let's dive right into it you've seen many of the big technological shifts over the past decades right from the internet mobile social media the cloud can you place AI in the context of these waves yeah so this is a big one um this is a really big one so there's this great book I'll recommend uh right off the back called rise in the machines by Thomas rid and he goes back and he reconstructs the very beginning of the computer industry the sort of computer Revolution back in the 1930s 1940s and it turns out back in the 30s and 40s they actually had a big argument at that time of whether a computer should be built the way that we understand them today which is what so-called Von noyman machines basically very fast you know calculating machines um you know they're very good at doing math at high speed but aren't very good at kind of dealing with with people or dealing with the real world um or whether computers should be built basically model on the human brain um and they actually knew they actually had enough neuroscience at that time where they understood the the the basic neuron structure of the brain the first paper on neural networks actually came out in 1943 uh you know 81 years ago now um and there were you know figures like uh you know Alan Turing and and John V noyman and others you know who argued uh you know that basically that you know that would be a superior architecture um and then basically what happened was right the the computer industry kind of forked and you know 99% of it you know was was was dedicated to building out computers the way that we've had them for the last 80 years um and then there was there was always this kind of rump you know kind of uh academic movement you know which was for you know to kind of build out neural networks um and you know that that proceeded you know decade after decade after decade and frankly you know if you talk to people in that field there wasn't much to show for it for a very long time um takes a lot of conviction to to Really persist there so there were there were you know people would get their phds in artificial intelligence um they they would be boring they would get their phds they would spend their entire careers researching it and they would die um before the field worked right so yeah this is like a high level a high level commitment when I was in college in the late 8s there had been actually an AI sort of Boom bust cycle uh in the ' 80s around so-call expert systems we we people thought we were very close to like automated like medical diagnosis all these things and it just it didn't work the technology wasn't ready yet um and so by the time I got to college in ' 89 like the field had been discredited and so you would meet these you know sort of sad figures you know who were sort of toiling away you know kind of wasting their lives um and then of course it turns out they were right um but but but it took us you know and how long did it take to figure out they were right really it took probably 70 years to figure out they were right because they started you know turning out to be really right starting in around 2012 2013 which is when the famous image net test happened and then the sort of pro that computers could recognized objects and images Better Than People um and then that led to the creation of the self-driving car which is now you know really working really well um and then that led to you know what what became like increasingly good voice recognition voice synthesis you know in in the mid 2010s and then the Transformer paper in 2017 and then obviously chat GPT and you know mid journey and Runway and and and all these things and here we are um and so the the the way I think about this this is like an overnight Revolution that's been like 80 years in the making right and so and so and it turns out it's it's incredible because it turns out we have this incredibly Rich you know kind of history of 80 years of ideas now that that turn out to work that are that are now going to get used in all kinds of ways well how do you see it unfolding now you know with um do you is the internet a good analogy if we're trying to look around the corner yes I think it's more just the I I think the internet is not actually as good of an analogy because it you know the internet was was sort of taking existing computers and using them to you know kind of stitch together networks I think this analogy is probably more to like the microprocessor or to the computer itself so the way the way at our firm the way that the way that we think about AI systems is there new kinds of computers um and the way we describe it is kind of traditional computers or so-called deterministic computers which is you know they give you the same answer every time they're kind you know they're kind of very you know very they're very literal in what they do and if they ever you know if you ever give them the same input twice and you get two different results it's because something has gone very badly wrong um and it's probably your fault as the programmer um AI systems we view them as as computers also about probabilistic computers you give them the same input twice they give you two different outputs which is amazing you know for those of us taught you know kind of the old model and then and then they do this incredible you know kind of thing which is you know they have different terms but you know they hallucinate right you know they like people describe it's like people describe to me it's like it's AI skeptic will say things like well you know ai ai you know is not useful because like when it doesn't know the answer it just makes something up and I'm like well have you met people um precisely what my 9-year-old does hm you know very interesting right and then you know as we just saw with you know with runways you know with these spectacular videos you know the you know there there's this like fine line between hallucination and in creation you know so we we have literal you know literally computers now that can create so it could be a feature or create art and so we we basically have sort of another bite of the apple right the the sort of magic Apple um you know we have all the