Billionaire BLOWN Away By Tesla FSD v12

Summary

このスクリプトは、Teslaの完全自動運転（FSD）技術の議論を中心としています。スクリプトでは、Teslaが従来のコーディング方式を捨て、ニューラルネットワークモデルにシフトし、最優のドライバーのビデオをアップロードして学習させているという革新的なアプローチについて語っています。この方法は、人間の運転を模倣する学習方式であり、これまでのバージョンとは大きく異なるとされています。また、Teslaが抱える膨大な車両データの活用と、そのデータの収集方法、そして他社がTeslaのFSD技術を追いつく困難さについても触れています。さらに、Teslaが将来的に人型ロボット分野にも進出し、さらにデータの価値を高める可能性があるとの予想も描かれています。

Takeaways

• 🚗 TeslaのFSD（完全自動運転）は、従来のC++の決定論的モデルから、エミュレーション学習に基づくエンドツーエンドモデルへと大きな転換を遂げた。
• 📈 新しいFSDモデルは、ビデオ入力から制御出力までをカバーし、より速く正確に運転を行えるとされています。
• 🤖 Teslaは、最高のドライバーからビデオをアップロードし、ニューラルネットワークモデルを使って学習させている。
• 🔥 Teslaは、以前のFSDのコードベースを捨てて、ビデオから学習する新しいアプローチに完全に移行した。
• 🚀 FSD V12は、過去のバージョンよりも大幅に向上し、改善のペースも目覚ましい速さで進んでいる。
• 🌐 Teslaは、600万台以上の車両を走行させており、そのデータから学習することで、他社と比べて圧倒的なリードを有している。
• 📱 Teslaのデータ収集は、エッジで行われており、意味のあるデータをフィルタリングしてモデルを向上させている。
• 💰 Teslaは、FSDの料金を半分に引き下げ、より多くの消費者による採用を促そうとしている。
• 📉 他の自動運転開発会社は、Teslaほどの車両数やデータの蓄積ができないため、競争において大きく後れを取っている。
• 🤖 Teslaは、人型ロボットの開発にも進んでおり、FSDの技術がそれにも応用可能である。
• 🌟 Teslaのビジョンは、未来を先取りし、他社が見据えられない領域での可能性を追求している。

Q & A

テスラのFSD V12がどのようなモデルに変わったのか説明してください。

テスラのFSD V12は、従来のC++で記述された決定論的モデルから、エミュレーション学習によって駆動されるエンドツーエンドモデルに変わりました。これは動画の入力と、制御の出力を通じて、より迅速かつ正確に動作します。

FSDの過去のバージョンに比べて、FSD V12はどのように進化していると述べていますか？

FSD V12は、過去のバージョンと比べて、能力が統合され、わずか的训练时间内に過去のバージョンの能力を凌駕しているとされています。また、改善のペースが以前よりも大幅に増加していると述べられています。

テスラが自驾動技術の学習に使用しているデータとは何ですか？

テスラは、最も優秀なドライバーからのビデオをアップロードして学習に使用しています。ビデオはモデルへの入力で、出力はステアリングホイール、ブレーキ、アクセルペダルとなります。

テスラの自驾動技術が他の企業と比べてどのような利点がありますか？

テスラは、500万台以上の車両を路上に持ち、それらから収集された膨大なデータを学習に使用しています。このデータの量と質は、他の企業が持つデータとは比べものにありません。

テスラのFSD V12の価格が変動した理由は何ですか？

テスラは、FSD V12の改善が顕著であると判断し、ペネトレーションを促進するために価格を半分に下げたとされています。これにより、より多くの消費者が製品を試し、データ収集が増えることで、モデルの改善が促進されると期待されています。

テスラはどのようにして自驾動技術の難しさと課題に対処していますか？

テスラは、コーナーケース（稀な状況）を解決するために、異常動作をとる瞬間に重みをかけたデータをモデルに入力することで対処しています。これにより、稀な状況での運転をより正確に学習することが可能です。

テスラの自驾動技術が抱える可能性の1つに、データの量が多すぎるとされていますが、これはどう対処する予定ですか？

テスラは、エッジでのデータの圧縮とフィルタリングを行って、必要なデータのみを抽出しています。これにより、必要なデータを見つけるプロセスが効率化され、データの量を効果的に活用することが可能です。

テスラのFSD V12が抱える可能性のもう1つの課題として、データの質とは何ですか？

データの質とは、稀なイベントやコーナーケースを含めた、意味のあるデータをモデルが学習できるかどうかです。テスラは、異常動作をとる瞬間に重視したデータを用いて、モデルの改善に貢献しています。

テスラは今後の人型ロボット開発において、FSDの開発からどのようなノウハウを活かす予定ですか？

テスラは、FSDの開発を通じて得られた視覚認識、計画、行動の技術を、人型ロボットの開発に活かす可能性があります。また、FSDのデータ収集と学習プロセスに関するノウハウも、人型ロボットの学習プロセスに応用される可能性が高いです。

テスラのFSD V12が成功するためには、今後どのような要素が必要ですか？

FSD V12が成功するためには、より多くの消費者が製品を利用し、データを提供することが重要です。また、データの質と量の両方を確保し、モデルを継続的に改善していく必要があります。さらに、ハードウェアの進歩も必要とされます。

