DBS Economics & StrategyKopi Time E126 : Gen AI with Microsoft's Zia Zaman
We hear about the subject of the moment from Microsoft’s Asia head of business development, Zia Zaman. We begin with the question if the world has been transformed in a comparable manner in the past year and a half as the smartphone and Appstore revolution achieved during 2007/08. Zia argues the transformation has been similar in influencing corporate strategy and productivity enhancement potential. We discuss the unparalleled cost and investment associated with operationalising and commercialising Large Language Models. Zia walks us through multiple stages of tech introduction, adoption, and value creation, with most still in the pipeline for GenAI. We talk about the compute needs and carbon footprint of running GenAI models, and the role of regulation in balancing business and public interests. Finally, Zia talks about the depth and breadth of the relationship between OpenAI and Microsoft. Fascinating insights.
Hello, this is Copy Time, a podcast series on Markets and Economies from DVS Group Research. I'm Tamur Bank, chief economist. Welcoming you to our 126th episode
today. It's all about A I and to be more specific about generative A I, we will hear about the subject of the moment from Microsoft's Asia head of business development. Zia Zaman. Zia has had decades of experience on growth and innovation, both at Microsoft and previously at metlife Lumen Lab and a whole bunch of other companies here in Asia, Zia Zaman. Welcome to Kobe Time.
Wonderful to be here, Tamar. Thank you so much.
It's great to have you Xia. Um I'm gonna start out a bit provocatively. Uh The iphone was rolled out in mid 2007 and I recall distinctly that by the end of 2008, from the hardware production ecosystem to the smartphone app development business. Not to mention the way we communicated fundamentally, there was a full blown mobile revolution in the world.
Now, it's been about the same span of time since Chad GP T came into our consciousness. Late 2023. Uh has the world been transformed in a comparable manner.
I think it really has Timor and thank you for asking. I think the iphone moment uh was for me, the uh biggest and maybe the only major disruption we had in the 2000 to 2015 time frame. I actually think that those, that decade, the the the Knots was probably um one of the least productive times for tech.
Um we had tremendous growth in tech um as a result of the transistor of mainframes of client server of the PC revolution. And then we hit the doldrums and the single bright light was mo the mobile revolution. And in many ways, I would liken what we're seeing with generative A I and we know A I has been around for 50 or 60 years, but the generative A I
to be even greater in terms of the change that it might have on work on the way in which we live and the way which we commute, it's still a tool, but I liken it perhaps more to the 1994 Netscape Moment. Um Or even I see the parallels back to the printing press in terms of it, enabling a different set of language, a different way of thinking about
um how we do things in a different type of if you will, intelligence.
Are we doing things fundamentally differently already? Or you are talking in aspirational terms?
I think that there are so many examples of companies that are doing things differently already. And yet we are still so early in this journey. If there's one thing that I would like your listeners to think about is that we haven't seen anything yet. There is a rule which is sometimes attributed to Bill Gates,
which he actually didn't say it's actually the Amara rule, which is that we overestimate the amount of change that will happen in one or two years and underestimate the amount of change that will happen over 10 years.
And when you think about this, this comes from the institutes for the future, it has proven to be true over and over again, we are in year two of 10.
And in that principle, the year one and year two change is sometimes overblown. I worked for Gartner for three years. There is a hype cycle in terms of the expectations. But what happens as any technology matures is that you start to see the benefits
through the second half of that decade if you will. And in that year of 10 principle, I do want to say say that we have, we haven't seen the amount of change that we will yet. But you have to remember that the 1st 1st iphone people said look like an ipod with a screen
and it was only until the app store unlocked a lot of value that we finally get the explosion, not just of users but of different ways in which we use it and the App Store itself was from N TT Docomo. Um They were the first to establish something like that. So it builds upon, build upon build
and eventually what we will get, um is um the kind of change that I think we're talking about.
Um Yeah, I mean, I completely concur with you, but I'm just going to play Devil's Africa for just a second. Um which is, uh in the last few years, we've had a few other hype cycles around and I think it's kind of unfair to say their hype cycles because I think there's something genuinely foundational about those developments, but they just didn't become as big or ubiquitous. I've been thinking about the metaverse or the Blockchain around cryptocurrencies and so on.
