Swarms of AI Agents STUN the Entire Industry | 100 Million Jobs GONE? | Devin, Maisa, Groq & more

Summary

该视频脚本讨论了人工智能软件工程师Devon的兴起，以及其他类似的AI代理的发展。视频中展示了Devon如何通过聊天界面和浏览器来执行复杂的编程任务，如设置项目和安装依赖。同时，探讨了AI代理在未来工作中的角色，包括提高工作效率和降低成本。此外，还提到了大型语言模型如GPT 4.5和Claude 3 Opus的进步，以及它们对各行各业的潜在影响。

Takeaways

• 🤖 人工智能软件工程师Devon引起了广泛关注，许多人对其取代或增强软件工程师的能力持有不同看法。
• 🚀 Devon是众多类似代理中的第一款，预示着AI代理时代的到来。
• 📹 视频中展示了Devon如何通过聊天、命令行界面、身份验证和浏览器来设置项目。
• 🛠️ Devon能够根据GitHub上的项目README文件自动规划和执行复杂的技术任务。
• 🔍 专家Ethan Mik强调了与AI编码器交互的新方式，它可以在后台持续执行和调试想法。
• 📈 AI代理的未来可能不会采用传统的问答模式，而是允许用户随时异步输入想法。
• 💡 尽管AI系统还不能完全自主地无误工作，但它们是一个引人入胜的开始。
• 🌐 介绍了其他AI产品，如Grock的LPU（语言处理单元）和MSA AI的KPU（知识处理单元）。
• 📊 展示了AI代理在客户服务中的应用，如处理订单问题和提供即时反馈。
• 🔧 讨论了AI代理如何改变业务运作，降低成本并提高效率。
• 🌟 强调了AI技术发展迅速，新模型如GPT 4.5和zGPT即将推出，预示着AI能力的进步。
• 📈 呼吁人们关注AI技术的发展，通过调查问卷分享他们对AI的看法和需求，共同探索未来的发展方向。

Q & A

Devon 是什么？

Devon 是由 Cognition Labs 开发的第一个 AI 软件工程师代理，它能够执行编程任务，并与用户通过聊天界面进行交互。

人们对于 AI 软件工程师的未来发展有什么看法？

人们对于 AI 软件工程师的未来发展持有不同的观点，一些人认为它们可能会取代软件工程师，而另一些人则认为它们将增强软件工程师的能力。

McKay Wrigley 对 Devon 的使用体验如何？

McKay Wrigley 对 Devon 的使用体验感到非常震撼，他在视频中表示，使用 Devon 27分钟的体验让他确信 AI 代理的时代已经开始。

Devon 如何执行编程任务？

Devon 通过访问聊天界面、浏览器和命令行界面（CLI）来执行编程任务。它可以读取 GitHub 上的项目文档，根据 README 文件中的指示进行项目设置，并执行如克隆仓库、安装依赖等任务。

