Economistwomen.com thanks inspirational place leaders and next generation mapmakers
IN 2009 two leaders who cared about their people met; prince charles knighted engineer Fazle Abed for his first 40 years of moving over from being regional ceo of Shell Oil to servant leadership of valuing the intelligence of the world's poorest village mothers and daughters; fazle abed died in 2019- much of abed mooc builds on the privilege of visiting bangladesh 15 times during his last decade of celebrating what bangladesh women develop -he was always asking is there a way to share knowhow faster, deeper, more lovingly?... but lets first look where King Charles has got to in 2023 as he launches a world tour of intelligence connecting commonwealth and indeed wherever the Kings English Language model helps shine light of education and all the goods people can spend time data and multiply goodwill and mother nature round
imagine you (with students, teachers, loving families, communities) could tour the world of 500 inspiringly good human intelligences guided by king charles and 4 others who see intelligence as magic when humans joyfully unite- who'd you choose - try our tour then help us chris.macrae@yahoo.co.uk improve- perhaps in a different language than the kings english
28:43
i floating Point numbers we call them tokens yep and so these machineries these AI factories uh are completely different than traditional data centers all the things that we've said just now talking about accelerated computing uh completely still still hold but using that same technology formulated in a different way um uh and setting up factories for a different purpose we now uh imagine this new Industrial Revolution producing tokens well where are you going to where are you going to produce tokens where you going to produce tokens all over the world and uh the countries with excess energy there are many countries with excess energy there are many countries with a great deal of renewable energy are going to have a a a tremendous Advantage you can produce the intelligence away from where you apply the intelligence and then which which which now brings me to uh The Edge Edge question uh there there are two ways to bring AI to the edge one of them is remote sensing of course just to understand what's going on in the in the physical world ; the second which is even harder to do much much harder to do and much more valuable to do is called physical AI which is the AI leaves the leaves the digital world and enters the physical world and there it it can't just understand uh English and the meaning uh of uh natural language or digital l information it has to understand physical properties it has to and in that world uh hallucination uh could result in performing something that's physically harmful and so you we have to make sure that that you you could see it in self-driving cars you could see it in robotics uh you generalize that basic idea AIS can't just you know learn how to generate next token like an image it has to be physically plausible it has to obey the laws of physics and so um future AIS will understand the laws of physics like energy conservation and mass conservation and understand invariance and you know those kind of things
=========
31:06
Switching gears right like you talked about intelligence as a service having been an entrepreneur we were just joking um I came up with a terminology
31:18
uh let's see whether Gartner adopts it or not but it's catching fire um rather
31:23
than calling intelligence as a service we said why not call it cognition as a
31:28
service because IAS exists infrastructure is a service for it pass
31:36
for developers SAS for business users and this is the way it will
31:41
delivered very similar to intelligence as a service but let me switch gears I
Overlap between Nvidia and first 3 Calls of King Charles AI Summit
strange but probably true- abed served up to 1 billion women from bottom up
hunag serves all fture inteligences from edge
Jensen Huang founded NVIDIA in 1993 and has served since its inception as president, chief executive officer, and a member of the board of directors, building a $2 trillion company from the ground up.
Navin Chaddha, Managing Director, leads Mayfield, a top-tier early-stage venture capital firm with over $3 billion under management. The Firm invests primarily in early-stage technology and bio companies.
Both men discussed the future of AI, among other subjects as part of TiECon Silicon Valley's 2024 Day 2 festivities.
Jense I want to help advance understanding of Sovereign AI - w one of the natural resources of country is its society's data; you know historically we've looked at the natural resources as the what's underneath the Earth that belongs to that country but NOW & into the future please see in this new world the data also belongs to the country and so key transparency question for what reason (it ever) should the data be exported to another country for it to be refined and then imported back to you to then pay a premium on at some you know at at at some first
26:03
pin rinciple thar doesn't make sense so eg India should of course Harvest its own natural resources in in the form of Digital Data its society's data train its own models use it internally um and connect it to let other people connect to it for a service and so
26:26many countries should do that, that's that's one layer of thinking about it -- another layer of thinking about it is to take a step back and observe what is happening at an industrial level you know that the first Industrial Revolution : steam powered engines the production of of Machinery uh the second was the production of something most people couldn't understand stand we produced at very high scale electrons AC power uh most people at the time could not imagine that AC power would be something of so much value and so much social value and Industrial value and therefore Industrial Revolution the third Industrial Revolution we produced something even within our generation within this is within your lifetime of mine and a lot of people couldn't understand it andin fact there are many compan in Silicon Valley that never saw the value of what what I'm about to say um in the beginning that that the idea that something that has no weight no Mass um and doesn't come in a box uh would be would be valuable software if you remember the computer industry started from a time where the valuable thing is the computer is the box is the system not the software and it changed with what Microsoft did ; so nvidia has done that with Cuda the lock end to end lock that's right and and so that that last Industrial Revolution
28:06
discovered the production of software and in fact whole bunch of jobs are created software Engineers the whole
28:12
methodology of developing software whole bunch of tools and ecosystems that support the production of software and
28:18
now we're in the fourth industrial re Revolution and we're producing something that most people at the moment uh uh
28:25
don't understand and um uh there'll be new factories being created and we're
28:31
going to produce uh intelligence at scale and so so the production of
28:36
intelligence we now know how to do uh how to manufacture it how to produce it how to make it better how to Market ,,
