NotebookLM just got 10x better… let’s run it

Notebook LM is the next big thing. If

you're not using it, you're falling behind. Notebook LM is an AI tool from

behind. Notebook LM is an AI tool from Google that makes you smarter.

Seriously, just using it for a month will make you noticeably more intelligent. But most videos on Notebook

intelligent. But most videos on Notebook LM are very outdated. And that's because in the last month, the tool has improved a lot. So, in this video, I'll show you

a lot. So, in this video, I'll show you how to use Notebook LM, we'll look at the new features you most likely missed, and I'll go over how to actually use it to become smarter. Listen, I was early

to AI agents. I was early to cursor. I

was early to finetuning. So when I say that notebook LM is going to be huge in 2026, just trust me. And if you want to have an unfair advantage over everyone else, make sure to watch until the end.

Real quick, I'm hiring a second video editor. So if you know someone or if you

editor. So if you know someone or if you are editor yourself, make sure to apply.

The link will be below the video. Okay.

So to access notebook LM, just type it into Google and click on the first link.

That will bring you to the landing page, which looks like this. And here is Andre Kpathy, one of the most respected people in all of AI, saying that it reminds him of the first time he tried CHBD. Yeah,

it is very, very powerful. So, let's get started. Click on try notebook LM, which

started. Click on try notebook LM, which will open the homepage where you can see a bunch of featured notebooks as well as your own notebooks below. Now, if you click on see all, you can see all of the featured notebooks that Google

suggested. But the best thing to do is

suggested. But the best thing to do is to create your own notebook. And you can do that either by clicking this white button in the top right or by clicking create new notebook right here. Now,

when you create your first notebook, Google instantly wants you to upload sources, which can be a bit overwhelming. So, what I recommend you

overwhelming. So, what I recommend you to do is either press escape or exit out so that you get to the main UI. And

don't worry, you can attach sources at any moment. Now, as you can see, the

any moment. Now, as you can see, the main UI has three parts: left, middle, and right. The left side is all about

and right. The left side is all about the sources. The middle of the UI is the

the sources. The middle of the UI is the chat. So, if you've used Gemini or Chad

chat. So, if you've used Gemini or Chad GPD before, you should be pretty familiar with this part. Now the right side of the UI is the most powerful part of notebook LM but more on that later.

Now before we go any further I must explain the concept of sources because this is the bread and butter of notebook LM. This is what allows us to do the

LM. This is what allows us to do the context engineering inside of the app.

Now notebook LM is a fine-tuned version of Gemini that's instructed to hyperfocus on the sources. So while a normal language model would give you some of its opinion and some from

sources, Notebook LM is all about the sources. So if you mess this up, nothing

sources. So if you mess this up, nothing else really matters. Notebook LM

supports a lot of different types of sources. PDF documents, Word documents,

sources. PDF documents, Word documents, Google Sheets, images, PowerPoint presentations, audio files, YouTube videos, website URLs, and just copy pasteed text. So a general rule of thumb

pasteed text. So a general rule of thumb is that the more sources you add, the better as long as they are relevant to the topic you're discussing. Now, one of the most common mistakes that beginners

don't realize is that the chat aka the middle part of notebook LM is not included in the sources, right? So, if

you say something or you ask it a question, that by itself is not considered a source. So, if you later then create an audio podcast or a mind map or video presentation, notebook will ignore your chat unless, and this is

important, you create a note. Now, you

might be thinking, "Okay, David, but what if I don't have any sources on hand, and I still want to use Notebook LM to upskill myself, to learn super fast, and to just become smarter overall." Don't worry, Notebook LM has

overall." Don't worry, Notebook LM has the solution for this, and that is this box on the left. So, as it says, it allows you to search the web for new sources. And there are two types of

sources. And there are two types of search, fast search and deep research.

The fast search will look for like 10 websites and give you results within seconds. The deep research on the other

seconds. The deep research on the other hand is a new feature that will browse hundreds of different sites and it can take like 5, 10, 15 minutes to do a really thorough analysis of the web to

find anything and everything related to your topic. So first let me show you a

your topic. So first let me show you a fast research and we can do something like look up promising research papers in AI ML from the last two months. Boom.

Hit enter and notebook LM is going to research archive, nature magazine, you know, and other publications where you can find scientific papers about AI and machine learning. And there it is. It

machine learning. And there it is. It

took like six seconds and we have 10 different sources right here. So you can click on import and boom, here we can see our sources on the left. And you can see that each source is distinctly separate from the last. And that's

because you can untick them. Right now

here is where the context engineering begins because anytime you untick a source, the chat doesn't know it. So pay

attention to this small number next to the set icon. It says eight sources. But

if I takeick these now it's 10 sources.

So notebook LM is very strict about what is inside of the context window and what isn't. Now anytime you add new sources,

isn't. Now anytime you add new sources, Notebook LM will begin analyzing them.