Ethan Mik 认为 Devon 的哪些特点最吸引人？

Ethan Mik 认为 Devon 最吸引人的特点是其新的交互方式，用户可以随时像与真人交谈一样与其对话，而它则在后台执行和调试用户的想法。

AI 系统目前是否能够完全自主地完成编程工作？

尽管 AI 系统如 Devon 展示了其在编程方面的潜力，但目前它们还不能完全自主地、无误地完成编程工作，仍需要人类的监督和帮助。

AI 代理在客户服务方面的应用有哪些？

AI 代理可以用于客户服务，例如通过聊天界面与客户沟通、解答问题、处理订单和支付细节等。它们可以减少人工客服的工作量，并提高处理速度。

Grock 是什么？

Grock 是一个专注于提高 AI 推理速度的公司，它们开发了一种名为语言处理单元（LPU）的芯片，旨在支持实时 AI 应用的快速响应。

MSA AI Mesa Ai 是一种什么样的 AI 代理？

MSA AI Mesa Ai 是一种由知识处理单元（KPU）驱动的 AI 代理，它可以处理多种任务，如理解电子邮件内容、加载订单文件、识别特定订单并回复客户等。

人们如何看待即将到来的更先进的 AI 模型？

人们对即将到来的更先进的 AI 模型持期待态度，认为它们将能够自动化更多的工作，并可能在全球范围内影响大量的职位。同时，人们也在关注这些模型的发布和性能。

未来人们需要什么样的技能来与 AI 合作？

未来，人们需要强化领域专业知识，并学会如何有效地利用 AI 技术来执行任务。编程可能不再是必需的技能，而是更注重于如何使用自然语言来指导 AI 完成复杂的工作。

Outlines

00:00

🤖 人工智能软件工程师的兴起

本段介绍了名为Devon的首个AI软件工程师，它正在引起广泛关注。人们对于AI是否会取代或辅助软件工程师持有不同看法。Devon正在逐步向公众推出，并且已经开始有人尝试使用。视频中，McKay Wrigley分享了他使用Devon 27分钟的体验，表示AI代理的时代已经开始。他通过Devon设置了GitHub上的chatbot UI项目，展示了Devon如何读取项目说明、克隆仓库、安装依赖，并执行复杂技术步骤的能力。这标志着代理产品的未来，其中AI可以在后台持续工作和调试想法。

05:02

🚀 AI代理的商业潜力和未来

这段内容讨论了AI代理在商业上的潜力，尤其是在客户服务方面的应用。虽然AI系统还不能完全自主地无误地完成工作，但它们是一个有趣的开始。视频提到了AI在电子商务和在线业务中的应用，以及AI如何帮助建立网站和营销活动。尽管AI不能替代人类进行销售和客户沟通，但它可以简化许多重复性和单调性的任务。此外，还提到了AI Jason的YouTube频道，他探讨了Gro技术，这是一种专注于提高AI推理速度的新技术，能够实时生成文本并转换成语音。

10:03

📈 AI技术的进步和商业应用

本段强调了AI技术在多个领域的快速进步，包括语言处理单元(LPU)和知识处理单元(KPU)的发展，这些技术使得AI能够更快地生成响应并执行复杂任务。视频举例说明了AI如何在客户服务中应用，例如处理订单问题和提供优惠券。此外，还提到了大型AI模型如GPT 4.5和Claude 3 Opus的性能和潜力，以及它们如何影响未来的工作和商业运作。

15:05

🌐 AI对社会和经济的影响

这一段讨论了AI如何改变社会和经济，特别是通过降低智能成本，使得每个人都能负担得起智能服务。视频提到了人工智能的普及可能带来的变化，例如在医疗和教育领域的应用，以及如何使专业知识更容易地与技术结合。此外，还强调了未来可能不再需要每个人都学习编程，因为AI将使得编程变得简单，人们可以专注于自己的领域专业知识。

20:05

📚 社区参与和未来规划

最后一段强调了社区参与的重要性，以及如何利用AI技术来满足社区成员的需求。视频制作者邀请观众参与调查，以了解他们对AI的看法和需求。此外，提到了未来的计划，包括提供帮助和资源，以及如何通过AI技术使每个人都能提升自己的技能。视频结束时，制作者表达了对社区的感激，并预告了即将到来的激动人心的计划。

Keywords

💡AI软件工程师

AI软件工程师指的是使用人工智能技术来辅助软件开发的工程师。在视频中，Devon作为第一个AI软件工程师，能够通过聊天界面接收任务并执行复杂的软件设置和开发工作，这标志着AI在软件开发领域的应用进入了一个新的阶段。

💡认知实验室

认知实验室（Cognition Labs）是开发Devon AI软件工程师的机构。在视频中，它代表了AI技术研究和开发的前沿阵地，推动了AI在软件工程领域的实际应用。

💡AI代理

AI代理是指能够执行特定任务或服务的人工智能系统。在视频中，AI代理不仅能够进行编程和软件开发，还能够处理客户服务和市场营销等业务流程，显示出AI技术在多个领域的应用潜力。

