Ultimate AI Masterclass (WARNING): Next 2 years will be HELL | 1% Club Show Ep 66

Category: N/A

Transcript

00:00

If you want humanity to be safe, we should let AI run the world. TCS just made the first bold move and now that the big daddy has made the

00:15

move, I would not be surprised if this would start a domino effect. So what do I need to do to be in that top 20 person who is not out of a job and is continuing to be in the top 1% of India?

I think there's a guy in our team who makes a million dollars a year making

00:31

30-cond video ads for brands all done end to end using AI. For somebody who want to get started in this world of AI, how much money do I need to spend to you can do that for free as well?

How? Install a software called as the country that has the most powerful AI would basically mean has the highest

00:48

touch in arms race. As a result, they will dominate the world.

What is the worst thing that can happen with AI? People will stop thinking and it is scary.

AIS will become so smart where will arrive to a AI which can build itself. And when AI can build itself, what takes a human 6 months of training

01:05

work will happen overnight. [Music] AI could be so powerful that it could come to a point where human extinction could happen by 2027.

How can they kill us? Well, because they are AIS, they can get into the passcodes of Donald Trump's

01:20

new click one button and Man, the things that you talk about is really really bonkers. It It's scaring me as an entrepreneur.

I'm sure it is scaring millions of people in India who is doing a job today. So the first thing

01:38

I want to ask you is what's on everybody's mind right now. TCS, one of the most safest companies of India has laid off 12,000 people.

Now the TCS management is saying, you know, it is just some routine, you know, efficiency improvement, nothing to do with AI.

01:53

What do you think? Are we going to expect more such layoffs happening?

What do you think is going to happen to the job market in India? Okay, before I get to the question, let me address this.

I don't try to scare people. You see, uh what I do is I run a company with 160 170 employees and every day with AI

02:10

coming into the picture, I'm seeing insane efficiency gains. And I have been reading a lot about this.

I've been applying a lot of this. So as a learner and as a founder who's implementing all of these things, I thought it is important for me to come out to talk about it and that is exactly why I came

02:25

back to content and I feel leaders of our country especially and also the world are hiding a lot of really important things from their people because if they say everything out people will get scared they might lose people right now in a short term as well and they're getting a lot of hate if

02:42

they say something that has anything to do with AI and I think that's exactly What is happening with TCS? Okay.

Okay. You see the news broke and everybody fed this was a first post.

Everybody spoke about it being AI and it

02:57

was very clear evident that it was AI. But then the CEO came back and said no you know what it was not AI.

It was skill gap. Now please explain me this.

What the hell is the skill gap that didn't exist for 20 years 30 years and suddenly existed right now? As a result a company

03:13

that is treated as a government company. It is the safest company in the world is what people used to say when I was in college.

Yeah. Let go 12,000 employees.

Boss, the reality is it is the skill gap which is true. But the skill gap that is coming because of AI because what does a

03:28

company like TCS do? They write code.

India's third largest export is what? Code.

We build product. We outsource it.

Yeah. It's the third largest export after jewelry and after pharmaceuticals if I'm not wrong.

Right. Why were companies outsourcing

03:44

stuff to us? Because the cost of a developer in the US to India is 1/5 to one/10enth in a few cases.

Okay. So they could charge a premium.

There was an arbitrage for these companies to charge a premium to a bank of America because the cost of labor was low in

04:00

India and we had a lot of people to do it. That is why all this TCS Viper and all these companies grew.

Now with AI coming into the picture, you're hearing people like Mark Zuckerberg, people Mark Zuckerberg literally saying we don't need middle engineers anymore or someone like Sam Alman saying open AI is 50% of

04:17

code being written by AI. So there are crazy coding agents that are coming out.

There are AI models like Opus 4 or 2.5 by Gemini or the O3 and right now the GPD5 that's going to come up by Chad GPT or OpenAI. It is not out yet.

By the

04:33

time maybe the podcast is out, it will be out. And these are incredibly good with code.

M and when these models are so good with code, you don't need a lot of humans to execute it anymore, right? And there's a lot of dead weight in a lot of these large companies.

People who have 10 years experience, 12 years

04:50

experience, 15 years experience. They're not as fast to adapt to start with.

They have been using old code stacks. For example, you would be using a React node, but they would be using some Pascal, some old Java stack, which are not relevant anymore.

So there is no way

05:06

that these companies can continue to pay salaries of these people. As a result, they are going to let go people.

And TCS just made the first bold move. And I think I'm I'm very certain a lot of companies were waiting for one company to make a

05:21

move. And now that the big daddy has made the move, yeah, I would not be surprised if this would start a domino effect.

The same thing happened in the US. Google was the first company to say, I'm going to let go people and then you saw a domino.

Today, if you walk in the streets of SF,

05:37

you throw a stone, you'll find someone who's unemployed. Okay?

And these are not random people, okay? These are software engineers, these are very high quality people.

And we teach people all over the world, right? I see so many people who are accomplished, who have done so well in the past not having jobs right now and that has started to

05:53

happen I believe in India right now and it's going to be it's just a beginning like some reports say that 70% of these jobs will not exist anymore. Is that number exaggerated or do you think it's somewhere close to that number?

I don't know. Uh but I will tell you

06:09

none of these numbers when I hear about them seems to be uh exaggerated. I'll tell you why.

How are teams working? Let's say in software specifically, right?

You have a leader who is like a CTO or a engineering manager and there are bunch of people who have a lot of tasks assigned on a tool like Jira.

06:26

Yeah, correct. Or a linear or whatever platform that you use.

These are all task management tools where a senior person assigns tasks to junior engineers and they go and execute the task and give the code and sub code merge and eventually it goes to production. Okay.

Now there are tools like Codex by

06:43

OpenAI, Jules by Google or Devon by Cognition which also acquired windsurf very recently right what are these tools see there are two kinds of AI tools that have come up okay just for code if you want to go deep into it one are assistants like cursor windsurf bolt all

07:01

these tools are assistants like you say I want to do this they'll quickly do that for you waiting for your command and then there are coding agents which are codecs jewels These are platforms where you assign a task to AI and the AI will go and do

07:16

that for you. Okay?

Right? It's like literally replacing an engineer to do the task for you and they do async.

So if you have 20 tasks to do on your code, you assign 20 tasks to these coding agents. They will work at 100

07:31

times the speed or 50 times the speed of any human can finish the task and push the code back to your repo. So basically what you're saying is earlier if I am a leader and if I had 10 developers reporting to me and if I have to give the responsibility to 10 different people now I need to have 10 agents probably nine agents and one more human

07:47

to maybe check everything. Got it.

So now I will have nine agents and one human to sort of oversee everything and I've reduced my cost by 90%. And and these people won't complain about working at 2 a.m.

4 a.m. No, they they work 24/7.

They don't care. They're happy to work.

And how how much how much do these cost?

08:04

Because a typical dev in Bangalore will cost me two lakhs a month. I mean one by 100 the cost maybe uh and we are moving towards a world where it'll be free like almost free like it will not matter you.

So you're saying you're saying now pursuing computer science in engineering

08:19

college is a waste of money and time. Very good question.

I don't know. I think there is still uh I was just talking to my team today when someone asked me the same question.

Weberh would you have done computer science or not? because I regretted not doing computer science because my friends when they graduated made 20 lakhs peranom.

I made

08:35

five lakhs peranom and I was like bro I wish I did computer science. Now the rational for learning code is very different.

See I also learned code but I never wrote code. But what it taught me is product sense.

What it taught me is design thinking. What it taught me is problem solving.

What it taught me is

08:51

original thinking. All these are very important traits that you end up learning when you learn how to write code.

What I believe is 2 years, 3 years from now, right? code will be taught at eighth grade or seventh grade at the same level that it was taught in engineering today.

I don't know what

09:06

engineering is going to look like. Quite frankly, it's very hard and I feel really bad for all the kids who are in colleges today because they're studying something so irrelevant.

Like when they come out, they'll be like, "What the hell is this?" They were always studying something very irrelevant, but the world

09:22

was not so dynamic. Now every week it's changing.

Yes. Every week it is changing.

So, it is absolutely madness. Absolute madness, bro.

Uh, I don't know if you've seen this. Sam Alman congratulated a person called a SIHO on his Twitter.

I was like, why is Sam

09:39

Alman and Greg Brockman, everybody congratulating the CIO guy and I went to that rabbit hole and there was this very techy looking geek looking leaderboard where CIO was rank one, rank two was OpenAI. When I geeked into it, I realized it was a Japanese coding contest.

It's one of the most toughest

09:56

coding contests in the world. AI ranked two in that coding contest.

They almost won. By the way, towards the end, this person Siho who was also an ex OpenAI employee beat the AI model to become number one.

And this is a new model. This could be

10:12

GPD5 is what people are saying or whichever that model is that will be out in public in maybe maybe 2 months max. M that would basically mean you, I and everybody else for $20 will have the best world's best second best software

10:27

engineer in the world writing code for us and we can have unlimited of these. So let me ask you this question.

You see of course there are some people who say that you know it's going to be the you know the the the job market is going to get disrupted a lot. But their argument

10:44

to that is if I go back in history, you know, there was the industrial revolution and the internet revolution. Eventually we found our way back and there were more jobs than what was there.

Why is this time different? Why is the AI revolution different from the internet revolution and the industrial revolution that it is predicted that

11:00

it'll lead us into a dystopian world where very few people will have jobs. See, I'll tell you what.

