$600 Billion AI Wave... software 3.0

Summary

Dieses Video skizziert die zukünftige Rolle von künstlicher Intelligenz (KI) in Unternehmen und für Schöpfer. Es wird vorgeschlagen, dass KI-Vertreter wie E-Mail-Adressen und Websites zur Selbstverständlichkeit werden. Der Fokus liegt auf 'Software 3.0', die auf großen, open-source KI-Modellen basiert und durch spezifische Anpassung für individuelle Anwendungsfälle optimiert wird. Der Upload zeigt, wie KI in Bereiche wie Factorio, einem Strategiespiel, integriert wird, um Prozesse zu automatisieren. Die Diskussion umreißt die enorme Chance, die KI bietet, um Geschäftsprozesse zu revolutionieren und individuelle KI-Modelle für spezifische Bedürfnisse zu entwickeln.

Takeaways

• 💰 Es gibt verschiedene Wege, Millionen zu verdienen, aber einige könnten potenziell zu Milliarden wachsen.
• 🌐 Es gibt unzählige kleine Unternehmen weltweit, die in Zukunft möglicherweise über einen AI-Agenten verfügen, mit dem Kunden interagieren können.
• 🤖 Software 3.0 wird als die nächste Evolution der Software angesehen, die auf großen, bereits trainierten KI-Modellen basiert und diese für spezifische Anwendungsfälle anpasst.
• 📈 Software 1.0 beinhaltet traditionelle Programmierung, wobei Software 2.0 auf maschinellem Lernen und neuronalen Netzwerken basiert.
• 🧠 Neuronale Netzwerke sind ähnlich den Neuronen im menschlichen Gehirn, wobei Verbindungen zwischen Neuronen stärker werden, wenn sie häufiger genutzt werden.
• 🔍 Software 3.0 konzentriert sich auf die Manipulation von Basismodellen, die bereits viele Fähigkeiten haben, ohne von Grund auf neu trainiert werden zu müssen.
• 🚀 Open-Source-Modelle wie LLM (Large Language Models) bieten die Möglichkeit, spezifische KI-Agenten für Unternehmen und Schöpfer zu erstellen, die ihre eigenen KI-Modelle trainieren können.
• 💡 Die Verwendung von KI in Unternehmen und im täglichen Leben wird sich in den nächsten Jahren stark erhöhen und wird letztendlich nahezu universell sein.
• 🌟 Es besteht ein großes Potenzial für Unternehmer und Investoren, die in die Entwicklung und Anpassung von KI-Modellen investieren, um spezifische Geschäftsbedürfnisse zu erfüllen.
• 🔧 Die Zukunft der Geschäftsführung könnte ähnlich wie das Management eines automatisierten Systems aussehen, wobei KI eine zentrale Rolle dabei spielt.

Q & A

Was ist der Hauptgedanke hinter 'Software 3.0'?

Software 3.0 bezieht sich auf die nächste Generation von Software, bei der Foundation Models verwendet werden, die bereits viele Fähigkeiten haben und keineswegs von Grund auf trainiert werden müssen, sondern durch spezifische Anleitungen und Verstärkung für den jeweiligen Geschäftsbedarf angepasst werden.

Was versteht man unter 'Foundation Models'?

Foundation Models sind große, allgemeine KI-Modelle, die eine Vielzahl von Fähigkeiten 'out-of-the-box' bieten und für spezifische Geschäftsanwendungen nur noch feinjustiert werden müssen.

Wie unterscheidet sich Software 3.0 von Software 2.0?

Software 2.0 basiert auf maschinellem Lernen, wobei Datensätze verwendet werden, um das Modell zu lehren. Software 3.0 hingegen konzentriert sich darauf, bereits trainierte Foundation Models für spezifische Aufgaben zu verfeinern und anzupassen.

Welche Rolle spielen offene Quellen in der Entwicklung von Software 3.0?

Offene Quellen sind entscheidend, da sie es ermöglichen, große, leistungsstarke KI-Modelle zu erstellen, die dann für spezifische Anwendungsfälle von Unternehmen und Individuen angepasst werden können.

Was ist der Zusammenhang zwischen Software 3.0 und der zukünftigen Geschäftswelt?

In der zukünftigen Geschäftswelt wird es wahrscheinlich sein, dass jedes Unternehmen einen eigenen AI-Agenten hat, mit dem Kunden interagieren können, ähnlich wie heute jeder ein E-Mail-Adresse, eine Website und eine Social-Media-Präsenz hat.