Uh What gives you confidence that this wave of technology adoption will be far bigger to your point that if you haven't seen anything yet, it's going to be far more consequential.
That's a great point. And uh many of us have thought about the promise of um sort of decentralized computing and, and I, and I've done a lot of work around uh distributor ledger technology and what that might do and yet the obstacles to roll out for that particular technology um hit a wall and when it came to um
the metaverse, uh there is potential in the future as well. But for now, I think that what we're seeing is two steps forward, one step back, they are big unlocks um as will be energy. I think that the big unlocks around energy will open up fundamentally new economies
and opportunities for us. They may drive marginal costs down to zero. But when you really think about how different it is decentralized computing gave us a different model for thinking about how we might organize. Um you know, Cryptocurrency was kind of a bet against the US dollar in many ways um as the the reserve currency of the world. Um and the metaverse, um I'm a big fan of Ready Player. One watched it, read it. And I do believe that there's AAA cautionary tale in there,
but there's something dystopian about where some of those technologies are leading us to. Whereas I fundamentally think of this tool as maybe more utopian and unlocking more uh types of value. And, and if you go to the printing press, it unlocked a democratic way of thinking about the dissemination of knowledge. It gave us tools, the internet that is at the beginning, um that would allow more people to democratize. Um If there's a second thing that I want the people of this podcast to remember is how inclusive
some of the technologies are that we are going to help develop. And I don't mean our company, I mean, the people that are our partners that are doing great work on top of the platform. And if I don't give enough examples by the end of the hour around the inclusive nature of tech today. Then uh then stop me and I will try.
Uh yeah, you've touched on a bunch of threads. Uh Utopia, dystopia. I certainly want to unpack with you later, but you also talked about cost. So I want to delve into that a little bit. The numbers that I see from your company and a bunch of other tech leading companies. It's just mind boggling tens of billions of dollars, not just this year, the plan is to spend tens of billions of dollars every year for the rest of the decade, if not longer. So the investment going into operational and commercializing large language model is just
astonishing. Why is it so costly? Give us a sense of where the money is going.
Um Yeah, thanks for asking. I think one of the things that we are recognizing is it is so prohibitively costly to uh build and run these models that it, it cannot be done at scale by hundreds or thousands of people. It will
probably be a handful under 10 companies that will put in the tens of billions of dollars in order to build the infrastructure to train and deploy these models. And you, you think about that as um just the, the minimum efficient scale to, to, to run the model and that, that
floor just keeps going up and up and up. And it's partly as a result of the exponential uh benefit that comes from scale when it comes to learning. Now, let's remember what the large language model does is it makes a prediction about the next best word that is statistical. And when that statistical prediction gets better as a result of having more parameters or inputs, it just simply means that, that
N plus one prediction is so much better when N is large. And therefore we are in a bit of a race to see how quickly we can build out. Um the smarter and smarter supercomputer for lack of a better word. And that scale is daunting. When you look at the financial statements, you think about how are we going to absorb the Capex investment having been a software company for so long
that is really built on opex. We are now an opex and Capex company and yet I look at this as a personal problem is really that we have the benefit of being one of those few companies that will spend that Capex in order to deliver services that we can't even foresee in the future that will deliver value.
Um One way of putting it. And this is something that Bill Gates said, I think originally, which is the value of a platform needs to be 10 times greater for our partners than the value that we create, capture from it. And in so many ways, I think that the, the value that or the investment that we're putting into building this platform should clearly unlock significantly more than 10 times the value when you think about the broader economy. Whereas previous computing waves may have been limited
to more of an it sector, the the thinking about and nobody really knows about the addressable market for um the ways in which A I will unlock value in every sector are significantly greater. So as much as this Capex looks daunting it, it truly is what's required for us to be able to deliver the value in the back half of that decade. I spoke of,
right.
Um Yeah, uh allow me to go on a bit of a tangent. It is related to this issue of the very high cost of rolling out gen A I models, which is the cost of innovation.