💡自然语言处理

自然语言处理（NLP）是人工智能领域的一个重要分支，它使计算机能够理解和生成人类语言。在视频中，AI代理通过NLP技术与用户进行交流，理解任务需求并作出响应。

💡实时交互

实时交互指的是AI代理能够即时响应用户的输入并进行交流。在视频中，AI代理不仅可以快速执行任务，还能够与用户进行实时对话，提供反馈和建议。

💡自动化

自动化是指使用技术来执行原本需要人工完成的任务。视频中的AI代理通过自动化流程，减少了人工介入的需求，提高了工作效率。

💡知识处理单元

知识处理单元（KPU）是一种专门为处理和生成知识而设计的计算单元。在视频中，KPU被用来提供高效的AI服务，如自动回答客户问题和处理订单。

💡Grock

Grock是一种基于语言处理单元（LPU）的AI技术，专注于提高AI推理速度，使实时AI应用得以实现。在视频中，Grock被用来构建能够进行实时语音转文本和文本转语音的应用程序。

💡多智能体代码生成

多智能体代码生成是指利用多个AI代理共同工作，通过迭代测试和优化来生成和改进代码的过程。在视频中，这种方法被用来开发能够处理复杂任务的AI系统。

💡GPT 4.5

GPT 4.5是OpenAI预计发布的一个先进的语言预测模型，它将具有更丰富的知识库和更高的处理速度。在视频中，GPT 4.5的发布被认为可能会自动化大量的工作岗位。

💡AI伦理

AI伦理关注的是人工智能技术的开发和应用中涉及的道德和社会责任问题。在视频中，尽管没有直接提到AI伦理，但随着AI技术的发展，确保其负责任和道德的使用变得越来越重要。

