years in tech and AI as a CEO,

board member, investor, building

billiondoll companies. And here's what

I'm seeing. The gap between people who

understand AI and those who don't is

getting wider faster. In this video,

I'll give you a clear sevenstep road map

to master AI like the top 1%. And the

best part is you can actually do it in

just 30 days, even if you're a total

beginner. Let's dive in. Week one starts

with learning what I call machine

English. Most people talk to AI like

it's a person. And that's a huge

mistake. Why? Because the generative AI

systems like Chad GPT don't actually

understand our language. They predict

it. And that's where most people get

stuck. If I said Humpty Dumpty sat on a

Your brain's going to fire wall, you

knew what was coming. Your brain

predicted it. You could have said Humpty

Dumpty sat on a roof. Now it's accurate,

but you knew wall was more likely based

on what you've seen before. Think about

Google search. It does autocomplete the

same way. Why? Because it has seen so

many search queries before. It has

learned from it and now is giving you

the most likely option. AI models like

Chat GPT or Gemini work in a similar

fashion, but they're different than

search engines because they don't store

any [music] pre-baked answers. They

generate the answer on the fly. How do

they generate it? Like at a very high

level, AI breaks your text into smaller

parts called tokens. Each token is a

word or sometimes a part of a word.

Humpty is probably one token. Dumpty

could be another token. Sat another

token. Wall another token. [music]

Then AI converts each token into a list

of numbers, also known as

multi-dimensional vectors. Those numbers

are placed inside a massive mathematical

space called an embedding space. And in

that massive space, similar ideas tend

to live closer together. The system has

learned from previous experiences. So,

it knows that the word Humpty, egg,

wall, and fall will be closer, [music]

but they're going to be far from words

like motorcycle or chocolate. Now, when

it's time to generate the answer, AI

looks at the context and predicts the

most likely next token. So, when it sees

Humpty Dumpty had a great, it weighs all

the options. Humpty Dumpty had a great

party. Humpty Dumpty had a great day.

Humpty Dumpty had a great chocolate. and

it sees that the word fall is the most

likely outcome. So the line is generated

and finished not from memory, not from

stored facts, but from probability and

proximity. That's why AI can feel so

smart, but also so alien. Now,

[clears throat] I'm skipping a lot of

details here, but the important takeaway

here is that when your prompt is vague,

[music] this guessing machine called

Chat GPT or Gemini will produce guesses

that are also vague. And if your prompt

is sharp and targeted, AI will come back

to you with sharp and targeted guesses.

That's what I call machine English. It

helps AI to compute your intent, not

just try to comprehend it. So, what does

a sharper prompt look like? I call it

aim. A for actor. Tell the model who

it's acting as. I is for input. Give it

the context and data it needs. And M for

mission. What do you want it to do?

Instead of typing, let's say, fix my

resume, try typing, [music]

hey, at GPT, you are the world's most

sought after ré editor and business

writer. You've reviewed thousands of

résumés that led to interviews at top

tech companies. You've told the AI what

its persona is, [music] what it's acting

as. A second line, I'm attaching my

resume and the job description for a

senior product manager role at a fintech

company. That's your input. Third,

mission. Review it and give me a bullet

list of 10 specific ideas [music] on how

to improve clarity, measurable impact,

align with the role. Your mission is to

help me build the best resume that gets

me hired. That's how you take aim. It

turns a prompt into a structure. The

model can understand, compute, and

reason with. You can use this three-part

structure in almost all prompts. And

from now on, you will start seeing the

results to be at least five or 10 times

better than before. Only when you learn

its language does AI finally start

working for you. Now that you understand

how to speak to AI, we're going to pick

your instrument. Here's the thing. Most

people start their AI journey the wrong

way. They Google top 50 AI tools. They

pick 10 and they jump from one to the

other. They skim through all of them.

That's a recipe for failure because

there's so much out there. My

recommendation, pick one, go deep. Think

of learning AI the same way you would

learn an instrument. You know, there is

a study in Frontier Psychology that

found that drummers pick up guitar

faster than complete beginners. [music]

Drumming is not even about melody and it

requires very different physical skills.

