About 50% Of Jobs Will Be Displaced By AI Within 3 Years

hu and I have been talking backstage and we were deciding should we give you guys some sweet dreams tonight or some nightmares and we decided maybe a little bit of both uh we have a lot to discuss what the future looks like um what we're building right now and how it will impact us in the future um so really excited for this conversation K thank you so much for being here thank you Alison so you've said AI is the biggest breakthrough for Humanity um it is also the final step to to understanding ourselves so I'm

curious Kyu why do you think that and what exactly does that mean to understanding ourselves I think I wrote that in my PhD application in 1983 and yes I at the time I thought uh once we figure out how once we build an AI we'll know how we think and and that was what drove me to go into AI at the time uh was basically AI winter but what I found out today is I think we can build something much more powerful than human not necessarily AGI but much more powerful than human and it doesn't

have to follow exactly our brain so so the good news is we can build something much more powerful than I thought the bad news it is that it may not give us as much insight about how the brain works we still have to rely on the cognitive scientist and the Brain uh scientist in the world so wait are you saying that there's something Beyond AGI I thought artificial general intelligence was like the end all be all end point when are we going to reach that and what's beyond yeah AGI is defined to be a superet

of human intelligence everything the human can do AI can do I think it's a very narrow-minded and narcissistic view because we are humans so when you watch movies you want aliens to be aliens want to become humans pets be want to become humans monkeys want to become humans but they don't you know AI is this giant machine that gets better with more gpus and computes and data it can do so so so many more tasks so much better than people but it doesn't mean it does everything people do because their brain operates differently from our

brain you know we don't compete we don't compare um you know marathon runners with cars right um and I don't think we need to compare AI with us even though it is exhibiting a lot of intelligence so I'm certain so if we look at what can humans do and what can AI do uh five years ago humans can do a lot AI can do a bit there's some overlap um and today what AI can do is gotten larger than what humans can do but but it's not a complete superet I think in the next two

to three years uh if the human circle of what human capabilities is this big uh you know the AI will be the size of Earth but it still may not do what everything we can do it it may not have awareness love um empathy compassion um or other skills so something super human greater than us all that has no compassion it sounds like a great great world we're building great and it can fake compassion oh god oh boy okay so we have a lot to discuss you have startups you have a startup you're running you

are a VC have been investing for years you've also worked at just about every major tech company Under the Sun um who's a leader in AI so lots and lots to dig in there into there but first I want to talk about 01 a okay uh your startup which you're the CEO and the founder spun up about exactly a year ago you had no team today your a billion dollar valuation no Revenue some Revenue some Revenue some Revenue some revenue on track for Revenue more Revenue um so what is it and what is its place

in this AI world uh yeah I I felt that um you know people in China cannot access chpt um actually CH many of you may not know this but um open AI blocks China from ability to access so I feel that China shouldn't be left out of this Revolution and um I felt that also I strongly believe as I stated in the AI superpowers that uh us will lead in breakthrough Innovations but China is better at execution and I thought a year ago the time has come for China to prove its prowess in execution and

rather than investing in someone I would do it myself this time so that was the initial attempt uh the other is kind of a frustration that the field is moving toward more closed despite its namesake open AI is not open at all it's probably the most closed company in the world even compared with Google and uh Gro and meta and and and others so I felt you know uh we need to work with academics we want to open with work with open source community and entrepreneurs and people like yourselves and that we should not only

make AI work great and make it create value but also make it accessible to everyone and if the best company in the bunch is closing all of its Technologies never open source don't publish then we can't engage and bring in all the brilliant people in the world so we decided uh we would also take an open approach which is a bit unusual because you know uh us is known to be you American companies are known to be more open than Chinese companies but we've taken an open approach uh to date every best model we have

built from tax model to multimodal model we've made available in open source their on hugging face and other sites because we felt we wanted to um change the way people think and make it accessible to more people and then maybe lastly what's unique is that uh I feel that this I'm I'm kind of seeing deu that what happened with um PC and the phone I'm seeing again with AI I know we're people love to talk about AGI but practically speaking about making money it's very analogous right PC was uh creating a stack from CPU to

