Microsoft Reveals SECRET NEW MODEL | GPT-5 DELAYED | Sam Altman speaks out against "Doomers"

AI Unleashed - The Coming Artificial Intelligence Revolution and Race to AGI
7 May 202420:48

Summary

TLDRThe transcript discusses the energy efficiency and societal impact of AI, emphasizing the importance of continued innovation and development in the field. It highlights Microsoft's new AI model, 'my1', which is set to compete with existing models from OpenAI and Google. The speaker also touches on the potential of AI to create a more prosperous and abundant future, countering the notion that technological advancement is inherently negative. Additionally, the script covers various AI initiatives, including Microsoft's investment in OpenAI, the development of on-device AI capabilities by Apple, and the ethical considerations surrounding AI training and content usage. The summary concludes with an invitation to join a private community focused on leveraging AI for growth and productivity, reflecting the belief that harnessing AI is not just beneficial but a moral imperative.

Takeaways

  • 🚗 **Energy Efficiency in AI**: The script discusses the energy efficiency of AI compared to traditional methods, like driving for information, highlighting that AI is a net saver in terms of energy consumption.
  • 💡 **Progress in AI**: There's an emphasis on the continuous improvement in AI algorithms and hardware, including the potential for breakthroughs like fusion power to sustain advancements.
  • 🌱 **Fighting Pessimism**: The speaker encourages a proactive stance against pessimism regarding the future, advocating for a mission to promote prosperity and abundance through technology.
  • 🤖 **Microsoft's AI Development**: Microsoft is developing its own AI model, referred to as 'my1', which is expected to be larger and potentially more powerful than previous models, indicating a shift from reliance on OpenAI.
  • 🔍 **OpenAI's Strategy**: OpenAI is considering postponing announcements and focusing on improving its models, including releases like 'I'm a good gpt2 chatbot', which is stronger in reasoning.
  • 📚 **Educational Outreach**: Sam Altman, CEO of OpenAI, has been visiting universities to discuss AI and the future, promoting a vision of a brighter future enabled by technology.
  • 🤝 **API Partnerships**: OpenAI has entered into an API partnership with Stack Overflow, aiming to enhance model performance for developers and create a mutually beneficial relationship.
  • 📈 **Content Creator Rights**: OpenAI is introducing a 'media manager' to respect content owner preferences on AI training, similar to the 'robots.txt' for web publishers.
  • 🛡️ **Cybersecurity Focus**: There's a significant focus on bolstering cybersecurity, with investments in AI for cyber defense and hiring specialists to address vulnerabilities that AI might introduce.
  • 📱 **Apple's AI Integration**: Apple is planning to integrate more AI capabilities into iOS, with on-device functionalities for text summarization and smarter search results, indicating a move towards privacy-centric AI solutions.
  • 💌 **Internal Perspectives on AI**: An internal email from Microsoft reflects on the company's initial underestimation of AI's potential and the subsequent push to catch up, highlighting the transformative impact of AI on infrastructure and business models.

Q & A

  • What is the main concern raised about the energy consumption of AI and computing?

    -The main concern is that despite the significant energy consumption of AI and computing, it is often compared unfavorably to the energy used for transportation. However, the speaker argues that the energy used for AI and computing has led to a net savings when considering the energy that would have been spent on traditional methods of obtaining information.

  • What is the speaker's opinion on the importance of developing more efficient AI algorithms and chips?

    -The speaker believes that developing more efficient AI algorithms and chips is crucial. They mention that as technology advances, we will continue to find ways to make these systems more energy-efficient, possibly even utilizing fusion as a power source in the future.

  • What does the speaker suggest about the attitude of some people towards technological progress?

    -The speaker suggests that there is a 'doom and gloom' attitude among some people who believe that the world's problems are insurmountable and that efforts to solve them are futile. They encourage fighting against this negative mindset and advocate for a more proactive and optimistic approach.

  • What is the significance of Microsoft's new AI model, referred to as 'my1'?

    -The significance of Microsoft's new AI model, 'my1', is that it represents a step towards creating larger, more powerful AI models that could potentially exceed the capabilities of current models like GPT-4. This model is part of Microsoft's strategy to develop both small, efficient language models for mobile devices and large, state-of-the-art AI models.

  • How does the speaker view the potential impact of AI on society?

    -The speaker views the potential impact of AI on society as overwhelmingly positive. They believe that AI can drive a better future by creating abundance, intelligence, and longevity, which expands our options and is a moral imperative.