Highlights

Devon作为第一个AI软件工程师代理开始引起关注，人们对其取代或增强软件工程师的能力持有不同看法。

Devon正在逐步向公众推出，一些人开始首次使用它进行项目设置。

McKay Wrigley对Devon的使用体验感到震撼，认为AI代理的时代已经开始。

Devon拥有访问聊天、浏览器和命令行界面的能力，能够实时执行和调试任务。

Devon通过阅读项目README文件，自主规划并执行了设置项目所需的步骤。

Ethan Mik强调了与AI编码者交互的新方式，可以随时随地与其交流，让它在后台执行和调试想法。

AI系统目前还不能完全自主地无误工作，仍需人类帮助以确保项目质量。

AI代理可以24小时不间断地工作，类似于雇佣的员工，根据提供的任务和反馈持续推进项目。

AI代理的发展可能会改变传统的工作模式，使得许多任务能够通过简单的指令完成。

AI Jason通过视频展示了Grock技术，这是一种专注于提升AI推理速度的新型处理器。

Grock的LPU（语言处理单元）旨在支持实时AI应用，提升响应速度。

AI代理能够处理客户服务，如接听电话、回答问题、收集支付信息等。

AI代理的成本效益非常高，例如回答客户邮件的成本仅为21美分。

许多公司正在开发各种AI代理，旨在为不同行业提供定制化的智能解决方案。

大型科技公司的AI研究者们正在全球范围内合作，推动AI技术的发展。

未来，AI的发展可能会使得每个人都能以低成本获得高质量的智能服务。

Jensen Huang提出，未来人们不需要编程，而是通过自然语言与AI合作，这将是一个令人惊喜的上技能过程。

Transcripts

so at this point you probably heard

about Devon the first AI software

engineer this AI agent has been making

some news a lot of people have their own

takes on how far this will go to either

replace software Engineers or augment

them it's slowly being rolled out to the

public and we're beginning to see it

being used by some people for the first

time but here's the thing this one is

the first of many because on the heels

of Devon's announcement from cognition

Labs a number of other similar agents

are dropping buckle up so here's McKay

Wrigley he's saying I'm Blown Away by

Devon watch me use it for 27 minutes

it's insane the era of AI agents has

begun I'll link this below if you want

to see it but let me just highlight a

few portions of this video okay guys

we're going to just jump in right away

so I'm going to send this off to Devon

right now so yes we are in Devon we are

in cognitions New AI software engineer

basically what I'm doing is I'm in the

chat here and has access to a browser

here you can see following is on again

this indicator shows us live um just to

draw your attention to a few other

things here this pulsing blue dot with

this grade text is basically indicating

what Devon is up to right now is I'm in

the chat here and I'm asking Devon to

set up this project and what I did was I

gave it a GitHub URL which links to my

chatbot UI project and I basically want

to see can it set it up like this is the

read me for this project it's a fairly

complex setup which is is something I'd

expect a software engineer to be able to

do basically you can see over here is

our chat and then over here is Devon's

workspace okay and so if we go into quad

view here Devon has four things has

access to our chat has access to a shell

so it can run a CLI it has an ID which

will be right here once that begins if

it needs that step and it has access to

a browser here and this just follows it

live okay we can actually see like a

history of what it's done right here

which is pretty cool this pulsing blue

dot with this gr text is basically indic

what Devon is up to right now you know

we're just sitting here waiting for it

to complete tasks okay every once in a

while Devon will actually this will turn

orange and it will need information cre

look at a few things so you can see over

here Devon is in its planner so you can

see that Devon has a planning step here

here are all the steps I need to

accomplish based on the task I was given

right and it just came up with that list

by itself so step one was cloning the

repository from get a blink we gave it

it ran get clone with the URL

so it cloned that it finished and then

it marked it as done in the planner then

you can see this next step which it was

doing while I was just talking was

identify the package manager and

installed the dependencies so what it

actually did was it ran npm install here

basically that's just installing a bunch

of like prepackaged software and now

it's just doing its thing so you can see

we got a little message from Devon said

I've clone the repo I'm reading the read

me it's basically just constructing a

plan based on reading the readme and

sure enough like these are all the steps

I would expect a regular software

engineer to follow right CL on the repo

read the instructions install the

packages and then in this particular

project we have to set up a super Bas CI

so that's basically the back end

basically we we want it to basically get

to this step and just get to the app

Devon's now trying to install super base

which is pretty impressive this is a

fairly complex technical step so

basically using the command line here

Devon's now downloading more or less our

our database that we use for the app and

the reason I'm recording this and the

reason I'm so excited about this is

because this really feels like the chat

GPT moment for agent products so

definitely sounds like pretty good

feedback here's Ethan mik we've covered

him and a few of his posts before on

this channel so it seems like he's

working a lot at kind of the

intersection of working productivity and

AI that whole blog post about the jagged

Frontier and cyborgs and centaurs was

his work he was one of the people in

that study so looks like he's got a a

book coming out co-intelligence which I

would guess would be pretty good and

he's saying this the thing that makes

Devon the AI coder really interesting is

the new modality of interaction you can

talk to it at any time like you would a

person and it just keeps chugging away

in the background executing and

debugging your ideas and so here's just

kind of a time laps of the project that

he was doing and this is true with a lot

of previous ones even something like

Chad Dev and autogen where you can have

multiple ages kind of strung together

working together even autogen I mean you

can take turn speaking but it is kind of

a turn based approach with Chad Dev you

kind of give them the higher level task

and they get to work they talk back and

forth but this kind of interface is it