uh to to operating system to applications and then to servers and to cloud and then to the end user and to business and to C and to be basically that whole stack made Microsoft the amazing company and made them incredible amount of money and the phone more or less did that for uh Google and apple and I think the same opportunity exists but but I feel a lot of the llm companies on out there are run by researchers who care only about making a great model and I think that science fair phase needs to end

um no matter how much Brilliance and great demos you make uh at some point there is a uh a point of Reckoning when investors are going to say what do you have to show for yourself what's your p&l what's your Revenue what's your growth when do you break even and I I've learned a lot of that myself having been a researcher from the beginning but now that I've been a VC for 14 years I feel like um I can make this thing make money so we're not spending a lot of time on it but we're

building the company with the ambition that this company will have have infrastructure model uh applications and also data working as a stack and we've come up with a number of ways which we think in the future will make money and we don't do it because we want the money but we do it because we see the need to continue to raise money to afford our gpus and we want to do it by not creating a bubble or promising the moon um uh or through demos we want to do it through through actual Revenue growth and

profit over time sounds very expensive sounds like something Venture Capital alone cannot fund I mean you have Google with gazillions of dollars at their disposal meta as well I how do you even compete in that World um to give you an example um when you think about gpus and models most of you probably just think okay he's they've got so many gpus then they can do great but actually Google has over 2,000 people just in its um deep Division and those 2,000 people are competing for gpus and resources and Google's uh approach is let you

know 100 flowers bloom and that's why open AI leaped ahead by betting on one approach uh with as much less GPU than Google had years ago now they have both have a lot so um I I'm think I I think the approach we think is we want a very small AI modeling team and a very large infrastructure team because if you have too many researchers and a culture where everybody can try ideas you'll quickly run out of money as a startup as you said but if we're very focused we want these researchers to read every

paper to understand every technology to discuss things um theoretically argue from first principles and then build some experiments to fight them fight it out and then before we really spend the expensive GPU we will have either reached consensus or I would have to make a decision so this is a very different culture than what Google and other companies try to do we're basically trying to gather lots of data and make um key decision so we don't take as much GPU as an example you may have read a paper by Steven Levy about Huawei and the

polar code I won't go into details here but basically Huawei asked this Engineers to read all the papers and they found this Turkish Professor who invented something called the polar code and that made all the difference and gave huawei's leadership position in 5G we're taking that same approach to be very very diligent to save GPU then the infrastructure is one of the least appreciated but most important Technologies because these these gpus don't work very well they break 4% 4% of the time each GPU breaks so if you have a cluster of 10,000 gpus half of

your month is gone because the cluster every time a GPU goes down the whole clust cluster goes down you've heard Jensen hang talk about how Blackwell has ways to recover well we can recover today by patching H H 800s that we have in in the company and also uh there's a metric called mfu which measures the uh model flop utilization um most llm companies are around 40% um you know Google Nvidia are slightly better but we're at 63% so we basically you know I think I truly believe that uh necessity is the mother of innovation

and the necessity for a poor company like us compared to a Google open AI Force us to be diligent about how the approach to take to be decisive about which approach to take and also to build a large infrastructure team to reduce the cost of computes because we just don't have that many got it okay um well big journey ahead um you're also called the tech Prophet so I want to get some prophecies here from you while I have you but let's look back at you've written two books um both sort of predicting the future

in their different ways superpowers um was sort of the US versus China and who will win um who will win uh where are we now where are both and were you right uh the our last last answer to your last question is of course yes uh the prediction I made in my book was was First Data is the new oil that makes all the difference and I think look what generative AI is right take all the data in the world to train using generative as a way to uh solve the objective function problem second prediction

in the book was that us was better at Innovation breakthrough China was better at execution and we saw that in the earlier AI days and in the recent Genai I think we're just about to see it I think we'll have to see how this works out but certainly uh taking my company as an example we were eight years behind a year ago now we're probably less than one year behind of the top American company so at least so far we've been executing of course the last one year is the hardest right because think about uh

what open AI has achieved in one year so we we don't view it uh in any kind of um um hubris or um uh taking for granted but we have closed the Gap because we did execute better in the past year uh in my company and in other Chinese companies um so I I believe in that still the other point I made was well when can us have an unsalable advantage over China and the point I made in the book was when the technology became invented by some corporate company uh by some corporate entity not