And you can see that some of them are in red and that means it wasn't able to get any data from that source. So for

example, this one at the bottom, it says this source is behind the pay wall. And

you can clearly see which sources failed by the red highlighting above them.

Anyways, once all the sources are analyzed, you'll see a message in the chat where notebook LM gives you an overview of what all these sources are about. So here we can see the provided

about. So here we can see the provided sources offer an overview of cuttingedge advancements and analytical methods within the field of artificial intelligence. Now let's say you only

intelligence. Now let's say you only want to focus on a single paper, right?

Such as this deep researcher paper.

Well, first off, you would unselect all sources right here. Then you would select just this paper. And by the way, another cool thing is that you can click on any of the sources and this will give you a source guide which is basically a

summary highle overview of that source what it is about so that you can decide whether you want to be using this source or not. And then below the source guide

or not. And then below the source guide you can see the full text of the source and yeah there is a lot in this paper.

So let's close that and let me ask a simple question in the chat. give me the main insight from this research paper and suggest five potential startup ideas

built on top of that insight. And by the way, this is a brilliant way of getting business ideas and startup ideas, is to read and research the latest research papers in AI, which most people never do. They just skip those and then think

do. They just skip those and then think what different products and services might be possible given the contents of the paper. Now luckily for us thanks to

the paper. Now luckily for us thanks to notebook LM we don't have to be a scientist or a researcher to understand these papers because it can explain all of them in a simpler and clear way. Now

still this is a bit too long. So I would say make your answer simpler and shorter. Boom. Now as I said earlier the

shorter. Boom. Now as I said earlier the chat by itself is not considered a source. So when later we use any of the

source. So when later we use any of the features in the studio on the right don't expect notebook LM to know your chat conversation. If you want to

chat conversation. If you want to include that, what you have to do is scroll at the bottom and below every message, every response from the AI, you'll see save to note. If you click

that on the right, below the buttons, it will generate a new note. And this is basically a saved version of the AI's response. However, if you want to take

response. However, if you want to take this note and convert it into a source, there's one more thing you have to do.

Click on these three dots, this more button, and click on convert to source.

And then on the left, we can see that this new markdown file was created. And

you can see that this markdown file is exactly the same as the message, which means that we've managed to convert part of the chat as a source. Anyways,

resuming from Ireland, the most OP thing you can do with sources and this concept is the search, right? So, you have the fast research and deep research. The DB

research is the new feature, but even the fast reach is super good and you can choose from either web or Google Drive.

So since Notebook LM is developed by Google, you can have it search across your entire Google drive, find different docs, sheets, even Google slides, even presentations and attach them as sources

right now. Obviously, web is going to

right now. Obviously, web is going to have more info in general. And this is really how you get sources fast. And

even just three or four great prompts can do wonders here. So for example, here we already have the 10 sources, but I'm going to do another prompt with slightly different uh intention. I'm

going to say research interesting and useful A IML papers about optimizations or inference and any new techniques in

token generation and GPU optimization.

Make sure the papers are from Q4 2025.

This is even more specific because maybe you want to start a business that's like competing with, you know, some of these neo clouds, some of these inference providers. This is how you get an edge.

providers. This is how you get an edge.

This is how you get an advantage. you

put in more time and more research than everyone else and you understand a topic deeply and then you put multiple things together right so if you're looking only at a single thing maybe you're expert at marketing there's many experts in

marketing but if you learn marketing and AI you're going to have a great time same with sales and AI same with software development and AI like you know just combine two different insights and that's how you get the best business

ideas and any advantage it doesn't have to be a business idea it could be a clever marketing campaign that makes your business uh seven figures, whatever, right? Anyways, the new

whatever, right? Anyways, the new sources automatically get added and now you can see that they're selected and we can see 19 sources above the chat. But

honestly, the chat is not the main thing. The main thing is the studio on

thing. The main thing is the studio on the right. This is what makes Notebook

the right. This is what makes Notebook LM special. Now, I already know some of

LM special. Now, I already know some of you are thinking, "But David, I don't even know what I would use Notebook LM for." And as I promised, at the end of

for." And as I promised, at the end of this video, I'll show you the most OP use cases I've been able to find. And

actually a big part of my content consumption in the education niche, right, because I don't really watch entertainment content has shifted from YouTube and human generated content to

Notebook LM, which is both scary and impressive because I'm also a content creator making educational content. But,

you know, hey, none of us are safe. The

future is uncertain. We're probably all going to get replaced. So, the best thing you can do is adopt fast. But

honestly, I might make a whole video on this topic because it's a huge topic.

And yeah, anyways, notebook alum has a ton of great use cases. As I said, later in the video, I'll show you some of the most OP ones. But first, if you join the new society this week, which means starting 17th of November, you'll

actually get all of my notebook alen projects for interesting AI and business topics. So, for example, lately I've

topics. So, for example, lately I've been studying game theory and how that applies to real life and, you know, business. Also, I've researched the best

business. Also, I've researched the best jurisdictions to start AI startups in.