Wie wird die Zukunft der AI-Verteilung in Unternehmen aussehen?

Die AI-Verteilung in Unternehmen wird sich von einem kleinen Prozentsatz, der derzeit seine eigenen AI-Lösungen nutzt, in den kommenden Jahren dramatisch erhöhen und möglicherweise schließlich nahezu 100% der Unternehmen erreichen.

Was ist der Unterschied zwischen traditioneller Softwareentwicklung (Software 1.0) und modernen AI-Modellen?

Software 1.0 beinhaltet traditionelle Programmierung, bei der Entwickler explizite Anweisungen für den Computer schreiben. Moderne AI-Modelle hingegen sind in der Lage, aus großen Datenmengen zu lernen und Aufgaben ohne explizite Programmieraufträge auszuführen.

Wie wichtig sind Daten und Compute-Power für die Entwicklung von AI-Modellen?

Daten und Compute-Power sind entscheidend für die Entwicklung von AI-Modellen, da sie die Grundlage für das Training und die Leistungsfähigkeit dieser Modelle bilden.

Was ist die Vision hinter der Idee, jedem Schöpfer und kleinen Unternehmen die Möglichkeit zu geben, ihre eigenen AI-Agenten zu erstellen?

Die Vision ist, dass AI in der Lage ist, die Interaktion mit der Community zu erleichtern, die Verwaltung von Geschäftsprozessen zu optimieren und die Kreativität und Effizienz jedes Einzelnen zu steigern.

Wie sieht die potenzielle Zukunft der AI-Anwendungen in Alltagsleben und -geschäften aus?

Die potenzielle Zukunft der AI-Anwendungen könnte so weit reichen, dass AI in nahezu jedem Aspekt unseres Alltags und in geschäftlichen Prozessen eingesetzt wird, von Hausautomation bis hin zur Unterstützung bei Entscheidungsfindungsprozessen.

Outlines

00:00

🤖 Die Zukunft der KI-Agenten in kleinen Unternehmen

Der erste Absatz diskutiert die Idee, dass jede kleine Geschäftsunternehmen in der Zukunft einen eigenen KI-Agenten haben wird, mit dem Kunden interagieren können. Der Sprecher vergleicht dies mit den heutigen Standardkomponenten wie E-Mail-Adressen, Websites und Social-Media-Präsenzen. Er betont die Bedeutung von Software 3.0 als nächstes Evolutionsstadium der künstlichen Intelligenz, welches die Schaffung von Programmen ohne explizite Anweisungen ermöglicht, sondern durch das Festlegen von Zielen und das Verlassen der KI, die Lösung zu entwickeln.

05:00

🔧 Evolution von Software 1.0 zu Software 3.0

In diesem Absatz wird die Entwicklung von Software 1.0 über Software 2.0 hin zu Software 3.0 erläutert. Software 1.0 bezieht sich auf traditionelle, von Menschen geschriebene Programme, während Software 2.0 auf neuronale Netzwerke basiert, die ohne direkten Code von Menschen trainiert werden. Software 3.0 wird als die nächste Stufe dargestellt, bei der KI-Modelle wie das von Mark Zuckerberg vorgestellte 'Llama' verwendet werden, um aus diesen großen Modellen kleinere, spezialisierte Modelle für individuelle Anwendungsfälle abzuleiten.

10:01

🚀 Die Vision von individuellen KI-Agenten für Business und Content-Creator

Der dritte Absatz konzentriert sich auf die Vision von Mark Zuckerberg, wie KI in Zukunft in Unternehmen und für Content-Creator eingesetzt werden kann. Es geht darum, wie Unternehmen und Kreative mithilfe von KI-Agents ihre Geschäftsprozesse optimieren und wie diese Agents für Kundenservice, Verkauf oder die Interaktion mit der Community genutzt werden können. Es wird auch auf die Bedeutung von Open-Source-Modellen und die Schaffung von Diversität in der KI-Entwicklung hingewiesen.