[music]

But I personally had the same

experience. I spent tens of thousands of

hours as a drummer. [music] And when I

picked up guitar, it wasn't easy, but it

wasn't uncomfortable because I already

knew [music] how to practice and my

brain was trained to see structures and

patterns. [music] The deeper you dig

into one foundational model, the faster

you will find the rhythm of all the

others. So, which one do you pick? If

you want the most mature one, pick Chat

GPT. If you're deep into Google stack

and Google's ecosystem, try Gemini. If

you want more business and projectbased

AI, go with Claude. But really, it

doesn't matter what you pick. In the

first week, spend time with one of them

[music] and learn its personality, its

cadence, its limits, its strengths. The

goal is to [music] start feeling the

rhythm. Once you get comfortable, try

using the aim [music] framework that we

talked about. By the end of week one,

you should be able to write a structured

prompt without thinking. All right, so

we've started using AI. Now, let's talk

about what actually makes your outputs

smart, and that's context. [music] The

world's smartest AI will sound clueless

unless you feed it context. Every answer

AI gives depends on how it understands

the question. If you don't give it

context, it has no grounding. Remember

[music]

that inside these AI models, there is

nothing but a crazy mathematical space

filled with billions of numbers. [music]

Context is the map that helps you

navigate that space to tell AI where to

look and what matters. And the best way

to build that map is with an acronym I

call [music] map. M is for memory. the

conversation history or the notes that

carry over from previous chat sessions

that you've had with the AI. Now, you

can repaste the thread or ask the model

to summarize before starting again.

That's how you'll start building

continuity [music] in your

conversations. A is for assets. The

files, data, the resources [music] that

you attach or copy paste in your prompt.

These assets help you ground the model

in reality. Second A is for actions. Now

these are the tools that the model can

call to do work. The action could be

search the web or scan your drive or

write this code or create a notion doc

and P is the prompt and the prompt is

the instruction itself. So the better

you get with memory assets and external

actions, the better context you'll give

AI in the prompt. And the richer the

context, the better the AI reasoning and

response. Once you start using these

frameworks like AIM and MAP, you have

joined the top 10% of AI users. But if

you want to hit that absolute expert

level, there is one more thing that you

really need. Debug your thinking, which

is step four. When you're not getting

the right answer, the problem is not the

AI, it's your thinking. [music] I

remember the first time I ever prompted

an AI. It was one of those earliest

models from OpenAI and I spent an entire

day trying to make sense of it and by

the end of it I was super frustrated

because it was random. It was

unpredictable. But back then no one

understood. The phrase prompt

engineering hadn't even existed yet

because prompting isn't typing. It's

iterating. When the output is weak, I

assume the fault is mine because it is.

[music]

Did I get it the right persona? Did I

provide the right context? Did I give it

the right goal? And sometimes I even ask

the model itself, what did you do? And

why did you choose that answer? [music]

It will explain its logic. He'll explain

his chain. And that's when the magic

starts. You're not just using AI, you're

learning how it thinks. There are three

cheat codes I use for that. The first is

the chain of thought pattern. When the

answer seems off, I would say think step

by step. Show your reasoning. Then give

me the final concise answer. The second

is the verifier pattern. I would say to

the AI, ask me three questions that

would clarify my intent to you. Ask them

[music] one at a time and then combine

what you've learned and try again. And

the third is the refinement pattern

where you're refining your input itself.

Before answering, propose two sharper

versions of my question. Ask which one I

prefer. So AI will tell me how to ask

the right [music] way. And then we

continue. And you have to keep iterating

with these patterns because these loops

can teach the model how to understand

you [music] and teach you how to

understand the model. test, tweak, tune

up, push until you can tell why [music]

something is working and why something

is off. That's when it clicks. You're

not talking at AI anymore. You're having

an ongoing conversation. You and AI are

learning together from each other. But

here's the thing, it's not enough to

just debug your mind. If your post

sounds like every other LinkedIn post I

see that's pasted from [music] chat GPT,

you still have a problem. And that's why

step five is to steer to experts. When

you ask Chat GPT [music] a question,

you're not searching a database of

answers. You're sampling from millions

of probable ideas that AI has learned

over time [music]

and is storing as billions of numbers.

is some are brilliant, some are average,

some are completely made up, [music] and

some are flat out wrong. If you prompt

vaguely, like explain how to make a team

more innovative, the model will give you

a superficial generic blah answer full

of buzzwords. And you'll read it and

think, "Yeah, I already knew that." So,

how do you [music] fix that? You direct

the model away from the middle and

toward the sharper edges of its brain.