テスラのFSD V12が他の企業と比べてどのような競争優位性を持ちますか？

テスラは、500万台以上の車両を所有し、それらから収集された大量のデータを学習に用いています。また、FSDの改善と学習プロセスを通じて、自驾動技術における競争優位性を築いています。さらに、消費者への販売モデルも、他の企業とは異なります。

Outlines

00:00

🚗 TeslaのFSD V12の進化とその意義

パラグラフ1では、TeslaのFSD（完全自動運転）V12の進化について語られています。2人の男性が、Savvy投資を通じて10億ドル以上の利益を得た後、TeslaのFSD V12について意見を共有する価値があるかどうかが議論されています。彼らは、FSDの11バージョン目までの開発を振り返り、新しいモデルがどのように人間の運転者のように振る舞うかについて話しました。また、Teslaが従来のコーディング方法を捨て、ニューラルネットワークモデルにシフトし、その成功への期待が高まっていることを示唆しています。

05:00

📈 FSD V12の改善速度とその意味

パラグラフ2では、FSD V12が過去のバージョンよりも急速に改善していることが強調されています。彼らは、FSD V12が短時間で過去のバージョンの能力を超え、さらに改善が加速していると述べています。このパラグラフでは、Teslaが過去のFSDモデルを捨て、ビデオから学習する新しいアプローチを取ることで、よりエレガントで効率的な方法を模索していることが明らかになります。

10:02

🤖 AIの進化とTeslaのFSD2のオープンソースモデル

パラグラフ3では、AIの進化とTeslaのFSD2が使用しているオープンソースAIモデルについて説明されています。彼らは、Teslaがどのようにしてニューラルネットワークを用いて自動運転を改善し、ハードウェアとデータのインフラを整える必要があると述べています。また、Teslaが最良のドライバーからビデオをアップロードし、そのデータを用いてモデルをトレーニングしているプロセスについても触れています。

15:03

🚔 自動車メーカーの遅れとTeslaの先見性

パラグラフ4では、従来の自動車メーカーがTeslaに遅れをとっていることと、Teslaが将来を見据えて技術を開発していることが語されています。彼らは、Teslaがどのようにして自動運転技術を他の分野（たとえば人型ロボット）に応用できると予想しているかについて議論し、Teslaのデータ収集の重要性と、そのデータが他の企業にはなかなか手に入れられないという点を強調しています。

20:06

📉 競合他社との比較とTeslaのリード

パラグラフ5では、Teslaの自動運転技術が他の企業とどのように比較されるかが議論されています。彼らは、Teslaが抱えるデータの量と質の優位性と、それをいかに駆使して自動運転技術を進化させているかについて話しています。また、他の企業がTeslaと同じレベルのデータを収集することができないという課題や、Teslaが人型ロボット分野でもリードを維持する可能性についても触れています。

25:06

💰 FSDの経済効果と価格戦略

パラグラフ6では、TeslaのFSDの経済的な側面と、価格戦略が議論されています。彼らは、FSDの料金を下げることの意義と、それがTeslaの収益性に与える影響について語ります。また、FSDの普及を促進することで得られるデータの重要性と、それが自動運転技術の改善に与える役割についても説明しています。

30:08

🚘 Teslaのビジネスモデルと競合他社との比較

パラグラフ7では、Teslaのビジネスモデルが他の自動運転企業（CruiseやWaymo）とはどのように異なるかが議論されています。彼らは、Teslaが消費者に車両を販売し、データを収集しながらソフトウェアを改善するというアプローチと、他の企業が行っているアプローチを比較しています。また、Teslaが抱えるリードをどのように維持し、競合他社が追いつくことが困難である理由についても述べています。

35:09

🎙️ AG1の効果とフィードバック

パラグラフ8では、話者がAG1というサプリメントの効果について語られており、視聴者からのフィードバックが紹介されています。話者は、AG1を摄入することでエネルギーが増し、生活の質が向上したと述べています。また、複数の証言が紹介されており、その効果について個人的な経験が語られています。

Keywords

💡テスラFSD V12

テスラの自律走行システム「Full Self-Driving」のバージョン12。このシステムは、従来のプログラミング方式からニューラルネットワークに変え、安全性と正確性を向上させた。ビデオでは、このバージョンが人間に近い運転を提供していると述べている。

💡模倣学習

ニューラルネットワークを用いて、データ（ここではドライバーのビデオ）から学習し、その行動を模倣する学習手法。テスラのFSD V12は、模倣学習を用いて最善のドライバーのビデオを学習し、その行動を模倣することで運転を行う。

💡深層学習

大量のデータを用いてニューラルネットワークを訓練させる手法。テスラは、深層学習を用いて自社の車が収集するデータを分析し、自律走行システムの精度を向上させている。

💡データインフラ

データの収集、保存、分析をするためのシステム。テスラは、自社の車から収集されたデータを効率的に処理し、自律走行システムの改善に利用している。

💡エッジコンピューティング

データの生成地点で処理を行うコンピューティング方式。テスラはエッジコンピューティングを用いて、車のデータの大部分をローカルで処理し、クラウドに送信するデータ量を減らす。

💡AIの進化

人工知能が時間とともに進化し、より複雑で具体的な問題を解決できるようになるプロセス。テスラのFSD V12は、AIの進化を体現しており、従来のプログラミング方式よりも優れた結果を出す。

