Microsoft's new "Embodied AI" SHOCKS the Entire Industry! | Microsoft's Robots, Gaussian Splat & EMO

Summary

この動画スクリプトは、AIの世界で起こった衝撃的な出来事、特にFigureという企業がNvidia、Jeff Bezos、Microsoft、OpenAIなどから莫大な投資を受け、人間型ロボットの商業展開を加速させるニュースに焦点を当てています。OpenAIはFigureと協力して、ロボットの認識、推論、および相互作用を拡張するための次世代AIモデルを開発します。ビデオはまた、中国からのSoraという革新的なビデオ生成AIの紹介、そして人工知能の分野での最新の進歩とそれが世界に与える影響についても触れています。最終的に、動画は、AIとロボティクスの統合がどのように現実の世界に革命をもたらす可能性があるかを示唆しています。

Takeaways

• 😊 人工知能の進歩についてのビデオです。
• 🤖 Figure社の人型ロボットについて説明しています。
• 😮 OpenAIとFigureが協力する驚くべき発表がありました。

Q & A

質問1

回答1

Outlines

00:00

🤖 AIと人間型ロボットの進化

AIの世界で衝撃的な出来事がサブレディットで話題になっており、特にFigureという企業がNvidia、Jeff Bezos、Microsoft、OpenAIなどから莫大な投資を受けていることが注目されています。OpenAIはFigureとの協力を発表し、ロボットの認識、推論、相互作用を拡張するためのマルチモーダルモデルの開発に取り組んでいます。FigureはMicrosoft Azureを活用してAIインフラのトレーニングとストレージを強化し、人間型ロボットの商業展開の加速を目指しています。この動きは、AIとロボティクスの統合が進んでおり、具体的なデモンストレーションが行われていることを示しています。

05:00

🌐 3Dシーンの革新的レンダリング

最新のAIモデル「Sora」は、従来のモデルを大きく上回る3Dシーン生成能力を持っています。従来のテキストからビデオへのAIプラットフォームと比べて、Soraは幾何学的一貫性において顕著な進歩を遂げています。このモデルは、実世界の現象をシミュレートする能力に関して活発な議論を呼んでおり、それが単なるビデオ生成ツールではなく、世界シミュレーター、物理シミュレーションの新しい形態であることが強調されています。Soraは物理的なルールや3D変換を学習し、高度なビデオ生成を実現しています。

10:01

🔍 AIによるデータアノテーションの未来

OpenAIの新しいプロジェクト「Feather」は、画像、音声、ビデオ、テキストなどのデータの自動ラベリングと注釈を可能にするシステムです。特に、GPT-4を使用してビデオゲームや医療データのシーンを解析し、その内容を正確に説明する例が示されています。この技術は、セキュリティカメラの映像分析や医療手順の確認など、多岐にわたる業界での応用が期待されており、次世代AIモデルの開発にも貢献する可能性があります。

15:01

🎭 ディープフェイク技術の進化

Alibaba GroupとIntelligent Computing研究所によって開発された「Emote Portrait Alive」は、単一の参照画像から表情豊かなポートレートビデオを生成する技術です。このモデルは、音声クリップと画像を組み合わせて任意の長さのビデオを生成することができ、リアルタイムでの表情の変化やリップシンクにおいて顕著な進歩を遂げています。この技術は、エンターテイメント業界やセキュリティ分野での応用が期待されています。

20:02

🕹 YouTubeのメタゲームとコンテンツ制作

YouTubeでの成功は、効果的なタイトルとサムネイルの使用に大きく依存しています。これは「メタゲーム」として知られ、最も効果的な戦略を採用することでコンテンツの視聴率を最大化します。クリエイターは、視聴者を引き付けるために衝撃的なタイトルや驚きの表情をサムネイルに使用することが多いですが、これは現在のYouTubeのアルゴリズムに最適化された戦略です。内容の質と視聴者からの反応が最終的にコンテンツの成功を決定します。

25:04

💡 クリエイター支援の重要性とYouTube戦略

YouTubeクリエイターは、視聴者に価値を提供するために努力していますが、成功するためには効果的なタイトルとサムネイルが不可欠です。これは、コンテンツが広く見られるようにするための「装備」であり、クリエイターの努力をサポートすることが視聴者にとって重要です。クリエイターを評価する際には、タイトルやサムネイルではなく、コンテンツの質とその提供する価値に焦点を当てるべきです。