Based on our understanding, based on all the data reports that are coming up, the world unemployment rate could go up to 30 to 50% in the next 2 3 years. Wow.

11:15

Okay. This is tracked unemployment.

For example, India's unemployment right now is 3.5 or 4%. You know that is not true.

Today in reality, it'll be 25 30%. Or whatever that number is, I don't know, but definitely not 3 4%.

Right. Right.

They are saying it's going to 2x to 3x.

11:32

Okay. What is that one job or what is those five jobs or three jobs that are going to get created which will make all these people busy?

Nobody has an answer. I have researched a lot.

Okay. I get these hypothetical answers saying every time a job has

11:50

gone, new job has come up, right? But this time it doesn't look like it.

Unless you want to go become a plumber or a carpenter, which is fair and square, you can do that. But not many people would want to go and become a carpenter.

But for my for that my argument would be uh Tesla robots. I

12:05

recently saw a video that these robots are now doing laundry. They cooking the Tesla diner.

Huh. So why can't plumbing be done by these robots?

Yeah. It's it is a little difficult.

It will take 15 20 years because the reason rational for that is robots are very good when the environment is understood,

12:21

right? Uh when they don't understand the environment, it's very hard for you to predict what a robot will do.

In plumbing, there's a lot of nuance that a robot needs to understand. Does that mean plumbers will never be replaced?

I think that is what the world of utopia is going to look like. But the dystopia

12:37

is going to happen over the course of next 5 10 years of whatever that is. So here's a different viewpoint.

Let us now look at the macroeconomics. If a lot of people lose their job, won't the economy self-destruct?

Because how is the business going to make money if there's not enough people to sell it to?

12:52

Well, if you think about it today, if you look at any business economics, right, except for things that are needed for everyday life, most of the economy runs around the top 5%, top 10% people and that is the consumption economy, right? Majority of the money is

13:08

revolving around those people. But that is in mostly in India, no, but in countries like US that could be 20%, that could be 30%, but there's always a pel.

Correct. Agreed.

right now like I think I saw some data on Swiggy that I think 80% of their revenue comes from the top five% of

13:24

high value customer it's the same for everyone will be the same for your business also will be the same for my business also correct right it's the 5% who pay the majority which covers all who can afford the gourmet restaurants where they charge 1,000 rupees every order high a order right that has always been the case

13:39

right now what will happen is when we say something like this happens right it all depends on how the government will actually react to this. I think on a long-term India will win because we have always figured out a way to make that work.

The only case this time that is different is we have to do

13:55

it fast and India has never been good with doing things fast. I'll give you a data point.

Okay, it is scary when I read it. You know, can you guess uh I'll tell you Elon Musk already has a million GPUs.

You know, Nvidia makes these graphical processing units on which AI is trained, right?

14:10

Elon Musk almost has a million GPUs. Okay.

Is that a large number to a million is a good number because each GPU cost $40,000 to $75,000. Okay.

Right. So the more GPUs I have the far better your power.

The more energy that you have and the better AIs that you can

14:25

run. So GPU is the hardware.

Without hardware you can't do anything about it. So he's bought it from Nvidia.

So yeah he has access to a million GPUs and he says he will go to a 10 million GPUs. Okay.

Okay. How much does an OpenAI or a Google have?

He says about a million GPUs. Open AAI says a million GPU.

uh I think Google

14:42

will Google will have a mix of GPUs and TPUs which is their own hardware that Google is building. Okay.

Now whatever let's say US has all put together 5 million GPUs let's say hypo or 1 million GPUs hypothetically. Okay.

How many GPUs

14:57

do you think India has? 10,000.

15,000. I was close.

Okay. Yeah.

I mean 15,000 whole country versus Elon Musk having more than a million GPUs. You see that difference?

Why don't we have more? We can't we can't buy them.

No, we have Yeah. So, it's funny that you can't buy

15:14

them directly right now because the GPUs are made by Nvidia. Nvidia is a US company and uh our very dear Donald Trump has put us in grade two in countries who can buy GPUs.

But I don't think we are doing enough there or if we are doing enough we we don't know about

15:30

it and I really really hope we do something about AI. So let us assume the probably the optimistic scenario that India is going to be slow and then other countries are going to develop their AI models much faster than us.

Uh the way I like to imagine this is that people are

15:45

not able to understand the difference between what a AI superpower country versus a non-AI country can look like. But the way I like to look at this is how is a civilized city like let's say Bangalore or Mumbai different from an un uninhabited island you know with tribal

16:01

people that I feel like that will be the difference it's going to be the difference between Wakanda and uh uh I think Namibia that's going to be the difference right like I'll tell you the significance of what you just said you said a very powerful line thank you

16:16

companies uh countries which has good AI will compound faster than anybody I'll give you rationalization of that. Okay, you've heard of AI right now which is artificial intelligence.

The next leg of it, it artificial general intelligence AGI. Now what is

16:33

AGI? AI is not as smart as humans in everything as of yet.

It is smarter than humans in a lot of things but it is not good enough than every other human being today combined. And when that happens that would be AGI which is smarter than every single human

16:49

in the world combined. when AGI arrives right the next thing that every company is optimizing for today Meta has been building something called a super intelligent team that is ASI AI will be able to build its own AI so there is

17:06

this paper dude AI 2027 it's by Daniel Koko who is a ex openAI researcher who has become a whistleblower right now he's predicting what will happen in the world for the next two years and his prediction is by 2027 AI will become so smart There's going to be insane amount

17:22

of compute. When there's insane amount of compute, it's like food.

Insane amount of food for these AI. When that happens, AI will become smarter and smarter and smarter over the course of time.

Where will arrive to a AI which can build itself? And when AI can build

17:38

itself, mean build better than itself, better than itself. Basically what the paper says is what takes a human six months of training work will happen overnight because there are half a million AIs that are flipped around and doing AI

17:54

research for you because they are all servers right nothing else. So the paper hypothetically says we will get to a point where AI will be misaligned.

What does misaligned means? Alignment is a very interesting concept in AI.

Okay. Alignment means if AI wants what you

18:12

want then AI is aligned. If AI doesn't listen to what you want and does on its own front then it's a misaligned AI.

All this dystopian movies that you see right it is all because of misaligned AI. It is not listening to you.

It is acting like

18:27

it is listening to you but it's doing whatever it wants at its own end. When that happens when AI makes its own AI, you don't have a control of it because you don't know the language it speaks.

You don't know how it works. You don't know where it is stored.

You know how it you don't know how it is built. So you have zero control on that AI.

As a

18:42

result, that AI could be so powerful that it could come to a point where human extinction could happen if we don't slow down AI by 2027 is what the report says. Human what will happen?

Extinction. How can they kill us?

Uh well because they are AIs, they can

18:57

get into your servers. They can get into the passcodes of Donald Trump's nuke because I'm sure by then he has a neural link on his head and get the passcode and click one button and there'll be nuclear missiles all over the world killing every single human.

So then why are we doing this? Because a bunch of

19:15

people, bunch of leaders realize that they can make outrageous amount of capital and on top of that the amount of power that they can hold if they build the first ASI is unfathomable. So they're taking an order.

Do you think

19:31

people make after after you after people make a billion dollars, do you think they work for money anymore? They work for power and control.

Exactly. Who do you think will have the highest amount of power?

Who is who has the ASI? who has the most super intelligent team.

And that's exactly why Mark Zuckerberg is spending $1.5 billion

19:47

to hire one bloody person, which is 20 times more than Messi, who's the best footballer in the world. But what what is so special about these people that have to pay 1.5 billion to have because they are the guys who are moving the research forward.

One little change in algorithm, one little discovery could propel the AI research 2 years faster, 3

20:05

years faster, 4 years faster. It's a race, bro.

See the thing is whichever company gets to ASI first will go so much faster that people can't keep up because what did I say the moment ASI happens AI will build itself 6 months of work will happen overnight so a company which doesn't have AGI

20:22

can never keep up because they will compound because years will become days will become hours will becomes minutes of research work my god the whole world works on code and that is how systems work as well Okay, now we've come to the conclusion

20:39

that everybody's not going to be saved. Yeah, it's a 20%.

So what do I need to do to be in that top 20 person who is not out of a job and is continuing to be in the top 1% of India? Yeah, I think for that you have to become a problem solver using AI.

You should have the capability to do to to

20:56

do a job of 10 people that you have replaced just now. For that I think you should become something called as an AI generalist.

Right? I think the era of specialist is over.

Specialist jobs are going to be given to AI and AI. Give me examples of specialist jobs.

Anything where you do only one thing repeatedly very very well.

21:14

Mention names. Let's say uh a chartered accountant.

Okay. It's a specialist role, right?

You do something very very specific. Lawyers.

Doctors. Lawyers.

Yeah. I mean doctors I think they'll be fine.

I feel and if you want to go deep into it, I have a hypothesis there. But most of the specialist roles in any company, let's say something,

21:31

let's say you're a software engineer. You just can't be a software engineer anymore.

You can't be a person who's just writing code. I've been saying this for the last two years.

Yesterday I saw a podcast of Vinod Kosa talking to Nicl Kamad where he said specialists are over, generalists are here. And my team was like web hub.

You've been saying for this for 2 years

21:47

now we listen you. I'll tell you how you can become a journalist.

No matter who you are, you could be a software engineer. You could be a marketer.

You could be a finance person. You need to become a problem solver with AI.