💡オープンソース

他の人々が自由に使用し、改変できるソフトウェアやソースコード。テスラは、オープンソースのAIモデルをカスタマイズして、自律走行システムを構築している。

💡ハードコーディング

ソフトウェア開発で、具体的な値や条件をプログラムに直接記述する方法。テスラは、FSD V12でハードコーディングを排除し、代わりにデータから学習するアプローチを採用した。

💡シャドウデータ

実際の動作データとして使用されるが、分析や学習の目的で使用されなかったデータ。テスラはシャドウデータを用いて、自律走行システムの改善に貢献している。

💡テスラのデータ資産

テスラが保有する、自社の車から収集された大量の運転データ。このデータは、自律走行技術の開発で非常に価値があり、他の企業では得られない情報源となっている。

💡自律走行技術のリード

自律走行技術における先進的な立場。テスラは、FSD V12の開発で自律走行技術のリードを確立しており、他の企業が追いつくのは困難とされている。

Highlights

Tesla has made significant progress in their Full Self-Driving (FSD) technology, moving from a deterministic model to an end-to-end model driven by imitation learning.

The new FSD model is video input and control output, which is faster and more accurate than previous versions.

There is skepticism about the true capabilities of FSD due to 11 different versions and mixed public feedback.

Tesla's approach to self-driving has completely shifted, abandoning the old model that was a complex codebase to a neural network model.

The new model uses neural networks and learns from the best drivers by uploading videos, which are used as input for the model.

The shift to the new model is considered a massive watershed moment for Tesla's approach to autonomy.

The rate of improvement for FSD V12 is significantly greater than previous versions, indicating a faster path to full autonomy.

Tesla's decision to use video from human drivers for training is similar to how humans learn to drive.

The new FSD model does not have a deterministic view of objects like stoplights; it learns from driver behavior instead.

Tesla's approach is more elegant and efficient, and it has the potential to solve the long-tail events that are crucial for full autonomy.

The infrastructure for data collection and processing is a significant advantage for Tesla, with millions of vehicles on the road collecting data.

The data collected by Tesla's fleet is not just vast in quantity but also rich in quality, capturing edge cases and long-tail events.

Tesla's ability to process data on the edge and only upload relevant information makes their data collection strategy highly efficient.

The value of the data that Tesla has collected and continues to collect from its fleet is immense and not fully reflected on their balance sheet.

Tesla's potential to license their FSD technology or data stream to other companies presents a new business opportunity.

The unit economics of Tesla's vehicles improve significantly with increased FSD adoption, creating a business case for driving more FSD subscriptions.

Tesla's strategy of selling vehicles to consumers and using them to train the software while collecting data is a more sustainable model compared to competitors.

The competitive landscape for autonomy shows that Tesla has a substantial lead, and other companies struggle to match their data collection capabilities.