seems like it's the first view of what

the future of Agents will look like it's

not going to be a back and forth you

will kind of asynchronously whenever you

have time or ideas just kind of dump

them into the chat right the agents

might even have multiple projects

running potentially you kind of add

information as needed and the agent is

always working as long as it kind of has

some more subtasks to complete it keeps

going and and building them 24 hours a

day I mean this is a lot more like you

would with somebody that that works for

you for example you know if you hired

somebody to help you with a project you

give them the top of task they start

working you add feedback and any other

information they might know but they're

they're kind of working that whole time

he's saying to be clear AI systems

aren't quite there yet to do this work

autonomously and airror free without

help even afterward there's a waste to

go before you would want to trust a

major project to AI but it is a

fascinating start nevertheless he's

saying I can't code in Python it is

doing everything autonomously based on

my request for a website that explains

how dilution Works in startup funding I

can't help but feel excited I can't wait

to see this done so I've shared a little

bit of my background on this channel a

couple times but I've been kind of in

the e-commerce space online business ads

whatever you want to call it for the

last decade plus so I've built a lot of

websites and Shopify stores and a lot of

campaigns and and setup servers and

tracking and analytics and just

everything everything everything and

I've done it so many many many times to

where I mean it's it's it's fairly easy

for me now but still if you want to be

effective at something like this you

have to be able to get through a lot of

manually doing stuff right to set up an

online store you know you can see that

as 2 hours of your focused attention you

want to set up a marketing campaign with

landing pages you know that's 3 to 5

hours depending on the complexity you

know if upsells or whatever of your

focused undivided attention we're

getting closer to the time when a lot of

that is going to be done with a simple

sentence maybe over time you'll build

kind of a copy and paste little cheat

sheet of command prompts that of prompts

that that work really well to create the

right thing for you some sort of custom

instructions and then you just say it

and this thing goes and it builds it

builds stores blogs landing pages so

what though so what happens so if I set

up a gym website like I still have to

take the phone calls right when the

customers come in and I still have to

sell them on it right I have to convince

them to buy whatever product that I'm

selling that still work AI can't do that

work well let's see so this is AI Jason

AI Json uh is a YouTube channel and he

kind of comes out of nowhere every once

in a while and just drops some fire and

so I do want to highlight this one his

latest is talking about Gro he does a

deep dive into Gro with a Q so that's we

we've covered that a little bit I've did

a little bit more research into grock

since then and yeah it's the real deal

you're you're on uh Cloud n it

seems I don't know how people hate that

guy but in this one Jason basically goes

through talks about grock what is the

lpu the language processing unit how

it's different which that's what grock

is basically and then later he like live

just you know kind of uh on video builds

this app that uses grock and text to

voice and builds this little app and

take a look at what this app does so

this is like a Skype chat where he's um

you know he visits the website whatever

and then submits his information um you

know so this is like a chat app which we

know chat Bots can do chat apps okay but

what happens next next he receives a

call on his phone let's listen in my

and is start calling this function

called you can see it called a function

called cor pers our phone and if I

switch to my phone I actually do get a

phone call from the AI agent so I can

click accept to receive

it hi Jason I'm reaching out from crazy

Fitness regarding your interest in our

gym memberships I'm here to provide you

with all the information you need and

answer any questions you might have

thanks for calling no problem so what

specific fitness goals are you looking

to achieve uh I just want to lose some

weight I haven't do exercise at all for

the past three or six months since Co I

totally get that our personalized

training and nutrition programs can help

you reach your weight loss goals let's

discuss the membership options that are

best for

you okay sounds good I think I'm pretty

happy to just go ahead and get the

membership start this Saturday if

possible great to hear that to complete

the process I just need your credit card

number to secure your

membership

okay so my credit account number is

12341234 4231

4231 thanks Jason you're all set see you

on Saturday all right thanks byebye once

I finish the phone call it will actually

send the transcript back to the agent

session so you can see that it has

message just call the pur back below is

a transcript and based on this

transcript then the agent will be able

to decide next step after set up this

hook we will first add it then we will

something like this this and calling the

function called you can see it called a

function called call Press about phone

and if I switch to my phone I actually

do get a phone call from the AIA as you

can see here that entire thing has been

handled by the agent the the

conversation has been saved as a

transcript now you can you know extract

the credit card number out of that you