  • What is the current state of Microsoft's involvement with AI?

    -Microsoft is actively involved in the AI space, having invested in OpenAI and now developing its own large language model. They are also focusing on creating small, efficient models for mobile devices and larger models for more complex tasks.

  • What is the speaker's perspective on the criticism of AI for its carbon footprint and potential biases?

    -The speaker acknowledges the criticisms but argues that they are not reasons to halt progress. They believe that over time, significant progress has been made in addressing both the carbon footprint and bias in AI, and that this trend will continue.

  • What is the role of OpenAI in the development of AI?

    -OpenAI plays a significant role in the development of AI by training frontier-defining models using supercomputers provided by partners like Microsoft. They are also involved in various initiatives to address concerns about data usage, content creation, and cybersecurity.

  • How does the speaker perceive the future of AI and its impact on productivity and e-commerce?

    -The speaker is very optimistic about the future of AI and its potential to improve productivity and revolutionize e-commerce. They are particularly interested in leveraging AI to automate tasks and enhance decision-making in these sectors.

  • What is the speaker's view on the importance of community and collaboration in the field of AI?

    -The speaker highly values community and collaboration, having created a private forum for discussing AI advancements and sharing knowledge. They believe that a community focused on learning and growth can be a valuable resource in staying at the forefront of AI developments.

  • What are the speaker's thoughts on the potential of AI to address cybersecurity concerns?

    -The speaker acknowledges the increased liability for cyber attacks due to AI but also sees it as a tool for improving cybersecurity. They mention the use of AI for cyber defense, audits, and compliance programs, suggesting that AI can be a powerful asset in safeguarding against cyber threats.

Outlines

00:00

🚀 Energy Efficiency and AI's Future Impact

The first paragraph discusses the energy consumption of AI and computing, comparing it to the carbon footprint of traditional transportation. It argues against the notion that AI is environmentally harmful, highlighting the potential for more efficient algorithms and technologies like fusion power. The speaker emphasizes the importance of combating pessimism and advocating for technological progress to improve society. The paragraph also touches on Microsoft's development of a new AI model, 'my1,' overseen by Mustafa Suliman, and the potential for it to compete with existing models like GPT-4.

05:01

🤖 AI Developments and Content Ownership

This paragraph covers various AI-related news, including the advancements in AI models by OpenAI and the introduction of a new, stronger GPT2 chatbot. It discusses OpenAI's partnership with Stack Overflow to improve model performance for developers and the potential impact on platforms like Stack Overflow. The paragraph also addresses concerns about content ownership and copyright, mentioning OpenAI's new media manager and its approach to respecting creators' choices regarding the use of their content in AI training.

10:04

🌐 Cybersecurity and AI's Role in Data Sensitivity

The third paragraph delves into the increased dangers posed by AI in terms of cybersecurity. It discusses how seemingly harmless data can become risky when processed by AI, highlighting a tool called GEOS Spy that uses AI to determine locations from images with remarkable accuracy. The paragraph also includes insights from Kevin Scott, Microsoft's CTO, about the company's investment in AI and the importance of building powerful supercomputers for training AI models.

15:04

📈 Microsoft's AI Strategy and Market Position

This section outlines Microsoft's strategy in the AI market, discussing the company's development of its own large language model and its potential to surpass current leaders like OpenAI, Google, and Meta. It explores the commercial viability of AI models and how companies like Microsoft, with its Azure platform, may have a more straightforward path to monetizing AI. The paragraph also reflects on Bill Gates's and Microsoft's historical positioning in technology waves and their recent efforts to catch up in the AI sector.

20:05

💡 Building a Community for AI Growth and Learning

The final paragraph is a call to action for building a community focused on AI growth and learning. The speaker, Wes, invites viewers to join a private forum aimed at discussing AI, sharing ideas, and learning from one another. He emphasizes the importance of not missing out on the AI revolution and positions the community as a valuable resource for those looking to stay on the cutting edge of AI technology.

Mindmap

Keywords

💡Compute

Compute refers to the act of performing calculations or processing data. In the context of the video, it is used to compare the energy consumption of Google's data centers to the carbon footprint of people driving to access information. This comparison is part of a discussion on the environmental impact of AI and technology.

💡AI Efficiency

AI efficiency pertains to the effectiveness and resource optimization in the development and operation of artificial intelligence systems. The video discusses the need for more efficient algorithms and chips to reduce the energy consumption of AI, highlighting the importance of innovation in sustainable technology.