15:03

🛠️ Anwendung von Software 3.0 in der Praxis

Dieser Absatz vertieft die Idee von Software 3.0, indem er die Verwendung von Foundation-Modellen beschreibt, die speziell auf die Bedürfnisse von Unternehmen oder Individuen abgestimmt werden. Es wird auf die Wichtigkeit von Anpassung und Feinabstimmung eingegangen, um diese Modelle für spezifische Geschäftsanforderungen zu nutzen. Zudem wird die Rolle von Open-Source-Modellen hervorgehoben, die als Grundlage für die Entwicklung kleinerer, anwendungsspezifischer Modelle dienen.

20:05

🌐 AI als allgegenwärtige Technologie in zukünftigen Geschäftstätigkeiten

Der fünfte Absatz spricht von der zukünftigen Allgegenwärtigkeit von AI in Geschäftstätigkeiten. Es wird eine Parallele zu dem Grad, zu dem heutzutage Computer in Unternehmen eingesetzt werden, gezogen und die Prognose geäußert, dass die Nutzung von AI in Unternehmen in den kommenden Jahren stark ansteigen wird. Der Sprecher reflektiert über die potenziellen Anwendungen von AI in verschiedenen Lebensbereichen und wie sie möglicherweise das Geschäftsmodell der Zukunft prägen wird.

🎮 Automatisierung von Factorio durch KI-Modelle

Der letzte Absatz präsentiert ein Projekt, das die Automatisierung des Strategiespiels Factorio mithilfe von KI-Modellen veranschaulicht. Der Sprecher zeigt die Fortschritte in der Entwicklung von Schnittstellen und Modellen, die es Bots ermöglichen, das Spiel ohne menschliche Intervention zu spielen. Es wird auf die Open-Source-Natur der entwickelten Tools und die Absicht hingewiesen, diese für die Community zugänglich zu machen, um die Verbreitung von KI in der Gaming-Industrie zu fördern.

Keywords

💡Software 3.0

Software 3.0 bezieht sich auf die nächste Generation von Software, die auf vortrainierten KI-Modellen basiert. Diese Modelle, die als 'Grundlagenmodelle' bezeichnet werden, haben viele Fähigkeiten von sich aus und müssen nicht von Grund auf trainiert werden. Stattdessen werden sie durch spezifische Anweisungen und Daten für den jeweiligen Anwendungsfall angepasst. Im Video wird dies als die nächste Entwicklungsstufe nach Software 1.0 (manuelles Programmieren) und Software 2.0 (maschinelles Lernen) dargestellt.

💡Künstliche Intelligenz (KI)

Künstliche Intelligenz (KI) ist das Feld der Informatik, das sich mit der Schaffung von Systemen befasst, die ähnliche Fähigkeiten wie der menschliche Verstand zeigen können. Im Kontext des Videos wird KI als Schlüsseltechnologie für die Zukunft dargestellt, die es Unternehmen ermöglicht, individuelle AI-Agenten für Kundenservice, Verkauf und Kommunikation zu erstellen.

💡Neuronale Netzwerke

Neuronale Netzwerke sind ein Typ von KI-Modellen, die nach dem menschlichen Gehirn modelliert sind. Sie bestehen aus verbundenen Neuronen, ähnlich wie die Verbindungen zwischen Nervenzellen im Gehirn. Im Video wird erläutert, dass diese Netzwerke durch das Verstärken der Verbindungen, die oft genutzt und korrekte Vorhersagen liefern, trainiert werden.

💡Gradientenabstieg

Der Gradientenabstieg ist ein Algorithmus, der in der KI-Forschung zur Optimierung von Modellen verwendet wird. Er ermöglicht es, die Gewichte in einem neuronalen Netzwerk so anzupassen, dass das Modell seine Vorhersagegenauigkeit verbessert. Im Video wird dieser Prozess als Teil des Trainings von Software 2.0 erwähnt.

💡Open Source

Open Source bezeichnet Software, deren Quellcode öffentlich zugänglich ist und von der Community weiterentwickelt werden kann. Im Video wird betont, wie wichtig Open Source für die Verbreitung von KI-Modellen ist, da es es ermöglicht, auf bereits bestehende, hochentwickelte Modelle aufzubauen und anzupassen.

💡AI-Agenten

AI-Agenten sind künstliche Intelligenzen, die als digitale Assistenten fungieren und in der Lage sind, mit Menschen zu interagieren. Im Video wird die Vision geäußert, dass zukünftig jedes Unternehmen einen eigenen AI-Agenten haben wird, um Kunden zu bedienen, ähnlich wie heute jeder eine E-Mail-Adresse und eine Website hat.