academics because academics were published and that corporate entity chooses to stop publishing which we are seeing now so it's definitely plausible that us can extend this leadership because open Ai and to a lesser extent Google have stopped publishing um but but we'll see so are China and the US headed on different paths here will they even be competitive in their a respective AI markets or will there be the China option for the world to engage with on on AI and the US version and then if that's the case how the heck do you regulate this

all if they're separate things yeah we're well beyond that and we're in our parallel universe this is not ideal none of us like it I think but it is what it is and given it's a parallel universe I think we're going to uh see a lot of interesting American solutions that won't make it to China and interesting Chinese solution that won't make it to us so as an entrepreneur or VC or just a curious mind it would make sense to look at both worlds to see what's exciting in each one because you'll while you compete

in your world whether it's the American world or the Chinese world if you have the advantage of actually having studied the other parallel universe you will have superpowers in the work that you do but we don't have have any unrealistic uh expectation that uh us and Chinese AI companies will actually compete um in the same country unless it's in a small number of countries that are friendly with both countries um when you look at the world's biggest tech companies today uh you've worked for at least half of them Apple Microsoft Google who is still on

top in 10 years or even 5 years Who falls and which uh startups kind of overtake them yeah um I think on the US side um I mean right now Microsoft is a darling and um uh Apple's doing stuff and uh Google is uh frustrated and a lot of negative uh comment and then open AI is the upst starter I would say that despite my earlier comments on open AI I am very bullish on their future uh they've really done an admirable and unbelievable job executing um even today GPD 4 is still the gold standard

you see Gemini Ultra and cloth 3 and make these claims but if you use these models GPD 4 and gbd4 turbo is unbelievably good and a great balance for uh performance and um uh cost so I I I I would say um you know open AI will like be a trillion dollar company in the not too distant future how distant or how not so distant two three years trillion dollars in two or three years I think that's the likely outcome of course they could Mis execute or some some other company could do a great thing

but I I despite my concerns about their lack of openness I have great admiration for them so I would if I could invest in any of them which I can't but if I could I would do open a ey uh Nvidia is another one that's a safe bet it's obviously extremely expensive but the processors they have and the Cuda and the um Technologies they have on um um the the ecosystem and the libraries it's very hard for a competitor to dislodge them they're obviously very expensive but um you know most of you know in secondary

stock you uh you buy high and sell higher rather than Buy Low you none of these companies are going to go low Microsoft is exe exec it brilliantly um I am a little bit disappointed at the Microsoft co-pilot because it is gluing gen to an existing product that is a dinosaur that should be thrown out uh but I understand why they did what they did but I would rather see a new product that throws away Microsoft Office and that has AI do most of the writing and human just kind of sit in the back seat

twe tweaking it a little bit uh but nevertheless I think my Microsoft's done an amazing job um on selecting open a and partnering with them and I think Sati has demonstrated unbelievable leadership in trying to bring in uh Sam Alman and then when that couldn't gracefully letting him stay and now getting um Mustafa it's just amazing to see uh SAA has been the most phenomenal uh CEO uh Google I'm still somewhat bullish uh despite all the issues we all know um it is still has currently has the largest density of AI talent in the world

by far uh more than open AI more than Microsoft more than the others so I guess the question is can they start to execute better so um be that as they may the incumbents are certainly in great position each of them um with their War chest of cash but it's a new era for entrepreneurs as well yourself included um if you were well you are an entrepreneur where do you start um if you want to get into this AI world and what kind of wealth creation are we talking about here I mean wealth in equity

is already so vast you're talking about this company open AI can be worth trillions almost overnight a couple years that's very fast what is the trajectory here um for wealth creation for capitalism in general for the Haves and the Have Nots well it's this is by far the most advanced and most amazing technology compared to anything by a factor of 10 right so if we Look Backwards electricity internet PC Mobile is nothing compared to this so if you subscribe to that view right then um then there's no reason why any of the companies we mentioned

large or small couldn't go up by 10x right including ours hopefully even more because we're cheaper but the IM mature companies U I don't see why they couldn't go go by 10x uh because I think um there this is just so and also many of the problems that plague these systems like hallucination we we I can predict uh that they will be more or less solved in a year and a half or so so um yes they hold things back but I'm very bullish on these Technologies moving forward um of course on your other question