How you can apply Miiamoto Mousashi's principles to business life productivity, the different strategies that successful content creators like Hermosi or Mr. Beast apply, and how we can replicate them and learn from them,

and plenty of other AI related notebooks. So, everyone who joins the

notebooks. So, everyone who joins the new society this week will get access to all of my notebook alms, and yes, even the private ones. But not only that, another bonus that I cannot do every week because it takes a lot of time is

that everyone who joins before Sunday will get a personalized short video from me. So this could be a question about

me. So this could be a question about your YouTube channel, business advice, help with your startup, or anything else. And no, this is not going to be AI

else. And no, this is not going to be AI generated. It's going to be me in the

generated. It's going to be me in the flesh consulting you on whatever you need help with. But again, this offer is time limited to this week only. So if

that sounds interesting to you, make sure to join the new society. It's going

to be the first link below the video.

Now, back to Notebook LM. So audio is what notebook LM got first famous for and recently it has been improved even more. So inside of notebook LM you can

more. So inside of notebook LM you can create custom podcasts on any topic right and this really depends on the sources. So as I said on the right you

sources. So as I said on the right you have the studio tab and if you expand it it's easier to access. Now audio

overview this is the podcast right? So

if you do it on your phone actually it automatically generates the medium length audio. Now I would recommend you

length audio. Now I would recommend you click on the edit button where you can do a lot more. And by the way, Notebook LM is great on phone as well, but the mobile app is a lot more limited. So if

you want to get the most out of it, make sure to use it on the web. Okay, so you can see that when you click the edit button, you get multiple things you can customize, right? Such as the format. Is

customize, right? Such as the format. Is

it a deep dive? Is it a brief? Is it a critique? Is it a debate? These options

critique? Is it a debate? These options

might be useful in different situations, right? So let's say you're preparing for

right? So let's say you're preparing for an important debate. Well, you would select debate. If you want to pitch a

select debate. If you want to pitch a startup idea to investor, you would select critique to know all the flaws and all the potential downsides of your idea. If you want to just uh have a lot

idea. If you want to just uh have a lot of sources and you want to quickly get up to speed, you would select the brief.

But the deep dive, this is universal, right? You want to learn a topic deeply,

right? You want to learn a topic deeply, deep dive is the default. Now, what's

just as important is the length. So, you

can see right here we have three presets, short, default, and long. U if

you don't select the edit button and you just click on audio overview, it will generate the default length. If you can see right here and it is going to be the deep dive type or format they call it

format but uh long is really OP. It's

like you know 20 30 minutes podcast that goes into the all the s sources you have attached super in-depth and the podcast is a format of two people talking back and forth right so a male voice and a

female voice and by the way one tip if you switch to a different language which there's a lot of languages like you know check Deutsch English espanol probably like 50 maybe more than maybe like 70

languages pretty crazy right however if you select a different language so check boom you only have shortened default English is the only language I believe that has the long format and then you can give it a system prompt right you

can tell it what to focus on so focus on papers that have real business use cases for example right you can do like a long and then generate keep in mind that the

long ones take longer to generate right because the default is usually like 5 to 10 minutes max usually around 5 minutes the long one can really be over 25 minutes so that's five times longer so

it takes five times longer to generate and as It says come back in a few minutes. Right? So, this is not instant.

minutes. Right? So, this is not instant.

This is not a chat GBT response.

Generating an audio podcast with two different voices. And this is actually

different voices. And this is actually really impressive because you'll see in a moment the voices complement each other. It's not just like one says

other. It's not just like one says certain sentence, the other one says certain sentence. It really is a two

certain sentence. It really is a two people. In this case, it's two AI agents

people. In this case, it's two AI agents or two AI avatars talking back and forth and having a discussion on this topic and explaining to you. So yeah it is completely different way of learning and

that's why I call it hyperarning. Now

unlike a regular podcast you can talk and interact with this one and that's how you can get an even deeper understanding of the topic. So as these two male and female characters are

discussing and explaining the thing to you can interrupt them. You can jump in and ask questions and it really feels like you're a guest on a Joe Rogan episode where he has an expert on. Now,

just to clarify, this is an AI image because I would be much taller than Joe here because Joe is very short. Anyways,

hyperarning with AI. This is a rough chart of what hyperarning looks like.