Transcripts

R me this if two guys who between them

via Savvy investment had made profits in

excess of1 billion were to share some

thoughts about Tesla FSD V12 would it be

worth listening we teased uh on the Pod

I think at the start last time that I

had taken a uh you know a test ride um

in in Tesla's new fsd2 and I said you

know it kind of felt like a little bit

of a chat GPT moment um but I think we

left the audience hanging we got a lot

of feedback hey tell you know dig in

more to that so you and I spent some

time on this both together uh and with

the uh with some folks on the Tesla team

so roughly the setup here background I

want to get your reaction to it is about

12 months ago the team pretty

dramatically forked their self-driving

model right moving it from this really

C++

deterministic um model to what they

refer to as an end to endend model um

that's really driven by imitation

learning right so we think of this new

model it's really video in and control

out it's faster it's more accurate you

know but after 11 different versions of

FSD I think there's a lot of skepticism

in the world like is this going to be

you know something different I you sent

me a video and I are tons of these

videos uh you know floating around at

the moment you know that really kind of

shows you you you know how this acts

more like a human than prior models out

there so Bill kind of just react um you

know you've watched this video react to

this video and give us your thoughts you

know I think you've been a long time

Observer of self-driving um I might even

describe you as a bit of a Critic of you

know or skeptic when it comes to full

self-driving so is this a big moment did

I overstate it kind of what are your

thought what are your thoughts here yeah

so you know one of the critiques and and

concerns people had about self-driving

is they would say

that yeah we're 98% of the way they are

99 but the last 1% is going to take as

long as the first 99 and one of the

reasons for that is um it's nearly

impossible to code for all of the corner

cases um and the corner cases are where

you have problems that's where you end

up in wrecks right and so um the

approach Tesla had been taken up until

this point in time was one where you

would um literally try and code every s

every object every every circumstance

every case in in in like a piece of

software this x happens then y right and

um that ends up being a patchwork kind

of a just a big nasty you know uh Rat's

Nest of code and it builds up and builds

up and builds up and and maybe even

steps on itself and and it's not very

elegant what we learned this week is

that they've completely tossed all of

that out um and gone with a neural

network model where they're uploading

videos from their best drivers and

literally the videos are the input and

the output is the steering wheel the

brake and the gas pedal extraord and you

know there's a there's there's this um

principle known as A's razor um which

has been around forever uh in science

but the the this the simplified version

of it is a a simpler approach is much

more likely to be the optimal approach

right and when I fully understood what

they had done here um it seems to me

this approach has a much better chance

of of going all the way and of being

successful so taking a quick pause here

I strongly agree as I've said in the

past with Tesla essentially having

burned to the ground their prior FSD

build which was riddled with code what

did Bill Gurley call it Rat's Nest or

something along those lines Fair call

with everything hardcoded completely

burning that to the ground and

rebuilding with videos of human drivers

in and incredible capabilities out no

multiple layers of [ __ ] in between

and now seems quite clear that the path

to full autonomy is clearly lit prior to

then there were still a lot of unknown

unknowns and until Tesa had made enough

progress and hit a wall they wouldn't

know there was a wall ahead but now we

can see all the way to the end it's a

massive watershed moment many of the

Skeptics and haters scored epic own

goals when Tesa made this complete

rewrite entirely new approach the haters

are Skeptics they laugh go don't know

what they're doing look they changed

that's so they're so dumb how far behind

can you be failing to understand that

the only reason do have made this

decision the complete rewrite to train

this software end to end on video only

was because they're so far ahead into

Uncharted Territory they have now

realized the optimal approach is no

heris no hard coding just train it on

good video the end it shows how far

ahead Tesla are versus everyone else

attempting to solve autonomy he was

still using very brittle approaches

now this clip was actually recorded more

than a month ago and in that time we

have seen so many updates on FSD V12

incremental improvements the rate of

change and Improvement on this software

is staggering the pace of improvement

has massively increased so if we put two

and two together here FSD V12 has

matched or exceeded the capabilities of

all past versions combined and did so

having been trained in a fraction of the

time and now the rate of improvement is

significantly greater than at previous

was of course no one knows for sure when

the problem is finally solved how long

does that take but if the rate of

improvement is accelerating and it's

already exceeded the capabilities of

Prior versions it's a very positive sign

to put it lightly and and certainly of

being maintainable and reasonable U it's

way more elegant um it requires them to

upload a hell of a lot of video which we

can talk about um but and and the other

thing that's just so damn impressive

is that this company which is very large

hundreds of thousands of employees um

made a decision so radical um to kind of

throw out the whole thing and start aesh

and it sounds like they the the the

Genesis of that may have been you know

three or four years ago but but they got

to the point where they're like this is

going to this is going to be way better

and threw the whole thing out and I

think um about four months after they

made the change um Elon did a drive

where he up loaded and and kind of

stream the drive so we can put that in

the notes and people can watch it um but

it's way way different it's way way

different and in my mind you know

basically with the sakom razor notion um

it's got a much higher chance of of

being wildly successful yeah let's dig

in a again I agree and just for the

record Tesla's approach now is

essentially the same way that humans

learn to drive photons into our dual

camera based system called eyes

inference fire our neural network our

brain and controls out as opposed to the

previous strategy which would have been

equivalent to hard coding a bunch of

instructions on how to drive a vehicle

what road signs are and literally

everything the [ __ ] else to do with

driving a vehicle into our brains

somehow at Birth a little bit into how

it's different right so and you you you

referenced a little of this so you know

like for example this model does not

have a deterministic view of a stoplight

right I mean karthy has talked about

this before you know before you you have

to label a stoplight right so you would

basically take the data from the car

that would be your perception data you

would draw a box around a stop light you

would say this is a you know this is a

stop light so that your first job on the