💡Schlüsselindustrie

Die Schlüsselindustrie bezieht sich auf Branchen, in denen KI eine entscheidende Rolle spielen wird. Im Video werden Bereiche wie Finanzen, Gesundheitswesen und Spiele genannt, in denen KI-Anwendungen enormes Potenzial haben.

💡Datensatz

Ein Datensatz ist eine Sammlung von Daten, die verwendet wird, um KI-Modelle zu trainieren. Im Video wird erwähnt, dass große Datensätze für die Entwicklung von KI-Modellen von großer Bedeutung sind, insbesondere für die Erstellung von Software 2.0.

💡Faktorio

Faktorio ist ein Strategiespiel, das im Video als Beispiel für die Anwendung von KI genutzt wird. Ein Entwickler hat ein Projekt gestartet, um das Spiel mit Hilfe von KI zu automatisieren, was zeigt, wie KI in Spiele und andere Anwendungen integriert werden kann.

💡Einfluss von KI auf Geschäftsmodelle

Im Video wird diskutiert, wie KI die Geschäftsmodelle von Unternehmen verändern wird. Durch die Möglichkeit, individuelle AI-Agenten zu erstellen, können Unternehmen ihre Kundenservice- und Verkaufsprozesse verbessern und neue Wege finden, um mit ihren Kunden zu interagieren.