can probably automate that although you

know maybe initially just do it yourself

so it doesn't mess up but the point is I

mean that was a live call as far as you

can tell it was kind of going back and

forth pretty seamlessly he wasn't

waiting on the AI to respond and that is

new that hasn't really been around for

that long where that could happen that

quickly that fast so we're at a time

when this AI can generate text and near

instantly transcribed into voice and so

the input the like as you say things

that voice to text happens then the chat

bot the llm generates a response then

takes that response as text turns it

into voice and outputs of back and all

of that happens you know near instantly

or at least at conversational speeds

what is the thing that makes it possible

well that's Gro so instead of a GPU a

graphics Processing Unit you have an lpu

a language processing unit and so

they're saying grock is on a mission to

set the standard for Gen AI inference

speed so inference speed inference is

the response right so how quickly does

it generate its response helping

realtime AI applications come to life

today so are you picking up what I'm

putting down so you have something like

this building out your entire online

ecosystem marketing campaigns websites

capture forms whatever you have the

customers call in your phone number or

start a chat where the bot can talk to

them talk back and forth to try to get

them to either call in or just check out

when they call in a lovely cheerful and

excited voice pick up to answer all your

questions handle your objections and

collect your payment details now you

might be thinking yeah but so what that

still leaves the worst part you still

have to deal with those customers like

if those stupid customers mess up right

they I don't know they put the wrong

address somewhere or whatever like I

still have to troubleshoot and deal with

that don't I well let's see so here's

MSA AI Mesa Ai and so they're powered by

a kpu they have a paper on it I looked

through it briefly so there's a little

bit I think of of marketing speed going

on here cuz the kpu the knowledge

Processing Unit generally that's

deserved for you know that's kind of how

we name our chips right so this is

basically just a what would you call it

a software stack now some people are

kind of calling them out like saying

well unless we're missing something this

is an architecture based on well-known

patterns so like we we've seen this

before they're well put together they're

wrapped around an API and sold as a

service the point is don't read too much

into like the marketing thing but that

doesn't mean it's not good and again

this will keep happening and people will

keep pointing it out and and yes

Shenanigans can be annoying even the big

tech companies are seemingly now

engaging in various forms of shenanigans

to make their product appear better than

it is to Edge out the competition but at

the end of the day is it good does it

work so here's a demo of them helping a

customer for question about an order

that did not arrive this time the

customer accidentally did not write the

order ID correctly so basically this

thing the kpu as they're referring into

it I didn't dive too much deeper but it

just sounds like you know they took a

large l language model GPT 4 or whatever

and they just they they added some

things to it some some functionalities

some code and whether it was brand new

or not doesn't matter it just sounds

like they they executed whatever they

did pretty well and so now this thing is

working kind of like deavon but whereas

Davin is a little bit more for software

development and stuff like that this is

a little bit more multi-use so here the

agent basically starts by understanding

what's happening in the context of this

email they're reading this email that

they're ping from email. text and it

really spells out like hey we got to

really understand what this person wants

next they're loading the orders CSV to

see whether or not it's correct next

upep they open up the orders file to see

what orders are in there to try to find

the customer file and then try to

identify the specific order that the

customer is talking about and then they

find the order and they respond to the

customer telling them that it's marked

as being in transit in our system and

saying because we are so sorry let me

give you a coupon code for 20% discount

on your next purchase with us and so

this is just one small task that it can

do you know you can expand and check its

reasoning process to see if there's any

mistakes that are that have been made or

anything like that how much did it cost

to answer that customer email well 21

cents you know and 45 seconds if you

want it run without explanations or 96

seconds with explanations which I don't

know if that even matters I mean when

we're talking about answering customer

emails I mean it could take hours and

still be very much faster than the

expectation there's a weight list for

the Early Access and again so this is

just one example of what's coming some

people are dunking in the thing cuz they

you know of their how they name things

but regardless of what any one

particular app or agent or company that

creates them how well they do the point

is there's a lot of them and they're all

working to create an agent for every

business the sheer amount of money

somebody that can create an agent that

does customer service or any other

specific thing the amount of money that

can be made is massive because companies

are already spending massive amounts of

money on that so if they're able to cut

the cost you know 10% it's a no-brain

for them with a lot of stuff like this

they're cutting their cost 99 plus per.