💡Fusion

Fusion, in the context of the video, refers to nuclear fusion, a potential future energy source that could provide abundant, clean power. It is mentioned as a possible solution for powering AI systems in a more sustainable way.

💡AI Bias

AI bias refers to the unfairness or prejudice that can be inadvertently built into AI systems due to the data they are trained on or the algorithms they use. The video acknowledges the need to address bias in AI, which is important for ethical and equitable technology development.

💡Microsoft's AI Model

Microsoft's AI model, referred to as 'my1' in the transcript, is a new, large-scale AI model being developed by Microsoft to compete with existing models from OpenAI and Google. The video discusses the potential scale and capabilities of this model, emphasizing Microsoft's strategic move in the AI space.

💡Open Source

Open source describes a type of software where the source code is made available to the public, allowing anyone to view, use, modify, and distribute it. The video mentions Microsoft creating effective small models and the potential for their new AI model to be open source, which would allow for community collaboration and innovation.

💡Synthetic Data

Synthetic data refers to data that is generated using artificial means, such as through AI models, rather than collected from real-world observations. The video discusses the use of synthetic data to train efficient AI models, which can lead to advancements in AI without the need for large amounts of real-world data.

💡Stack Overflow API

The Stack Overflow API is a set of programming tools that allows developers to access and interact with the data on the Stack Overflow platform. The video talks about an API partnership between OpenAI and Stack Overflow, which aims to improve AI model performance for developers and enhance the development of Overflow AI.

💡Content Providence

Content Providence, as mentioned in the video, likely refers to content provenance, which is the practice of tracing the origins and ownership of digital content. This is particularly relevant in the context of AI models that generate content, as it involves respecting copyright and creator rights.

💡Cybersecurity

Cybersecurity involves the protection of internet-connected systems, including hardware, software, and data, from attack, damage, or unauthorized access. The video discusses the increased liability for cyber attacks in the age of AI, highlighting the need for robust security measures to safeguard against potential vulnerabilities.

💡On-Device AI

On-device AI refers to AI processing that occurs on a user's device, such as a smartphone or computer, rather than on remote servers or in the cloud. The video mentions Apple's focus on on-device AI for features like text summarization and smarter search results, emphasizing the importance of privacy and speed.

Highlights

The comparison of Google's computational energy usage to the carbon footprint of traditional information seeking methods highlights a net energy savings with AI.

The importance of developing more efficient algorithms and chips to reduce energy consumption in AI.

The potential of fusion power to sustainably fuel future AI advancements.

Addressing the negativity and hopelessness some feel about the world's problems, advocating for a proactive and solution-oriented mindset.

Microsoft's development of a new AI model, internally called 'May one', to compete with OpenAI, Google, and others.

Mustafa Suliman, former Google AI leader, overseeing Microsoft's new AI model, indicating a strategic move in AI development.

Microsoft's pursuit of dual AI strategies: small, efficient language models for mobile apps and large, state-of-the-art models.

The speculation that Microsoft might not need OpenAI and could potentially reveal a new model at the Build Developer Conference.

OpenAI's potential postponement of a new announcement, suggesting strategic adjustments in the competitive AI landscape.

Sam Altman's visits to universities to discuss AI and the future, facing opposition from those skeptical of a technology-driven prosperous future.

OpenAI's release of two new models in the LM system, showcasing advancements in reasoning and capabilities.

OpenAI's domain change from 'chat.openai' to 'chatgpt', reflecting a rebranding or strategic shift.

OpenAI's API partnership with Stack Overflow, aiming to improve model performance for developers and create a mutually beneficial collaboration.

OpenAI's introduction of a media manager for creators and content owners, respecting content usage preferences in AI training.

The discussion of a potential 'robots.txt' equivalent for AI, allowing publishers to opt-in or out of content usage for training AI models.

OpenAI's efforts to address cybersecurity concerns by investing in infrastructure security, AI for cyber defense, and hiring specialized personnel.

The demonstration of AI's evolving capabilities, such as GEOS spy's ability to determine precise locations from images, indicating increased potential for data aggregation and analysis.

Apple's plans for AI integration in iOS 18, focusing on on-device processing for text summarization, smarter search results, and other features.

Kevin Scott's email to Satya Nadella and Bill Gates, detailing the strategic importance of investing in AI and the potential benefits of partnerships like Microsoft's with OpenAI.

The potential for Microsoft's large language model to surpass current models like GPT-4, indicating a shift in the AI competition landscape.