For that I split this up into five levels on how you can get there. I'll make it very quick so that it becomes a road map.

And

22:03

in fact I'll give a road map to you. You can link that up in the document as well for people who want to go deeper into this.

Sure. I think level zero is something that we are making all our employees do already today which is list down all the tasks that you are doing on everyday level.

What are the things that you don't enjoy doing? Figure out what are the right AI tools

22:19

or right AI workflows which can start doing that for you. That is level zero.

Bare minimum things that you can do can open up your time and space. Okay, level one is where you figure out your one of the most important skill set for every human being is going to be how I can communicate with AI in the best way

22:35

possible and that is going to happen with prompt engineering and context engineering. No, but here I have a doubt you can also use AI to write prompts.

You can. So then that solves the no but you will only what can AI give you as an answer based on what has been trained.

Your original thinking is what

22:51

will matter in the world that we are living in. And a lot of times most of the answers are not coming from one prompt.

They come from something called as a workflow. You get to AI to think like you.

For example, let's say I want to get AI to write like me. Okay.

Yeah. This is a problem that I've been

23:07

facing. Uh let's say you want to get AI to write like you.

What do most of the people say? Most of the people say, "Hey, my name is Shahan." Right?

Uh I want you to write content like me. This is how I write my content.

Or maybe you'll give a couple of samples. Write like this is what you'll say.

So I tell you what I did, right? So I've

23:23

written almost 500 LinkedIn posts in the last like 3 four years. Correct.

So I took my top performing LinkedIn posts about 100 of them and I put that into a document and I gave that as a training model for my chat GPT writer. Then I told it that you are my LinkedIn writer so and so.

This is the way I

23:39

write. This is the context and this is a structure that you need to follow.

You know high emotional hook body with a lot of information and then a nice CTA. This is how I've trained it.

Now every time I come up with an interesting topic I will do the research for it. Let's say I find an article or whatever I put that

23:55

into it and say that write it in my way of writing. Perfect.

I think it's not a bad workflow at all. But you realize it was not a prompt.

You did a workflow. H you did multiple steps.

This is because this is how I thought in my head. Exactly.

Because this is how I would because this is how I would explain to a employee if

24:11

that person is working for me as a human being. This is how I would explain to them.

go through my recent post, study it and any new research that comes write like this. Perfect.

So that's what I told you. I think it's perfect.

But my process is slightly more refined. I'll give you context.

This is where a generalist comes into the picture. You use the right AM model to write

24:27

solve the right problem here. First thing that I would do for getting AI to do things like me is just like you pulled up 500 of your LinkedIn posts, I would pull up all my top performing LinkedIn posts or Instagram res or whatever that is.

I would make a document to it and I will not attach it with AI. I will not build a rag of it or

24:43

I'll not build a custom GP of it but I'll actually upload that document into a reasoning model like 03 on chat GPT okay or Gemini 2.5 pro which is a reasoning model which can think I'll give a reasoning model which is different to other models because reasoning models can think

24:59

can reason before they give you an answer rather than just spitting out an answer and I'll tell you the technical difference between two also okay whenever there is something for AI to think you give it to a reasoning model okay like 03 03 03 03 or 2.5 or uh claude

25:15

opus 4 or anything like that right you give a model and I write a prompt you are a expert when it comes to finding patterns in content go through my content you have done a lot of uh you know writing for a lot of authors go through

25:32

my writings and tell you tell me how I write give me a extensive content DNA analysis of how I write what are the kind of words that I use what is the what is how do I frame my sentences how do I structure my emojis what is the kind of

25:48

emotion that I invoke in my content because I don't know about it I write the way I write because I just do like that right I you can't explain it so you ask AI to tell you then AI will give you a report saying Sharon these are the findings that I found from your writing when I did an analysis like

26:04

a consultant you know consultant does research work you have done that before at your job you do that consultant work on the content you get like a extensive analysis. Then I take that analysis.

Then I go to a normal 4.5 or a 40 model and I say

26:20

convert this into a reusable prompt so that I can use this prompt every time I give a topic and research. You will spit out a post for me by following everything that I've given on the research report.

So you're making the

26:35

research report as a base, not your content, because this is a much more cleaner form of knowledge for AI. So you're using my earlier context to create a prompt.

Yes. For future AI purposes.

I'm using your context to get a report.

26:51

I'm using that report to build a prompt. I'm using that prompt inside of a GPT or a custom GPT or a Gemini gem or something so that it can mimic your style of writing.

And this works better. Yeah, this is a workflow, right?

You are giving it in a form that AI understands the best

27:08

because think about this, every time AI needs to write an article, it has to go through 100 articles of yours and every time it'll pick a different nuance. But now that you've given a report, you have given a consolidated report of what works.

You're making AI think less. You're cutting the ambiguity of AI.

As a result, you get much more

27:25

better outputs every single time. Yeah, I follow this process and I keep making this better obviously.

Right? last two years, every single piece of content on my LinkedIn has been written by AI.

I can't tell if it is written by me or by AI. Yeah.

Because the kind of words, the sentence formation, everything that it does, it's

27:41

impossible for me if you tell me I'm being frank. I can't tell if it is written by me or AI.

Of course, I add do some touch-up works here and there. But every time do I touch up work.

I go back to that prompt and give it update saying, "Hey, this is how my style has changed. Here's an updated instruction." And it updates the prompt.

And that

27:57

prompt goes into the thing as well. And then you built this as a custom GPT or is it like a single thread that custom GP custom GPT.

So it's repeatable. No.

And a lot of people will realize that custom GP is paid. You need a paid chat GP.

So what is the difference? Like a lot of

28:12

people use the same thread and they keep using that same thread again and again thinking that all the things that I've chatted with this chat GPT in that thread it has context. So what is the need for me to build a separate custom?

I'll tell you why. I'll tell you why.

Because all these AI models have something called as context window. Huh?

But that's expanding. No, that

28:29

no, but that's at 100,000 only. Not today.

And two, you're confusing the AI model. I'll tell you what I mean by that.

Let's say you got AI to write a post for you and you gave it some feedback which is very specific to that post, right? Which is not relevant for everybody.

Everything else it's very specific to this post. But AI understand

28:45

that as an instruction and your 100k context window will fly like this without even you knowing. And you have not got to a point where it's unlimited context window.

And you're giving wrong context. You're making AI remember about random things that you have spoken with it in this which is

29:00

specifically to a different post which is not learnings. So your outputs are not get and I've done this test.

Basically it has a DK effect. The longer the thread becomes the worst the post will becomes.

Interesting. But there is one way to hack this is that every time you have a new

29:17

conversation with it go back to the first conversation you click on edit and drop that research again. So that for AI, every conversation is a brand new conversation and you're not breaking the thread flow and you're not breaking the context.

Level two is where you figured out these workflows, right? Like I just told you, you automate those workflows

29:33

using something called as either you build your own rags. What is rag?

I'll get to that. Or you do something called as MCP.

Rag custom GPD is a rag. Okay.

Custom GP is rag. So level gem is a rag.

Yeah, that is level two. But that is foundations of level two.

So in level two, what you do is you use something called as MCP. Have

29:50

you heard of MC? I've heard of MC Claude launched it first, right?

That's what I've heard. Yeah, Claude only released the protocol called as model context protocol.

It is bloody gamechanging. Okay.

And it's scary AF. I'll tell you what MCB does.

Humans use tools to do their job.

30:06

Correct? If you want jobs to be automated, AI needs to use the same tools that a human is using.

H so MCP replaces you as a human with an AI assistant and it can do all use all the tools that you're using

30:22

today. Okay.

Let's let's talk about let's say you're a marketer. Okay.

Okay. As a marketer, let's say you have given your social media team a job of coming up with 10 great ideas for content.

What do your teams do? They look at all your competitors or who create similar content.

30:37

Go through their content. Look which one has got higher views than on average.

Download those videos. get the transcripts, upload those transcripts into your custom GPT, rewrite the whole script and tell you Sharon, we have found a script for you.

Yeah, it's exactly I know that. I know that we do.

Everybody does the same. There's nothing

30:54

wrong about this, right? We all take inspiration in the world, right?

For this, they must be using a Instagram scraper to scrape the content. If they're smart, they would be doing it automated by downloading all the content, putting it on a Google sheet, doing a data analysis on that, finding the top 100 pieces of content.

Once they

31:10

find the top 100 pieces of content, they'll probably using a different tool to download those videos. Then find a different tool to get the transcript of those videos.

And once they have the transcript, they'll probably dump it inside of your custom GPT or Gemini gem to rewrite the script. This is a manual process.

One question that scraping thing you

31:25

talked about, what is a good scraping tool to extract my competitors? Let's say ampify is a mother of all scraping tools because ampify is a marketplace of scraping tools.

So for everything that you want to scrape they have micro apps inside of that

31:41

right or you can use a tool like phantom buster or you can use a tool like textu or you can use any scrape you go. So you're saying let's say let's say I want to analyze a a content creator like Mr.

Beast and I need to analyze every video that he's posting out.

31:56

So you're saying I can use the scraping tool to download all his videos to download all his videos data of all his videos. data of all his title, likes, comments into an Excel file.

Excel file and then you do basic data analysis. So that can be automated.

Now this is what downloading can be automated. Data analysis is something that you will do once you have the top

32:12

once there is some manual labor still involved because you have to still click on the buttons to download the video then get the transcript using a different tool. Then once you have done all the transcripts then you have to start uploading it each one of them on your custom GPD.