Highlights

未来每个企业都可能拥有自己的AI代理，与客户进行交流，就像现在拥有电子邮件地址、网站和社交媒体一样。

软件3.0的概念，即利用基础模型进行微调和特定领域的应用，可能成为AI领域的最大机会。

Andrej Karpathy提出软件2.0的概念，即通过数据集和神经网络权重而非传统编程语言来定义软件行为。

软件1.0和2.0的比较，展示了从人类编写代码到通过训练神经网络来实现特定行为的转变。

介绍了神经网络如何通过类似巴甫洛夫条件反射的方式进行学习和适应。

解释了软件3.0如何使用大型基础模型来创建特定用途的小型定制模型。

Mark Zuckerberg讨论了开源AI模型Llama 3.1，以及它如何允许社区创建定制的AI代理。

讨论了如何通过微调大型模型来创建适用于特定业务需求的小型AI模型。

Sarah Guo分享了她的AI创业投资理念，强调了软件3.0的重要性和对特定领域AI应用的关注。

预测AI将像计算机和互联网一样普及，最终几乎所有企业和个人都将使用AI来优化和自动化任务。

展示了如何通过开源模型训练小型AI模型来服务于特定用途，如客户服务或个人助手。

讨论了AI在教育、游戏、金融和医疗等领域的潜在应用和影响。

强调了AI的普及将如何改变我们管理业务和个人生活的方式，类似于工厂自动化。

介绍了一个使用大型语言模型自动化游戏Factorio的项目，展示了AI在特定领域的应用潜力。

讨论了AI如何帮助用户在游戏中自动化任务，提高效率和游戏体验。

展示了如何通过微调和训练来改进AI模型，以适应特定的游戏策略和行为。

讨论了开源AI模型和工具如何使更多人能够创建和使用定制的AI解决方案。

Transcripts

there are like ways to make a million

bucks and then ways to make a million

bucks that could turn into a billion

bucks right there are hundreds of

millions of small businesses in the

world a business can basically you know

few Taps um stand up an AI agent for

themselves I kind of think that every

business in the future just like they

have an email address and a website and

a social media presence today I think

every business is going to have an AI

agent that their customers can talk to

in the future and we want to enable that

if you always knew that there was a huge

opportunity with AI but didn't quite get

exactly how to take advantage of it I

think this video might clear it up the

thing that I'm talking about is software

3.0 it might sound a bit weird but give

me just a second to explain because I

think that if you're an entrepreneur or

you always wanted to get into business

for yourself or you just want to stay on

The Cutting Edge of AI well this might

be the single biggest opportunity of our

lives so really fast this is Andre kpy

former open AI highly respected AI

researcher noticed the date this was end

of 2017 he's talking about software 2.0

what is that well the classical stack is

the software 1.0 that's kind of what we

think of a software it's code it's

written in languages like python C++ Etc

and it consists of explicit instructions

to the computer written by a programmer

right so so basically a human that's a

human if you can't tell and he's telling

a computer what to do sort of explicitly

right letter by letter typing it in and

and giving it instructions by writing

each line of code the programmer

identifies a specific point in program

space with some desirable Behavior right

so in other words we want something from

the computer we want this thing to do

something for us we make that happen by

writing each line of code he

specifically uses that terminology there

right writing each line of code that's

important so then what is software 2.0

can you guess well Andre karpathy he's

saying sof 2.0 is written in much more

abstract human unfriendly language such

as the weights of a neural network no

human is involved in writing this code

because there are a lot of Weights

typical networks might have millions

keep in mind this was written in 2017 so

you might imagine that there might be a

little bit more than that nowadays and

coding directly in weights is kind of

hard he tried he would this is kind of

what a neural network looks like the

neurons are connected by way there are

these uh lines here and this is somewhat

similar to a human brain we also have

neurons and then the connections between

those neurons that wire together

stronger when they are used more often

when they are predictive of something

happening that was that whole Pav's dog

the response to the Bell right so over

time as we smell something right so the

part of our brain that smells that is

able to capture smells like we smell

food and we respond to that by for

example with a dog salvation right

because we formed those neural

connections that helped us predict that

certain smells mean that there's tasty

food about however if we ring a bell the

dog has no response there are no neural

connections to indicate that a bell

means that there's food but if we start

giving them food and ring a bell over

time the bell ringing produces you know

him salivating and thinking that he's

about to get food that whole episode on

the office where Jim feeds Dwight breath

mitts every time a bell rings if you

didn't catch that that was literally

what that whole thing was about and

that's kind of how neural networks work

all these connections create some sort

of outcomes but we don't program them

ourselves well unless you're Andre karpy

actually how we approach it is we

specify some goal or desirable Behavior

right like win a game of Go and yes

there's some you know code a rough

skeleton of of the code AS audre put

said right so sort of the neural net

architecture and then that neural

network is trained using gradient

descent back propagation we're not going

to really get into it because we don't

quite need to to understand what's

Happening Here by the way if you are

interested in learning more highly

suggest Andre Kathy's latest thing so he

just announced it within the last few

weeks looks like this is dated July 17th

open co-founder Andre karpathy announces

Eureka Labs an AI education startup keep

this in mind you might need it later

just a hunch but kind of the big

takeaway about software 1.0 and software

2.0 I think is this this is a again from

Andre Kathy's blog right s sort of like

this distance represents the complexity

of the program right something very

simple to something very complex and

software 1.0 is here that's what

software can do again if we're defining

software 1.0 as a human being coding

something up for a computer to do that's

sort of the extent of it I mean it's

quite a bit the entire world that runs

on it it's done quite a bit but it's

limited compared to software 2.0 and

with software 2.0 instead of typing in

line by line we have a sort of process

right so that's training the AI right I

just kind of make it into a black box

and that process trains the AI it trains

the neural net to do the thing that we

want it to do we just say what we want

the alcome to be we tell it make cute

pictures of dogs and it trains the

neural network the brain that can make

cute pictures of dogs we tell it learn

to codes stuff and it figures out how to

make the thing that learns to code stuff