car would have to be to identify that

you're at a stoplight then the second

thing is you would write all of this C++

that would deterministically say when

you are at a stoplight here's what the

controls should do right um and so for

all of that second half of the model um

you know the heuristics the planning and

the execution that was all driven by

this Patchwork that you're talking about

and that was like you would just chase

you know every one of these uh Corner

cases and you could never solve them all

now in this new model um it's pixels in

so the model itself has no code it

doesn't know this is a stoplight per se

in fact they just watched the driver's

Behavior so the driver's behavior is

actually the label it says when we see

pixels like this on the screen here's

how the model should behave which I

thought is just an extraordinary break

and I don't think there's a deep

appreciation for the fact that you know

again because we've had 11 versions of

what came before it those were just

slightly better Patchwork models um in

fact I think what what you know we

learned was that the rate of improvement

of this is order of magnitude five to

10x better per month as a model versus

the rate of improvement of those prior

systems and facts extremely important

facts as well again I mentioned this

prior the rate of improvement here has

significantly cranked up now In fairness

eventually Tesla I believe would have

solved autonomy with the patchwork model

it's just inelegant messy a total

cluster [ __ ] inefficient but hey with

enough time and enough data I believe

they would have done it however this end

to end approach pixels in controls out

as Brad described is just so elegant and

efficient and as we're seeing the rate

of progress has ratcheted up so much now

but a lot of people still aren't quite

appreciating this yet once again the

audacity to throw out the whole old

thing and put a new thing in is is just

crazy one thing for the listeners well

actually two things I would I would

mention one um in terms of just how they

got this going you know a lot of people

I I fear equate AI with lolms um because

it was really the arrival of chat GPT

and the llm that I think introduced what

AI was capable of to most people but

that those are language models that's

what the L one of the L's stands for and

these AI models that Tesla's used for

for fsd2 are these generic open- Source

AI models that you can find on hugging

face you know and and and they obviously

customized them so so there's there's

some proprietary code there at Tesla but

but you know ai's been evolving for a

very long time and this notion of neural

networks was around before the llms

popped out which is why you know they

had started on this four years ago or

whatever right right um but the the

foundational elements you know are there

and by the way they use they use the

hardware that we're talking about right

they use the big Nvidia clusters to do

the training um they need some type of

GPU or TPU to do the inference at at at

runtime um so it's the same Hardware the

llms use but it's not the same type of

code I just thought that was worth

mentioning and yeah no it's a it's a to

me if we dig in a little bit to um you

know the model itself you know the

Transformers the diffusion architecture

the convolution neuron Nets those are

all like these modular open source

building blocks right like the thing

that's extraordinary to me and we're

going to get later in the Pod to this

open versus closed debate but like this

is just this great example you know you

talk about ideas having sex I mean these

these open-source module uh you know

kind of uh modular components those have

been worked on for the last decade and

now they're bringing those components

together and now all of their energy and

I want to dig into this a little bit

that is is really going they're taking

all these Engineers who were writing the

C++ these deterministic you know patches

effectively and now they're focusing

them on how do we make sure that our

data infrastructure that the data that

we're pulling off of The Edge comes in

and makes these models better so all of

the sudden it becomes about the data

because the model itself is just

digesting this data brute forcing it

with a lot of this uh you know Nvidia

hardware and outputting better models

you know it's such a classic um Silicon

Valley startup thing where you need all

the pieces to line up if you go back and

watch if if you haven't watched if

anyone's watching the general magic

video um which is fantastic it's on the

internet um about why General magic

didn't work and Tony Fidel who ended up

building the iPod and ran engineering

for the iPhone talks about how the

pieces just weren't there so they were

having to do all the pieces right the

the network and the chips and it just

wasn't there yet and so these models

have been around maybe ahead of the

hardware and now Nvidia is bringing the

hardware and these pieces start to come

together together and then the data like

and and I think one of the most

fascinating things about this story of

Tesla and

fsd2 is when you understand where they

get the data so they are tracking their

best drivers with five cameras and the

drivers know it they've opted into the

program and they upload the video uh

overnight and so you know talk about the

pieces coming together um we've found

Reddit forms and stuff we can put we can

put links to in the in the notes where

um users are are Tesla drivers are

saying they're uploading 10 gigabit a

night and so you know you had to have

the Wi-Fi infrastructure the like like

like how would it be possible to upload

that much here here's here's someone

who's whose Tesla uploaded 115 gigabyte

in a month right so just taking a quick

break here fascinating conversation I

know it's a bit nerdy and Technical but

the point about needing all of the

pieces to a line is so critical the

reason we didn't have iPhones in the

1970s or ' 80s is because it wasn't

possible in theory it could have been

attempted but the cost to produce

something like that would probably have

been a million plus dollars one of the

things Tesla does so well and I really

want to underscore this does so well is

look into the future understand where

things are heading EG with electric

vehicles pattery cost will decline as

they scale up we eventually reach price

parody with ice vehicles and then

continue to become even more affordable

over time the electronics in the

vehicles will continue to become more

affordable these kind of things oh wait

okay well today they won't be that

affordable but if we produce them Drive

profits into the Next Generation scale

of production they will be cost

effective in the future we can reach

high volume and so on autonomy is

another example and humanoid robots are

another example Tesla Has This brilliant

capacity which comes straight from the

top to look into the future 5 10 15 20

plus years think what is possible then

and start putting in place the Stepping

Stones to reach reach that end

destination when everyone else thinks

they're insane what destination that's

not going to be POS that isn't possible

you must be high an incredible example

of not getting this not seeing the

future is Automotive manufacturers today

who temporarily thought oh [ __ ] people

are buying Teslas maybe we should make

electric vehicles too these same morons

who by the way were about 15 years late