And this is already a 10x process for most of the people.

32:27

Correct? Because they didn't know scraping existed.

Now we are living in a world with MCP where the same tool that I told you which is ampify can be connected to my claude and I can say go through this Instagram profile of Shahan here's the URL

32:44

go through his top 500 videos on Instagram res find me likes views comments and everything else that he has got on this okay and then analyze that data find me the top 50 winning videos and then use the appropriate Ampify MCP tools to download

33:01

those videos, find the transcripts of those, and then use my Claude project to rewrite the whole script and give me 50 brand new scripts. It'll do it in like 30 minutes.

Wow. Something that would take you maybe a couple of weeks.

And and how easy it is to set this up on

33:18

It doesn't need a single line of code. Okay.

Interesting. So that is MCP.

This is like I would say like 2 days worth of job. I know.

I mean I can do this in like 30 minutes. Set this up in 30 minutes.

I'll tell you one more very good example. You might have come across this.

Zerodha

33:35

released something called as Kite MCP. Yes.

Yes, I saw that. Yes.

What is Kite MCP? All of a sudden they have created a mic which is the MCP protocol for your kite account.

So your claude or chat GPT can now connect to your kite account. So every day you open your stock stocks and see yeah what is

33:52

working, what is not working. If something is down, you quickly Google about them.

Should I sell this stock? Is it doing well?

This is what humans do. Correct.

And you are, if you are in the market, if you are in the business of buying and selling stock, you're always on it. But for someone like me, no way.

I mean, I lose all my money

34:08

purely because I'm not on top of the stocks that I've invested in. You will tell on a phone call, bro, this stock is crazy.

Bite off. But you'll not tell me when to sell it or you'll forget it.

Or you'll drop me a message, but I'll forget it and I'll lose money. How do I automate this?

So today, I've connected my Kite account to Claude.

34:24

Every day there's a task that runs in the morning where it will check my portfolio. What all stocks have I bought.

If any of them is on red or has lost money, it will do a research Google research about those stocks from top sources and it will tell me should I hold the stock or sell. And it will

34:41

because it's already reading the news and I can give it sources saying watch all Axhat's videos, watch all Sharon's videos, right? And get all the context if they're saying something new.

use that information also to tell me what stock should I buy. As of today, you cannot buy or sell today.

34:57

Those trades that you can still have to do it yourself. But you all of a sudden have an AI which you can make sure is reading your reading your portfolio every 2 hours and telling you what needs to be done on your WhatsApp for almost for free.

And this is again power of MCP.

35:13

So you built the system again. It's it sounds like a fancy system, bro.

It's a connection and a bloody prompt. So every day on WhatsApp you get these kind of I get on cloud only because I'm very active on claude.

It comes in as a report and I can get it on WhatsApp. It's not rocket science at all.

Level three is where you jump deep into

35:29

images, video, audio because as an AI generalist, you'll be like web of but this guy is a CA. He doesn't need to do anything with images, videos or anything.

He's not a content creator. But you're getting up to something which is called as an AI gendist.

You're a problem solver. And as a problem comes, you cannot say this is not my territory anymore.

Everything is

35:46

your dietary because AI has made it possible for you to do anything anything and everything. So here you go deep into diffusion.

How can you train a AI model to talk like you? How can you train a train a AI model to generate images the way you want?

How can you train a video model like you've seen the clones and all of that

36:02

as well? How can you get AI to create any kind of video, any kind of images, any kind of audio?

You go deep into the world of diffusion. You go beyond of text in level three and whatever is needed you need to have a tool stack in a right way so that every time there's a problem you know how to approach it using the right tool

36:18

and the right model. Now if you go deeper into level three today when we look at level three right we can think of AI tools like you would be like hey Jen you would be like Higsfield you'll be like creai P3 right these are all basic models but that is very very basic that anybody

36:35

can use it's an equivalent of chat GPT when you go deeper into the space right you go into this world of open-source diffusion which is comfy UI where you can build your own models train your own content to make it look like you to make It feel like you imagine Amul's creative

36:53

amount of effort that goes into creating that one art is so high. Yeah.

Imagine I can train a comfy UI model and you don't need to write any right of any single line of code for all of this. Huh?

If I train 50 of AMU images and upload it and I just say today's independence day because it knows what

37:10

needs to generate. It will show me 10 options of Amul branded creators for Amul as a brand.

M and the amount of money in this Sharon is in please be kind. But why why is it insane?

Like for what? Like for the you know I have people who I know who

37:26

have built micro apps. Bro, you know what the app does?

If you upload an image, any image, any AI image, it adds a human-like texture to your skin. That's all the app does.

M the guy makes $30,000 a month just selling that app because lot of

37:42

companies want to solve that specific problem of AI. You have told me AI clones look very unnatural.

Yeah. To make it look natural, you need two more steps which most of people don't know.

And those two steps can be productized using comu and you can start selling

37:58

that as a service to people. Got it?

And there are so many people who are creating I we we have a lot of teachers, right? There's a guy in our team who makes a million dollars a year making 30 secondond video ads for brands all done end to end using AI.

38:16

Okay, workflows. Something that was taking production companies like crows could be done for a lack today.

You saw all our ads also, right? All our ads are AI generated and looks stunning.

Now that you've learned all of this, it's time to systemize things. Right now

38:32

in level four you learn how to build AI agents. Okay.

So anything that you want to do in a repeated format and you want it to work autonomously, you have to build an AI agent for that. For example, a good AI agent could be instead of doing this work every time you click on a button,

38:47

it does the research for you, finds the best topics that all your competitors are creating, finds the best winning one, also looks for data signals from the internet, finds that content, writes the script, generates the audio, generates a video, stitches them

39:03

together, makes a video, and also publishes it on your profile every single day at 7 p.m. It does the task end to end.

That is an AI. I just have one confusion.

How is this different from what you said about MCP? Yeah.

Because MCPs is still an assistant for you. You click on a button, you still

39:20

have to take action on it. When I get stock data, I'm looking at that data and then I'm taking a call.

But there are lot of tasks that you don't want to take a call on. You want AI to take a call on.

That is where agents come into picture because they do task end to end. Automation and agents difference.

39:36

Automation automates you and gives you the output by following a workflow. A agent executes the task end to end.

For example, you had deep research. Before this on chat GP, you were able to search for any topic and you were able to get an output.

Correct?

39:51

Why was deep research called as an agent? Because deep research had the ability to read up hundreds of articles, make sense of it, understand your problem statement, and give you the detailed report or a presentation that you wanted as an output.

But with chat GPT, you were doing all that work to

40:08

eventually build a report yourself. But a deep research is doing that work end to end.

So deep research is an agent. Agent.

Yes. Productized agent where you don't have to build it.

It's already built for you. So going forward these companies will start building more and more agents for every is building right now.

No. Chat GB

40:23

launch chat agent which has access to it is bloody crazy if you want to go into the world of agents right the kind of agents that are already out there is insane. For example, so it's like I don't even need to learn how to build an agent.

These companies are already building the agent. So there are two kinds of agents.

One are generalized agents. These are called

40:39

super agents which can do very very commonly done tasks very well. Like recently charad agent I think I can go and order sneakers online.

I can order some food, book an Uber, all of that. It doesn't can't do any of those.

I buy No, no, it can't do any of those. It

40:55

needs your supervision. Supervision in the sense you will say go find the best sneaker for you.

So it'll go look up sneakers, find the best sneakers, come to the checkout page and wait for you to enter your card details. So it's not end to end task yet as of yet.

Okay. But there are agents which can do end to

41:12

end task as well. For example, the browser that I use today is called as comet.

Uhhuh. Perplexity is perplexity is comet.

Insane like just insane. I'll tell you things that it does for me.

What the way charg works is let's say you you tell chad GP agent saying buy find me the best Nike Air

41:29

Jordan original in Bangalore which could be shipped to me in next 5 days. So it'll go do the research find everything eventually get to a point where it'll be like Sharon please login please enter your details because what it is doing is it is every time you give a task right

41:46

it's opening a computer on the cloud h and it is opening a chrome browser on the cloud to do that task for you so it doesn't have your access to anything but if the same thing was running on your browser your Amazon is already logged in

42:02

your card details are already added the address is already added. Yeah.

So on comet it does the same stuff that a chat GP agent does but on your browser. As a result you already have everything signed in there right.

I can come to comet and say buy me exactly the same

42:19

thing right? Buy me a Logitech MX or buy me an AirPods and ship it to my home address with cash on delivery enabled.

It will do the end to end task and order it for you. It's doing that.

Yeah it can do it for you. Oh my.

If you have a card details if you're in the US your card is already attached. It

42:34

can make the payment also right and it is safe because using your browser and you use your browser saves all the data on the client so it is not shared with servers shared with AI. So that is what comic does which is an agent.

Now bro I run a newsletter you know that

42:50

and we run it on beh I used to bother my team guys saying yeah open rate how much was the open rate of the last newsletter. I will have this urge every single time.

I'm too lazy to log in and do all of that. Now on comment I say what was the open rate of my last newsletter.

It'll open behive. It'll look at the last

43:06

newsletter tell me the open rate and ping me. But how does it have access to your email ID?

No I've logged into behive on my account. If you have logged into any account on your computer it can use it because it's using your Chrome only instead of you using it as a human.