traditional computers and now we have this new kind of computer and then we have this like and then of course a big part of the story is the massive payoff right why why does AI work today part of it is you know experts kind of got the algorithms to be really good uh but a big part of it is mors law actually provided the compute power and it just turns out right you just need a lot of compute power and then the internet provided all the data yes right um and so you know why why you know why do why do you know why why do computers do a better job now ai does a better job recognizing you know cats you know in in photos it's because the internet is filled of photos of cats yeah we take the these scaling laws for granted now but uh that was not obvious you could not 20 20 years ago 30 years ago 30 years ago when was in school like even if you had the right algorithms you could have never made these systems work um and all of a sudden now they work and and of course now the pace of change is just absolutely extraordinary I mean I see things every week now that just put my jaw on the floor so I I think we're in for years of just incredible creativity I mean you have a unique vantage point at a6c seeing many different Industries right I'd be curious what you think about the impact that AI is going to have in a variety of different domains whichever ones you want to talk about but I'd be curious about AI in biotech or AI in finance or AI in defense yeah so a couple things so one I mean one is we think it's going to be transformative in actually every single one of those sectors and we're we're investing across all of those uh we can talk about each of those um um so I think it's going to be a really big deal um in all of those I think there's a really good chance that it you know really reinvents almost every product category that we think we understand today and again just take Runway as an example like are people in the future really going to be doing video editing or photo editing like you know using like Photoshop or Final Cut Pro or these things or are they going to are you going to be talking to the computer and the computer does does the does the image generation editing and so I like I think there's you know these huge Industries where people kind of assume things work a certain way where they're going to work very different and of course that means it's kind of it's going to be prime time for startups right that that have the huge advantage of being able to uh you know to start from scratch so yeah no we think this is going to be a really a really big deal um you know in the tech industry you know this classic Dynamic is playing out where every incumbent tech company is adding AI to their product right um and then you know and then we have we have startups starting in every conceivable category from scratch you know the the the the metaphor that we use to kind of just if you know because we go to our our investors and say we're going to invest in all these AI startups what we say is like look like if you have an existing product and you add AI to it it's a little bit like you know adding flour into a cake like after the Cake's already been baked like it doesn't like doesn't work well you you if if if there's a really transformative technology you have to like get you have to get the flower into the recipe from the very beginning you have you have to build from scratch um you know knowing knowing you have the new capability uh and by the way there's there's a bunch of companies the sort of later stage startups are like halfway in the middle we we have this phenomenon we call bullet point number six um or we'll get a sort of a you know 5-year-old company or whatever will come in and pitch and and they'll have their product slide and it'll have you know the five kind of things their product does and it'll have bullet point number six is oh it also does AI right and it's always bullet point number six because of course they added it after the fact right they you know they couldn't quite you know the Smart Ones goad and redo the slide um and so you know there's a lot of adaptation happening now look in the mobile to web you know in the in the web to mobile wave a lot of the big actually incumbent companies like Google and Facebook actually did just fine right and so it it's it certainly may be the case that a bunch of incumbent companies actually adapt to this just fine I think it's going to be challenging it's going to be a lot of work to do that because this is such a different way of operating um you know I tend to think that startups have a have a have a really uh great run here but you know it's going to be a real fight you know kind of in the industry and we'll see you pay a lot of attention to biotech yeah right can what does this look like I understand for software products but what do you think the role of AI will be in biology yeah so look so the the the big thing the big thing in biotech biology is sort of we've always had this concept of sort of drug discovery um which is this kind of amazing term and and basically you know the way that sort of the modern world of kind of drugs treatments um you know has emerged is you you sort of have you know basically a lot of scientists basically trying lots of different compounds and kind of seeing what works so you know you're literally kind of discovering what works um uh you know the the you know the sort of overall giant transition that that field wants to go through is to go from Discovery to engineering right so where where you're actually you know designing designing designing things from scratch you know including literally designing new drugs from scratch uh and you know we're starting to see examples of that you know mRNA platform that was the basis of the covid vaccine you know is a good example of that U you know madna had the co it a very interesting madna is you know madna and fiser are the two companies that had the first covid vaccines with similar technology uh madna had the first covid vaccine two days after they were emailed the Genome of the covid virus from China like in a Word document um you know which is first of all like a fairly