Keywords

💡シンギュラリティ

シンギュラリティとは、人工知能（AI）が人間の知能を超える瞬間を指し、それによって技術的な進歩が加速し、予測不能な未来がもたらされる概念です。このビデオでは、AIの世界での衝撃的な出来事を紹介する文脈でこの用語が使われています。これは、AI技術の急速な進化とその社会への影響を示唆しています。

💡フィギュアロボット

フィギュアロボットは、ビデオで言及されている特定の人型ロボットを指します。OpenAIとMicrosoft Azureを活用し、次世代AIモデルの開発によって、商用展開を加速することが目指されています。このロボットは、AIの身体化された応用の一例であり、自律的な動作能力を示すデモンストレーションが強調されています。

💡マルチモーダルモデル

マルチモーダルモデルとは、複数の種類のデータ入力（テキスト、画像、音声など）を統合して処理できるAIモデルのことを指します。ビデオでは、OpenAIがフィギュアロボットとの共同で、ロボティクスの知覚、推論、および相互作用を拡張するためにこの種のモデルを開発していることが言及されています。

💡オートノミー

オートノミー、または自律性は、ロボットやシステムが人間の介入なしに独立してタスクを実行できる能力を指します。ビデオでは、フィギュアロボットが完全に自律的であり、プリスクリプトされた動きや遠隔操作なしに動作することが強調されています。

💡AGI（汎用人工知能）

AGIは、あらゆる知的タスクを人間と同等またはそれ以上のレベルで実行できるAIの形態を指します。ビデオでは、OpenAIがAGIを単一のエンティティとしてではなく、複数のピースが組み合わさることで形成されるというアプローチを取っていることが説明されています。

💡Sora

Soraは、ビデオ内で言及されているテキストからビデオを生成するAIプラットフォームです。このモデルは、他のプラットフォームよりも優れた性能を持ち、現実世界の現象をシミュレートする能力が特に強調されています。Soraは、3D空間をリアルに再現する技術として詳細に説明されています。

💡ガウス分布

ガウス分布は、ビデオ内でSoraが3Dシーンをレンダリングする際に使用される技術的な概念です。3D空間の表現にガウス（正規分布）を応用し、シーンの各部分の位置、色、透明度を定義する方法として説明されています。

💡物理シミュレーション

物理シミュレーションは、実世界の物理法則をコンピュータ上でモデル化し、シミュレートするプロセスを指します。ビデオでは、Soraが実世界の物理現象を模倣する能力、特に3D空間と物理ルールを暗黙的に学習することで、現実に近いシミュレーションを実現することが強調されています。

💡メタゲーム

メタゲームは、特定のゲームや環境内で最も効果的な戦略や行動様式を指す用語です。ビデオでは、YouTubeのタイトルとサムネイルの作成における現在の「メタ」が言及され、視聴者の注意を引くために使われる手法が説明されています。

💡ディープフェイク

ディープフェイクは、人工知能を使用して人々の顔や声を操作し、実際には存在しないビデオや音声クリップを作成する技術を指します。ビデオでは、この技術の進歩が議論され、特にリアルな映像生成能力の向上が強調されています。

Highlights

Figure is making headlines again with massive funding from Nvidia, Jeff Bezos, Microsoft, and OpenAI.

OpenAI and Figure signed a collaboration agreement to develop Next Generation AI models.

Figure leverages Microsoft Azure for AI infrastructure, training, and storage.

Investment from big firms and support from Microsoft Azure will ramp up Figure's timeline for humanoid commercial deployment.

Jeff Bezos invests heavily in robotics, not just through Amazon but also through Bezos Expeditions.

OpenAI is seemingly building AGI in pieces, which will eventually come together to form a greater whole.

OpenAI focuses more on building autonomous AI agents and a web search product, possibly similar to Perplexity.

Sora, an OpenAI video model, generates videos of stunning geometrical consistency.

Google's model Lumiere is anticipated to compete with Sora, but comparisons are yet to be made.

Sora's video generation capabilities suggest it's more of a world simulator, learning physical rules implicitly.

OpenAI has collected pieces of AGI, including technologies for creating more realistic and autonomous AI models.