to the party had to see consumers buying

Tesla vehicles hand over fist before

they thought about it who are now

suddenly slowing their EV plans and

quadrupling down on hybrids because

these morons are not looking into the

future not seeing battery cost

continuing to decline they're not seeing

the pieces of the puzzle and by the time

they realize what's possible it's going

to be too late so what we're seeing

happening now with autonomy couldn't

really have happened sooner there was an

available compute I mean imagine the

idea let's say in the 1990s you're

trying to pursue autonomy and you have

to figure out some way to have vehicles

on roads covered in census that are

uploading in a single month over 100

gigabytes of

data Madness by the way special shout

out to the absolute high quality human

beings on Reddit especially the mods

what a great great online resource

without any bias and zero ban happy

basement dwellers and so these are

massive numbers and the infrastructure

five years ago your car couldn't have

done this and you know we I think we'll

talk about competition in a minute but

like you know who else has the capacity

to do this right it's unbelievable to

like the footprint of cars they have and

then the notion that oh yeah we could

just go upload this data and it is a

buttload of data it sure [ __ ] is a

buttload of data and as Bill asked who

else has the capacity to do this the

answer of course is no one no other

company has 6 Plus million vehicles on

roads collecting and uploading data no

one the answer is literally no one

that's not hyperbole the answer to that

question who else could do this is no

one unless of course they just press the

magic catch up to Tesla button which

apparently a lot of folks seem to think

exists actually so you just do the 5

million cars 30 m a day I think eight

cameras on the car 5 megapixels each and

then the data going back 10 years right

this amount of Shadow data you could

comine the clusters of every hyperscaler

in the world and you couldn't possibly

store all of this data right that's the

size of the challenge so what they've

had to do is process this this data on

the edge and in fact I think 99% of the

data that a car collects never makes it

back to Tesla so you know they're using

video compression these remote send

filters they're running you know neural

Nets and software on the car itself so

basically they you know for example if

80% of your driving is the highway and

it's there's nothing interesting that

happens on the highway then you can just

throw out all that data this is super

important what matters here is sorting

out filtering for the data you actually

need which is a fraction of everything

you collect but the challenge here is

you must collect everything in order to

get the tiny specs of data you actually

need imagine for in for gold in order to

find the specs of gold the stuff you

actually want you need to fos through a

fuckload of material that you don't need

and if you're not collecting vast

quantities of data you will never get

the stuff you actually need the stuff

that actually matters the edge cases so

what they're really looking for is you

know what is the data that is a long way

away from the mean data right so what

are these outlier moments and then can

we find 10 tens or hundreds or thousands

of those moments to train the mod

so they're literally pulling this

compressed filter data every single

night off of these cars they built an

autonomous system so before they would

have Engineers look at that data and say

okay what have we perceived here now how

do we write you know this Patchwork code

instead this is simply going into the

the model itself it's fine-tuning the

model and they're constantly running

this autonomous process of fine-tuning

these models and then they're

re-uploading those models back to the

car okay this is is why you get these

exponential moments of improvement right

that that that we're seeing now which

then brings us back to bill this

question you know Tesla has 5 million

cars on the road they have all this

infrastructure they have they they are

collecting this data we know they're a

couple years ahead think about wh for

example they're still using the old

architecture it's geofence I don't know

they have 30 or 40 cars on a road and

they're only running the so do they have

any chance does weo have any chance of

competing or even adting this

architecture H let me think for a moment

here uh [ __ ] no it'd be it'd be it's

such an interesting question and and by

the way just on one quick comment on the

previous thing you said it's genius

actually that they are um they've taught

the car what moments it should record ex

and so they they they mentioned to us an

example of um anytime there's you know

well obviously a disengagement so a

disengagement becomes a moment where

they want the video before and the video

after the other thing would be any

abrupt movement so if the if the gas

goes fast or if the brake has hit

quickly or if the steering wheel jerks

that becomes a recordable moment and the

part I didn't know um which they told us

which is just Fascinating People with

lolms have heard you know about

reinforcement learning from Human

feedback rhf and they've talked about

how that could make it even with Gemini

they said maybe that was what caused

that um what what we were told is that

those moments these moments where like

the car jerks or whatever if it is super

relevant they can put that in the model

with extra weight and so it tells the

model this is this if if this

circumstance arises this is something

that's more important and you have to

pay extra attention to and so if you

think about this corner case um these

Corner case scenarios which we all know

are the biggest problems in self-driving

um now they have a way to only capture

the things that are most likely to be

those things and to to learn on them so

so the amount the amount of data they

needed to get started was this

impossible amount of data with the

millions of cars and now the way that

plays to their advantage is they're much

more likely to capture these these these

Le these more severe less frequent

moments because of the bigger footprint

and so you say to yourself you know you

ask the question who I don't know who

could compete it it's that's a general

way of saying the so-call competition is

[ __ ] and they may as well just give up

of course bill would not say that

directly but that is the actual truth no

one can compete no one has the fleet no

one has the existing data and certainly

to the earlier analogy no one now has

Auto fosk for edge cases ain't no one

catching up certainly couldn't if if

let's let's make an assertion if this

type of neural network approach is the

right answer yes and I once a reason

once again you know aam's Razer seems

that way to me then who could compete

and one of the companies or to you know

several of the companies who would be

least likely would be Cruz and whmo and

these things because they just don't

have that many cars Bill Gurley out here

is spitting facts please don't tell that

to the autonomy leaderboard published by

some hack from Legacy automotive

industry who obviously knows a lot about

things other than autonomy now this is

really important I've said it before

before I said it again Cruis Ando what

they're doing might look impressive but

they glorified party tricks in a little

sandbox with training wheels and guard