The discussion on the most straightforward path to monetizing AI models, with Microsoft's Azure and product integrations being a strong contender.

Transcripts

play00:00

yeah I read this thing about the compute

play00:01

Google used once compared to the amount

play00:04

of carbon that people used to spend

play00:06

driving in their cars places to get

play00:08

information and you know you have people

play00:10

saying Google's so horrible we should

play00:11

shut it down it's like spending El

play00:13

energy and it's it intellectually a very

play00:15

dishonest thing to say because it was a

play00:17

net Savings in energy you know for like

play00:19

AI yeah it's going to need a lot of

play00:21

energy we're going to keep figuring out

play00:23

way more efficient algorithms way more

play00:24

efficient chips we're going to get

play00:26

Fusion we're going to power the stuff

play00:27

this way um so I think it is like

play00:31

important to address this issue um but

play00:34

we will in all of these fantastic

play00:37

ways but I think this points to

play00:40

something else which is the you know you

play00:41

open asking about P doom and the level

play00:45

of

play00:49

dorismar our young people that the world

play00:51

is totally screwed that it's hopeless to

play00:53

try to solve problems that all we can do

play00:55

is like sit in our bedrooms in the dark

play00:57

and think about how awful we are is a

play00:59

really deeply unproductive streak and I

play01:02

hope MIT is different than a lot of

play01:04

other college campuses I assume it is

play01:06

but you all need to like make it part of

play01:08

your life mission to fight against this

play01:11

Prosperity abundance um you know a

play01:13

better life next year a better life for

play01:15

our children that is the only path

play01:17

forward that is the only way to have a

play01:18

functioning society and there will

play01:20

always be people who want to sit around

play01:21

and say we shouldn't do AI because we

play01:23

may burn a little more carbon or we

play01:25

shouldn't do AI because you know we

play01:27

haven't fully addressed bias and it

play01:29

turns out a couple years later we made a

play01:30

lot of progress on both of those things

play01:32

and the anti-progress streak the anti

play01:34

like people deserve a great life streak

play01:37

who are usually the people that have

play01:38

quite a lot of privilege in the first

play01:40

place um is something I hope you all

play01:41

fight against

play01:44

yeah there has been a ton of activity in

play01:47

the AI space one of the big things to

play01:50

drop is that Microsoft rolls out its

play01:53

very own AI model to compete with open

play01:55

AI Google and the rest the new model

play01:58

internally referred to as May one that's

play02:01

ma

play02:02

ai-1 is being overseen by Mustafa

play02:05

Suliman the ex Google AI leader who

play02:07

recently took over as the CEO of

play02:09

Microsoft AI I'm going to call it my one

play02:12

not sure it's May one my1 whatever my1

play02:14

will be far larger than any of these

play02:16

smaller open source models that

play02:17

Microsoft has previously trained it's

play02:19

important to note that Microsoft created

play02:20

some very effective small models so even

play02:23

though my one will have roughly 500

play02:26

billion parameters that is much smaller

play02:28

something than something like GP 4 that

play02:30

has more than 1 trillion parameters this

play02:33

certainly does not mean that it will be

play02:35

bad meta llama 3 of course the currently

play02:39

biggest released model is 70 billion

play02:41

parameters and that is on par with GPT 4

play02:44

so Microsoft's new model could very well

play02:47

exceed far exceed GPT 4 potentially we

play02:50

we still don't know yet but it would not

play02:52

be out of the room of possibility

play02:54

they've been publishing a lot of

play02:56

research on how to create new models by

play02:59

taking

play03:00

the outputs of other models so taking

play03:02

synthetic data produced by other a

play03:04

models and creating smaller but very

play03:07

efficient models trained on synthetic

play03:09

data warut 2 for example is a great

play03:12

example of that so Microsoft is pursuing

play03:14

a dual trajectory of sorts and AI aiming

play03:17

to develop both small language models

play03:19

that are inexpensive to build into Apps

play03:21

that could run on mobile devices as well

play03:23

as the larger state-of-the-art AI models

play03:26

then looks like they're willing to chart

play03:28

A New Path that's separate from from

play03:30

open AI up this point we can assume that

play03:32

open aai and Microsoft would be kind of

play03:34

tied together their Fates interl but

play03:37

this is implying that Microsoft just

play03:39

does not need open Ai and we could

play03:42

potentially see this new massive model

play03:45

at the build developer conference later

play03:47

this month now if you saw my previous

play03:49

video we were all expecting that open AI

play03:52

would have a new announcement coming up

play03:54