Uh You know, in, in economics, we, we have this long standing sort of, you know, almost comical observation that technology is everywhere except in productivity that from the 19 seventies onward, we've had a bit of a downshift in productivity few times here and there, perhaps around the dawn of the internet age age, we saw some pickup in productivity. But by and large in the industrial world productivity has been disappointing. And one argument among many arguments is that it's just that, you know, we have sort of
exhausted the low harvesting fruits and now any tangible innovation is very, very costly. Uh Does it have to be like that? I mean, do you have to always spend hundreds of billions of dollars to bring prosperity to the world going forward because we're gonna scrape the bottom of the barrel already.
I don't think so. I think that this particular roll out is at a infrastructural layer which then enables people who are building on top of the platform to use less money to deliver productivity gain. Um I I just saw this morning an example of how we can take people who have lower no vision and give them through the tools that we have with GP T four and beyond
the ability to interact and see the world, unlocking a lot of that productivity gain that doesn't require very much in terms of additional cost on top of it. So once we've built out the railroad or the interstate system or whatever it might be that creates the fluidity in the economy, then the productivity unlock I think happens in terms of the innovation in every sector. So in financial services, it might be, you know, better Ekyc
or uh micro moment lending. It might be a deeper understanding of thin file customers. Now I'll use my first inclusion example, which is a company called Trusting Social. A partner of ours is thinking in Vietnam and beyond about how it might deliver those micro moments of credit through a generation of A I interface time war that gives us a deeper understanding about where we can do lending or provide protection products in the moment that's significantly better than the way
we've done it in the past. And I think that that ability to create value through productivity or thinking differently about our problem is where the initial investment um will seem like, OK, now, I understand why we built that. And now the things that we need to do to unlock in the future are going to be less. But I I your point is extremely well taken, which is, it still seems like we haven't gotten to the point where productivity is
accelerating. In fact, you might argue that it has been decelerating and that does go back to my earlier point about 2000 to 2015. I shrug when I think about the true technological events in science and engineering. But since about 2015 to this moment, these are really the golden years. And I'm very keen to talk about the idea around gen A I in science and in particular around the idea of discovery and design.
We don't know how the design of new things will improve as a result of the discoveries that Gen A I now enables us to make. And a simple example was around batteries. And if you don't mind, I'll, I'll share this story even though it was a bit of AAA Microsoft story, which is that we were working um on
how we could potentially figure out new battery discoveries and that would have taken years. But fundamentally with the PNNL, which is in, um, Washington State, we are testing a new battery material, right? And a fundamentally what we were thinking about doing was thinking about, um,
5 32 million different possibilities. The A I narrowed it down to about about um, 500,000 stable materials and then narrowed it down even further to 18 promising candidates. So fundamentally, it did all that trial for us
to go from 32 million to 18. And then those 18 could be built in the real world in the atomic world and then tested. And then one of them turned into a breakthrough battery technology that will then design something that will help unlock out
and that could have been yours. But it's now happened in days as a result of the A I basically saying, no, we're just gonna, you know, push our way through this particular battery design and our competitors at Google uh with the deep mind effort have always been one of the um the the shining lights in terms of the use case of protein folding salt significantly faster than we all thought. And we and a number of the companies are working at the protein level and then even at a a nano level to figure out how we might design
um new drugs and new therapies that will improve health outcomes. So I go back to this idea that the productivity gain that we might see in all the other fields when it comes from energy or education, from banking to health care are unlocked by this $200 billion of Capex that we're going to see
hit um the financial statements of companies like ours for the next little while. But we see this happen in history before and it has worked before and I'm not just being an optimist. I truly do believe that as a tool. This is something that has been greater than many of the other things that we've seen in a long time.
So it came to Eisenhower's Interstate Highway program in the fifties or laying submarine cables to connect the world through internet in the nineties and two thousands.
Yeah, I think that it, those are, those are good metaphors. Um And uh I, I think that the railways is also a good metaphor and, and any other infrastructure, the internet itself, you know, the dest development and, and to go back to the printing press because it is a tool right in the hands of humans to do things that we could not predict and education. And um and, and maybe even art
is, is super interesting and how it might be unlocked, uh which we didn't think of two years ago that the creative
side of it.
Right. Uh Yeah, I mean, I, as you know, I have an 11 year old, I'm excited about an infinitely patient. A I became a hard math skills which, you know, be unmatched by any teachers. Uh even the most patient teacher, that's what I'm looking forward to on the education side of uh Ja I application. Uh Yeah, tell me about the five phases of gen A I.