now this is greatly simplified obviously

there's a lot of things that go into

this data compute a million other things

but the point is instead of explicitly

telling what to do we're just kind of

telling it what we want and it then

creates the thing that does the thing

that we want right so we're kind of like

a layer removed all right so pop qu Hot

Shot what does software 3.0 look like

right in terms of what we're talking

about what is the next Evolution if this

is explicit directions this is telling

it kind of what we want and it figures

out how to create the thing that does it

what does the next evolution of that

look like what is 3.0 I'll add this last

piece and just that a year ago it wasn't

even obvious that this thing existed it

wasn't obvious that it would be

available to us keep in mind Andre

didn't even mention software 3.0

anywhere in this blog post that was in

2017 it it wasn't even obvious at that

point how this whole thing would shake

out but so at this point you might be

thinking okay okay so what's software

3.0 why don't you just tell us so first

of all here's Mark Zuckerberg talking

about his latest release llama 3.1 the

open source AI giant that is at GPT 4

level now if you saw that interview feel

free to skip forward I'll have the video

Chapter set up so you can skip to the

next part but whether you've seen it or

not pay specific attention to what he's

saying that you're allowed to build what

you're able to build with the big bad

model that he released the model that he

refers to as the teacher model so I'm

really excited to see what people do

with that especially now that we're

making it so that our the community

policies around llama allow people to

use it as a teacher model to distill and

fine-tune and um and basically create

whatever other models they want with it

and here's another Quick Clip where he's

talking about what he believes will be

the final result of people using these

open source models to create agents Etc

is he talking about everything running

on top of the sort of Base llama model

just sort of one model to rule them all

or is it more fragmented than that I

think we just have the ability in the

business model to basically build in the

most advanced models in the world and

offer it to everyone for free so I think

that that's a you know kind of a huge

Advantage um it's really easy to use

from all of our apps um so I'm pretty

excited about how that going so that's

yeah we have the basic assistant um and

and I think that that's going to be a

big deal but even more than that a lot

of what we're focused on is giving every

Creator and every small business um the

ability to create AI agents for

themselves um making it so that every

person on our platforms can create their

own AI agents that they want to interact

with and if you think about it these are

just huge spaces right so there are

hundreds of millions of small businesses

in the world and one of the things I

think is really important is basically

making it so with a relatively small

amount of work um a business can

basically you know few Taps um stand up

an AI agent for themselves that uh can

do customer support sales communicate

with all their people all their

customers I kind of think that every

business in the future just like they

have an email address and a website and

a social media presence today I think

every business is going to have a um an

AI agent that their customers can talk

to in the future and we want to enable

that for all those that's that's going

to be hundreds of millions maybe

billions of of kind of small business

agents similar deal for creators um

there are more than it's more than 200

million people on our platforms who

consider themselves creators who

basically use our platform um in a way

that is primarily for you know building

a community um you know put putting out

content feel like it's it's kind of like

a part of their job is is doing that and

they all have this basic issue which is

that there aren't enough hours in the

day to engage with their Community as

much as they'd like and likewise I think

that their communities would generally

want more of their time but um but again

not enough hours in the day so I just

think it's a there's going to be a huge

unlock where basically every Creator can

pull in all their information from

social media can train these systems um

to reflect their values and their

business objectives and what they're

trying to do and then people can can

interact with that it'll be almost like

this almost artistic artifact that

creators create that that people can can

can kind of interact with in different

ways and then and that's not even

getting into all the different ways that

I think people are going to be able to

create you know different AI agents for

themselves to do different things so I

think we're going to live in a world

where there are going to be hundreds of

millions of billions of different AI

agents eventually probably more AI

agents than there are people in the

world and um and that people are just

going to interact with them in all these

different ways so that's part of you

know that's the product Vision um

obviously there's a lot of business

opportunity in that that's where we want

to go make money so we don't want to

we're not going to make money from

selling access to the model itself um

because again we're not a public Cloud

company we will make money by building

the best products an important

ingredient to the best products is

building is having the best models which

having the best kind of ecosystem around

open source will help us do so that's

why it's kind of all aligned for us and

why this is going to end up I think

being really valuable for us to build

the the highest quality products that we

can um and and have the best business

results by by kind of building out this

open source Community but but it's also

why it's all philosophically aligned

right we don't we just don't believe

that there's going to be kind of one big

AI whether it's a product or a model

that everyone uses we kind of

fundamentally believe in having this

broad diversity and different set of

models and that you know every business

and um you people are just going to want

a lot of their own stuff that they're