sometime next week or this week about

play03:57

something new that was dropping

play03:58

potentially a search product something

play04:01

that would compete directly with Google

play04:03

as well as perplexity AI but right now

play04:06

it does seem like they're considering

play04:08

postponing that announcement we're not

play04:10

too sure why now recently Sam Alman has

play04:12

been visiting multiple universities and

play04:14

other places to talk about Ai and the

play04:17

future in general he visited Stanford

play04:19

MIT Etc and it looks like some people

play04:21

don't like this idea of a brighter

play04:25

better future driven by technology s

play04:28

posted this using technology to create

play04:30

abundance intelligence energy longevity

play04:33

whatever will not solve all problems

play04:35

that will not magically make everyone

play04:37

happy but it is an unequivocally great

play04:39

thing to do and expands our option space

play04:41

to me it feels like a moral imperative

play04:43

which one thing I like about Sam is he

play04:46

he's not afraid to say things that some

play04:48

people might not like to hear Prosperity

play04:50

is a good thing actually ddg growth and

play04:53

then he says I'm a good gbt2 chatbot and

play04:56

looks like opening ey just released two

play04:58

new models in the LM system Arina last

play05:01

night we have I'm a good gpt2 chatbot

play05:05

and I am also a good gpt2 chatbot the

play05:08

new one the also a good gpt2 seems to be

play05:12

stronger in reasoning and is even more

play05:15

capable than the previous gp2 chatbot

play05:19

release last week also looks like

play05:21

opening I changed the domain of Chad GPT

play05:23

from chat. openai to just chat GPT and

play05:28

just keep in mind as I'm going through

play05:29

this this this is all in the span of

play05:30

just a few days this is all like about

play05:33

48 hours worth of news so there's this

play05:36

flurry of activity on the open AI front

play05:39

so we're not quite sure what's happening

play05:41

are they preparing for this massive roll

play05:43

out are they delaying it semon's m

play05:46

making the tour across the country

play05:47

talking to people I mean something's

play05:50

bubbling up something's happening we're

play05:52

not quite sure what but yet another

play05:54

announcement is openi has an API

play05:58

partnership with stack overflow stack

play06:00

Overflow is the world's leading

play06:03

knowledge platform for highly technical

play06:04

content and they're partnering with the

play06:07

large language models for AI development

play06:09

so what exactly is happening well open

play06:11

AI will utilize stack overflows overflow

play06:13

API product and collaborate with stack

play06:15

Overflow to improve model performance

play06:17

for developers who use their products

play06:19

and stack Overflow gets to use openai

play06:21

models as part of their development of

play06:23

overflow AI which of course sounds like

play06:26

a win-win for both but some people are

play06:28

suggesting that opening I might be

play06:30

getting the better deal here cuz once

play06:33

they have all the data and models that

play06:35

are trained on all of this data as you

play06:36

can imagine AI model that can code very

play06:39

well would be very disruptive to

play06:41

something like stack Overflow open AI

play06:43

also post this our approach to data niai

play06:46

and they're launching this interesting

play06:49

thing they're calling a media manager

play06:51

for creators and content owners and the

play06:54

goal here is to respect the choices of

play06:56

creators and content owners on AI of

play06:58

course a lot of people have complained

play07:00

that opening eye trains on everyone's

play07:02

data things like mid Journey The Works

play07:05

of artists to then replicate the art at

play07:08

least that's their stance they're saying

play07:09

that's goes against copyrights and

play07:12

they're talking here about something

play07:13

similar to the robots.txt which the

play07:17

internet ecosystem sort of adopted for

play07:20

web Publishers right so if you have a

play07:22

website generally you're going to have a

play07:24

robots.txt file and there you kind of

play07:26

say do you want crabots like Google

play07:29

crawlbot crawling your site and indexing

play07:32

it to use in the search engines or do

play07:35

you want to just opt out now of course

play07:37

if you're running something online if

play07:39

you're a publication if you're a new

play07:40

site of course you want to opt in you

play07:42

want the traffic from Google so most

play07:45

people decide to opt in and here the

play07:47

same ideas for open AI to enable

play07:50

Publishers to express their preferences

play07:51

about the use of their content in AI do

play07:54

you want to use your content to train

play07:56

this AI what's interesting to me here is

play07:59

once they have something like this and

play08:00

they have and they have people opting in

play08:03

like let's say the New York Times who is

play08:04

suing open a for copyright infringement