If it already logged in it just needs to open it without writing the password

43:22

because it's already logged. Yes.

Exactly. If it is not logged in it will ask you to login.

For example my LinkedIn messages. I used to get hundreds of messages every day and I always miss out on my LinkedIn messages.

Now everyday morning I run a task automatically. It will look up for all the messages in the last 24 hours.

Read all my 100

43:38

messages. Close out the ones that are not important.

The ones that are important it will say Sharon dropped you a message inviting for a podcast. This this company dropped you a message doing this and all the links to all the messages are ready.

So that I can in the morning I can go out click on four links go through them reply to them and close

43:55

them. Now none of the important messages are missed because it can use my LinkedIn.

My god. So this is this is what comet is basically doing that it's basically doing end to end everything for you anything.

Yes. Let's say you have to uh set up a

44:10

Facebook ad. What can do that?

Yeah. What is there?

It's clicking of 10 different buttons only. No, it can click buttons very well.

No problem. You give it if you say that figure out ad creative also.

will generate ad creative and it'll upload ad creative also by yourself by but if I need to pro I didn't prompt it

44:26

well yeah you say this is my product uh this is my com let's say I want I'm starting a finance workshop Sharon's ads are the best ads that I've seen so go through Sharon's ad library go through the ads that have been running for the last 60 days pick up the top three that you think

44:42

will work well use uh chat GPT to generate similar ad creators with my name on it instead of Shahon and you have access to my ad account run ads for $50 per today starting from tomorrow build a landing page using bolt as well it will go build a landing page using bolt as well and build it it's all

44:57

about prompt so this is what Sam Alman said that we are not very far away from a day where there will be a oneperson unicorn because I can do 100 people's job by just doing all of today it's like 10 people's job maybe but I'm sure we'll get to a point that is why in level four building agents is

45:14

important so here you're doing one-off task by asking comet do this for me do this for me do this for me what if everyday morning when you wake up wait for example I'm just thinking with comet can I sort of um figure out calendar booking with important people

45:30

without me actually you know physically doing it like give me an example like how like say for example I'm thinking from the perspective of a salaried person also today who is looking for a job at a company that they want now a lot of people will just go to a you know normal job posting and wait for some application all that right but

45:47

both of us know that the real way of doing it is through networking on LinkedIn finding the people who you want to work with and tell them how you can help them and then they will figure it out. But very few people have the time and patience to follow up and all that So can I tell um comet that I want to let's say for

46:04

example get into Zomato find out all the product leaders in Zomato also they are one level junior and then I tell them and write them a very customized message of how my skill set is going to help them and set up and if anybody's interested set up a calendar meeting in my Google

46:20

I can take it to a different level also I can say I'm looking for a job in the top consumer tech companies in Bombay look for LinkedIn jobs. Look for a marketing manager role in any consumer tech companies that you find.

Okay? Make a list of them.

Update them on this Google sheet. And once you've updated

46:36

them, find five key decision makers that you think would be the right person for that specific role. Find them on LinkedIn.

Connect with them. Send a connection message with a personalized note by reading their latest LinkedIn post that they have written or the latest LinkedIn like

46:52

that they have done and personalize that message and say that hey, I'm keen for this role. Would I can I get a referral and send them?

So, comet do that can do this. Can do this, bro.

Then what work do you do, bro? Everything is done.

You delegate. That's the whole point, right?

You delegate. No, there's no work left, you know.

You

47:07

just tell comet do this, do that, do that, do that, and then you just chill the whole day. Yeah.

But there are a lot of things it cannot do. The moment it goes beyond your browser, the moment it becomes intelligence coming into the picture, the moment you want multiple tasks here, you have to give it a command every time you want it to do anything.

You don't

47:23

want to give it a command anymore. Let's say every day at 4 p.m.

it should do this for you by finding right jobs and all you to. And if your computer is off, comet will not work.

But if you're able to build your own AI agent, you can do this without writing a single line of code. A specialized AI agent exactly to

47:42

apply for jobs for you every single day. You can do that using a tool like NA10 make or a crew AI where you can build these agents connect all these APIs to it without writing a single line of code again and that will work much better

47:57

than what a comet can do because it can work on cloud it can work 24/7 you don't need to keep your browser open and look at it so there is a massive knowledge gap that exists and that is what is keeping people behind it's not rocket science it is

48:14

something that everybody can learn and the products are built in a way that millions and billions of people can actually use it. Okay.

So now I can't imagine what is level five. Level five is what you actually told right you built an agent here right you're automating stuff it's about

48:29

productizing your workflows. So build products using AI use bold use lovable use cursor use your own business build convert your workflows convert your agents into full-fledged end to-end products that becomes a business then you can use it as a business.

For example, I've built a lot of internal AI

48:45

tools for my company. For example, my graphic designer was leaving.

She decided to leave and my head of content came to me and said, "Bro, we have to hire a content or hire a graphic designer." I was like, "But you can do uh designs with chat GPT. I've taught you already you can do prompts." He's like, "Those technical prompts are very

49:01

hard. So, not everybody can do it." And then I built a tool using bolt where you tell the name of the product details.

It will automatically generate five different prompts which you can copy paste inside of chat GPT and it will generate a ad creative similar to ours.

49:16

So you built a tool to help your team generate prompts because they don't know how to write good prompts. Yes.

Now my team was even lazy. They were like it's too much of work for me to copy and paste it.

So I built that tool in a way they they say 10 I want 10 ad creatives. They write what the creative is about.

They click on one

49:31

button. It will generate 10 prompts for them at once.

Then if they're happy with the prompts they click on one button. Then it'll open 10 10 different chat GPT windows with all the prompts injected inside of them.

They can just go have coffee, come back and all the creatives are ready. Now did I build anything

49:46

crazy here? No, I built a tool on top of chat GPT using a tool where I don't know how to write code using a tool called as bolt which is similar to lovable where I just told AI this is what I want and it figured out a way and built a solution

50:03

for me which basically we call AI create strategist. So wait what in bolt is a website.

It's a product creator. Yeah.

It's like glovable where you can text prompt it for it to write code. It will write a code and build a product for you.

So what did you tell exactly? Initially I told I want a prompt

50:19

generator. Right.

So here but there are already prompt generators available. But I created my own prompt generator because it is supposed to generate prompts.

I love that question by the way. Nobody asked me that.

Uh for you to generate great ad creatives. I wanted winning ad creatives to come up.

50:34

Okay. And I already had winning creatives in the past.

So what I did is I went through 03. I uploaded 10 ad creatives of mine because 03 has something called as OCR.

It can read the image. It can understand what is on the image.

So I said here are the 10 images of ads

50:50

that has won for me like static static images. Take those images and understand what you have understood about the image.

It gave me a big list of things. Now I said if you have to generate an image like this every time I tell you a name of a product what would

51:06

be the prompt that I need to use. So it in itself wrote a technical prompt for me so that it can replicate an image similar to this.

Interesting. I used that prompt inside of the tool that I built inside of Bolt and I said every time my team gives you

51:22

an instruction saying personal finance masterclass by web of Sicinti right this is the copy on the ad use this prompt use the chat GPT's open AAI 03 model it will generate that number of prompts

51:38

to create an ad creative inside of chat GPT okay it's a prompt that is created to generate different prompts. It's a father of prompts.

Then my team said, "We have we want we don't want to copy paste." Then I built a tool which takes

51:55

the prompt right on bolt. Every time my team gives a copy, it uses this prompt and uses the API of 03.

I'm not using chat GPD anymore. I'm using the API of 03 to generate 10 different prompts in a nice little UI.

52:11

The UI was built by the tool called as Bolt. Now if they click on one button I have to open that up open 10 different windows for 10 different creatives.

How do I solve that? I I went to perplexity and I said when I enter a prompt in chatgptt and I click on enter I see a

52:28

new URL gets formed. Is it possible that I can send the prompt in the URL that is is it possible I can type chatgptt.com/ prompt is equal to and I enter that prompt there.

Can chatgy take that? Is there a URL structure for that?

52:45

Is there? There was.

Okay. I copied that information and I gave it to Bold and I said this is the URL structure.

So every time you generate a prompt, append the prompt that you generated next to chatgptt.com/ whatever that was. I don't remember anymore.

Prompt is equal to colon model

53:01

picker is equal to colon whatever that was there. I didn't understand it.

So I copied and pasted it and I said every time you generate a prompt, put it inside this URL and keep the URL ready. And then I asked it I want to open 10 tabs at once.

Is that possible with Chrome? It said yes it is

53:17

possible. And again I copied that instruction on how to do it.

I asked it I want to implement this product on Plexity. I on Plexity I went and I said I want to open 10 different tabs at once.

M how do I write the code? How do I approach this problem?

I have to tell my engineer you tell me. So it told me

53:33

these are all the steps you have to take. I copied it and I gave it to my engineer which is a bolt which is a tool and I said this is what you have to implement so that on one click it can open 10 tabs and you know what it did I think this is level five for a reason man this this is this is a little too

53:50

complicated trust me it is no it is not bro you have to play with it is not it's difficult to wrap my head around what's happening so because I jumped to a product that I build which is being used by my team you feel like that initially I started with content generator which is very simple I got an

54:05

aha. I said, "Okay, this is possible.

I want to do more." Then I built a graphic designer said, "Oh, you know what? I can give a prompt.