Great thanks. And I don't think this is necessarily the five phases perhaps, but it is again a way of thinking about this decade that we might be in and it has five steps to it. And without sharing this on the screen, I think people can visualize something that, which looks like an exponential curve where uh the X axis is time
or some sort of level of um of automation. And then the Y axis is value like just how much value do we deliver from it. And we are still in like I can call it a 532 model, but we're still in that first five years, we're in the current phase of a
uh the first two steps, there are conversational A I and maybe business knowledge, then we will move into the 2nd 3rd, if you will, where skills for work and then sort of reasoning and problem solving will start to kick in.
And then finally, we'll get to the, the last phase where we'll get self learning. Now, I'll step through these pretty quickly because, you know, I just, I recognize as a podcast and not a presentation, but look, the interfaces of the past have usually been mechanical or punch cards, keyboards, gesture interfaces like the mouse trackpad, touch screen. We are already starting to see ja I use conversational interfaces that are more natural and more human like.
And in many ways, we in Asia have known this for quite some time. It isn't just about typing into a window and then generating a response. We want our conversational A I to be Multilingual, multimodal, spoken in one language, you know, responded to it in another um over an Android phone, over a weak connection that's underpowered um on the back of a motorbike.
So the ways in which we in Asia demand A I to be used so that it is applicable and useful require a conversational approach and not full literacy or numeracy in English. And so we're getting there and we saw a couple of things that have happened the last couple of years, a couple of days even or weeks around how we've tried to think about this to be more conversational.
So how will it help? I mean, just basically makes it more accessible. Uh Someone once said English is the hottest new programming language. Where does this go is really in terms of the the idea around workforce, right? And the idea is that we are using this as a tool to improve our productivity. But in many ways, our time
hours of of the day are spent on doing tasks that can be routinized or can be done to summarize, to analyze, to a log uh to shorten creation time. And this idea that we're doing some of the same things, but in shorter periods of time
or with the assistance of some other type of tool, perhaps a little bit like a calculator. If you think of it that way, or another type of tool in whatever field that you're in, basically transforms and accelerates and assists the way we work to succeed in the modern work face, right?
But then as we go into the phases three and four, what we really get into is this idea that you get domain specific skills for work. And that is when an agent maybe takes over some of the things that you might have had a group of humans or tools
do and that agent can act on behalf of a company, a function or an individual to perform a task. It could be something like credit scoring or, or, or underwriting. It could be something as simple as as uh organizing a set of information and those skills for work will then start to be more advanced in terms of how they might reorganize the or chart. For example,
it gets exciting for me in that fourth phase where we start to do planning. And I don't mean we, but I just generally think that the tools start to do planning about what it is that they're gonna need to learn in order to solve the problem up till now, we've given them a problem. If they have that skill, they, they, they basically figure out how to do it faster than we did, you know, 99 faster than 99% of humans. But when you give them the ability
to plan out and figure out how they learn and reason in order to solve an open ended problem, then that's when you start, start to get to more of an advanced phase. And this is again, phase four of reasoning and problem solving. Finally, the last phase is self learning is when people, when, when the agents and when the artificial intelligence learns new skills that we didn't tell it to learn,
not just because we gave it a task, but because it realizes that it can and it develops a more generalized intelligence. So the G moves from the, from the beginning of the word to the center of the world and we enter into the artificial general intelligence space or
G I and that will unlock potentially a lot of different ways in which the marginal cost of many things drops down to close to zero. And I'm a big believer that humanity will still persist even in the presence of A G I. There's one thing I'm going to end with is that when we even get to that point where we have processes and tools and agents that do some of the more menial things,
um We'll be outsourcing these things to A G I but still valuing time more and unique human abilities. We may have a re appreciation for craft, for artisan recognition and reward for creation, empathy and mental, mental and physical health will spike. People will still need to seek help and seek support. We love competition, we love love and the enjoyment of drama. I think that we're going to have an adaptation of a
solution of what it means to frankly, to be human and maybe our, our standards for what we want out of humanity will rise in 10 years. Another way of putting it is that the things that we thought were just sort of very difficult to do 2030 years, maybe not in our lifetime or our careers. Now, all of a sudden they could happen
57, 10 years earlier and we might be living in a time where we solved nuclear fusion. One of my favorites either through a new Telcom Ma reactor design or through some other types of means we accelerate the process by which we come up with the discoveries we need in climate tech to solve some of the issues around energy sooner and that,
that optimism that we unlock some of these things that we don't even know about yet. Through A I and through the combination of humans using A I in interesting ways in each of these verticals, gets us to that point where we're solving big, big problems like education, like health, health outcomes like inclusivity and of course, like energy uh renewable energy sooner rather than later.