Sam Altman discusses the need for a $7 trillion investment to build the infrastructure for AI services.

GPT-5 rumors hint at significant advancements in AI, automating many tasks and providing powerful analytics capabilities.

Feather, an OpenAI project, focuses on automated labeling and annotation, potentially revolutionizing data processing.

The integration of OpenAI and Figure highlights a significant step towards embodied AI, with Figure becoming a crucial part of OpenAI's AGI development.

Transcripts

so the other day I'm scrolling through

the singularity subreddit and a headline

jumps out at me they're talking about

something shocking and stunning that

happened today in the world of AI now of

course I click on it and what they

revealed in there shocked me to my core

well come back to that in just a bit but

in other news figure is making the

headlines once again we've already

talked about the massive amounts of

money that it's raising from Nvidia Jeff

Bezos Microsoft openi Etc in addition

openi and figure signed a collaboration

agreement to develop Next Generation AI

models here's open AI open AI plus

humanoid robots we're collaborating with

figure robot to expand our multimodal

models to robotic perception reasoning

and interaction now again we already

knew that there's a lot of money going

into this company from the big players

but here's the new plot twist if you

will so figure will leverage Microsoft

Azure for AI infrastructure training and

storage we are excited to partner with

open Ai and Microsoft to to bring

embodied AI to the real world and

they're saying that this investment from

these big firms as well as the added

support from Microsoft Azure open AI

this will ramp up figures timeline for

humanoid commercial deployment and will

be used for AI training manufacturing

deploying more robots expanding

engineering headcount advancing

commercial deployment efforts here's

February

2024 this is the robot this is what that

looks like so you can see it here up a

box walking around so he's carrying an

object this I got to say is an

incredibly good demonstration of its

abilities the coffee maker demo was it

was a little bit hard to tell how

complicated that was this is definitely

a lot more compelling they're saying

that this is fully autonomous so there's

no teleoperation there's no prescripted

movements now Jeff Bezos has been

pouring a lot of money into robotics not

just from kind of the Amazon fund the

Amazon side but also through other

investment arms of his as well so

bezos's Expeditions is one of them and

so this Amazon industrial Innovation

fund is a 1 billion Venture investment

program that's putting money into

supporting robotics but definitely

figure is the one that a lot of people

are buzzing around there's definitely a

lot of attention on it and the fact that

opene is going to help develop the

models behind it and Microsoft will use

Azure to kind of power a lot of the AI

infrastructure training and storage this

kind of seems that figures now the deao

of open AI robot the GPT robot in a

previous video we talked about how open

AI seemingly is building AGI not as one

thing but rather in pieces these pieces

they will come together to become one

thing there's a similar Concept in these

card games right where you have one one

part representing an arm a leg the Torso

and head Etc together when you get all

of them on the board then the big thing

sort of emerges so each of these things

has its own use its own stats it's its

own thing combined it kind of merges

into this greater thing and that's kind

of my take on how openai seemingly is

building the AGI pieces openai focus

more of its efforts on building

autonomous AI agents Peter Willer open

AI vice president of product remarked on

X that the product new house described

will change everything this thing that

will change everything that really seems

to be this autonomous agentic structure

that they're developing right now call

it this arm of AGI at the same time

they're developing a web search product

probably something similar to perplexity

that may or may not be part of sort of

Chad GPT it might be its own Standalone

thing we of course seen Sora open AI

video model capable of producing some

pretty stunning footage that's uh almost

near lifelike at this point here's a

paper out of China just recently

published Sora generates a video of

stunning geometrical consistency

stunning they say these are Gen 2 and

paika the other two text to video AI

platforms probably the most exciting

ones up until now and here this massive

massive thing is Sora right it's this

massive Pentagon that's maxing out all

the abilities that it's measured on

speaking of pentagons or I guess in this

case a hexagon but why is why is the

North Pole on Saturn a hexagon it's a

persistent hexagonal Cloud pattern

around the North Pole of the planet

Saturn the size of which is longer than

the diameter of Earth probably nothing

right but back to Sora it's

significantly better than everything

else out there now Google recently

announced their model Lumiere so I'd be

curious to see how that compares to Sora

but so far we don't have access to them

so here they're talking about gausin

splatting renderings G in here you can

think of as like little units from which

the image is made so like pixels on a

screen but instead of like 2D pixels for

2D space we're using kind of a 3D

representation so so in computer

Graphics where in video games often

times we use triangles to represent