amazing thing because it's like spell Che is pretty aggressive in Microsoft Word and like you certainly hope like it didn't like you know screw things up but it turns out it didn't it was fine uh but then like literally had that vaccine two day you know they had the covid vaccine in January of of 2020 and then the rest of it was just testing and validation and that was the consequence of mRNA being an National engineering platform for for for biotech and and so the minute you have an engineering platform you can you can bring AI to Bear do you think we have enough data in biology today so it's a real challenge um it's a real Challenge and I would say it's a real challenge in particular because access you know what you really want like if you wanted the perfect biotic industry what you would want is you would want to basically get you know get Everybody's Health Data you know pull it all together you know you'd want to get everybody's genome data pull it all together you know that in the US that would all be completely illegal you know you would never be allowed to do that you would go straight to jail um you know interestingly in China that's actually just fine um uh so it is actually this very interesting opportunity for societies like China to maybe maybe Le leap progress um uh on that point so yeah there are big challenges on that but the the payoff is also going to be very big um and so I think there's going to be a lot of a lot of positive pressure to figure out at least how to get an updated to be able to do the work speaking of China yeah how do you think AI will shape geopolitics or defense or some of these spaces yeah so this is an area where actually the sort of plan the sort of theoreticians in defense were actually ahead they they were on this several years back so right right around the mid-2010s uh there was actually a sort of major change of military Doctrine in both the US and China um so in the US uh the the the Department of Defense has a thing called called offsets and it it it basically an offset is like a new technology that will fundamentally change how how how Wars are fought um and they declared that Ai and autonomy in combination were the what they call the third offset U and it's a big deal because like the first offset was nuclear weapons the the second offset was what they call maneuver Warfare and then the third offset is is AI and then China similarly has been very public actually what is maneuver Warfare maneuver Warfare is is sort of like small unit tactics Precision guided weapons um you know sort of specific applications of force as compared to just like launch the nukes or like launch the bombers um and so maneuver Warfare was like a consequence of advances in like communication technology GPS was a big part of of of maneuver Warfare um but but you know AI is the big one and and and so anyway so the US kind of forecasted that and the the def defense department you know look we have like the Predator you know we we have the predator drones you know that that have you know that fly autonomously that that have played a big role in counterterrorism so that started to happen about a decade ago and then China similarly embarked on a major they've been very public about this kind of a major rebuild of of of all of their systems you know based on AI thata goes pretty early for this shift given that you know the excitement around AI now what what triggered that oh I think it was B Bally they started well when is they they you know self-driving cars self-driving cars started to work you know the the the original DARPA Grand Challenge for self-driving cars was like in 2005 and then so you started to see those basic systems started to work and then they you started you know obviously autopilots on on airplanes started to get really good and they started thinking about taking the pilot out of the airplane and if you take the pilot out of the airplane all of a sudden you can build very different kinds of aircraft that you know maybe are like much more maneuverable much faster because they don't need keep a pilot alive and so you know these kinds of things and you know then they keep they keep people safe right um and so so so those changes started and then of course what's really accelerated all this thinking in the in the National Security world is the the specifically the the UK the battlefield in Ukraine um where it's it's like this like you know and and start by saying like it's an incredible tragedy that that that this is even happening but what's happening on the Ukrainian Battlefield is shaping a lot of thinking around National Security what is happening what is happening well it's it's like it's like 20th century versus 21st century it's like it's like a it's like a it's like a cyber Punk movie is fairly amazing which is you know the the the Red Army you know the Russian military basically is you know driving tanks down um and the ukrainians are responding with you know first person you know basically literally professional video game players piloting V in VR you know piloting autonomous drones um you know and then coming in on top of the on top of the plane and and and and blowing up and and cooking it on cooking it on the inside um or uh you know these these shoulder mounted you know weapons like javelins um that again are kind of you know they call them fly by wire they sort you know they sense where the the thing is so so there's a tank and you have a javelin you fire the you sort of wave the Javelin in the direction of the tank you fire it and it automatically home comes in on the tank and then drops down uh through the hatch um and so you know the Russians have taken devastating losses you know as a consequence of this um and and and and there's always for the Russians there's always a guy in the tank for the ukrainians you know there's there's no guy on the Drone so so so there there's an there's an asymmetry to it um and so you know Ukraine has punched way above the weight that all of the planners expected when this whole thing first started because they've been able and they've been very aggressively innovating in the field you know for how this stuff works and then of