play08:07

let's say they opt in or rather opt out

play08:09

of having their data used openingi could

play08:13

use that to just make sure that chbt

play08:16

doesn't output anything that resembles

play08:18

that data so if it starts producing a

play08:20

paragraph that's very very similar those

play08:22

outputs could be blocked which certainly

play08:24

would reduce the liability for open AI

play08:28

they are also introducing a new

play08:29

classifier to help researchers identify

play08:32

content created by Dolly 3 and joining

play08:34

the Coalition for Content Providence and

play08:37

authenticity so again a lot of this is

play08:40

seemingly in response to a lot of the

play08:41

criticisms that they've had for using

play08:44

content or creating things where we

play08:47

don't know what's real or what's not and

play08:49

this is their answer to various cyber

play08:51

security concerns and this is a beefy

play08:54

blog post that goes pretty deep into how

play08:57

certify various portions of the

play09:00

infrastructure for security including

play09:02

gpus making sure they're authentic

play09:05

testing them for integrity testing

play09:07

networks data centers using AI for cyber

play09:10

defense as well as AI specific audit and

play09:13

compliance program and their building

play09:15

and investing to achieve these goals and

play09:17

hiring people to work towards those ends

play09:20

they have a grand program as well as

play09:22

hiring people for certain roles in that

play09:24

field a few weeks ago we covered a

play09:26

presentation by one of the C eia ex

play09:30

directors I believe saying that AI has

play09:33

really increased the liability for cyber

play09:36

attacks the various vulnerabilities that

play09:39

we might have for cyber attacks not just

play09:41

because the ability to conduct those

play09:44

cyber attacks is growing with the help

play09:46

of AI but also and this is an

play09:48

interesting thing to think about that's

play09:49

not very obvious but back in the day you

play09:52

might have a bunch of random data out

play09:55

there on the internet without having to

play09:56

worry too much about it but now with AI

play10:00

a lot of Fairly harmless things that we

play10:03

would put out there in the past has

play10:05

become a lot more dangerous we covered

play10:08

this post by Andrew gal who used GEOS

play10:11

spy this brand new tool using AI That's

play10:14

able to determine where in the world you

play10:15

are just by looking at a picture it's

play10:17

able to guess that this is Slovenia and

play10:19

the specific church that you're looking

play10:21

at this it guessed as being next to

play10:24

Mountain View Campground and it's

play10:26

correct to within 15 miles notice

play10:29

there's no landmarks or anything to

play10:31

really identify this there's no text no

play10:34

road signs not too much information here

play10:38

it gets this location in Germany within

play10:40

40 Mi uh picture in Germany so this it

play10:43

was 4 hours off but still if you look at

play10:46

the picture there just isn't that much

play10:49

to go on here it's just a field of grass

play10:52

figured out this was Singapore Etc and

play10:54

many many more but as we put more and

play10:57

more information online that information

play10:58

can be aggregated fed into neural

play11:01

networks into Ai and potentially it

play11:04

would be able to come up with certain

play11:06

insights that could be used to conduct

play11:08

maybe attacks or scams or whatever other

play11:12

nefarious activity that even 5 years ago

play11:15

we didn't think about it because it was

play11:17

impossible meanwhile looks like we have

play11:19

some more leaks about Apple Ai and what

play11:22

apple is planning to do with AI in the

play11:24

future kind of upgrades are heading to

play11:26

the iOS 18 and key features include AI

play11:29

power text summerization across Safari

play11:31

and Siri smarter search results Etc so

play11:35

the big thing here is that a lot of this

play11:37

will be on device right so Ajax will be

play11:39

able to generate responses to some

play11:41

prompts entirely on Apple devices

play11:43

without sending anything to the cloud

play11:45

we've covered a few research papers

play11:46

posted by Apple they're really going

play11:48

hard after on device agents they're

play11:51

building these tiny Nano models these

play11:54

very small AI models that are some

play11:57

fraction of 1% % of what GPT 4 is for

play12:01

example that are able to interact the

play12:04

screen that are able to understand the

play12:06

context if you say select the second

play12:08

website on the screen it'll be able to

play12:10

kind of figure out based on the location

play12:12

on the screen what you're referring to

play12:13

so kind of that context of what you're

play12:15

talking about and everything because

play12:17

it's so tiny will be able to run quickly

play12:19

and on device and then there's this this

play12:23

little email that we have access to

play12:25

called thoughts on open AI so this is

play12:28