various scenes to represent various 3D

models so here instead of triangles it's

Gans here's kind of what that looks like

this border is just for clarity but it's

basically made up of the position like

where it is covering so how it's

stretched over some distance color and

how transparent it is so here's multiple

gossans drawn at once I God I hope I'm

pronouncing that word right now what do

you think 7 million gossans would look

like well here that's what it would look

like here's how that would look like if

they're kind of more fully opaque but

what's the point well the point is We

Can Make Scenes like this we can create

3D scenes where everything is kind of

render on screen we're able to float

around zoom in zoom out it's able to

render these 3D spaces kind of from

images you can give it a 2D image and

it's able to kind of try to recreate

that 3D space but p and Gen 2 don't

really produce great results when this

technique is is used to kind of render

that c in 3D there's limited

reconstruction scope which suggests that

P and Gen 2 they kind of have a poor

understanding those models have kind of

a poor model of the 3D space they don't

really get how it works Sora on the

other hand seemingly does here's for

example that uh that Coast as you can

see here that 2D image that Sora renders

is easily made into a 3D gausian sort of

3D space Splats as they're called here's

another example as you can see like the

buildings kind of remain consistent and

I mean you can tell what you're looking

at you can see the different levels of

of elevation that are shown in the image

here's an example of that sort of

western town I think it was so you can

see here as we're kind of floating

through it I mean you can see 3D space

you know where the mountains are where

the buildings are where the little creek

is here's that sort of uh Museum scene

right so the important thing to

understand is that the video videos that

sort of generate yeah I mean they're 2D

images right or pixels in a 2d plane but

when we try to recreate the 3D space

we're able to recreate a lifelike

realistic accurate 3D space and of

course if you've been following up

what's happening of Sora behind the

scenes you kind of this is not a

surprise right CU it's not just a little

video producing machine it's a world

simulator here they match sort of the

geometric consistency between two views

red represents discarded matching

results green represents high quality

matches Sora is really good Gen 2 and P

not so much paa seems to be really good

at keeping its logo in the same place

and this paper points out that this

newly developed Sora model has exhibited

remarkable capabilities in video

generation sparking intense discussions

regarding its ability to simulate real

world phenomena their point is we don't

really have established metrics to

evaluate its Fidelity how well does it

create the real world physics the 3

spaces Etc again the reason I bring this

up is because I think people who think

of Sora as just video generation might

be missing the bigger picture it's more

accurately something like a world

simulator like a physics simulator but

instead of something like an Unreal

Engine where we kind of explicitly tell

the engine we program the engine to

create something some 3D structure or

whatever other physics engine might be

used in video games where we

specifically kind of coded what what

needs to be done sores soft physics

simulation is an emerging property it

learns implicitly the various things it

needs to create those scenes 3D

Transformations R trace renderings and

physical rules so that's yet another

part of sort of openi pieces of AGI

which again some of openi employees kind

of hinted at this but still this is kind

of like conjecture right but if you're

looking at what they're doing it

certainly seems like they're like

collecting all the various pieces and

then Sam alman's quest to build much

more compute much more various gpus and

other the chips that are needed to power

all the AI services that we're going to

need right we've we've heard that7

trillion doll figure quoted uh recently

he came out he kind of explained that

that's that's probably the number that

the entire world all all the countries

will need to invest over a period of how

however many years right to build up the

infrastructure and that whole entire

sort of tech stack that we need to

produce everything that we need so not

just chips but everything everything

everything that we need to sort of fully

deploy all the AI services that we want

to have right so that might be this

piece then the much awaited GPT 5 let's

say that's the head and torso so the

rumors are that it will be able to

automate a lot of work it will be

smarter and better it's not going to be

this revolutionary thing that will

change everything right but it will be

another powerful Leap Forward for those

models and if you've seen if you've

played around with Advanced data

analytics you can imagine what that

might look like for I mean anything you

can do with Excel a lot of programming

tasks a lot of writing tasks analytics

Etc I've mentioned open eyes feather

that they're still stubbornly refusing

to let me in so this is what that looks

like feather open.com so you need a

feather account to continue but what is

it here's the trademark application for

it it's a systematic process systematic

service using automated labeling and

annotation of images audio video text