[music] So instead of that vague prompt,

you can say this. Explain how to make a

team more innovative using ideas from

Pixar's brain trust, Satya dea strategy,

[music] and Harvard's research. Now you

pull the model from mediocrity into

mastery by navigating it toward experts,

[music]

frameworks, depth. What if you want to

learn about black holes and you don't

know who the experts [music] are? No

problem. Ask AI first. List the top

experts, researchers, and [music]

research papers and current thinking on

black holes. Then feed the same thing

back to [music] the model and prompt

using these experts and sources

synthesize the original framework that

fills the current gap on the science of

black holes or whatever it is that

you're after. That's [music] the way you

make sure AI is not an echo chamber

anymore. But remember, you're going to

need to verify what you get. That's our

step six. Sometimes AI will tell you

things like 68% of Americans are getting

divorced. I mean, you know, it's not

true. But the scary part is AI will

sound just as confident when it's wrong

as when it's right. So, you can tell AI

100 times, stop making stuff up. [music]

But all models are essentially

generative by design. [music] Making

things up is why they exist. So, what do

you do about that? You simply verify.

[music]

Don't just consume. Critique. There are

five ways to separate intelligence from

illusion. Assumptions, sources, counter

evidence, auditing, and cross model

verification. Let's take one at a

[music] time. Assumptions, ask. List

every assumption you made and rank them

each by confidence. Second is sources.

[music] Ask. Site two independent

sources for each major claim that you

just made. Include title, [music]

URL, and a oneline quote. Now you can

check it yourself. That's the [music]

scaffolding behind the answer. Counter

evidence. Push it. Find one credible

source [music] that disagrees with your

answer. Explain the dependencies. That's

where real reasoning lives. Auditing is

the fourth one. Ask. [music]

Recomputee every figure. Show your math

or code. You'll be shocked how often the

numbers change once you make it slow

down and [music] start auditing. And

finally, crossmodel verification. This

one's my favorite. I run the same prompt

in ChatgPT and Gemini and Claude.

[music] I take the output from one model

and ask another to critique it. Or

[music] I feed the claims of one model

into the other and say, "Verify this."

That's how you separate [music] noise

from knowledge. By the end of your third

week, you'll start feeling more [music]

in control of your output. But here's

the problem. The best AI output aren't

the ones that sound the most original,

[music] they're the ones that sound like

you. That's why step seven is about

developing tastes. Most people use AI

like a vending machine. They push a

button, grab the same junk food output

everyone else gets, and call it a day.

If you did that, most people will know

you just copy pasted it. But you are

past that now, right? It's your fourth

week. It's time to step into the ring.

Treat AI like your sparring partner.

Argue with it. Push back. Sharpen your

thinking. Sharpen its thinking. That's

where the ocean framework comes in. Is

how you turn generic answers into

tasteful insights. Something that sounds

like you. Oh, original. Look at the

response. Is there a nonobvious idea in

it? If not, push it. [music] Ask, give

me three angles. no one else has thought

about. Label one as risky and recommend

the one that you like the most. C

concrete. Are there names, examples, and

numbers that make sense? If not, ask.

Back every claim with one real example.

E is [music] evident. Is the reasoning

visible? Is there enough evidence? If

not, ask. Show your logic in three

bullets. [music] Provide evidence before

you provide final answer. A assertive.

Does it take a stance? you could agree

or disagree with. If not, push it again.

Don't tell me what I want to hear. Pick

a side. State your thesis, defend

[music] it, and then address the best

counterpoint. Narrative.

What's the story? [music] Does it flow?

Is it tight? Guide it. Write it like a

story. Hook, problem, insight, proof,

actions, whatever you want in that

story. So, that's the ocean framework to

add taste to your output. Now, as you

apply this over 30 days, you will start

noticing something [music] deeper. Every

prompt you write, every revision you

push, every judgment you make, you're

not just [music] training the model, you

are training you. AI is coming whether

we like it or [music] not. To some, it

might be triggering lots of deep fears,

but I remain a perpetual optimist.

[music] I think AI is not here to

replace human work. It's here to restore

human worth. If you like this video,

[music] don't forget to subscribe and

check out my most recent video here.

Thank you and I love