rails they do not have scalable

solutions they compensate for a lack of

data and a lack of intelligence by

putting guard rails training wheels

safety nets and massively massively

overcompensating wi complexity in order

to create a general autonomous solution

that is generalized scalable widely you

need the data and if you don't have the

Fleet you don't have the data now there

is there is a bit of a curveball

possibility I don't think it's that

likely but it is possible if companies

are too dumb to just license FSD from

Tesla directly which is the obvious

thing to do when they realize that

they're not going to solve it themselves

it is technically possible the Tesla

could license their data stream to other

companies to train on but why add the

unnecessary steps now here's an

interesting thought what is the true

value of the data that Tesla has already

collected cuz I can guar [ __ ] key

that is not listed as a current asset on

Tesla's balance sheet though it should

be more to the point what is the value

of Tesla's data coming in every single

day from their Fleet of vehicles the

answer is not zero that's for sure and

remember at the end of the day what

tesra is really solving is Vision seeing

perceiving planning and acting in the

real world a huge amount of what Tes is

doing with FSD translates directly or

mostly into a humanoid Rob robot maybe

it would be a good idea for Tesla to

start developing a humanoid robot oh

wait they are hm so here's a thought

imagine a few years from now and Bill

and Brad are discussing Tesla's Fleet of

X number of robots which is multiple

orders of magnitude more than the next

largest Fleet the equivalent today would

be Cru and weo versus Tesla Fleet of

five to six well technically 6 Plus

million vehicles but at least five of

them are capturing and returning data

could the same conversation be playing

out again regarding humanoid robots bit

of food for thought Cru and wh and these

things cuz they just don't have that

many cars and their cars cost

$150,000 so if they wanted to have like

the math just doesn't work you can't

build the footprint um you know and so

who could I don't know could you I don't

know could you put a could you what

would it cost to build a five camera

device to put on top of every Uber I

don't know like a lot it would be weird

but they're not going to they're not

going to do it I mean like and that to

me is um you know when you look at these

alter alternative models right if this

really is about data and remember bill

just said an important point which is

it's not just about quantity of data

Something Magic happens around a million

cars yes you've got to get all that

quantity of data but to get the longtail

events right these are events that occur

tens or just hundreds of times that's

where you really need millions of cars

otherwise you don't have a statistically

relevant pool of these longtail

instances and what they're uploading

uploading from the edge Bill He said

Each instance is a few seconds long of

video and you know uh plus some

additional vehicle driving metadata and

it's those events if you only have

hundreds of cars or thousands of cars

you can get a lot of data quickly it's

not about Quantum of data 100 cars can

produce a huge Quantum of data driving

a, miles it's about it's about the

quality of the data those Adverse Events

yes and and and I guess the other type

of company that maybe could take a swing

at it it would be like mobile eye or

something the problem they have is they

they they don't control the whole design

of the car and so this part where Tesla

has the car in the garage at night and

uploads gigabytes and puts it right into

the model like are they going to be able

to get that done working with other oems

like are they going to be able to

organize all that you know do they have

the piece on the car that says when to

record and when not to record and and

and like it's just a massive

infrastructure question I would probably

if I had to handicap anybody it would

probably be BD or one of the Chinese

manufacturers right and if you think

about they're they have a lot of M miles

driven in China right much less so

outside of China um I imagine you're

going to have some of this nationalistic

stuff

that you know that emerges on on both

ends of this but like one of the things

I asked our analyst bill is like if we

just step back I think the these guys

have network advantage they have data

Advantage they're clearly in the lead

they have bigger h100 clusters than the

people they're competing against I mean

they have all sorts of things that have

come together here but if you think

about like what's the so what to Tesla

right and just in the first instance and

we'll pull up this slide that Freda on

our team made if you look at the unit

economics of a Tesla right with no FSD

they're making about 25,000 bucks uh on

a vehicle if you look at it today about

7% penetration of FSD that was let's

call it through fsd1 and those people

paid

$122,000 incrementally for that FSD and

as we know you can go read about it on

Twitter people are like yeah it's good

but it's not as good as I thought it

would be so now we have this big moment

of a step what feels like you know kind

of a step function the model uh getting

better at a much faster rate so I asked

the question what if we reduce the price

on this by half right what if what if

Tesla said this is such a good product

we think we want to drive penetration so

let's make it 500 bucks a month not not

a th000 bucks a month so if you assume

that you have uh you know penetration

you know go from 7% to 20% give it to

everybody for free they drive around for

a month they're like wow this really

does feel like a human driver I'm happy

to pay 500 bucks a month you know if you

get to you know 20% penetration then

your contribution margin at Tesla right

is about the same even though you're

charging half as much now if you get to

50% penetration all of the sudden you're

creating billions of dollars in

incremental eitaa now think about this

from a Tesla perspective why do they

want to drive even more adoption of FST

well you one reason would be more data

get a lot more information and data

about disengagements and all these other

things so that data then you know uh

continues to turn the flywheel so my

guess is that Tesla seing this

meaningful Improvement is going to focus

on penetration my guess is that they

they want to get a lot more people

trying the product and they're going to

play around with price why not right

maybe a hundred bucks a month is the

right you know uh intersection so as I

mentioned earlier this video was

recorded more than a month ago guess

what since this was recorded Tesla has

literally cut the price of FSD in half

so credit where it's due here the dude

[ __ ] called it Legend between

adoption or penetration and price um but

again I think that all of these things

are occurring at an accelerating rate at

Tesla and when I look around you know I

still hear people saying wh's worth 50

or 60 billion bucks but you could be in

a situation on that business where it

just is you know gets passed really

quickly and they have a hard time

structurally of catching up well we you

know people have said that uh and and if

someone has data once again that they

want to correct this I I'd be glad to

state to recorrect the data but but you

know we've been told they have a

headcount similar to Cruz and the Cru