and various other forms of data so I

think it has something to do with so

this is Microsoft an Interactive agent

Foundation model so Microsoft Stanford

UCLA but at the end here they described

this agent sort of providing synthetic

data labeling of these pictures right so

they they show them a picture and it

says this is what's happening in that

picture the patient is awake and calm so

this is for medical data for example

they also do it for Minecraft what is

happening in this Frame of Minecraft

what do you think the next action will

be here's another game that they're

playing with it's a multiplayer game

right and the model predict what's going

to happen and they're using GPT 4 with

vision to create these descriptions so

here's a video a many frame video from a

game called bleeding edge and we're

asking GPT 4 Vision to give a simple but

precise description of what the player

is doing and GPT for vision outputs this

so it describes what the player is doing

it's running around interacting with

different checkpoints fights against

enemy players and captures the

objectives so think about that for just

a second here's a whole video that is

completely sort of narrated by AI the

descriptions the annotations are

provided by AI the only sort of human

involvement is writing this prompt right

how useful would this be across many

different Industries to review security

camera footage to review what's

happening in hospitals to make sure that

the doctors and nurses are doing the

correct procedures not only that but it

creates really valuable data for

training future AI models for example

with Sora it sounded like what they did

is they used something like this some

something very very similar to annotate

a bunch of various videos that Sora was

trained on so they had a large amounts

of video so they could have a human

being sit there and kind of write here's

what happens in this Frame here's what

happens in this Frame right that would

be slow super expensive might be not

very accurate instead gp4 goes through

and annotates everything then that data

those those pairs right the video and

the description that's given to Sora to

train on and now it's able to do sort of

the the opposite cuz before give it

video with text descriptions and now

we're able to give it just the text

description and it'll spit out a video

it will kind of make that video appear

so all these pieces that open ey is

building and collecting not only is it

going to be automating a lot of things

that we do helping people complete task

faster being more effective at work but

a lot of it also helps to build the next

generation of AI models but opening eye

has been kind of missing that that final

piece the actual physical embodiment of

of AI in the world AKA robots so this is

from July 16th 2021 opening ey disbands

its robotics research team after years

of research into machines these hands

that can solve a Rubik's Cube for

example open AI said it shifted its

focus to other domains where data is

more readily available so this is a

piece that they were going after but

decided to not pursue it by the way

since then since 2021 I think this piece

became more readily available there's

been a lot of progress on this front I

think it's fair to say but now it seems

like open Ai and Microsoft both are

pouring more of their resources more of

their talent all the tech that they have

available into figure robot so open AI

they've entered into a collaboration

agreement to develop the Next Generation

AI models for humanoid robots combining

open AI research with figures deep

understanding of robotic hardware and

software and Microsoft Azure for various

AI infrastructure training storage which

kind of reminds me of this quote by run

azure the endless stretch of digital

ocean where leviathans swim the

birthplace of AGI civilization in other

news our ability to create deep fakes

just got a whole lot better check this

out so this is the reference image and

that's the generated video here's that

sort of image from Sora right where the

ladies walking down the streets of Tokyo

and here's I believe Mira Madi sea level

executive at open AI we collaborate with

and uh maybe we have sever of them and

maybe they all have different

competences and maybe we have a general

one that kind of follows us around

everywhere knows everything about uh you

know my context what I've been up to

today um what my goals are um sort of so

I got to say that's pretty good so they

use a lot of popular music in this so

I'll link this page if you want to check

it out cu the music is excellent I just

can't I can't show it right cuz then

I'll get hit with some copyright issues

but you kind of got to see to believe it

it's very accurate it's really good all

right so there's no way I can play this

but yeah it's realistic video it's

realistic songs very realistic speech

the thing you heard the term don't cry

you don't need to cry just for from from

a simple image which is really

interesting because it captures her

expression like that's a kind of a

unique expression that she has and

translates it to that entire sort of

speech that she's doing the conversation

crying is the most beautiful thing you

can do I encourage people to cry I cry

all the time and basically what to me

this means is you can create images like

this in mid journey of whatever

characters you want and pretty much have

fullblown videos stories that revolve

around them and the paper behind this so

it's the Alibaba group and the institute