And really, like guys, I'm not over exaggerating. If you're not using it to

exaggerating. If you're not using it to upskill yourself, at to learn, you are falling behind. I don't care what the

falling behind. I don't care what the topic is. I don't care if you're

topic is. I don't care if you're improving at sales, at content creation, at programming, at at your job, at dating, social dynamics,

whatever you want to get better at. With

Notebook LM, you can learn much much faster. If there's a complex book that

faster. If there's a complex book that you've read and you fully don't understand it, load it into Notebook Element. And actually, pro tip, a lot of

Element. And actually, pro tip, a lot of I've watched some videos obviously preparing for this video on YouTube and a lot of YouTubers, they think there's like a secret if you combine sources, right? Because here is a limit on 50

right? Because here is a limit on 50 sources which most of the times you will not even hit that but let's say for some reason you hit uh 50 sources right obviously you can copy multiple of them into a single dog and attach that as a

new source and you can do more sources but this is actually only on the surface only on the first glance this is like a fool's errand because it's not good what you actually want to do is want to you want to split sources into chapters

right so let's say you have a book you know PDF format such as this one psycho cybernetics you would do uh typically a single source for this, right? But this

is a missed opportunity because what you can also do is you can split it into chapters, load each chapter as a separate source and then focus on the chapters that you do not understand. And

this is one of the biggest secrets that I've never seen any other content creator talk about is instead of merging sources into together, you actually want to take a big source and split it into

chapters or different chunks and then attach each of those chunks as a separate source. then you can select or

separate source. then you can select or unselect you know the ones you want to focus on and have notebook LM explain just those chapters to you so that you get an ultra deep understanding of the

material. Now we have the podcast

material. Now we have the podcast generated so let me minimize these and let's play it because uh this is really if you've never heard this it's actually really good.

>> Welcome to the deep dive the place where we take the hardest newest research in tech and well we distill it down to what you really need to know. So today we are wrestling with this this core paradox in

advanced AI. We need models to be

advanced AI. We need models to be smarter, you know, capable of real research and complex reasoning. But

getting there just burn.

>> This is just the female voice. You'll

see in a moment that the other the male AI interacts with her and goes back and forth. And this is really the magic of

forth. And this is really the magic of what makes it interactive and what makes it easy to listen to and understand. And

it's not just running two AIs in parallel, right? You might think, okay,

parallel, right? You might think, okay, maybe you can recreate something similar with 11 Labs. The problem is that Google has really built it well where the AIS interact with each other and it sounds natural. Just take a listen.

natural. Just take a listen.

>> We're diving into a whole stack of new research papers that attack this efficiency problem from well, every angle you can think of.

>> And what's so fascinating is just the sheer breadth of it. We're covering

everything from new training methods that create genuinely smarter agents like this deep researcher framework. and

then we'll dive into the algorithms that you know compress those long thinking processes.

>> Now on the right you can see this interactive mode where you can actually jump in and ask questions. So I'm going to click on this. I'm going to hit play audio.

>> Welcome to the deep dive the place where we take the >> and if you want to join you have to click the join button.

>> So today we are wrestling with this. Oh

hey our listener wants to join in.

What's up?

>> Yo I have a question on which of these papers we should focus on. Which is the most interesting?

>> That is a fantastic question.

>> Oh, we love that you chimed in.

>> You know, we're covering a whole stack of new papers.

>> They're all interesting for different reasons, >> but if I had to pick the one that feels the most like a breakthrough.

>> I think we're both leaning toward the deep researcher framework.

>> Yeah, that's where the real AI capability leap is. Yeah, guys, uh there is going to be a lot of people who develop parasocial relationships with this because there's, you know, chatting

with LLM is one thing, but having realistic voices talk to you with like perfect inonation. Yeah, that's going to

perfect inonation. Yeah, that's going to oneshot so many people and especially women because women are the ones who are really developing this addiction to chatbots. Recently, there was in Japan a

chatbots. Recently, there was in Japan a girl who married an AI chatbot, legally married with a ceremony and everything.

People think like, "Oh yeah, it's going to be the AI girlfriends, right?" But

actually, AI boyfriends is a much bigger market because what's the saying is like men fall in love with what they see.

Women fall in love with what they hear.

And uh having custom voices that are personalized, that can do emotions, that can be convincing and persuasive. Yeah,

guys, u there's going to be some very not so good use cases, but luckily this one, Notebook LM is for learning. It's

for educational. It's for making yourself smarter and these podcasts I just you know as I was flying here to Cork Ireland on the on the plane I was listening to multiple notebook LMS and

yeah this is the future of education absolutely personalized AI generated education that's like highly specialized on the topic. It's going to replace a lot of like lazy content creators and I think the only advantage you can do as a

human is to generate unique thoughts.

This is what language models cannot do.

This is what we humans can do. So if you are considering, you know, either making your own YouTube channel, building your personal brand, you know, writing LinkedIn posts, creating Twitter threats, if you're just going to

regurgitate other people's content, you're going to replace by the notebook LM 2.0, 3.0, whatever in the future. The

only way you can compete is to have unique insights, is to consume different books, different inputs, and distill them in your head. Think about them deeply and then have unique thoughts and

share unique perspectives. Otherwise, uh

tools like Notebook LM are going to replace you. It's pure that simple. Now,

replace you. It's pure that simple. Now,

as I said, splitting sources is great, but another thing you can do in Notebook LM is generating personal mind maps. So,

again, on the right, we have the studio.