from Kevin Scott this is Ken Scott you

play12:30

probably have seen his face before he's

play12:32

the chief technology officer at

play12:34

Microsoft and he's writing to sich Nela

play12:36

the CEO and Bill Gates the the the Bill

play12:40

Gates I don't think he even needs a

play12:42

title he's just Bill Gates and he saying

play12:45

the thing that's interesting about what

play12:46

open Ai and deep mind and Google brain

play12:48

are doing is the scale of their ambition

play12:51

and how that ambition is driving

play12:53

everything from data center design to

play12:55

computer silicon to networks and

play12:57

distributed system architectures and all

play12:59

this other stuff but he's saying that in

play13:01

the beginning it seemed like all these

play13:03

people were just trying to play this

play13:04

game to see which reinforcement learning

play13:06

system could achieve the most impressive

play13:08

game playing stunt he kind of dismissed

play13:11

all those efforts now he's saying well

play13:13

that was a mistake when they took all of

play13:15

that infrastructure to build NLP models

play13:18

that we couldn't easily replicate so he

play13:21

was dismissing their little games but

play13:22

then they built something that made him

play13:24

realize that we're falling behind we

play13:26

can't replicate what they're doing then

play13:28

he started to take things more seriously

play13:30

and as I dug into try to understand

play13:32

where all of the capability gaps were

play13:34

between Google and us for model training

play13:37

I got very very

play13:41

worried turns out just replicating Bert

play13:44

large wasn't easy to do for us so this

play13:46

is one of the earlier Google models even

play13:49

though we had the template for the model

play13:50

it took us around 6 months to get the

play13:52

model trained because our infrastructure

play13:54

wasn't up to the task there were many

play13:57

years behind the competition in terms of

play13:59

machine learning scale sa comes back and

play14:02

says a very good email that explains why

play14:04

I want us to do this and also why we

play14:06

will then ensure our infra folks execute

play14:09

very interesting style of writing it's

play14:11

very compressed so this document is from

play14:14

us V Google in

play14:16

2024 and so here's Kevin Scott the

play14:19

person that wrote that original email

play14:21

responding I think he's responding in

play14:23

part to those emails and the news

play14:26

stories about this he's saying I'm not

play14:28

sure why this is news but just to

play14:30

summarize the obvious we build big

play14:33

supercomputers to train AI models our

play14:35

partner open AI uses these

play14:37

supercomputers to train Frontier

play14:39

defining models and then we both make

play14:41

these models available in products and

play14:43

services so that lots of people can

play14:45

benefit from them we rather like this

play14:47

Arrangement smiley face we've been added

play14:50

for almost 5 years now each

play14:52

supercomputer with build for open AI is

play14:54

a lot bigger than the one that preceded

play14:56

it and each Frontier Model they train is

play14:58

a lot more powerful than its

play14:59

predecessors we will continue to be on

play15:01

this path building increasingly powerful

play15:04

supercomputer for open AI to train the

play15:06

models that will set pace for the whole

play15:09

field well into the future there's no

play15:11

end in sight to the increasing impact

play15:13

that our work together will have and

play15:15

he's saying they'll be building their

play15:17

own models and adding them into products

play15:20

and there will be more of this going on

play15:22

in the future too some of these models

play15:23

have names like Turing and my some like

play15:26

fi for example we even open source so

play15:29

that letter thoughts on open aai was

play15:32

sent just weeks before Microsoft

play15:34

announced its 1 billion investment in

play15:36

openai and now about 5 years later

play15:39

Microsoft is building its own large

play15:41

language model which assuming it's more

play15:43

efficient than GPT 4 which a lot of

play15:46

these new models are they're very

play15:49

effective even though they have less

play15:50

parameters we could potentially see

play15:53

Microsoft supersede open Ai and Google

play15:57

and meta and correct me if wrong here

play16:00

but out of those other AI companies open

play16:02

AI meta Google Microsoft seems like it

play16:05

has the most straightforward path to

play16:08

making money with these AI models Google

play16:10

search engine we're still not sure how

play16:12

that cash cow for Google how it's going

play16:15

to intersect with AI is AI going to

play16:18

diminish how much search we do it

play16:20

certainly seems like that could be the

play16:22

case meta initially started using AI

play16:25

just for their internal purposes for for

play16:27

WhatsApp and Facebook and insta just to

play16:29

improve functionality they're not

play16:30

necessarily selling those Services

play16:33

Microsoft with all of their products