course you know the the again the other stry playing out but the you know similar kinds of developments are going to happen with in the Middle East now um and so I mean we get these amazing asymmetries and so the uh you know the the the the houthis for example are firing these I think they're Iranian drones you know they cost you know whatever you know thousands of dollars you know and we'll we'll liter US military will use like a$2 million Tomahawk missile right to like take down a $10,000 drone right so you know you have these like moment it's like slippage in time right these things shouldn't exist in the same era um and so if you talk to people in the US military they're like okay now we need to basically reinvent the whole thing like the the the way that we fed everything from aircraft to submarines like all these things is is is just going to is going to completely change war in the future is going to look totally different is the battlefield itself the place where this will have the biggest impact or do you expect it uh to shape policy or deterrence or other other uh aspects of geopolitics well it changes a lot I mean so the the sort of more the sort of most advanced version of the thinking I think is basically it you know defense historically like you know sort of Warfare has been an issue sort of of mass and you know sort of number of men amount of material right so how many how many soldiers do you have and then like how how many tanks do you have you know sort it's Napoleon I think once said God is on the side of the big battalions right and so it's like you know just just like mass and force um and you know argu if you project this forward enough what you kind of see a picture emerging is actually in the future successfully be based on who has the best technology and the most money right and and and and there's some connection between the best technology and the most money because if you have the most money you can either develop the best technology or buy buy the most technology um and so it it it sort of changes the the and by the way it means like if you want to then be the strongest military force you then want to be the strongest economy with the most advanced technology it means it's like really strategically important for your country to actually win in technology which is you know something that's sort of hotly controversial these days um and so like that you know that that you know that that matters a lot you know will you even have human soldiers you know at risk in the field or in planes or in submarines in 20 years like maybe but I think probably not what does it take to win in this technology yeah you so this is a very inter so so this is sort of very interesting topic right now for kind of us us politics you know which is like so I have these you know technology has become a very politically loaded topic over the last you know decade for a bunch of reasons and so I've been at our firm we're much more engaged now in politics than we used to be and I've been spending a lot more time in DC than I used to and I sort of have these two very different conversations in DC um I call it's called the Tuesday conversation Tuesday conversation is it's it's US versus China and then you know Iran Russia and so forth and like we and this this revolution is here and the US needs to be the dominant technology Global superpower because that's what's going to lead us you know being being strong and pro protecting democracy and protecting our allies and all these other things and so what can you and Silicon Valley do to advance technology as fast as possible um and then on Thursday I go back and they forgot about the China topic um instead they're just focused on the US and they're like technology is like weird and scary um and like it's like freaking us out and like AI is really scary and like we think we ought to stop this whole thing and like we need to regulate and we need to shut the stuff down and like can't you people please slow down right and then and then you know I kind of say well like you know I I I I repeat the Tuesday conversation on Thursday and I'm like you know could you please reconcile you know these two points of view for me and they say no that's your job um so it there's a real tension underway right now um and uh you know one of the you know are the same people or different people so some sometimes sometimes they're different people sometimes they are the same people um maybe may it may may surprise you to hear that sometimes politicians are not completely consistent um um from dayto day and and look and look both of the you know kind of give the devil is Du like both both of these impulses are very emotional right these are very emotional topics because if you picture us losing a war with China you know that's just devastating emotionally if you picture AI you know completely revolutionizing everybody's jobs and you know changing Society that's also very scary um and so you know we we we do we we as an my conclusion is we as an industry are going to exist kind of between these two very I would say you know sort of emotionally valid you know kind of impuls where people have a lot of questions and concerns you mentioned that technology has become far more political over the past years um what led to that yeah so look I think basically so I so start by saying I think it's our fault um so I I I blame us and by us I mean myself and my firm and my companies and and all I think it's all your fault uh everybody here shares the blame so um look like a lot of us who kind of grew up you know in technology grew up you know engineering you know I you know my degrees in in computer science um you know we we sort of were like wow like we think the stuff that we're working on is like really important and like we wish the rest of the world would understand how important it is and and now they do right like like they didn't for a long time and now they do and so I I use the metaphor of the dog that caught the bus right which is you know the dog chases the bus you know never expecting to catch it and then God forbid you know the dog ends up with the tailpipe in his mouth like what does it do