I'm also worried that uh a few big players will dominate more than ever before um and um and that if if one company dominates it would be a really horrible thing if five or 10 companies are doing better that's a lot better but even in that case uh it does create an accelerated pace of um job displacement and accelerated um set of difficulties for smaller entrepreneurs if you could only raise five or10 million and you decide to do an app um take a look at Jasper right they built a great app at the time but

the foundation model sucked in all the learnings and the app became of much more questionable value and that is not even by evil uh Planning by the platform provider is just the natural power of the fundamental models to suck in all that you feed it so that creates um I think uh serious challenges for Have Nots the and also researchers uh if you're talking about lack of gpus I mean we I I try to cry to you how much how few GP CPUs we have but you know we have 100 times more than 99.9% of

universities so what will professors do so the halves and Have Nots have become at the Historical um Gap where professors entrepreneurs and people who have um don't have the skill sets and people who are working on White Collar relatively routine jobs I worry a lot about the uh disparities uh for them and that's again why on 01 I chose uh by you know making this great technology accessible as our mission and and the the underlying emphasis on is on the word accessible because I think all of us should do what we can to avoid the

case of an extreme halves and half Nots and job displacement I think we all know this is coming um you had said around 2017 you thought in 10 to 15 years about 40 to 50% of all jobs would be replaced by AI is that still an accurate timeline in your opinion um what the heck is everyone going to do when they don't have a job in three years if so it's actually uncannily accurate people have criticized me for being too aggressive in the 2017 17 18 19 and I was a little nervous at the time

but when J ji came came out I think everybody's on the bandwagon and believing that is the correct Pace um and I think the white job collar jobs will go a lot faster blue collar job maybe a little slower because more people are shifting to the software only displacement and and I think it's a very very significant problem and I think finally some governments are waking up and realize they have to do something about this and in my AI 2041 I outlined a number of um creative maybe not necessarily workable Solutions that will that was

intended to get people thinking so get a copy of the book um well uh so we've got a lot ahead somehow we're out of time and we need to let these people eat but the one question I do want to leave everybody with is can we get some hope here um how should we prepare our kids to live alongside machines right um if this is what's coming and it's coming fast you know let alone all of our jobs we need to be thinking about how to work this in and and help employees and help everybody

but what do we do for our kids when they say what should I be when I grow up yeah I I think first thing we all have to do and influence all the people around us is to stop this nonsense about kids are using chadd to cheat right this is not cheating any more than using word or photoshop the kids when they go into the workplace are going to be measured based on the final output of their work they're not going to be measured on what did you use chbt did you use Google search so

I think we need to en encourage people to to harness Ai and use all the tools so that they can be the best that they can be and also it's a great guide to uh what things they can aspire to and what things are not worth uh following I think that's incredibly important is to harness Embrace Ai and stop trying to catch people cheating this is not cheating this is um producing great output it's not any more cheating than um Fortune journalists using Microsoft Word Spell checking or a um um a a fortune photographer using

Photoshop this is chbt is a tool everyone should use it and learn it learn from it uh and embrace it and and the second thing is that um I think there needs to be a belief that there is something unique about our Humanity I continue to hold out that we have a soul the machine every will that we have compassion and empathy we have emotions and the ability to love we have the ability to connect to other people and create trust and win trust and that all of you know as successful people in your companies

your success is built more more than anything more than your technical skills more than your business skills the most important skill that I tell any young people for the last 20 years is that winning trust from other people is most important and winning trust is about uh authenticity about teamwork about sharing sharing um and it's about high EQ not just the IQ so I believe in that holds out something that we can all Embrace uh you don't have to be a genius to have a high EQ you don't have to be a genius to to

love and be compassionate and in both of my books I talk about that as the essence of our humanity and I continue to believe it uh do I think AI can fake it yes do I think people will accept the faking AI at least for the next 50 years no so that's long enough for your kids to survive and figure out the next step for their kids okay our kids will survive and they'll figure out something else later um thank you so much Kaiu I wish this had been an hour it was such a pleasure

chatting with you thank enjoy your meals