And by the way, I just collapse these, right? So, you can resize them however

right? So, you can resize them however you want. Um, so it is very flexible.

you want. Um, so it is very flexible.

The UI is nice. On the right, one of the new features is also the mind map. And

this is uh very visual, right? If you

are a visual learner, like all types of learning preferences, notebook has you covered, right? You might be an audio

covered, right? You might be an audio learner, then you would listen to the podcast. You might be a textbased

podcast. You might be a textbased learner. You can just chat with the with

learner. You can just chat with the with the LLM. This is powered by Gemini

the LLM. This is powered by Gemini obviously. Or you might be a visual

obviously. Or you might be a visual learner which in this case you're going to love these mind maps. And this is what it looks like, right? We have the mind map here. And we have inference engine optimization survey. And then you

have five sub points from this one. Like

let's see which one is interesting, right? Context challenges. Maybe we want

right? Context challenges. Maybe we want to do optimization techniques. Well, you

would click on this one. Boom. Not only

does it open in chat and then Gemini is generating response specifically on this but we can see in our mind map it's extending right so the tree is growing and it has generated five new

subtopics based on optimization techniques. So maybe you understand some

techniques. So maybe you understand some of these maybe you want to improve yourself at parallelism. Well, let's

click on here and we have another former options and the mind map is growing. And

this is really OP for education because let's say you get like, you know, 10 layers down. You can backtrack to see

layers down. You can backtrack to see how you got there and really like you're make a strong neural connection in your own brain. You know, the most powerful

own brain. You know, the most powerful neural network in the world so far. And

uh yeah, I mean this tool is like actually crazy for learning. And this is a clear example like maybe you understood maybe you thought you understood parallelism but then it gives you four types of them and maybe you

haven't learned about tensor parallelism right or fully shard sharded data parism fsdp right and now we're getting real deep here and actually this is the last

level here because it again depends on the sources so depending how many sources and how well the sources build on top of each other you can create different mind maps and uh chat with

different points of the mind map and again any of the points. If you like them, you can save them as a note and then convert that note into a source and generate an audio, you know, do more

with it to understand it even deeper.

Oh, and by the way, two quick tips at the top. You can share the notebook with

the top. You can share the notebook with other people. This is very useful

other people. This is very useful because uh you can like prepare this for somebody and deliver them, right? Like

for example for a higher up at your company you can do a research and just give him the notebook with a ready mind map ready podcast for him to get up to speed or you can see the analytics once you have people shared but you need at

least for other users. So if you share it with your team you can see like different metrics on this but this is only available on the pro plan. Now

another thing I want to show you is the videos. So this is the cherry on top

videos. So this is the cherry on top rang, right? Because a lot of people

rang, right? Because a lot of people know about the audio podcast, but the video presentations, man, it can it can create a whole presentation for you. So

there's uh three more things I want to show you. Video presentations,

show you. Video presentations, flashcards, and quizzes. And again, this is like more things, more tools for you to do hyperarning. And again, I need to reiterate how important it is that you customize this. This the future of

customize this. This the future of education is personalized education, right? It's like asking a prompt that's

right? It's like asking a prompt that's like in your own writing style, in your own medium. So it could be text, video,

own medium. So it could be text, video, image, mind map, quiz, in your own difficulty level, in your own conciseness, in your own language, in

your own speed. I mean, you can customize all the variables and learn much faster than you could ever before.

I I don't think we guys understand learning because all of us associate learning with school or like reading books. All of these are such low

books. All of these are such low leverage. learning is going to take a

leverage. learning is going to take a completely new form over the next five years. This is a prediction that I'm

years. This is a prediction that I'm willing to make. Bookmark this, you'll see uh there's going to be some crazy types of learning even three years from now. So, for example, you can create an

now. So, for example, you can create an entire guide on anything. And once

you're happy with it, you can transform it by saying make this for my team who are beginners. So, if you are a manager

are beginners. So, if you are a manager on the team, you can understand the concept deeply and distribute it across everybody on your team and train up your team much faster than ever before. So if

you're working on a company and especially if you have people below you do not ignore this because you can get such a high leverage of unlock right even if you make those people 5% productive more productive that

distributed among 10 people is like 50% u you know relevant gain if that was to your own time instead of working eight hours a day that would mean 12 hours a day right so obviously that's much

harder to do than boosting productivity by 5% of multiple people and with notebook LM you can easily do And one of the reasons is that it uses nano banana to generate these visuals. And keep in mind this is this is the worst it's

going to be. It's only going to get better. Let me show you how the videos

better. Let me show you how the videos work. Right? So if you go into studio

work. Right? So if you go into studio again, we can see that we have the video overview. And again, if we click on edit

overview. And again, if we click on edit option, we have a multiple options. We

have the explainer, we have the brief.