going to make and I think that's kind of

going to be interesting it's going to be

a lot of what makes this interesting so

this is Sarah gou she was recently on

the my first million podcast where she

discusses some of the startups and and

AI feels that she's very interested in

investing and she's got a very specific

sort of investment thesis AK what she

thinks will really move the needle in

terms of AI startups and this is where

she kind of spells out what software 3.0

is take a listen and I think the next

level of like value and impact is

definitely going to be um fine-tuning to

specific voice and what's your

overarching investment thesis so you

have this thing called software 3.0

which what is software 3.0 yeah yeah

okay so the seed for that phrase

software 3.0 it comes from actually an

essay that Andre karpathy wrote years

ago about software 2.0 and the base

premise here is that like you had to

write uh a lot of software by hand in a

prior generation before machine learning

and then software 2.0 Andre you know

worked at Tesla was working on autopilot

was really about data set labeling right

you know you you are

teaching a machine learning model by the

data you choose to put into the pipeline

um how to do new tasks software 3.0 is

the idea that the next generation of

software a lot of it is about

manipulating Foundation models and

they're called Foundation models because

they have a lot of capability out of the

box you don't need to train them from

scratch you just need to give them like

guidance reinforcement the information

specific to your business and so an

example would be like Sean was is

talking about for his lead capture

intake form voice like he doesn't need

to go train a model he doesn't need to

go like collect data for that software

application like the voice agent is a

software application he just needs to

like make sure it's plugged into his

scheduling system and his database of

candidates and be able to retrieve the

right information about the business and

like you know respond consistently to

customers in a certain tone right right

and so that's more about like

manipulating a bunch of this Bas work

that people um like Labs have already

done for you and the premise here is

like that last mile of getting a

foundation model to be like something

that serves all these use cases in the

real world that you know maybe the

research Labs think of as niches like

the world is composed of very large

niches and so I think it's a I think

it's a really big opportunity for

entrepreneurs and for us by the way

that's not the only investor and fund

investing in this idea we also have a16z

that's betting it big on some fields

that intersect with AI games apps

infrastructure growth Etc they're very

interested in things where AI intersects

Finance for example as well as games

Healthcare Etc but getting back to our

kind of original question if 1.0 is US

humans creating the code to make the

computers do what we want 2.0 is kind of

this idea of where we tell this sort of

process what we wanted to do and then it

then goes and kind of creates the AI

neural net brain to the thing that we

wanted to do to me I'm seeing 3.0 kind

of the next iteration of that as taking

this AI brain let's say it's the big

base model let's say it's l with 3.1 the

405 billion parameter model this thing

that's extremely expensive to train to

create but it's open source and it's

very very good it's also kind of

unwieldly you can't just run it on your

laptop but you might want to use it for

a million different abuse cases for your

business for your personal life Etc what

we're able to do with this sort of

teacher model is used to create

synthetic data to train smaller models

that are customized for our specific use

case so this thing kind of bottles up

its thoughts and creates some sort of a

training set for a smaller model a

smaller model to run customer service or

help you with your homework or run your

calendar or answer your text messages or

or anything where having that

intelligence might be useful to you this

is kind of what the research behind Orca

2 from Microsoft kind of showed us early

on is we're able to take this big huge

unwieldly expensive model create a sort

of uh data set from it and train the

kind of tiny model on top of it and this

model is very very effective at doing

that thing so let's call that software

3.0 now none of this is official we're

kind of making these uh numbers and

stuff up as we go along but it's just to

illustrate a concept but so what's the

point of all this basically think of

this as a line from 0% to 100% here and

this is kind of the penetration of AI

into all of the businesses in the United

States in the world wherever you are

what percentage of businesses currently

are using kind of their own Homespun AI

to automate everything that they can

with AI to add intelligence to any of

the tasks that they're doing where it's

applicable right so I don't know what

that number is but it's low I'm going to

say 1% although I doubt that's anywhere

as high as that and similar how to back

in the days when computers came out

right 0% of businesses at some point 0%

of businesses use computers once they

became good and useful and in expensive

you know maybe 1% of businesses use

computers and then over time it

approached 100% nowadays close to 100%

of businesses use some sort of computers

desktop phones whatever with microchips

in it to help them run their business in

one way or another we are about to see

the same thing happen with AI we're

going to go from some tiny percentage of

businesses use this to run their

business and over the next X number of

years whether that's 5 years 10 years 15

that number is going to start climbing

towards 100% this is also going to

happen for students in school for a lot

of our personal stuff and all that AI

isn't going to be developed by coders

coding it line by line no is it all

going to be developed by these big AI

Labs using their massive training

clusters to create these giant models or

maybe even smaller models but the point

is yeah they'll they'll create once for

big use cases the profitable use cases

but there'll be a million custom special

jobs where AI will be needed for that