I can imagine what I want. It can generate 10 creatives instead of one because Chad GD generates one." So, is this the future of work that people operating at level five solving problems in the company, building agents

54:23

and productizing future of work, every level is future of work. The reason why I put them in levels is because you progress.

You can't do level five without level one. Mhm.

But at your work, it doesn't mean that if you're not in level five, you can't do your work. If you're very good with level one, you can solve a lot of

54:39

problems. If you're very good with level two, you can solve even more problems.

And every bloody level is monetizable. Let's say you have learned how to write prompts great way in a great way.

You can you can be a prompt engineer or a freelance prompt engineer just like you hire a consultant to do XY Z. You'll be

54:56

like, "This is the product that I want to make. I don't have the time to write the best prompt.

Here's $1,000. Build a prompt for me.

You know what? I spent over $10,000 with freelance prompt writers already.

They're good. Very good.

They work for they write meta prompts three, five, 10 page long. I

55:11

don't have the I know how to write it. But what are you trading?

Why do you pay money for? Not because you can't do it.

Because you don't have the time and you're not an expert at it. So prompting can be outsourced, right?

And there are a lot of people making ton of money. Where where do you hire this freelance

55:26

prompter? I hired from my LinkedIn, right?

I actually the way I did it is I went to someone who was working at Microsoft copilot team. Okay.

He was the guy who was writing prompts from Microsoft copilot. Huh.

That is why it was so expensive. So how much does he charge?

He charges $1,000 per prompt.

55:43

Per prompt. Yeah, bro.

That's $10,000 for 10 prompts. But with those 10 prompts, I can build a $10 million business.

And this is the heart and soul of it. M and there are a lot of people who are willing to pay a lot of money for this because one prompt, one good data analysis prompt

56:00

can help your team to do attribution of your data which used to take them 10 hours in a minute and you don't need 10 people. You need one person to run that prompt.

Isn't that worth $10,000 for you? Yeah, that is why people can pay a premium for that.

So tell me this. I've understood about all these levels and it is

56:16

understandable that in the current world that we are living in if I'm a level four or even a level five there is no way I'm losing my job right I'm an asset to any company the CEO uh I think if you are very good with this forget about losing your job you will be sitting in a position where

56:33

you'd be working for four different jobs at once because the kind of you think that is the future that people will be having four different jobs because they can do so it's very much possible it's very much possible and this is not me saying Again this is very big uh people who who have been very good with predictions of the future talking about it where you know

56:50

Nal Nal Ravikant yeah of course basically says this that in the world of in the world of AI that we're going to live in people who are real knowledge workers who can work with AI the right way will mostly be gig workers. The reason why they'll be gig workers is

57:06

company can't afford to pay them because they're that expensive and they're very good with what they do. As a result, everybody, most of the people will wake up in the morning with a ping on their phone with five tasks to be done.

They pick what they want. They bid for it.

Wow.

57:21

They do that task and chill at Tahiti for the next two weeks and come back and pick a task and do it again. The real independent, the real, you know, fire is going to happen for a lot of people who actually embrace AI the fastest.

Interesting. This is Nal Raviant talking about it in

57:37

a pod. So let's talk about a day in your life, right?

Like what what kind of setup do you have on your computer? Can you tell me like in the morning when you wake up, how does your day go?

Because all the things that we've been doing and talking about maybe 3 years back can now

57:53

probably be done in half an hour. So I'm just curious to know what do you do in a day?

So my tool stack is very simple right now. Uh I have something called as Jerry which is an AI executive assistant of mine which resides on my Slack which has access to my emails which has access to my calendar which has access to my

58:08

meetings. Right?

So every day morning I wake up I say give me a rundown of all the emails that I've received in the last uh last 24 hours that have not opened and tell me if there's anything important to attend. Next is if I have calendar let's say I have to block a calendar with you.

I just say block a calendar with Sharon.

58:25

block my calendar with Sharon at 7:00 p.m. today for a podcast.

It does all of that. The other thing that it does which I use it quite a lot is I don't attend meetings anymore.

My AI agent attends most of my meeting takes notes. I ask Jerry which is Jerry.

Jerry attends all

58:41

the meetings on my behalf and I ask what happened in the last engineering standup. I had asked them about this update.

Did do we have an update on this product on this product spec? it can go through the meetings, tell me if that was discussed or not so that I can have that conversation with my team.

And then

58:56

I have uh also a Limitless which is on my in my bag right now which is a physical device that I wear which records everything that I'm talking about. It's called Limitless.

It's a physical device. I'm not wearing it right now because podcast.

I only use it in the office. I don't use it outside.

59:11

Every conversation that someone in my team is having is being fed to AI. So there is nothing that you can tell me and I'll forget about if you tell Weber I'm like hey Shahan like uh when are we doing the collab and you say bro day after tomorrow

59:28

I'm not forgetting that because my AI is not forgetting that it's my second brain right Jerry has access to it and what about your tool stack you told me about Jerry you told me about that physical device which reads everything what about actual execution because yes you have meetings but you also want to tinker around and play and come up with

59:43

yeah so my go-to content creation tool today is Claude my goto research tool or brainstorming tool is O3 Pro. O3 Pro.

O3 Pro which is by CH in Chad GPT open AIS model. This is good for what you said reasoning.

It's the best reasoning model. Reasoning.

Give me an example. What is

59:59

reasoning? So bro, so what these things do is they can do end to end thinking.

So what a reasoning model does is let's say I want to redo a landing page. I'm giving a very basic task which everybody will understand.

There's a landing page. I want to redo the copy of the landing page.

I can take a good

00:15

screenshot of the landing page and I can upload it on 03 Pro and I can say act like a expert conversion rate optimizer go through the content go through the structure of the landing page and recommend me the landing page which will double my conversions. H so it's going to analyze it has media it has OCR so it can read

00:33

images. I've uploaded a screenshot, not a link.

Understand understand the structure. Go and do research of what are the top traits of a good uh CRO right now or what are the new things that have been working really well with new experiments.

Pull that information,

00:48

implement that information. If I upload a sheet of data with it saying this is how my landing page is performing and can do data analysis of that to see what are the metrics that it should optimize for and finally can give me a copy new copy that can improve my conversion rate two times.

So a reasoning model can

01:04

think before it answers. So when you go to a 40 model which charged is open AAI 40 model which is a free model right it is a prediction model.

What it does is it predicts the next word one at a time. It's just that it predicts the

01:20

next word really really well. As a result the outputs are great.

So for text generation it is very very powerful but it is not thinking. It is just blurting out words based on what it thinks is right.

So basically 03 is for for me to outsource my thinking. Thinking thinking model

01:36

because I want to think faster. Yes.

So I'm outsourcing thinking and I'm seeing what it giving me and I want to think about it more and tell you think. Yes.

Come back you. So I'll tell you one good example.

Right. Let's say I have a podcast today with you.

I literally did this by the way. Okay.

I was a little concerned that you'll ask

01:52

me tough finance use cases which I don't understand like you know and I was like okay. I told uh 03 that hey I'm doing a podcast with Shahan.

Uh this is his profile. He's going to be hosting me for a podcast.

I've asked him to keep it on AI but I'm sure he'll jump into finance.

02:08

So ask me questions as if you're Sharon and do the podcast with me and I'll give you the answers and also ask very fundamental AI questions so that I don't want to look like an idiot because a lot of times foundations screw up because you think you know but you can't define it well. You know

02:23

what I mean? So, let's do a mock podcast.

And I'm on a voice mode. So, Jerry will act like you.

Sorry, I mean Chad GPD acts like you. Uh, and it started to ask me questions.

So, it asked me questions on hey, how do you do what is that back trading or whatever

02:39

that is? Back tested back testing or whatever.

I said if Shahon asked me this question, I said don't ask me this question because I don't know about it. Right?

So, it asked me a lot of questions like, hey, how if I have a portfolio, how do I manage? Have you built a personal assistant for yourself to manage your personal finances?

How do

02:55

you track your expenses? Is there anything like, "Oh, these are all very interesting." How how do you track your expenses?

I have an app built on Bolt for that. Okay.

It has access to your financial transaction, bank statement. No, I just upload the PDFs of my bank

03:10

statements. Okay.

because uh HDFC and all don't give you uh there are already apps today which sort of it's fun no for me to do things and the advantage of you being able to build a product is that I can connect see my world I want to revolve around Jerry I

03:26

want Jerry to know everything about me so I want to build everything that can connect to my Jerry you get my point because my world has to revolve around asking a question and getting an answer so this is what Mark Zuckerberg says build your own personal super intelligence bro personal software personal software.

03:42

There's a massive rise of personal software that's going to happen which is an insane bloody opportunity. So recently I read I I read this story on Reddit where somebody has used some AI agent to u set up dates for him on

03:58

Tinder every week set up 10 dates by talking and swiping to girls and he's figured it out and it's working for him. Yeah.

I mean, I can't imagine. I mean, I'm sure uh uh a AI can talk much better than an average human.

No, I don't know about your game, but my game is terrible

04:13

with that. So, I'm sure an AI can do a better job than that.

So, it's like um you know, it's as much as your imagination, right? Your imagination is the extent to which what you can do using these AI agents.

Correct. So, now let's talk about monetization when it comes to AI, right?

Because we've talked about all the different

04:30

crazy things that AI can do in terms of making me more productive. But I think every single person who is watching this podcast has the aspiration of becoming financially free very quickly, right?