So this is like the concise option. Then

we have the language classic. And the

reason for that is it comes with audio.

So it's not just video, it's a whole video presentation. Guys, uh this really

video presentation. Guys, uh this really might take my job soon. So I really need to double down on these unique insights and especially touching controversial topics. That's another way. So having

topics. That's another way. So having

uni unique insights and original thoughts that's the first way you can differentiate yourself and make yourself irreplaceable from AI. But the second

second way is willing to touch unique um not unique topics controversial topics okay especially stuff that is like uh against AI which you know since it's developed in San Francisco is usually

super woke super liberal. If you have any opinion that's even remotely right or off center, that already is going to be something that AI models are going to disagree with you on. And it's an easy check to see if the person you're

talking to is a human or is a AI. You

know, you can just say a slur. That's

another quick verification. These AI

models there don't really like to curse, right? So that's how you can prove your

right? So that's how you can prove your humanity. But again, have original

humanity. But again, have original ideas. Really try to get unique

ideas. Really try to get unique insights. And number two, be willing to

insights. And number two, be willing to discuss uncomfortable and controversial topics which the LLMs are afraid to do.

Anyways, back to the presentation here.

So, next we can choose a visual style, auto select, custom, classic, whiteboard, some anime styles, retro.

We're going to do classic and then the prompt you can see make a very very short presentation explaining these aim

papers. use lots of graphs and math

papers. use lots of graphs and math style visuals generate and again just like the audio this uh takes some time

right text generation is the cheapest and fastest audio generation is the second I guess maybe image generation is the second fastest audio is the first but video generation is the most expensive right it takes the most

compute and so don't expect this to be instant now as this is going let me show you the remaining types in the studio because we have reports this is uh probably the most boring out of these because it's just a basically Google

doc, right? It it saves it like a mark.

doc, right? It it saves it like a mark.

It's a markdown style and it's just text. So, we're going to skip it. We're

text. So, we're going to skip it. We're

going to do flash cards. This is much more interesting. Uh this is amazing way

more interesting. Uh this is amazing way to learn flash cards. So, number of cards, let's do fewer or let's do standard level difficulty. Let's do easy because these papers are complex. Uh

explain the core breakthrough ideas in these papers. Okay. Generate. So, we're

these papers. Okay. Generate. So, we're

going to generate some flash cards. As

you can see, you can be generating multiple things at a time in Notebook LM, which is very very powerful. And

really, like if you sit down on any topic with Notebook LM for 60 minutes, you're going to learn that topic faster than anywhere else. Like maybe the only exception is that if you had a 60-minute private tutoring call with the world

expert on this topic, right, which would cost you probably like $500 an hour.

Notebook LM you can get started for free, right? I have the pro version, of

free, right? I have the pro version, of course, because I want to show you guys all the features, but you can get started with Notebook Alm for free instead of paying a professional tutor, you know, $200 an hour, $500 an hour,

whatever. All right, so here are the

whatever. All right, so here are the flash cards. Uh, that took like a

flash cards. Uh, that took like a minute. So, longer than I would like.

minute. So, longer than I would like.

Video obviously takes a little longer.

So, let's click on that. Boom. And as

you can see, it's nicely designed. So,

in LM inference, what are the two main phases of processing a user's input?

Generate a response. Okay.

So we need to tokenize it because LLMs don't see words, they see tokens.

And then we need to run the actual inference of the model, right? So run

the like have the weights generate a response. Let's see if that's correct.

response. Let's see if that's correct.

Are the prefill phase and the decode phase. Okay, so it's not really talking

phase. Okay, so it's not really talking about the same thing, but whatever. Next

flash card. What's the primary function of the prefill phase in LLM inference?

So I guess that's prefilling the weights with default values and then it does back propagation. Let's see if I'm

back propagation. Let's see if I'm correct. Process all input tokens into

correct. Process all input tokens into parallel compute initial KV cache. Okay.

So my knowledge on inference is clearly lacking. I should probably actually go

lacking. I should probably actually go through all these cards. But I'm also going to show you the quiz because uh quiz is more interactive than the cards, right? So you can do number of

right? So you can do number of questions. Let's do fewer. Let's do easy

questions. Let's do fewer. Let's do easy difficulty. And let's generate a quiz.

difficulty. And let's generate a quiz.

Now, while the video presentation and quiz are generating, let me show you some of the most OP use cases of Notebook LM. First of all, as I said,

Notebook LM. First of all, as I said, debate prep to destroy anybody in a debate. Notebook LM is a superpower. Not

debate. Notebook LM is a superpower. Not

only to understand the counterpoints, but just overall to get the best arguments for and against whatever topic you're arguing, right? Sales.