special handcrafted use case and

developing that special sort of uh right

that homemade AI that Artisan AI right

whatever you want to call it well it'll

be made from these big open source

models trained created fine-tuned

whatever you want to call that this is

kind of the process that we're referring

to as software 3.0 so be trained and

then deployed into those businesses and

people's lives and everything from

thermostats to monitoring security

cameras to I mean everything you can

possibly think of and this process will

probably not be quite as technical as

you know for example writing code

definitely not as technical as

understanding deep learning and neural

Nets and doing kind of like this sort of

work my guess is that 20 years from now

you have kids kind of spitting up their

own custom little AI models to help them

I don't know keep track of their Pokemon

whatever now I'm not going to unpack how

big of an opportunity this is because

I'll probably unell it and in fact me

trying to sell it to you isn't the point

of this maybe you see it maybe you don't

maybe I'm right maybe I'm wrong but I've

said this before in the future running

businesses will kind of look like

managing your factorio base everything's

more or less automated and you're just

kind of tweaking the different things

the different processes and systems to

optimize the performance I might have

been wrong about that because there's

literally a guy on Twitter X that's

using a large language model to can you

guess to automate factorio now I've been

tracking this project with great

interest he just posted a 3-week update

and I do want to kind of highlight this

project that he's doing because I find

it just endlessly fascinating so the few

minutes that he's doing the update I'll

post this right now to kind of close off

this video check it out I'll leave links

to his profile below but as you watch

this ask yourself this as this field

progresses and it becomes easier and

easier to add this intelligence whether

that's large language models or maybe in

a number of years we'll have some other

sort of type of Frontier models anywhere

we could use help making decisions or

keeping an eye for something or

automating something it seems like these

these AI neural Nets could be trained to

do that specific thing for whatever

specific use case you want it seems like

this is going to be everywhere in every

facet of Our Lives helping us with

pretty much everything that we're doing

what do you think let me know in the

comments am I crazy or is this going to

be bigger than the computer the internet

the cell phone let me know in the

comments check out this automating

factorio of large language of models

video my name is West Roth and thank you

for watching guys it's been 3 weeks

since I'm made the I want to automate

all the factorial tweet uh and I thought

it made sense to make a progress video

about everything that happened in

between so uh the first week essentially

just spent learning Lua learning the

factorial moding libraries and so on and

then the next two weeks I worked on this

uh Library if you go to the GitHub link

on the tweet it should take your this uh

page and here you can find a bunch of

remote interfaces that like the mod that

I creat um provides you so you can walk

to entities like go you can mine

entities you can put things inside boxes

you can put things inside furnances um

it has a bunch of features it still

needs a lot more so it is like work in

progress right now um but people someone

did raise like a p request to fix like

my pathing library a pathing code um

yeah I think it's just going to improve

over time um so this code let me show

you how to install it so essentially you

just need this one file it's like a 700

line uh Lua mod for factorio so if you

go um just clone the repo go to your

sort of game folder so is why my game is

stored and then go to your mods folder

within that and just clone um the that I

showed you once you have this you can

just open up factoral uh like normal and

it should have the mod installed um so

let me let me show you yes you just

start a new game um just with normal

steps skip the intro and now uh if you

go to if you go to sort of the items

here let's just go and walk towards the

coal entity so I'm going to do that

right

now and you can see here it sort of like

finds the code entity um then you can do

the same thing to mine it um

yeah so this is super cool because now

your agent doesn't have to like actually

look at the screen look where the goal

is it can just focus on thinking on how

to play the game how to win the game um

it should make it a lot easier for

people to make agents on this um yeah so

uh my plan for the next week is to sort

of create a bunch of models uh which is

going to be my attempt to beat the game

I know that the library doesn't have

enough things so I'm going to keep

adding more interfaces for agents to

play um but where I'm creating data sets

right now is called glaive it's a start

that I work at uh it's glaive doai and

here I'm creating a bunch of models so

this one I'm sort of um using to sort of

take an input uh like this one says I

want to walk to the nearest iron or and

then the command is like the command

that I need so I'm currently fine tuning

a bunch of models so like this one I can

fine tune right now I'm just going to

use f because it's like it's going to

run easily on my 4090 um so yeah this is

training right now but I do have like a

T of sort of uh I have so many uh

data sets right now that I need to find

tune on but I'm going to open source all

these models eventually um to sort of

make it a lot easier for everyone to use

and um right now this is not public yet

but I'm sort of working on a bunch of um

scripts so this is using grock's

function calling tool that uh we had GL

also worked with them to create and then

um I sort of have um just go to like go

ahead and all tab essentially into

factorio uh press the command button and

then sort of tag commands in so yeah

basically um it's going to be a lot

better uh next week which I'm going to

give you like a bunch of models and like

um yeah just going to opens a bunch of

stuff any I'm just going to keep yapping

so I just going to I'm just going to

leave and uh see you guys whenever