They want to sort of have multiple sources of income. And the current world is mostly designed in a way that you are

04:46

only dependent on a salary, right? Because 8 to 9 hours of your time is going over there.

And what we are learning with AI right now is that because of these you know efficiencies in productivity, I can do a lot more with the 8 to 9 hours a day that I have. Maybe I can do two to three jobs at the same time.

Maybe I can start my own

05:01

business. I can be a freelancer.

and so on and so forth. So can you give us a road path you know like a pathway for someone to have multiple sources of income because of the power of AI in the 8 or 9 hours that I have in my Monday to Friday.

Yeah, I think uh for this to happen it

05:17

depends on who you are and uh what you do right now. Let's say you are with someone with some years of experience right you work in let's say healthcare or finance industry hypothetically right?

Yeah. So if you have worked in that industry for a couple of years, you know what are the top problems that this industry faces today

05:33

as a problem, right? And if you are an AI generalist, you know how you can solve those problems using AI as well.

So when you have very good understanding of some understanding of business and very good understanding of AI, there are very few people who fall in this cross-section

05:48

and that unlocks a massive arbitrage of an opportunity of becoming a AI consultant. Okay.

And now you have context of the industry. You have worked there for a couple of years.

You know the top problems that they face. If you are in a fintech business, there are hundreds of

06:05

companies which fall into the fintech business. You become a consultant.

You understand the top problems. Use the right AI agents or uh building the right solutions or right prompting whatever it takes to solve that problem.

You take that problem, you take that solution and sell that to multiple companies as a

06:21

consultant. Can you give an example?

uh for example, let's say creative because again I'm going back to marketing which is simple and everybody would relate back to it. A lot of companies spend a lot of time doing ad creatives.

Yeah. Let's say I'm here.

I figured out a way

06:38

to automate my graphic design by building this tool. Yeah.

Right. It is personalized to my company because I told you that it generates creatives on my brand.

Yeah. It is built on top of my learning.

Now that I've built a base solution, I can go to 100 different companies and say,

06:54

I'm going to build a solution for you which is going to generate thousands of ad creatives for a fixed cost for you. You pay me $1,000 a month.

Are you doing this? No, I'm not doing it.

Okay. There's an opportunity, of course.

Right. I'll pay you.

You have to pay me $1,000 a month, which is like cost of one graphic designer or even even less.

07:10

This can generate unlimited ad creatives using your own chat GPT account. I will come and do it as a service.

Initially, you'll come write your own prompt, optimize it. So it'll work for you.

Also train your team for that. I'll charge $5,000 one time.

After that, I'll charge you $1,000 every single month to use my

07:26

product and and this is personalized to you. I I don't think anybody's doing this in India right now.

There are so many solutions like this. Is this just one solution that I've given you?

Give me another example. Let's say voice agents, you know, today you have a sales team.

Yes. How big is it?

30 people. How many calls can they take per per

07:42

day? I think 10.

I'm not sure. Yeah.

Connected calls could be like 20. Okay.

Right. They might be making 50 calls but connected calls is 20.

Okay. 20 into 3 is 600 calls per day by a 30 member team into 25 working days.

600 calls happening per month.

07:58

600 people are per day. 600 calls are happening per day.

Able to reach out with 30 people. Yes.

Only 600 people. Yeah.

Right. I what if I say I can deploy a voice agent Sharon and all the 30 people will not convert at the same levels.

One will be a high performer, one will be a low

08:14

performer. Yeah.

I will tell you that you know what if your human does 10 lakhs of revenue per person per month I will build a voice agent which can make unlimited calls can work 24/7 and can operate at 80% efficiency of

08:31

your best salesperson. It can talk like your best salesperson.

It is more knowledgeable than most of your people and can convert sales for you. As a result, you just need three people in the bottom to just collect the money.

But the selling will be done by an AI. That's a voice agent.

08:47

Today a voice agent can do 100 calls in a minute. So it can do 100 simultaneous calls in a single minute.

Wow. And you know what?

They don't take a fixed salary. They charge you based on number of minutes they spoke.

H. So your fixed cost is down.

You can you

09:03

can all of a sudden have the ammunition to make unlimited calls. Yeah.

Because one minute may if you can make 100 calls that means almost unlimited. It's like having an army of 10,000 people.

Correct. Correct.

They can work 24/7 and your lead to call time, which is a very important metric. If your lead came now, the sooner you

09:19

call, the higher is the conversion rate. Correct?

Humans have break times and everything. AIS don't.

So, if you get a lead at 2 a.m., the call will happen at 2:5. And it's going to be an AI which will talk like a human.

It's the most knowledgeable. It's been trained to talk

09:34

like the best salesperson. Has ability to do objection handling by like your best salesperson.

How big is your sales team? Uh at peak it was 100 plus today it's 25.

Wow. And then it's all and we make 10 times more calls than we ever did before.

09:50

Wow. So that's like AI people AI agents calling AI voice agents calling and then after that it comes to a human people to close needed.

So so I'll tell you how we have set up this process. I'm not relying on AI to make a transaction yet.

Right? I'm relying on AI to qualify the person to get to the point where this

10:07

person says yes I want it. Once the yes I want it comes in then a human picks it up.

So instead of having 150 or right now we would need 250 people I have 30 people. My average connection rate connection rate if I have 100 leads how many people am I able to connect with?

10:23

Before this used to be at 40. Right now it has gone to 67 to 70.

You know why? A human will call once twice and give up.

A AI can call two weeks later also four weeks later also. can say first call and because you we are able

10:38

to call right away already the connection rate is much higher because what happens to a lead bro you find something very nice you give a lead saying I want a call back and your life happens 4 hours later if someone calls you back you don't want to talk to that person because you've moved on correct but if the person gets if the person if

10:55

someone gives you a call in 2 minutes won't you talk because right now is when you saw it and you want it and that's why you gave a lead it takes your connection rate H so you can call them immediately because in 2 minutes in 3 minutes in fact we're trying to build a workflow where you

11:11

submit the lead and it will show up a 90cond timer to you. We are giving you a call in 90 seconds so that the user is hooked.

He's expecting a call to come. Keep your phone handy.

We are calling you from this number and the call will come from this number. Oh, so keep it handy because a lot of

11:27

times people get it blocked by true caller and all correct. So you want them to pick it.

As a result the connection rate has skyrocketed. So for somebody who is uh you know want to get started in this world of AI, how much money do I need to spend per month to get started?

Zero. No, but to at least get the good

11:42

subscriptions of these AI models, I have to spend two cable. You can do that for free as well.

How GPD plus is 2,000 a month? No.

Install a software called as Olama. Okay.

Once you install a Olama, if you have a basic computer which has basic GPU computer, laptop, if you have 50,000 laptop or Windows or Mac, doesn't

11:59

matter. Install a software called as Olama.

M once you install the software called asama you can almost run any decent open-source AI model directly on your computer for free without internet access unlimited you can run a local model on your computer to do most of the

12:15

tasks if you need 03 in a few cases well you know what reasoning models Gemini models are free you can use that if you have if you're a student or if you're not a student also right you can generate an ID card using AI and

12:30

upload that and Gemini is free for you for one year. If you are a subscriber, yeah, perplexity is free for you.

And a lot of people don't understand the power of perplexity. You know what?

Perplexity is not just for research. It of course can do all the research work, but perplexity also has the ability to switch models.

12:46

So you can do everything that you're doing with chat GPT with Perplexity by going to settings and playing with all the bloody models that are out there. Perplexity has something called as perplexity labs which is again for paid users only.

But if you have a account, you can get Perplexity for free. If you don't have an account, get one.

It will

13:02

cost you 100 rupees. Right.

Interesting. Interesting.

Now, now let's talk about the risks with AI, right? What is the worst thing that can happen according to your head?

Like why do they say the dystopian world is just going to finish everything? What what do you think can go wrong?

13:18

Well, there are two things, man. Like I think one that I think will definitely go wrong and one which is very scary as well, which I'll talk about.

The thing that will definitely go wrong is people will stop thinking and it is scary. Huh?

I saw some recent MIT report. Yeah.

The MIT report says people are

13:35

getting brain rotten or brain dead as they used. They show the different brain scans because you're outsourcing your ability to think to an AI.

So you stop thinking your cognitive ability is going to donkeys. Yeah.

Right. Forget about us.

What about the kids who are 6y old, 5y old, 7y old who

13:52

are already using chad gutty for everything they're going and asking chip and parents are very proud. Oh, look at my kid.

He's using Chad Gupetic. The kids are outsourcing thinking.

How will their cognitive ability build if they're not thinking? What will happen to their you understand

14:08

right? If you stop that is the only thing that is left to us original thinking because thinking makes you smarter and intelligent.

You have to push yourself to think more. But if there's AI, I will not keep thinking.

I'll just ask AI to think and it is so smart because it has so much context. Today meta knows everything about us.

It knows more about

14:25

us than we know about us. Correct?

But they are only using that for recommendation. If AI knows more about them more about us than we know about ourselves, you'll never go to a friend, bro.

Who why do you go to a friend for recommendations and such? Because you feel good about the answers that they

14:41

give cuz they personalize it for us. They personalize it for you.

But AI will personalize to a level that a friend can't. If a friend is in bad mood, he will but AI will be the nicest person.

Oh,

14:56

Sharon, I feel so bad for you. You know what?

You've been working very hard. First of all, you should take a break.

You get my point, right? It'll always be nice.

It'll understand your mood as well. I mean, imagine this dystopian world, right?