Sure. And you know, when you were talking about the final phases of A I, I was thinking about Iron Man and how Tony Stark interacts with Jarvis uh where Jarvis is not necessarily all knowing but is an incredible complement to Tony's
brilliance and, and helping him solve problems in, in speed. That is not humanly possible. So I'm not really sure Jarvis needs to have a G I or a very advanced form of J I is good enough. But uh but that's what I was thinking about when you were talking about that.
You read one of my previous slides, I have a slide that says, and it's got, it's got Tony Stark saying it will be more like Jarvis and will be like terminator.
And so you, you, you read my mind on that one and I'll throw out to the team like I love to end this particular phase of this discussion around genius. All right. And in, in, I'll ask the audience to, to think about what is the next best word that comes to their mind. When I start with this particular line, the first two words are misery loves
his company,
his company. So I've done this in a variety of contexts and some context of younger people don't know what the word might be. But the suggestion that it is company is a moment of genius
from when it was first written in 1600 by William Shakespeare. And since then any tool, any reference, even any A I is just holding a mirror up onto what it is that we have done and saying that the next best, most likely word for Misery Loves is company. But until
the Great Bard wrote that word, it would not be suggested amongst the 10,000 English words as the next best answer. So we will still have moments of genius that will then embed themselves into the corpus from which the A I can then say maybe the next word is company.
Very good. Uh Zia, you talked about one of your dreams being, you know, uh clean nuclear fusion and so on. Well, we're not there yet and therefore, right now, we're still burning a lot of fossil fuel and some renewables to power our current and aspiring world. So let's talk about the compute needs and the carbon footprint of running gen A I models.
There's no doubt that they are high and that the potential for them to grow at less fast a pace is clearly there.
So we care very much at Microsoft about our scope three emissions. Of course, scope one and two is super important and those have dropped. But scope three continues to rise and it is as a result of the fact that many of the places in which we we we get components for the supercomputers and the things that we need to build are still using grids that are based on non renewable energy. And in particular, more than 50% of most companies scope three emissions actually come from right here right here in Asia.
And as you look at the different grids in China is doing super well. Um We are starting to see an inflection point when it comes to renewable energy. And um recently, we've seen in the West and in countries like China, um that tipping point where the cost of renewable energy gets to the point where it starts to be more affordable,
but we're by no means anywhere close to that point yet for the rest of Asia. And so what we are hoping to do is to ask our partners in markets where we need to build a data center to think about the ways in which renewable energy can unlock the decision to put a data center nearby.
Put it differently if you were making a decision about where to put a data center that's gonna suck up some of the earth's resources. And there was no sovereign data issue and we had zero latency and we had the ability to transport electrons from one part of the world to the other but not electricity.
You would always locate your data centers next to a place that could generate uh energy renewable and cleanly rather than not.
And yet it is the distortions of sovereign data and obviously geography that make us think differently. But in the competition for where the compute and where the intelligence of what it takes to take advantage of A I uh occurs. The solution is find a way to create more renewable energy. And you will find a way to be more convincing as a destination,
not just for D CS, but for this whole A I wa and I firmly believe that um A I is unlocked faster with renewable energy and we are consciously trying to hit our 2030 commitments. We and many other companies like ours. And the only way we're going to do that is in partnership with large public sector entities
and private sector entities that are delivering renewable energy at scale so that we can eventually get there. I think we may be the first company to ever sign up a power purchase agreement with a nuclear fusion company. Um That's starting, it's due to becoming online in 27 and, and I I may be wrong in terms of whether it is fusion or fission. But it's super interesting to see that if a large buyer of an energy says I'm only gonna buy renewable energy or a mix that's mostly renewable, it changes behaviors, it changes economics of the market.