with the bus right um uh and you know of course the reality is the bus just keeps on going and the dog gets dragged down the street and so that's us right uh we're the dog um and again it's a consequence of like okay we we won like the the stuff that we're doing turns out really matters it turns out software really matters AI really matters microchips really matter the internet matters like the these things these things all matter and by matter I mean these things don't just you know they don't just matter to us they matter to everybody they matter economically they matter culturally they matter you know socially uh you know the runways you know spectacular Runway demos is like you know people in Hollywood are like freaked out right cuz like you know what does it mean to be entering a world in which movies can be rendered by AI right um and what does it mean you know for example I give you an example like you know movie actors like if you know the you know if the you know great movie actors of the past can can have their images you know being rendered in AI you know for the next thousand years you know literally is the future of the movie industry going to be Tom Cruz movies for the next thousand years you know kind of rendered over and over again in AI like will there ever be any new actors right or would just all be watching Tom Cruz movies um and you know technically we're going to be able to do that and so people in Hollywood are like super freaked out um uh you know lots of people are are are are are just getting you know I say increasingly upset there's a long Shan strike actually that's that kicked off today you know where the the the guys you know basically people who load and unload boats on docks and it's just like it's like a flat out just anti-tech it's just flat out no more automation like we're done literally that's the strike because no more automation no more machines we're done do you think there's an angle here where AI levels up the discourse helps improve the discourse helps with governance or is that you know further out yeah look it it does I mean one of one of the thing look people see one of the so start by this one of the great things about AI is it's rolling out in actually a very Democratic way right so so and if you think about this like the way the computer technology rolled out like 50 years ago it showed up in the form of like you know $20 million mainframe computers and then it trickled its way down over 30 years to personal computers and smartphones one of the most amazing things about AI today is the best AI systems are rolling out to Consumers first right and so you know C GPT and Claude and all these things are rolling out to Consumers first um and so you know and this is one of the great things things about technology is like you know by the way the usage numbers are like you know on fire and so the the number of just ordinary people in regular jobs in regular professions um you know who are able to use chat GPT or any of these systems today and are able to you know ask questions and get guidance and figure out how to like deal with complex situations or deal with the government or navigate you know laws or you know figure out like you all of a sudden everybody has like a an AI they could talk to to kind of help help help them through all these things um and so I I think there's the prospect for sort of a general lift of in of of intelligence you know basically of of of people having you know a level of intelligence available to them at their fingertips that we haven't historically had you know that that certainly includes you know politicians policy makers you know um that certainly includes you know people like military planners like everybody's going to have this sort of additional level U we used to call decision support systems um and so like you know there's sort of a chance for everybody to kind of get smarter uh in a in a very interesting way but you know that I would say that with that with hopefully that means the arguments can get a little bit more rational but what that doesn't do is I don't think that dampens emotion at all right and and I think that I think there's going to be just like just tremendous amounts of emotion as this as this technology rolls out as as as is already happening um and so I I just think like we as an industry are going to have to increasingly step up and explain ourselves and articulate what we're trying to do and you know engage in the policy process and in politics um and really help people understand this and and and by the way also make sure that this technology stays Democratic and stays open one of the reasons like I think open source is so important is like I think it'd be very bad if this technology centralized in the hands of a handful of companies I think you know it's very important societally culturally that this is something that everybody has access to On the Open Source topic right do you think that how do you see that playing out do you think it could genuinely go in multiple different directions or there's a you know kind of an inevitable outcome here yeah know so there's a real fight so there's a real move underway to ban open source and that and specifically open source Ai and specifically the math um under under under AI which you know is is fairly mindboggling for me as as an engineer you know just in talking about linear algebra it's not that complicated right like it's it's right it's linear algebra it's it's you know gradient descent these are fairly basic algorithms they're taught you know broadly in schools today my nine-year-old is starting to you know learn this stuff um uh there are videos on YouTube that teach you how to do this there's textbooks um you know it's like half what is it like half the kids at the Stanford freshman class are taking the intro to AI course they're learning this stuff um and so but like there there are serious like the discussions in Washington are like disconcerting on this topic because they they seriously uh you know about this um and then you know there there's like outright bans and then there's also just like sort of what I call sort of the potential for like a this