You have a neural link in your head. You're wearing a rayband meta

15:11

glasses where everything can be seen. By the way, what is what Neuralink for those of you who don't understand?

What is it? Is it a way for AI to directly go into my brain?

I mean, no. Neuralink is a device which detects your you know the way your brain works is all using electricals electric signals right neural link reads those electric signals

15:27

so that it knows what you're thinking it translates your thinking into words okay by translating those electric signals that is what neural link is doing okay that basically means if your neural link which will be connected to everything in

15:42

the world and your AI is going to be a single signin today as you click with Gmail email right and you sign into everything tomorrow there'll be an AI sign in like this and the moment you sign in using that AI platform or whatever that is and that is what chat GP is trying to become

15:58

by the way they want to become the single sign on of your life so that every piece of data that you have is with them so the show that you watched on Netflix yesterday the Google search that you made 3 days back the conversation that you had with your

16:14

girlfriend yesterday or today the fight that you had with your father. Everything is behind a login today.

Yeah. Of some login.

Yeah. If that login is happening via chat GP and you allow it to and you allow it to get access to your information which you will by the way because you want your

16:30

results to be personalized. So you will today you will say I will not give but tomorrow when that option is there you will give.

Yeah. It will be so bloody personalized.

Yeah. That it'll be addictive.

You think about oh you know what uh oh it's 6:30 right now 7:00 I have to get down. Uh I need

16:46

to get home. You're just thinking in AI comput 7:00 uh Sharon has to leave.

Sharon is in office right now because I can detect the location. Uh Sharon doesn't have a car because I see his car at home.

So he probably needs an Uber. So it books an Uber for you.

Bang at 7:00. Steady and steady right below

17:02

because you cognitively thought about I need a car. I need to get home and AI just does the task for you.

Wow. Because it's all interconnected, right?

In the same way based on the show that you watched yesterday on Netflix, it knows the kind of mood you are in. H so when you go to ask a question back to

17:19

AI a suggestion AI knows the mood that you're in as a result it can translate the answer in a way you feel nice about it so you not go to humans my friend you go to an AI so that is what neural link can allow neural link can do like a meta glass right

17:34

basically it's like I don't need to even put in effort to tell AI what I want AI can understand what I want just by reading my brain brain signals and it's going to happen maybe in a couple of years right have you have you seen these is even reality glasses. Huh?

So, yeah, you were saying about um

17:49

the neural link and then you were talking about meta rayband glasses which I have, right? So can you tell me what that futuristic world looks like because um a lot of these tech billions I tell you it's not even future it's possible today a lot of people have already hacked the device to do all of

18:05

this okay there is a technology called as OCR Sharon optical character reading yes so if I take that's old no that's not new that's old but it's powerful so I have a glass you are in front of me I've never met you but everybody is behind you taking

18:20

pictures right so I know you're someone important M and I'm supposed to do a podcast with you. Hypothetical situation, right?

I'm supposed to do a podcast with you or have a conversation with you. Yeah.

I have zero idea about who you are. I just look at it look at you with my

18:35

glasses. The glass has camera ray.

Yeah. Right.

It takes a picture of you. It sends it back to a let's say a platform like Perplexity.

Perplexity reads about who you are because with OCR it can know who you are. Google images already has that data.

almost has data of most of the

18:50

human beings today because everybody has a Facebook profile or an Instagram profile. So that data is available with Google.

Google identifies that you're Sharon Hag Plexely does a research about who you are. Gives out all the information understands the context that I'm having going to have a conversation with you because I just told I'm going to have a

19:06

conversation with this person. I just thought about it.

I need to know what who this person is, what I should talk about him, what I should talk to him. So it knows the context, sends that data to Perplexity.

Perplexity reads about you. does a script and throws it right in front of my screen which you can't see.

So I'm literally reading a

19:23

teleprompter while I'm talking to you through my glasses and this technology is there today. Even realities is this company which has made a sun which has made a glass.

Bro, if you look at that glass now I've ordered I've not got it yet. I'm very excited.

What is this called? Even realities.

Even realities. Okay.

19:39

They make glasses. Huh?

Rayban Meta when you look at it. No, you know it has a camera.

This doesn't have a camera by the way. This doesn't have a camera, but it has a bloody screen that others can't see.

I can only see and I can literally read whatever I want

19:54

on that screen without you even knowing. No, but then it can't read the world around me, you know?

Not yet. But merge that with a Rayban Meta Glass.

Do you think Rayban Meta is not going to be that? It will.

It is going to be that because it already has voice. So, we can probably assume a world where smartphones won't exist anymore.

20:10

I don't know about it. I don't know about it.

But that is what Mark Zuckerberg Zuckerberg is betting on. Is there a way now that I'm thinking about it is there a way for like you said if AI is not aligned with you and if we are always connected to AI with neural link is there a way that AI can slowly and

20:26

gradually lead the entire humanity in a manner that leads to each other. Yeah.

Like without even them knowing about it. Yeah.

Right. Much possible by creating conflicts and problems between different human beings in a very subtle manner and then they destroy.

So isn't that what most of the politicians in the world are already

20:42

doing with media? Yeah.

I mean imagine personalization to this level where AI decides exactly what to show you based on even more data points. So actually this is already happening through media correct because you're controlling the media now they can control at a individual level

20:57

input level individual output level a misaligned AI can make you think whatever it wants you to think but so this is what they mean that if if this technology goes into the wrong hands they can control human beings. Do you remember the case of Camry

21:13

Genetica? Yes.

What happened there? They used a lot of data.

They got a lot of understanding of how humans think and all of that and they created very specific campaigns for the right users. But that was done on a one to many level.

Correct. AI can operate on a onetoone level.

Interesting.

21:28

So they can even personalize even further to get you the Yeah. Who is who is uh who are we supposed to you know trust our uh humanity into like who is supposed to be that person who like you know I won't let this happen.

You know what? There is this very interesting read that I had recently which is very very

21:44

counterintuitive. It's fascinating.

The read says if you want humanity to be safe, we should let AI run the world. Let me tell you what I mean by that.

Okay. If your AI is aligned, that is AI is

22:01

right now aligned. We have not got to a point where AI is making its own AI.

As a result, we don't even know what is being built. Correct?

As long as AI is aligned, we slow down the progress slightly and let AI take the word ordered decisions. When you tell AI, an

22:19

aligned AI will do the right thing for you. Let's say the king of the world, the leader of the world, not of the country, the leader of the world is an aligned super intelligent AI.

H you tell AI listen you tell the world based on

22:35

the country based on the person what they should do so that the whole world is prosperous every single human in the world is happy make a prediction of that make the permutation and combination of that and you rule the world in a way the

22:51

whole world is prosperous the odds of this working out the abundance is better than Donald Trump doing it no human will ever do it because humans always have bias. Yeah, humans are greedy.

AIS are not. AIS are

23:06

a bunch of code. If they're aligned, they would be a much better leader than any human.

So this paper or this reading by this very interesting person, I think it's Mogot only but I can't remember exactly where he says the world should be run by AI. If you want the world never to get

23:23

into dystopia, we are saying 10 years of dystopia, right? M if you want the world not to get into dystopia we have to make sure in the next 2 years the world leader is an AI that's scary but I don't think that will happen yeah I don't think that'll happen I think what will eventually happen is that because the the greatest

23:38

advancements of AI is happening in US and China that'll eventually be under the control of the presidents of those countries and these guys going to use it for world domination and they're going to use it for manipulating the for human beings right that's what's going to happen exactly why sovereign AI or AI for your

23:54

own country has to be a mandate for a country like India. You need to be able to work out an AI which is personalized and sovereign to our country.

M but if we are able to figure that out and I think we should be able to figure that out right if we put our right energy that's

24:11

the only way to not allow this you see what Trump is doing bro% tariff what the hell bro you wake up in the morning you're feeling your stomach is upset and you think it is because of India and you say 25% tariff these are the kind of world leaders you have around you

24:27

China is supporting Pakistan with missiles when there's a fight against India right if you're not sovereign about this, it's going to be painful because like you rightly said, the country that has the most powerful AI would basically mean

24:43

has the highest touch in arms race. Yeah.

As a result, they will dominate the world. All right, on that note guys, let's end the conversation.

Uh VB, thank you so much. I think uh it's it's a

24:58

beautiful rabbit hole that we went on today. I have learned so many things about AI and I'm I can't wait to start off, you know, start implementing these things.

And guys, if you've been watching this for so long, uh I'm sure you are as shocked and surprised as I am. But this is just the recent updates that Vib is

25:15

talking about. I think if uh 1 month later when you analyze this world again, it'll be completely different, right?

Very very dangerous, you know, unpredictable times that we're living in. But um hopefully you take the right steps and start tinkering with AI

25:32

and start implementing it the right way. The only way out of this is to actually focus on this and actually start using the little bit of free time in your life to change your life around.

Folks, that is the biggest takeaway of this podcast and I keep saying this to everyone. Take 2 hours

25:48

out. Take one hour out.

That's all you need. But like every I just got into this 3 months back and whatever I've learned till now is only because I've been tinkering around with from 8:00 p.m.

to 2:00 p.m. 2 a.m.

instead of watching Netflix. I'm just playing around with it for 4 hours in the night.

And you've been watching a lot of my res

26:04

and I've been watching a lot of your right. Okay.

Anyways, on that note guys, I'll see you in the next one. [Music]