for intelligent Computing and it's

called emo emote portrait alive

generating expressive portrait videos

with audio to video diffusion model

under weak conditions and they're saying

using a single reference image this is

the reference image that you put in

there they can create any duration video

depending on the size of the audio clip

so audio clip Plus image is turned into

any length video so we don't have access

to stuff like this yet like we can't use

it yet but it's coming fast and it's

probably very quickly going to be open

source in the paper they actually

mention in the related work kind of uh

they mentioned stable diffusion so

that's an open- Source model so I feel

like once these once people see what

it's capable of I mean they kind of

describe what they did here so they talk

about sort of what kind of data they've

used 250 hours of footage and more than

150 million images across multiple

languages such as Chinese and English I

would not be surprised if we see

something like this you know either as a

closed model with a monthly fee or

perhaps even open source you know maybe

before the year is out pabs has this

kind of lip sync technology right now

but this seems much much much better

much more accurate able to run for

longer periods of time I mean this to me

is mind-blowing when I was a kid I feel

like you heard the thing you heard the

term like that looks like you know

footage shot in whatever year that was

shot in somebody doing an interview with

just really high fidelity sound so AB

absolutely stunning this isn't quite as

good if you notice the lip movements me

watching takes a night but it's not bad

and keep in mind that this is probably

difficult to replicate just because of

the mascara and all that stuff this is

like near perfect reality now you might

be wondering okay but what was that

shocked stunning thing that you saw that

you clicked on that was so crazy well

here it was right shocked and stunned

clickbait titles with my name on it so

my first thought was I am become meme

shocker of Worlds and my second thought

was why aren't I Schwarzenegger in this

image but okay so I just wanted to very

quickly kind of touch on this just to

give my perspective because obviously

this this is becoming a meme more and

more people are sort of talking about it

the reason I was trying to kind of avoid

talking about it is because I think that

that might actually make it worse but I

kind of just wanted to tell people

what's going on behind the scenes but

okay I'll try to explain what's

happening so a lot of you might have

heard of Tim Ferris he was kind of one

of the original Tech influencers I guess

his 4-Hour workweek book was kind of

groundbreaking at the time but in 1999

he won the gold medal at the Chinese

kickboxing National Championship while

he had a martial arts background he

probably wasn't the best person at that

Championship but what he did was is he

brought a technique that was fairly well

known in the US in the wrestling and

taekwan doe community and the trick was

that the weigh-ins were the day prior to

the competition so he used these

dehydration techniques that I mean my

understanding is it's very common in

most martial arts competitions that you

see on TV for example and so what he did

was he he lost 28 lb in the 18 hours

prior to weigh-in so he clocked in at

165 lb at the weigh-in and then showed

up on the day of ready to fight at 193

lbs so of course it's pretty hard to

fight somebody from three weight classes

above you and then he proceeded to win

by technical knockout and went home as

the national champion now what do you

think of that would you say that he

cheated I mean technically everything

was within the rules do you think what

he did was unethical or do you what he

did was you know disruptive and genius

and wonderful and does your opinion

change if I tell you that this spurred

most of the fighters in the future

championships of this nature to start

using those same techniques it became

the common practice now different people

will have different opinions about it

but there's actually a name for this in

gaming it's called The Meta game so

basically in essence a meta in gaming

terminology is the generally agreed upon

strategy by the community the strategy

is considered to be the most oble way to

win and has the best performance at a

specific task some people even said that

the acronym that it's that it's an

acronym for most effective tactics

available you've probably seen it if you

played games different seasons have

different metas League of Legends DOTA

and stuff like that different

combinations of characters can be great

for particular times for example this is

James alcher so he used to be I think he

was like a master Grandmaster in chess

as a kid then stopped playing for a long

time came back as an adult and so he's

saying that it's really different now I

don't know if he used the word meta but

I think the reason why it's different

now is the meta game has changed a lot

of it due to ai ai Unearthed new

strategies new ways of playing and the

new generation of chess players learn

from that and the game changes with

YouTube there's sort of two parts of how

content gets shared distributed

recommended Etc and the first is the

title and thumbnail so kind of the

packaging if you will the cover of the

content right and the second part is the

actual video kind of the long form

content and if you nail this and this

then your video does well but if you