sort of stifling blanket of Regulation um and you know the precedent for that is what's happened in the EU where they have actually done this they've actually implemented a stifling blanket of Regulation including most recently on AI and the result is you know they've effectively outlawed you know to a large extent they've outlawed technology startups in in in the EU and the market has responded and they're they're you know there's a handful that are really good but there's not very many and in Europe used to be like amazing like half the Great Tech compan in the world used to be in Europe and now it's it's down to you know a very small number and so we you know we have this like real life cautionary example of how this can actually go bad and then we have a fair number of people in the US and in California which is amazing uh who basically want to adopt the EU model and they're you know they're lobbying very hard to do that some some of those people by the way are True Believers you know they they really believe that there's like a profound societal threat or you know existential threat or various other kinds of threats you know they really believe it and you kind of see it in their eyes um and then um you know there's a bunch of other people that have commercial self-interest right and you know they want to establish monopolies or cartels of different kinds and and they want uh there to be um you know basically government protection uh for their business and so there's this like Unholy Alliance that sort of formed up between these two groups so there's and there was a real you know we just had this real fight in California yeah yeah and it's a very important role for policy to play here yeah um so this is the ray Summit this is an AI infrastructure audience largely what does this wave mean for um people building AI infastructure providing those capabilities what can we learn from previous technological shifts that you've seen yeah yeah so I mean look like it's it's going to be prime time like you guys are in the right field um because you know this is it like this is going to be the big buildout um and it's just starting and you know this going to be one of those things where we look at the stuff that we have today you know we look at you know five years from now 10 years ago and it's it's going to look like it was you know belongs in a museum as compared to what's coming um and so and you know all of us here are going to play play a big role in building that um yeah and so the the this is going to be like absolute prime time um you know there's an unlimited number of of technical issues issues to solve obviously uh you know but look on top of that we're on this incredible Voyage of Discovery I think as an industry and of a world of you know which which of these domains of AI are going to be the ones that really take the fastest you know just give you this sort of here's like an amazing question that's playing out right now which is the whole history of the you know kind of AI robotics you know kind of field broadly is that everybody assumed for a very long time including when I was in school you know that you would kind of get you would get the lowlevel stuff first and so you would get like robots that like pack your suit case and like clean your toilet and cook your breakfast first you know and then maybe decades later you would get robots that could like compose songs or like draw artwork right yeah um and and it turns out it turns out to be the reverse it turns out it actually turned to be quite easy to get a robot to like compose a song or like draw artwork it still can't pack your suitcase it turns out math is actually really hard yeah yeah exactly yeah it's it's and especially right that the minute you intersect with the real world you know things get very complicated um but look I I you know the the the AI Revolution is playing out today is Main one at these sort of higher levels you know things involving you know basically software information bits um but I think the robotics you know Revolution is very close and I'll just give you my my favorite example of that there's this uh company in in China called unry and they make a robot they make a robot dog if everybody's seen the demos of the Boston Dynamics kind of robot dogs but you've probably never seen one in real life uh because just because they're just too expensive they're like you know $50,000 and up um you know this company in China has $1600 basically that robot dog so do you think the uh and yeah so the the the there's this robot supply chain that's basically forming up in China like really quickly um and and and you know and drones drones were kind of an example of that you know where China just like swamped the world with drones and I think you know they may well swamp the world with robots at least on the hardware side and then there's all these questions around how to actually build all the software for robots and you know Elon and others are and lots of startups are working on that and so it may be that we have the sort of bits version of AI Revolution it may be followed like almost immediately by the Adams version that's incredible so there's a chance that the um humanoid roll out will look more like a $20 subscription democratized versus a $20 million main frame yeah look so this company unry has um they have a I think it's a it's I think it's a60 $50,000 humanoid now um hardware and you know if you if you just watch on their website like they're advancing the the the sort of Hardware capabilities like super quickly um and so you know it looks like the human robots when they roll out I don't know Mass Market 20 you're talking about to Consumers 20 yeah to Consumers well the the hardware cost looks like it's going to come in I don't know $20,000 $110,000 and and at that point you can afford to have a lot of these things right and and and yeah and you can afford to have right and you can afford to have subscription models yeah right and so I I like I I don't I don't know I'm not I'm not positive how fast this happens but it could actually happen very quickly now yeah one of the one of the very funny things right now is these robots