Yeah, for sure. Um You talked about Asia's energy needs. Can we just talk about the US energy needs for a second? Because my understanding is a lot of the data centers, a lot of the compute will be in the US. And I've been reading articles about how in vast parts of the US energy demand is,
is outstripping supply. California is an exception. But there are others where this issue is becoming a bit of a, you know, conundrum for those who want to pursue as fast expansion of gen A I related models and compute and those who want to sort of balance it out from a carbon footprint perspective.
Yeah, I, I'm not as familiar with the state by state potential for renewable energy as I probably should be, but I will go back to the idea that it is lumpy, right? Uh The future is already here. It just isn't evenly distributed yet. And if you think about the simple logic of hours of sunlight
and draw that as a curve, it may closely match the potential for data center or deployment uh with the exception of, of, of course of wind. But fundamentally if you look at the southern states and the ones that have more um sunlight,
it tells you, I think it gives you a clue as to where PV, which is actually getting more efficient at a close to a rate if not faster than the uh compute power required gets us there. Um And I think that gives us, gives me promise for the global South Time War. Um places like where you, you and I and our parents grew up places that we have visited where you think about wars are going to be
potential and you know, we just announced a billion dollar investment through G 42 in Kenya. Um And there are places like Iran and Pakistan, there are places that are, have many, many days of sun that could potentially turn that into
uh future power. So answering the question simply, it's probably PV, that is a, a clue as to where the potential will be. But what you said about America is probably true about every other nation in the world with few exceptions that just simply have exceptional renewable power potential.
Um Earlier, uh you, you made that point about Jarvis versus Terminator. I want to come back to that point uh because there's a broader discussion about the dystopian A I scenarios and the utopian scenarios. So what state of future do you think about uh Zia on a day to day basis?
I, I do believe that
we will have a guarded stepwise approach to how we think about these bigger things around A I principles around fairness, which I've somewhat alluded to this idea around transparency and accountability. If you will, we, whenever we put out a product at Microsoft, we goes through a second review around reliability and safety.
Plus, we have more safeguards around privacy and security than we ever have. So the principles that we have around how we deploy
are going to act as a natural governor for how we do it. And why we do it. And I think that that is something that we will share with the industry, with our partners and with other tech companies and how we'll do there is we're going to do skilling and we're going to do testing and tools and training to make sure that the way it's implemented is the right way. And of course, there's oversight around compliance and auditing, reporting, monitoring and we're going to build tools to help people figure out just how they're using A I
I think though that where it's going with the root of the question is um this idea of optimism that the optimism in terms of how a new group of agents that's a billion or 2 billion strong that you don't have to pay any money. But then basically deliver work that nobody wants to do that just simply creates value in places that then allows us to shift
to other activities. It's really the, the, the most interesting part for me and I will go back to health outcomes because it's something I've thought about over the years and my personal family issues around health and how we might improve the way in which we look at orphan diseases or even simple things like um early detection of cancer. And we've seen so many companies that look at and see the hardware infrastructure we have today when you apply smarter software and A I to it will provide an assist. It's a great word and assist
to a well meaning thoughtful technology oriented doctor to be able to say, you know what, I don't see a shadow in the JPEG version of that file if you will. But if you look at the raw version of the of the file of that image that was created, it went back to like the DS LR metaphor in that raw image 99 times out of 100. That thing that you see is actually a shadow and A I thinks that it could be something that is worth investigating further. Doctor, here's an assist
and that level of Jarvis like augmentation of what you could potentially identify at scale through simply imaging and diagnostics around health care outcomes starts to change. Um Your not just the way of work but the way in which you think about maybe even longevity and mortality and
the the delivery of health care in variety of different contexts. These are the reasons why I'm just, you know,
much more optimistic than I am any other adjective. We can't quite understand how some of our best partners. And it's actually the biggest customers are thinking about using these tools to come up with something that's really quite exciting. And I'm going to go back to creation just for a second because I think that the most exciting part of around one of the most exciting parts about 8 to 10 years from now,
it's gonna be just a wide variety of content and craft that we will be able to see. So there was a Korean web lit, think of it like a like a animated web drama. They used to put out an episode every two weeks immediately after the, the launch of gen A I, what they were able to do is to use background shading and other types of tools to go from an episode every two weeks to two episodes per week.