You can massively improve at sales using Notebook LMS as like your what's it called? Roleplay buddy. That's that's

called? Roleplay buddy. That's that's

what I'm trying to find. But even before that, you can do a deep research on a prospect, right? So, let's say you need

prospect, right? So, let's say you need to cold call an important uh client and you only have like five of these clients that you want to cold call. Well, you

can either just be completely uh lame and just cold call them without any knowledge on their company or you know their problems or their strategy or their team. Or you can do deep research

their team. Or you can do deep research with Notebook LM. Turn that into a podcast. Listen to that for 10 minutes

podcast. Listen to that for 10 minutes before the cold call and suddenly you know a lot more about his company, about his industry, about his team and what are the problems, what are their

objectives and that will give you a massively higher chance of closing the deal. So if you're in sales, notebook is

deal. So if you're in sales, notebook is a must have sensitive topics, right? So,

anytime there's something in the news, let's say like the Iran war, you don't know who to listen to, right? Like

there's literally people uh saying on both sides there are like people are lying left and right, especially anything that's policy sized. And if

you're not expert on the topic, you really don't know what's true. Well,

guess what? With Notebook LM, you can attach 50 sources where 25 are for on one side of the topic and 25 sources are for the other side of the topic. And

then you can have a really deep understanding of a sensitive issue that otherwise you would have to watch like multiple different podcasts each of which is like multiple hours probably to

even make up your own opinion. But with

notebook LM you can cut down that time from like five to 10 hours of research to sub one hour of research and while having a deeper understanding that's the crazy part notebook LM not only allows

you to learn faster you will also have a deeper understanding than if you do traditional analog slow learning another use case as I mentioned earlier is the AI research papers and as a reminder if

you join the new society in this week you'll get all of my notebook alms so make sure to join first link below but research papers are usually complex, right? If you're not a scientist, if

right? If you're not a scientist, if you're not a researcher, you will have a trouble understanding a lot of these.

And since now notebook LM can accept image sources, right? So this is a again a lot of people don't realize this because most of these YouTubers skip over it. You can do like for example the

over it. You can do like for example the transformer architecture or any of these you can just take a screenshot. Boom.

There we go. Copy this. Go into notebook and paste this as a source. Oh, by the way, just notice the source limit is 300. Okay, maybe that's uh depending on

300. Okay, maybe that's uh depending on if you have a free plan or not, or they just upgraded recently. Let's verify

that. Yes, notebook increase the source limit from 50 to 300, but only for paid users. Okay, so that's another reason

users. Okay, so that's another reason why you might consider upgrade updating.

This text below my webcam is the number of sources you can have in a notebook.

And if you have a paid version of notebook LM, you can do six times more sources. All right, so back to image.

sources. All right, so back to image.

You can attach that as a source. And you

can see that I just drag that in here and it's scanning the screenshot. And

yeah, that's another massive update to notebookm you can attach images as sources. So while something you know

sources. So while something you know like this might be difficult to understand if you're not a AI researcher with notebook that's no longer the case.

You can attach as a source and also it can explain this right this visual illustrates architecture design of a large language model specifically comparing a standard transformer encoder decoder structure with the more

specialized DCV3 model. Boom. You can

each source as the source guy as I said and then you can see the source in full.

Really OP now ability to add the images.

Anyways, we have the quiz. So let me show you how the quiz looks like.

Collapse this. Boom. So you can see that instead of a flash card where you click it and you see the answer, it tells you four options, right? So LM inference process parameter divided into two

distinct. Okay, so it's the same

distinct. Okay, so it's the same question. So let's see if I remember it.

question. So let's see if I remember it.

The prefill and the code. Boom. That's

right. All right. Next. Which

paralization strategy involves replicating the entire model across multiple GPUs and splitting the data badge among them? It sounds like this.

No, data parization. Okay. Yeah, I need to up my knowledge of inference. But

this is pretty deep stuff. Unless you're

working at a LM inference company or like Nvidia or chip maker, you probably don't need to know all of this. But hey,

if you want to educate yourself, now you know how. But we also has a video brief.

know how. But we also has a video brief.

So okay let me show you a few more use cases before we get to the video brief chatting with biographies. So you can basically talk to dead people right you can talk to people like Isaac Newton

Carl Gaus the geniuses that have lived throughout history and you can discuss with them and understand their idea deeply. You can oh Isaac Newton was a

deeply. You can oh Isaac Newton was a prolific writer. He wrote over a million

prolific writer. He wrote over a million words. So maybe you cannot get all of

words. So maybe you cannot get all of his sources. Maybe you can if you have

his sources. Maybe you can if you have the problem, but basically you can get all of his writings and then really understand his thoughts on a level that

you won't get by listening to others who simply interpret his ideas. Another op

use case is traveling. So let's say you're traveling to a new city or you're planning to move to a new city. With

Node LM, you can just do like two prompts and you can learn everything about the city, right? Best restaurants,

history, random facts, the must visit spots. And again, you can do that while

spots. And again, you can do that while you're walking at the airport because you can listen to it in your AirPods since it's a podcast. Now, I know guys, you're going to find this one valuable.