Mayfield's Navin Chaddha, at TIE Nvidia is the dominant company - how do entrepreneurs connect? you have an Inception program you have a venture program what would you want to tell the entrepreneurs where you are playing and since you are a platform where do you want to grow the ecosystem which you talked about and if those layers get fixed of course application will get built and digital twins digital teammates will get built so these four layers where are you partnering to enable accelerated Computing to get to the next level and they're both one of your team members is leading who works with us the Inception program another team member is leading the Venture staff how are you going to help entrepreneurs right l
33:16
jensen we are recognizing that on the one hand um Ai and accelerated Computing uin the way that people want to enjoy Computing today is in the format of a data center um whereas hyperscale is about many small microservices um sharing and many users sharing one Computing resource in the future an AI Factory is one giant AI using all of the data center as a as a resource and so so this this way of doing Computing is is um uh radically different than the past and so we became a data center company however everybody's data center is different and we had the good we had the good wisdom of Designing our data centers in such a way that it could be be then disaggregated taken apart and then integrated into Azure and integrated into gcp and integrated into AWS so that our platform is everywhere the the great achievement of our company is is one of course pioneering this approach called accelerated Computing but making it available literally everywhere is one of the the great things that we've done is to enable architecture compatibility and software compatibility uh throughout and so we're in every cloud we're on Prem we're in PCS we're everywhere yep and and so first thing is for a developer is to realize that we have toolkits for you and you can run your software anywhere the second is uh reach out to our developer
35:00
program we have a developer program called Inception uh uh we have some 20,000 startups that we're working with
35:05
around the world uh we provide our capabilities to augment yours so that you have superpowers of a very large company and um it could be it could be uh infrastructure access it could be technology access uh we have lots of Engineers that that develop codesign algorithms with startups and large companies and startups and uh uh so we're we're delighted to help you learn about accelerated Computing learn about deep learning learn about U how to how to um train models and uh deploy models cost effectively across all these clouds uh so there there are a lot of resources including including of course go to market resources um how to celebrate you at GTC provide you a platform to tell your story uh to um
37.30 60 Direct reports so let's think about that - the reason why organizations were historically created the way they look is um so that militaries can go to war - information systems are incomplete, commands don't want to be questioned ie you want you want the general to be far away from the battlefield because you don't want them to see the pain and suffering of the battlefield and to make decisions that are strategic and you want people who are on the battlefield to not know how the decisions were made so that they could sacrifice their lives give their lives to country and Mission but that's
38:21
not our goal ; with intel accelerated by million times more tech - I don't want anybody to die in the field of battle for me and and I need all my employees and so the way you think about the w organizations therefore should come from first principles and and the question becomes : what is it that this organization intended to produce : what is the nature of the product and what environment is it in? how do you want this box to thrive? is the environment such that is fairly stable? is the environment Dynamic? is it hostile ? what are the conditions of this environment and what is the maturity of this box and so so you have to think about it from first principles and what I came to was was that information flowing in the company has to be incredibly High because we're in a in a highly volatile environment; if technology is changing by a million times every 10 years you know we've got to make sure that we're we're incredibly agile and people are up to speed or understand the nature of the decisions we're making because time matters a lot and so that's is to realize that that the people that are reporting to the CEO are all really worldclass experts I'm surrounded by amazing people, world class in their field they're incredibly talented and they're excellent managers and they require very little management ..if you you know they've achieved incredible results themselves and if that's the case then the way the CEO manages uh if you will the team surrounding the CEO should be very different
x
12.30
applications of accelerated computing - it doesn't make any sense that you could create One processor to do so and so we had to create a whole bunch of whole bunch of domain specific libraries that sit on top of an acceleration platform; and so life sciences uh manufacturing uh entertainment of course video games of course artificial intelligence of course robotics self-driving cars they all have their own domain specific libraries data; processing scientific Computing climate Tech' life scienes tech - they they all have these uh dsls that sit on top of our platform we have to go create them and once we create them we have to help the ecosystem take advantage of it and if they take advantage of it they get the benefits - so nvidia depends on world class parters in any context we connect inteligence (milion times more comouting deep data to)
13:11
technology that we've developed and it all occurs to them uh we know we know now uh uh in the
13:19
beginning the reason why people used Nvidia as accelerated Computing is it enabled them to do something they
13:25
couldn't otherwise do mhm uh today in in fact almost everything has
13:31
to be accelerated and the reason for that is because uh CPU scaling has really reached the end of the road and
13:38
yet Computing demand continues to double almost every year unless your your CPUs
13:45
are doubling in performance at the same power or the same cost every year you're going to have computation inflation and
13:52
and in fact we're seeing that all across the world uh people are starting to see computation inflation and so the thing
13:58
that we have to do is go refactor the application accelerate everything we can now one of the applications that we
14:04
accelerated uh discovered the the extraordinary cost benefits of
14:11
accelerated Computing in the final analysis what we do is we drive down the marginal cost of computing by Insane
14:17
amounts over time if you look at Mor's law uh over the course of over the course of the last uh 40 years in the in
14:24
the best three decades um it it uh reduce the cost of computing by a factor
14:30
of 10 every 5 years one and a half times every couple years right or a couple of
14:35
times every one and a half years so 10 times every 5 years um 100 times every
14:41
10 in the 15 10,000 so on so forth in the last 10 years we reduced the we
14:47
through accelerated Computing we reduced deep learning cost by nearly a million
14:52
times wow to the to the extent that that um uh people uh realized why don't we
No comments:
Post a Comment