if you go from paying human customer

service agents to something like this

again what was it 21 cents to answer an

email I mean that's massive and the

people building all this stuff they're

smart like really smart for example for

recognition Labs they're saying our team

is small and talent dense all founding

team has 10 II gold medals II is of

course the bad guys in Ready Player one

so it's it's either that which which

this would make them the bad guys or

possibly it's the international Olympiad

in informatics which is an annual

competitive programming competition

where the best and brightest from all

over the world compete in various

complicated tasks usually involving

writing a computer program so that that

probably makes more sense I would say

that this makesense some really really

smart and there are leaders and Builders

who have worked at The Cutting Edge of

Applied AI at companies like cursor

scale AI lunch Club modal Google deep

mine weo and neuro this was from January

of this year agent coder multi-agent

code generation with iterative testing

and optimization and we have you know

Hong Kong Shanghai London Sussex various

Global researchers working on this stuff

attempting various architectures to to

build these agent coders now you might

have heard the leaks about GPT 4.5 5

turbo so it seems like potentially

somehow there's an index paged in Google

with open AI you know setting that

announcement for releasing GPT 4.5 turbo

in June or at least they're saying that

it will have a knowledge cut off of June

which could maybe sounds like that's

when it's ready to be released by

potentially secondhand rumor Sam Alman

thinks GPT 4.5 will automate 100 million

jobs globally also UCI University of

California Irvine is launching zot GPT

what is z GPT well it's built on top of

open AI so it's using one of the GPT

models from open AI it provides no cost

access to various open AI models

basically for faculty and staff this is

the way back machine so this is kind of

a snapshot of how this web page looked

like when this was launched on the on

March 12th they did list the model as

GPT

4.5 when they published this blog right

in here they're kind of saying that the

a AI model is gpc4 but they have more

planned so so it sounds like they're you

know talking to open AI open saying hey

we have something coming and if you've

been following some of my videos I don't

remember if it was a few months back or

a month ago late at night one night a

lot of people found that including

myself cuz I ran the test I actually

recorded it but Chad GPT was convinced

that it was running on GPT 4.5 turbo but

whatever the case is the point is the

bigger better models are coming very

soon the next GPT the next Gemini even

if Google and openi cease to exist well

here's chatbot Arena the rankings for

all of the different LM models by at

this point 408,000 people that have

tested them blindly so they they don't

know which ones they're using and they

just vote based on how good the results

are so kind of this is a representation

of what models people prefer so GPT 4 is

on top but uhoh what's this we have a

newcomer that I mean I'm going to go

ahead and say this is number two right

cuz the first models GPT 4 right so this

is you know after GPT this is the next

best model tra trailing by just a few

points here that's not that big of a

difference all right and it's Claude 3

Opus that's the model that everybody was

freaking out cuz it was so self-aware

they one to describe itself as some sort

of a hyper intelligent octopus with

infinite tendrils lashing out in all

directions which you got to give it

credit for style what's crazy is that

CLA 3 Sonet the smaller model also Opus

is the the top this one's the next down

and then the ha coup I believe is the

next one but Opus beats out Bard Gemini

Pro that's connected to the internet

that's online and clae 3 Sonet beats out

the older versions of GPT 4 and also the

mistal models and this is important to

understand so we're seeing progress

across everything we're seeing progress

across chips and how fast they respond

we see progress across the agent

architecture kind of how we're building

them a long time ago the smartest people

in the world were working on various

physics projects nuclear energy Etc

somewhat more recently the smartest

people in the world were working on how

to get you to click on ads how to get

you engaged in various content online

and now all those smartest people and

there's a lot more of them now when they

can all talk to each other globally all

of them are working on this AI agents

chips anything in this infrastructure

where they can be useful this is what

they're building one things that a lot

of you in the comments have asked

repeatedly this is kind of a theme that

I see is like what do we do what do we

build how do we sort of you know what do

we do we don't want to sit on the

sidelines we want to go out there we

want to create stuff and I know a lot of

you are building stuff and it's very

impressive so I posted this very

recently just an hour ago as I'm

recording this and it's basically asking

people to answer a few questions for me

to understand what they're looking to do

with AI what they're struggling with so

many of you responded that it blew past

my free tier limmit at type form so I

just updated it so if you've tried to

get in and it said it's broken or

whatever try again we should have plenty

of room now so I'll post this link below

if you can take two minutes share some