So it's a four times throughput of the amount of content that they could create. So you could be more imaginative and focus more on the story, more on the narrative, more on
maybe even the merchandizing. But the truth is that that productivity gain, the forex just comes from basically doing things that humans used to have to do in background shading and now get delivered so that he releases the people as it takes to build that story that that drama into doing things that are perhaps even more creative. So we will live in a in a horn of cleaning when it comes to content, we'll have a variety of different ways of thinking about um
how we might use this tool to create more. And yet there's gonna be supples along the way where we ask ourselves, wait, are we being inclusive? Is this fair or is this, you know, we have the right accountability. So I don't think that it's wrong for us to constantly check to make sure that we're doing all the right things but that the optimism I have for how it might unlock in the real world is rooted in things that we're already seeing in this year to upend.
Sure. Yeah, I, I don't think we have a problem of too little as far as content is concerned. We probably have a problem of plenty already, but I, I hope that we get a quality quantity optimal mix uh going forward. And while I fully appreciate that, you know, companies like Microsoft are trying to instill without any nudge necessarily from positive
a system of governance and internal uh sort of, you know, cross checks. Uh It is true that, you know, through waves of transformative technology, when the disruption takes place, regulators and regulations sort of struggle to keep up with the pace of change. And I think the last 20 years with the internet
revolution to the social media proliferation, I think there have been failings at times to protect consumer privacy, fake news, election interference, et cetera. So how should we balance the imperative of the business to go out and innovate and come up with products for the customers and public interest which are not necessarily always aligned?
Yeah, I think that you refer to some really big disappointments and you know, speaking personally about what the internet could have been. And uh I think that there is uh an element of being thoughtful around safeguards that as an industry plus the public sector we could have done to make sure that we protect, um, Children or, or, uh, identity
as we go forward. I think that working together on a set of standards with the industry and governments in lockstep is important. And there's this data that came from Oliver Wyman. I'm looking at it right now. It was, it was done with across 16 countries and I don't mean it to refer to less regulation is good. But the, if you look at the question of how often are you using generative A I in your current job?
Three out of the top five are right here. India, Indonesia and Singapore are 13 and five on that list with UAE in Brazil, um being two and four. And so you look at that, you say the, is it the global South? Is it a light touch regulation or is it optimism again? Is it an optimism that I hear in India can basically improve my pro
activity by using a tool in Indonesia? I can basically become Multilingual immediately by using tools or Singapore or UAE leaning forward in their governments and saying we need to experiment to think about ways in which we can embrace this tool versus I hate to say it dead last on this list. Do you have any guesses? Tw
don't
say us
Canada
really?
Us middle of the pack 46 Canada is the worst Australia. Close second.
UK third. Why, why so much suspicion I asked my Canadian relatives all the time. There's just this
worry about. Oh, you can't use that at school. You can't just instantly. Do you use a calculator? Do you think about research? Do you ask it some questions? I have a Canadian friend. We, we, we both work for, um, big tech companies now outside of Canada. But we asked the question about something that we've been thinking about since we were teenagers. And it's so interesting how general de comes up with a better mathematical answer to that today than when we were young. And we're using simulation and other tools and just full disclosure, he said, Google
and he asked the question, do you think this is 90 better than 99% of humans? I said, I'm not sure it is yet. And then he said, you know what it must be because it's better than you at solving this problem. And I laughed and I said, yeah, you're probably right. And so the problem
is something that you might see at work, it might be something that you see in school, it might be something you see in your research lab. But if you're not using a tool that just hopefully helps you solve that problem a little bit faster than what are you doing? So you can't shut it off. You can't say no, we're not going to have generated A A I, you have to think about this tool in the right context with the right safeguards in order to unlock the economy. So, you know, I'm gonna get a lot of like flame from friends in Canada. But we need to think about why is it that
Indonesia are one and three? What is it, the UAE is doing? That's gotten our attention and certainly Singapore as well. And let's not forget, let me just count this all China is uh eighth which is pretty good out of 16. Um And, and there seems to be a cliff after China where 60% of the people use it according to this Oliver Wyman study. So not sure I fully answered your question, but I say embrace it. Have safeguards, don't be afraid of it because it does actually help you do your job
fascinated by that survey. Very interesting. Um I want to ask you the question. Maybe I could have started this whole conversation with because these days, Gen A I open A I and Microsoft are almost synonymous. So tell us the depth and the breadth of the relationship and what's next in this collaboration.