Understanding a codebase. For all of you who are building software with AI, Notebook LM lets you reduce technical debt, right? You can paste in all of the

debt, right? You can paste in all of the dogs, all the markdown files and learn the codebase 10x faster and really eliminate your technical debt. Something

that most vibe coders will never do.

Okay. So, let me show you what this video presentation looks like because uh a lot of people don't realize that notebook element is capable of this. So,

let me make this full screen and let's play this. Ever feel that lag with your

play this. Ever feel that lag with your AI chat? Well, today we're digging into

AI chat? Well, today we're digging into the secrets of AI speed. So, that delay you're feeling is not just the AI thinking. It's a pretty complex process.

thinking. It's a pretty complex process.

>> Okay guys, I have to pause it. This is

already insane.

This is going to speed up so much productivity, but at the same time, it's going to replace some jobs, right? If

you're Yeah, I mean I really need to rethink my whole strategy because uh it's not clear at all. See, not it's not only about

at all. See, not it's not only about like the products and services, right?

Because okay, if you're in business, you're worried about your business being replaced, right? If you have a job,

replaced, right? If you have a job, you're worried about your job being replaced. But it's not just that. If

replaced. But it's not just that. If

you're in business, you the only you you cannot only worry about like whether your product or service will be relevant. It's also the money model. For

relevant. It's also the money model. For

example, SAS software as a service subscriptions don't work nearly as well as they used to because now with inference each token scales linearly

with the amount of users. In the past, you develop a software and you distribute it to millions of users and it's like basically pure profit, right?

It's like the margins can be 98% because distributing that code to another person was basically free. That's not the case with AI SAS because each person is

generating new tokens. So your cost is growing with the amount of users which is absolutely eating your margins.

So not only do you have to reinvent what B like what your business does, what products and services you provide, you also have to invent new ways to monetize. So yeah, the future is highly

monetize. So yeah, the future is highly uncertain and if you're feeling the pressure and uncertainty, trust me, you're not the only one. I'm feeling it just as much. And uh again I probably will make a whole video on this because

it's a massive topic. But for now the best thing you can do is be as adaptable as possible. You know be like water as

as possible. You know be like water as Bruce Bruce Lee said as Miiamoto Mousashi would say be formless. Do not

fall into any rigid tactics or you know past habits.

Be formless. Be adaptable.

Constantly reinvent yourself and try new stuff because stuff like this wasn't even possible year ago. And now we have AI generating whole whole videos on any topic. Insane.

topic. Insane.

>> You see, AI inference has two parts.

Prefill reads your prompt and decode generates the answer. We measure this speed with a few key metrics. And yeah,

both.

>> What? This is crazy, guys. I know I'm making this video, but like I'm still getting shocked by how good Notebook LM is. This is OP. This is like too good.

is. This is OP. This is like too good.

This is almost like disturbingly good.

Both of >> those phases can create bottlenecks.

So, what's the solution? It's a

specialized piece of software called an LLM inference engine. Unlike your

general purpose stuff, these engines are built for just one job, maximum speed.

Okay, so how did they pull it off?

They've got a whole toolkit of tricks.

Let's look inside. First up, continuous batching. It grew.

batching. It grew.

>> All right, you get the point. So, I'm

not going to go through the whole video presentation, but yeah, this is good.

This is uh this is scary good.

Not only are the visuals nice, easy to follow, uh varied, dynamic, but uh you know there is two voices

again to make it more engaging and the explanations are clear on topic. There

is no glitches, no artifacts. Yeah,

Google, shout out to the team at Google because you guys really cooked with this one. Notebook LM is OP. And if you want

one. Notebook LM is OP. And if you want me to make more videos on notebook, comment below because uh there's so much you can do with this, especially the use cases is really crazy. And uh yeah, I

could make a whole 1 hour video of just different esoteric and non-obvious ways to use Notebook LM to get an unfair advantage over everybody else. So if you want me to make not more notebook alam

content, number one, subscribe because if I see a lot of videos, a lot of people are subscribing from this video, that's a very strong signal that I should make more content and number two comment below. With that being said,

comment below. With that being said, hopefully you guys found this video valuable. If there's one thing I can

valuable. If there's one thing I can encourage you to do, go and create your first notebook. Try this. Push through

first notebook. Try this. Push through

the initial friction and use notebook LM for the first time. It takes five 10 minutes max and it will absolutely change the way you see learning and yeah it will make you smarter. That's one

thing I can promise you. If you use notebook LM you will be smarter. That

being said thank you guys for watching and have a wonderful productive week.

The