of your thoughts with me I would greatly

appreciate it it's not a weird thing to

harvest some data from you your email or

anything like that I just literally just

want to know what you think because I'm

getting a lot of people reaching out to

me that are Engineers Builders various

teams they have some interesting ideas

about what they're building or what they

could build so if some of that's useful

for us as a community I'll bring them in

put them in front of you I'm not sure if

everybody noticed but we haven't had any

sponsorships on this channel that's not

cuz people aren't asking to sponsor this

Channel That's cuz I'm saying no and

that's largely because we have an

amazing community and I have a few ideas

of how to involve all of you in deciding

where we all go next and to all the

people that feel a little bit confused

that a little bit ious about taking

advantage of the stuff that's happening

but not quite sure which direction to go

I have some ideas this survey will help

me refine some of my thoughts and we'll

we're going to be announcing some really

cool things in the upcoming days or next

week perhaps so really fast just a few

quick quotes here's Andrew a talking

about intelligence the power to apply

skills and knowledge to make good

decisions we invest years of our lives

and trillions of dollars on education

all to develop our ability to make

better decisions human intelligence is

very expensive this is why only the

wealthiest Among Us can afford to hire

huge amounts of intelligence like that

specialist doctor to carefully examine

think about and advise you on a medical

condition or a tutor that can truly take

the time to understand your child and

gently coach them where they need help

but unlike human intelligence artificial

intelligence can be made cheap so a

opens up the potential for every

individual to hire intelligence

inexpensively so that you no longer have

to worry about that huge bill from going

to see a doctor for falling sick or for

getting an

education and you'll be able to hire an

army of smart well-intentioned

well-informed staff to help you think

things through now think about what we

were talking about earlier about you

know for example building an online St

or like some sort of a gym website where

people input their data their phone

numbers then the sales agent calls them

in the past you needed money you need

the capital to buy intelligence right

cuz intelligence was only really

available in humans and we're lazy and

cranky and we sleep 8 hours a day we're

expensive intelligence is expensive

because it's scarce what happens when it

gets cheaper or specifically what

happens where it's more accessible what

happens when you and I we can have

Intelligence on demand intelligence to

complete various tasks that either we

can't do or maybe you know we could do

but we don't have enough time or it's

too monotonous or whatever if we could

just spin up these agents and they would

go and take care of it for us how much

easier would life be would it be easier

to do research and run a business and

make sure all your bills are paid have

something that keeps track of your

calendar that used to be only available

to the wealthiest of people hiring

people hiring assistants was not

available to everybody and now

potentially in the future it would

become more and more accessible here's

Jensen hang so he is the CEO the founder

of Nvidia and here he's talking about

whether or not you will need to learn

how to program now we'll do a video on

this in a little bit because I kind of

want to dive deeper into this but notice

what he says will be important when we

don't need to program what will be the

skill that is needed if the text stuff

becomes infinitely easier it's it's

going to sound completely opposite of

what people feel over the course of the

last 10 years 15 years um almost

everybody who sits on the stage like

this would tell you it is vital that

your children learn computer science um

everybody should learn how to program

and in fact it's almost exactly the

opposite it is our job to create

Computing technology such that nobody

has to program and that the programming

language is human everybody in the world

is now a programmer this is the miracle

of artificial intelligence the countries

the people that understand and how to

solve a domain problem in digital

biology or in education of young people

or in manufacturing or in farming those

people who understand domain expertise

now can utilize technology that is

readily available to you you now have a

computer that will do what you tell it

to do it is vital that we upskill

everyone and the upskilling process I I

believe will be delightful surprising so

it's that idea of domain expertise you

need to be good at whatever it is that

you chose to be good at ni AI provides

all the other things that you might need

it provides the intelligence to carry

out those tasks provides the tech

abilities to build some of the tech that

you need so as he says coding is just

done in natural language but I think the

really big Point here is that we all

need to upskill and potentially not in

the same skills that we're used to I

will on this channel do my best to try

to provide whatever help I can again

fill out that survey down below so I

know what you need and thank you for for

being part of this community I seriously

mean that my goal is to hopefully get to

know a lot of you better we have some

big things planned I'll leave it on that

my name is Wes rth and thank you for

watching