mess up the title or your video or if

you have a great title and your video

kind of sucks then it it doesn't do too

well and I got to give credit to YouTube

for this I think their algorithm while

you know maybe not perfect is pretty

good at serving up interesting stuff to

watch especially if you're interacting

with it quite a bit right if you're

clicking on videos and you're watching

them over time it figures out okay I

know what you like here's some of the

stuff that other people like that are

similar to you so I got to say they do a

pretty good job and if you if you're on

YouTube watching I'm sure you can point

out some issues with it but feel like

overall they they kind of nailed it but

here's a thing that I think people might

not understand who are not in the sort

of YouTube ecosystem the people that

make the videos we have a lot of control

over the content of the video I can say

or do pretty much whatever I want in the

video and as long as you like it you get

some value out of it then it continues

to get shared and and shown across the

network like if I sit here for an hour

and draw like a little like a little

bear or something right and if as long

as you're enjoying that Google's like

yeah dude whatever that's fine as long

as people are watching liking it

commenting subscribing you can do

whatever you want within limits

obviously but so here we have a lot of

control over what we do what we say Etc

but on this side the title and thumbnail

not quite as much we don't have as much

control here there's a very specific

metag game why do you think all those

YouTubers put those stupid YouTube faces

on their thumbnails do you think it's

cuz they love making these faces can't

they just smile and make a normal face

instead of this sort of face why is this

popping up in my feed why do Mr Beast

one of the top YouTubers why does he

have the same glazed over eyes and weird

smile in every single video is because

those little weird faces and whatnot

they'll pull a little bit more clicks

and that'll kind of tip the balance in

favor of that video it'll do a little

bit better and what happens if you don't

play the game well then Tim Ferris comes

along and just kind of manhandles you

out of the Arena those are the rules of

YouTube and they are unflinchingly rigid

now I wasn't the first to do those shock

titles I saw other people doing it and

it was extremely effective and the

reason that a lot of people are doing it

you know the goal isn't to deceive you

or annoy you or anything like that the

goal is to just make sure the video gets

seen and these shocking titles is just a

kind of this fun way to do that to make

sure that it's that it gets out there to

the people that find that annoy just

understand that's the current meta and

it too shall pass but then there's going

to be another meta and it's probably

going to annoy you still right maybe the

next one will be us making those stupid

surprised faces going I can't believe

this this just happened or whatever but

the point that I would like to make is

this for all the creators that you like

whether that's me or somebody else the

people that are informing entertaining

you you know I just ask this judge them

on this the long form content they

produce that's what they have control

over that's what they make make for you

you might notice I don't have too many

sponsorships I don't think I had a

single sponsorship on this channel right

I'm not pitching you a paid course I'm

not pitching you a paid Discord Channel

my stuff is just 100% content would you

like it better if these were you know

toned down but I spent half the video

pitching you some product probably not

so my take on this and this is not just

for AI but anything else you know for

the people that you like what they do

the content engage with that content

judg it based on that long form content

this is what we make for you this the

title and thumbnail this is more like

our lone little Gladiator right it's

this Gladiator with a sword and shield

and we have to give it all the resources

and equipment that it needs to go out

there and fight all the other YouTube

videos that have their thumbnails and

titles and if we make a boring thumbnail

and a boring title that's like sending

this guy out there with no armor or

weapons it has no no chance it would

break my heart to do that so I armed

them with the sword of shock and the

shield of stun and then send them out

there to battle all the other YouTube

videos and this makes sure that the

video that I made gets seen it would

break my heart to send them out there

without equipment so please do not ask

me to so to all the people that were

super cool about this and just having a

lot of fun with this I really appreciate

you thank you so much to the people that

maybe didn't like this so much but we

still kind of cool about it I hope this

kind of explains why we're doing it it's

not to be deceptive it's just how the

game is played which is giving our

little guy the best chance to win and I

think just the best sort of thing that

you can do is just judge us by the

content right don't judge the book by

its cover right that old cliche like

just don't judge the video by the

thumbnail those are two separate things

and to those of you that say that me

using the word shocking in the title

makes me the wor YouTuber you've ever

heard of I would just like to point out

one thing but you have heard of me my

name is Wes Roth and thank you for

watching