if you see the demos like so these humanid robots like they've got their like they've got their like control system which is like the the state-of-the-arts like this reinforcement learning algorithm to try to get it to like crack an egg without destroying everything on the table or whatever um but then they've got they've all got like llm user interfaces with like voice voice now and so it's like struggling to crack an egg but it's like teaching you quantum physics right like like in like a plumy British accent like at the same time right and so there's this software kind of wave happening right now to try to like merge those approaches together right uh and have you know sort of multimodal models that are capable of reasoning and planning and and you know that you know the new llama release just dropped I think last week that has multimodal open source um and then you know open AI just released their 01 you know the planet the reasoning and planning engine which they you know they say is going to get a lot better from here and so like it may be that in a year we actually have a sort of single multimodal model that can do reasoning and planning that can power a robot and then We're Off to the Races it's interesting to see how important all the multimodal capabilities are for Robotics and real world stuff yeah and and look that then what you're dealing with then it goes back to the topic of this conference then what you're deal you know so what made Tesla self-driving cars work is they put a million cars in the road they gathered all the imagery from all those cars they used that to do deep learning and then they made all the cars you know really great at driving themselves and they're they're and they're they they roll that forward they compound it but like we might be in a world within a few years where we've got a billion humanoid robots that are also Gathering all that data and feeding it back and getting better and so it it may be one of the largest scale Technical Systems that we've ever built it I'd be very close um let's talk a little bit about investing in AI so it's maybe a little too early to say who the eventual big winners in AI will be has your perspective on that changed over the last few years yeah so there's this glorious amazing difference between running a company and running an investment firm and I'm going to I'm going to I'm going to really he's going to be haunted by this for the next for the next month what I'm about to say um because he actually has to run run one of these companies um uh when you're running a company you have to have a strategy um and like it has to make sense right and you have to it has to be like an asual like you have to like lay it out and you have to like have a have a conference like this you have to show everybody what the strategy is and it has to make sense and if does if it doesn't make sense it doesn't reconcile how people get upset um the the one Saving Grace about on the investment side is we we don't we don't need one of those um and and the reason is because we we invest across a portfolio and so we we can invest we can invest across basically different approaches yeah right and so we and it just we can invest in infrastructure and apps we can invest in open source and closed Source we can invest in you know different technical approaches um and so we you know we kind of fully take advantage of that so so I would say like a lot of the questions on the future of like the the investing side you know a lot of those questions become economic of like where's the value going to be where's the revenue going to be where's the profit going to be I'll give you one of the very big questions right now which is you know are the large language models like open AI are those businesses basically erased to Monopoly and infinite profits because of scale right the the biggest company is going to have the best model it's going to answer all the questions better and then they're going to charge whatever they want for it it's going to have super high margins be the world's best business you know and basically and that's like the track that like you know that's what happened with search right where you know kind of Google ended up in that position um and the answer is maybe and then or the answer is actually are all those companies actually in a race to the bottom right in which it actually turns out you know selling you know intelligence is like selling rice or something where you just like it just turns out anybody can make an llm there's open source llms um you know there's new LM startups every day um and it just turns out anybody can go scrape the internet anybody can buy the gpus and then anybody can basically get the same result um you know there was a famous Google memo that leak called you know we have no moat and neither does anybody else where they kind of articulate basically saying if you have the same data you get the same results everybody has the same data um and so in that case is it AAS to the bottom and by the way there's some evidence that it's AAS to the bottom because the price per token right of generated llm output has dropped 100x in the last year right which is like much faster than Mo's law which is the exact opposite of what you'd expect if somebody was going to you know get a lot of profit out of that business so you know maybe that's the race to the bottom and the term sometimes use there as a profitless Prosperity which is you know like like selling rice or any commodity these companies will have a lot of Revenue they'll just have very very low margins I think sitting here today nobody quite knows the answer to that you know it also depends on things that happen you know in the ecosystem think like you know meta you know where I'm on the board the the release of the Llama I think open source models has really I think has really changed how the future is going to unfold you know because having fully competitive open source I think really changes things and so so there there's really big questions around where the value capture is going to happen and we're we're we're actually trying very hard to not prejudge that because I think we actually