Yeah, I think the big takeaway here is that Microsoft has multiple partners in generative A I. And uh we have made um
partnerships with companies like Mistral in Europe. We saw G 42 UAE, not just the language model companies but companies that are doing great work on top of the model as well. We made an investment in a company called hidden layer that does cybersecurity to help protect your model, to make sure that it is not assailed and corrupted by uh bad actors who will try to taint what it is that you tell people is the next best word.
But it does all start with our first um uh relationship with open A I many years ago. Right. I mean, it wasn't just a year and a half ago. Um Our first um partnership with them was more, let's gather a signal from this group of people that are uh coming out of Y Combinator. And, and then Sam started to run the company and things started to take off with the Transformer model to the point where we had a moment where we had to make a decision as to whether or not we make that big bet.
And that was just before Chat G BT, we made the big bet that big partnership with Open A I. And then since then, we've made other bets which are also very big around how we might think about generative A I more broadly. But certainly the chat GP T moment should not be underestimated for what it unlocked. And we continue to be not just on their board but also the beneficiary that every time you do a search on uh or a query or uh prompt
uh or on open A I, it hits our servers in the back end. So we obviously a beneficiary of being the one that does the supercomputer. But that's the case for so many other models as well so that we can be in a not quite frenemy but in a collaboration and cooperation's like competition with even
other models. And so it isn't perhaps as synonymous as you might say. But um we are clearly uh AAA strong partner of theirs and a beneficiary of when they innovate, we benefit and everyone benefits. But there are other sources of innovation uh back to, you know, it's just unevenly distributed
a year from now. Will you and I be able to do a podcast in English and have it simultaneously capacitated in Bengali and
it already
is in place. So we invested in Saran, which is a company that um
we put some thought into Sara. I'll be a little bit more clear. Uh and some resources into a company that basically helps us translate into 19 indic languages, I think right now, indie Bengali Urdu uh are probably already covered by our live teams app. So this is on Zoom, but it had it been on teams. It would have been real time Bengali Bangla uh translation with voice if we wanted it to be
um I use this tool perhaps more than any other tool, just the ability to say. What did I miss? Uh when was the name Zia mentioned? Did anyone tell a joke?
Did anyone tell a joke about Zia and then like, it's just so interesting to think about how we can very quickly query in, um, what has been said and, and, and understand it independent of language now. So we're already there. Um, if the, if you say what's the one year from now thing that you think might be unlocked that we won't be able to do? I'm gonna go back to teaching. I think there's a possibility that something that our 11 or 14 year old learn
will through the tool will far exceed
what it took us many, many years of education and living on this planet to get good at. And we will be shocked at how quickly
our kids are able to do specific things for which we have absolutely no clue how they learn how to do that. It is an element of magic to all great innovation, right? And uh the ability to use this tool to do so many new interesting things that I think is unwritten.
I am waiting for that infinitely patient and infinitely empathetic instructor. I'm really, really waiting for, not just for my son but for myself as well. I'm sure there are a bunch of things that I could learn. Uh And I would need a very patient teacher for that. Yeah, I so much appreciate you coming on the show and this is a very, very valuable insightful conversation. Thank you so much for your time and insights.
Thank you so much for having me. I really appreciate it. It's long overdue. Thanks again. Thank
you and thanks to our listeners and viewers as well. Kobe Time was produced by Ken Delbridge at Spy studios, Violet Le and Daisy Sherma provided additional assistance. It is for information only and design
represent any investment advice. All 126 episodes of Cookie Time are available on Apple, Google and Spotify, as well as youtube as for our research publications, webinars and live streams. You can find them all by Googling D BS research Library. Have a great day.