play16:36

with Azure seem to be like the most

play16:39

straightforward Revenue generating

play16:41

company when it comes to having their

play16:43

own AI model again correct me if I'm

play16:45

wrong am I missing something it's

play16:47

interesting to see kind of the Arc of

play16:49

Bill Gates who rode the personal

play16:51

computer wave kind of missed out a

play16:53

little bit on some of the internet

play16:55

Revolution certainly other businesses

play16:58

took much more advantage of though of

play17:00

that maybe missed out a little bit on

play17:02

the mobile Revolution right the top

play17:05

phones are Android and Apple Microsoft

play17:08

phones maybe not so much and looks like

play17:11

they were really falling behind in AI 2

play17:14

until five years ago Kevin Sai and Bill

play17:18

all got on a horn and said hey let's

play17:20

throw a few billion at this Sam Altman

play17:23

character and yeah let's see how well

play17:25

that bet will be paying off I feel like

play17:28

that may be the highest Roi investment

play17:30

Microsoft has ever had and if you've

play17:32

made this far then you might be

play17:34

interested in this and yes this is a

play17:36

product pitch so feel free to tune out

play17:39

but if you've been following this

play17:41

channel for a while you know that I am

play17:43

really interested in building these AI

play17:46

agents in building Automation in

play17:49

learning how to wield V various AI tools

play17:52

and AI capabilities to improve my

play17:55

productivity to improve my ability to

play17:57

have impact I mentioned before that I've

play18:00

been in e-commerce for the last 10 plus

play18:03

years so certainly using AI to help me

play18:07

run various Ecom businesses and help

play18:09

other people in that field I'm very

play18:11

interested in that as well as

play18:13

productivity as well as things like this

play18:15

this channel where I talk about Ai and

play18:17

try to teach Ai and how to use it so

play18:20

this forum is just a few days old and

play18:21

we're already approaching a thousand

play18:23

members hopefully we'll cross into a th

play18:26

in the next couple of days and the

play18:28

overall sense that I'm getting from

play18:29

people in The Forum is that most of us

play18:32

see that there's big things coming and

play18:34

we don't want to miss it we want to be

play18:37

on the bleeding edge of AI we don't want

play18:40

to look back on this you know in 5 to 10

play18:42

years and go whoops we missed it

play18:44

certainly I know if I did that I would

play18:47

feel pretty foolish I'm very fortunate

play18:50

that this channel is watched by and

play18:52

listened to by many incredible smart

play18:56

very highlevel people and for a long

play18:58

time time people in the comments have

play19:00

been asking me to organize something

play19:02

like this so that we can all get

play19:03

together and discuss and form ideas and

play19:07

network and think over this stuff

play19:09

together a mastermind group if you will

play19:12

and I've attempted a few things and I've

play19:14

kind of took my time of some of this

play19:16

stuff unfortunately but but no more here

play19:19

it is it's live it's working the

play19:21

discussions are popping off I'm rapidly

play19:24

releasing tutorials on how to build AI

play19:27

agents on how to use the various AI

play19:29

tools that I've learned and hopefully

play19:31

mastered and that might be a good

play19:33

jumping off point for people that are

play19:35

just getting into the space but my real

play19:37

big goal is to have the community be the

play19:40

really valuable and and rare resource

play19:44

the goal is to make this the biggest and

play19:45

most valuable private Community online

play19:49

the reason it's private is because my

play19:51

first attempt at creating this was

play19:53

creating an open form and there's a lot

play19:55

of problems with that spam Bots it's

play19:58

really hard to moderate and you have a

play20:00

lot of people that are necessarily all

play20:02

focused on the same goal the goal of

play20:05

this community is to learn to level up

play20:07

to grow I think that's why this thing

play20:10

that Sam said resonates with me so much

play20:13

to grow to create abundance is

play20:16

unequivocally a great thing to do and he

play20:19

says that to him it feels like a moral

play20:21

imperative and to me it does too so if

play20:25

you want to join in if you think that

play20:27

this is something that would be useful

play20:29

to you number one and number two that

play20:31

you can hopefully provide some value

play20:34

back to the world or to the community

play20:35

based on the things that you learn the

play20:37

things that you know then I will ask you

play20:39

to join me help me let's build this

play20:42

thing together with that said my name is

play20:45

Wes rth and thank you for watching

Rate This
★
★
★
★
★

5.0 / 5 (0 votes)

Related Tags
AI DevelopmentEnergy EfficiencyEthical AITech GiantsMicrosoft AIGoogle BrainOpen AIAI ModelsCybersecurityProductivity ToolsE-commerce AICommunity Building