don't know yet makes sense I mean the fact that a lot of AI startups need to spend a lot of money on gpus right how does that factor into your investment decisions yeah so that's the pet by the way the Peter teal thing the Peter teal thing is um you know we're all wasting time with all these software companies right we should just put all of our money into Nvidia um you know and just and call it and by the way for the last like you know 10 years that would have been a very good you know very good strategy um you know and so the right and the model there basically is nvidia's uh Nvidia will soak up basically the entirety of all venture capital invested into into uh into AI startups you know plus all revenue that those startups generate um and they'll just capture the whole thing and and by the way there's there's a a picture of the world where that actually La you know that's what's happening today right for sure and then that there's a there's a version the world where that happens for many years to come and that argument basically says Nvidia is way ahead on the technology for the chips and then on top of that they've got all this other technology like Cuda so they've got this incredibly you know kind of strong you know kind of Technology lock in um so that's one argument the other argument though is what traditionally has happen in the chip industry is leads like that last for a while and then they generate a huge profit pool that draws competition and Market forces kick in and then you have every other chip company in the world from you know AMD to Qualcomm to Samsung to you know the Japanese companies everybody else and all the startups and cerebrus I think is going public this week um um and so you have lots of startups and you just have like lots of people who are like wow I want a piece of that pie uh and and by the way another argument on that would be gpus gpus were not originally developed for AI they were developed for graphics right um and if you were developing chips from scratch to just do AI they might be very different chips by the way Google has their own chips Amazon has their own chips seeing a explosion there explosion of chips and so you know it may be in you know five years from now Nvidia is still a great company but they just they they you know they're they're part of the ecosystem and it turns out there's you know five other big companies and 10 other startups and a whole bunch of other approaches um uh you know that that that make just as much sense and then and then again you might have big Revenue growth in the chip industry but much smaller profits um and so but again what these are none of these things one of the things it's hard to forecast these things is it it depends on what people do right and so what you know what decisions are made today at the big at the other big companies and what decisions are made by the startups and VCS and you know by the way customers and users um you know and all the people building around this you know what what do they all want and so we're we're we're kind of on a on a collective mission of Discovery to figure that out maybe to wrap up for all the future Founders and current Founders in the audience do you have advice you'd give them about starting businesses or running their business in today's environment yeah so I think the big thing on starting a business I think the big thing is um a starting company the advice is kind of always the same which is it it it very rarely makes sense to just say I want to start a company therefore I'm going to go in search of an idea the great companies at least in Tech usually what happens is they're started by somebody who has been deep in the field of what the startup is doing for probably many years earlier uh before they're probably very deep domain experts and and and and by the way this gets glossed over a lot a lot of these a lot of these successful tech companies in the valley they they kind of have these mythological founding stories where it's like oh you know there was just I was just a little kid and I just had this idea and it all worked and it's like well no like I was in a lab for 10 years before that right trying to get trying to get this stuff to work I was I was at school you know I worked for a big company um I worked for another startup and I I learned everything um and so most of the successful startups come from people who have spent 5 or 10 or 15 years deep in the trenches really deeply understanding you know the nature of a problem and figuring out a better way to solve it um and so I I and look I I think there's like a probably a large number of great startups that are going to come just just out of this audience in the next in the next decade um you know in terms of of of how companies operate in this kind in this kind of rapidly changing environment our big advice always is basically run experiments um you know there there's always this kind of daunting you know thing whenever you're sort of responsible for an existing business I've got all my existing customers my existing employees I've made all these commitments you know doing a new thing always feels kind of weird and scary because it's like what if it doesn't work and I you know going to really break what I have um and so I we always sort of advice is the right way to do that is try try to figure out you know try try to run the the smallest possible experiments to kind of try to prove a hypothesis you know take a small number of customers or a small segment of the market and explore and then by the way try to make that a habit inside the company try to run lots of experiments like that over time seems like generally a good practice what's that it seems like generally a a great practice yeah exactly yeah exactly and and basically build build option value you know learn as you go uh you know but without without having huge downside of risk as you run the experiments and then then when you know when you tap the vein and you think you figured it out you know that's that's when you commit the business mark thank you so much really appreciate all of your insights good awesome thank you everybody thank you thank you all right