financials came out and they were

horrific and so I don't I don't have any

reason to believe that the weo

financials are any different than the

cruise ones and right um I've always

thought this model that we're going to

build this incredible car and and our

business model is going to be to run a

service like the capex like if you just

build a 10-year model the capex you need

like they would have to go raise 100

billion and and there's another element

that's super important point and a great

opportunity to once again discuss the

420 IQ decision by Tesla their model so

absurd it's just we have to say it out

loud method one Cruise wayo using some

round numbers spend $150,000 on a

vehicle as in lose

$150,000 per vehicle on roads then pay

an engineer call it another $100,000 a

year to drive the thing to train it

additionally pay people to remotely

supervise your vehicles in other words

lose a couple $100,000 per vehicle on

roads Tesla's approach sell a vehicle to

Consumers and make a few th000 per

vehicle sold and then for free consumers

voluntarily while using the vehicle

train the software to get better as the

vehicle itself collects [ __ ] tons of

invaluable data these two business

models are not alike so is it just me or

does anyone else get the sense that

these two guys possibly believe that

Tesla May in fact have what some could

describe as an unassailable lead when it

comes to autonomy and that companies

like Cruz and wh are completely and

utterly [ __ ] am I the only one getting

that impression want more content Early

Access bunch of perks click the links in

the pin comment ag1 giv me a massive

meaningful boost in energy allowing me

to do a lot more every day including

using my brain more and using my body

more highly recommend you guys and girls

check it out is an excellent way to fill

in nutritional gaps it's got 75 high

quality vitamins minerals and Whole Food

Source nutrients plus prebiotics and

probiotics and digestive enzymes and

adaptogens to help you deal with stress

plus if you click the link at the pin

comment or head to drink a1.com smmr you

can get yourself a 1-year free supply of

vitamin D3 and K2 but don't take my word

for it here's what some of you guys and

girls have to say ag1 has changed my

life I was as you described treating

myself like a circus ate like trash

rarely exercised use alcohol as a stress

crutch Canabis also ag1 is what gave me

the kick in the ass got me back to the

gym motivated me to do more for myself

family my business Etc keep doing what

you do now I know there's some Skeptics

the same kind of people who think El

musk is a fraud reading this G what do

you thought there's no way that's

possible bro it must be a placebo effect

believe it or not this is a recurring

theme if you give your body everything

it needs to feel and performance best

including having a lot more energy

you'll need ways to use that energy for

me personally that includes more

exercise moving my body more more social

activity and more cognitively demanding

tasks including producing a [ __ ] T of

exclusive content over on Twitter and on

patreon Plus my daily YouTube uploads

the proofs in the pudding onto another

testimonial from a viewer of this

channel SMR you asked me to provide

feedback on ag1 here it is it has helped

with mental acuity stamina and

intestinal Waste Management uh can't bre

between the lines certainly helps with

regularity and digestion that's what the

digestive enzymes are for it has also

dramatically reduced my cravings for

sugar you guys need to stop eating sugar

it's [ __ ] poison I'm 50 59 and

overweight AKA a fat [ __ ] I

think that's a technical term for

overweight isn't it is it fat

[ __ ] or obese I can't remember I

average 100 hours a week in the west

Texas oil fields as a safety supervisor

Jesus Christ dude no wonder you're

struggling to keep your weight under

control 100 hours a week brutal it has

helped me lose weight it is not an

appetite suppressant it can help fat

people suppress cravings and motivation

to be healthier is critical for changing

your diet love you brother again this is

a great point and something people

really don't seem to grasp if you have

more energy everything becomes easier

it's like turning on easy mode for Life

few years ago before I was taken ag1 my

health was trash I was struggling to get

through the day had afternoon fatigue

the last thing I wanted to do was either

use my brain or move my body didn't have

the energy now my biggest struggle every

day is figuring out ways to use that

energy I'm exercising way more doing a

lot more with my friends and family and

of course my work output has increased

substantially and you can fact check me

check out the average length of my

videos I was posting on YouTube 3 years

ago need I say more and one final

testimonial love this one okay here's

the deal for me with this ag1 [ __ ] I'm

41 years old and not the type to eat

drink smoke or or sleep healthy so I was

skeptical that being said here's what I

experienced day one me day two afternoon

fatigue was about 45 minutes late day

three zero afternoon fatigue day four

zero afternoon fatigue plus extra energy

day five again zero afternoon fatigue

plus energy wondering what the f really

see this is the thing right the results

for many people are just almost too good

to be true this this is same experience

I had my afternoon fatigue just vanished

out of nowhere I'm like what the [ __ ]

why am I not tied in the afternoons

anymore surely it's not that ag1 is it

turns out it was day six and 7even same

thing day eight same thing plus I had

the want to get things done around the

house that I normally would slack off

and not get done again the point extra

energy you'll need to use it you'll find

ways to use it day 9 10 and 11 and today

is day 12 I [ __ ] love it so however

you managed to get me to buy it I'm so

glad you did thank you so much SMR it

really changed me so far guys this [ __ ]

really works just try it by the way this

is the reason I continue to relentlessly

promote ag1 a lot of people get realing

mad in the comments oh my God snake all

Sal sold out oh my God he's a scammer

this is fraud but constantly I'm pretty

sure everyone making these comments is

also currently short Tesla stock I'm not

particularly concerned about people

having a negative perception those folks

suffering from small brain syndrome

still living in my bums basement

syndrome Etc writing mean comments

claiming a1's a scam or it doesn't work

I mean bro when I get feedback like this

this is what keeps me going just try

this stuff for a month and if you don't

get these results get your money back

see it's a literal no-brainer it's an IQ

test at this point point in time

testimonial after testimonial after

testimonial like this get your money

back if it doesn't work just try it for

a month and if it doesn't work get your

money back today's the day it's finally

time be like this guy who was a massive

skeptic but finally after a thousand

promotions in a row caved in tried ag1

and has results like this head to drink

a1.com smmr or click the link at the pin

comment and please let me know how

you're feeling in a few weeks time now

if you'll excuse me time to put my extra

energy to good use I'll be recording

some more exclusive content for patreon

and my Twitter subscribers so click the

links up comment see you over on Twitter

and on patreon and don't forget to grab

your ag1 love you