Fireside Chat with Marc Andreessen and Ali Ghodsi Marc Andreessen Andreessen Horowitz and Ali Ghodsi

mark thanks so much for making it we're super excited to have you here good morning so you've been around you've seen the seen this movie before and there's a lot of excitement around AI these days is there an AI bubble so AI is a really interesting concept kind of from from beginning to end so so AI is a very old idea the the idea of AI actually traces all the way back to the invention of the computer and there's one of my favorite stories in the history of our industry as Alan Turing the you know the great inventor and a co-inventor of the computer in the in the 1944 Alan Turing was at Bell Labs in New Jersey and he was having lunch with Claude Shannon who was the inventor of of information science and they got in a big argument about AI and because the Turing was proposing that we could we could make AI just a pretty advanced thought for computers at that time and in Claude Claude Shannon was skeptical and he's like I don't know if we can we can get computers to think the way that people do and Turing got more and more frustrated and finally stood up and yelled in the middle of the cafeteria I'm not talking about making a genius computer brain I'm talking about making making a mediocre brain like the president of AT&T a big hit in the executive dining room and so you know that sort of by the way that actually began the debate about like what is the definition of AI right is it replicating human intelligence or is it actually creating a sort of a new form of intelligence so that that's like 1944 you know this is like what 7080 years ago over the what over that period of course many many people have spent their careers trying to build AI I think we've counted there were five boom bust cycles and AI leading up to where we are today so there were there were there were sort of there were be tons of AI hype and enthusiasm and then there would be kind of these AI winters when I actually arrived in the valley in the early 90s there had been an AI boom bust in the 80s right that didn't pan out expert systems was a big thing at the time and it didn't really work our assessment is this time is different and the reason this time is different is because the technology has finally started to work in a real and deep and serious way and we think you know the breakthrough was around 2012 you know and specifically right it was the imagenet competition in 2012 the kind of demonstrated computer starting it starting to advance human beings in terms of image recognition and then of course generalizing from that of computers can recognize images you know then they can do lots of other interesting things right get increasingly smart and so we like we think primetime has arrived got it so it's actually the technology actually works now that's the difference make sense that's a lot of we're not disagreeing with that so you see a lot of startups you know you see you hear a lot of pitches of the hundreds of ones that you see every year what stands out in the ones that really succeed and what stands out in the ones that sort of fail and what would you see most common mistakes are that that you know startup founders did yeah so it actually turns south thousands a year so it's it says it's the rough numbers it's so for sort of us venture us high tech so this is not counting right europe or asia just us it's about four thousand startups a year that are it's a venture fundable we see about 2,000 of those we narrow that down to about thirty bets a year so we invest at a rate of about 1.5 percent which means we're in the world's most weird optimistic pessimistic job right because you you never know when one of the 30s gonna walk in the door but 98 and a half maybe eight and a half percent of the time it's not and so your day job is crushing out to pronunce and dreams you know with the occasional exception right and so it's it's it's it's a heart it's a it's a harsh it's a harsh job the reason is such a harsh job is because the idea is so hard to build a successful startup Elon Elon Musk has the the best line on on the nature of high-tech startups he said doing a high-tech startup it's like chewing glass eventually you start to like the taste of your own blood I always loved the reaction of that that content gets from people because that's like yeah that's the feeling like that's what it's like it's just like oh my god and so it is an extraordinarily difficult thing to do the the winning companies you know there's some kind of magic or alchemy the way things come together but it always kind of resolves down to three fundamental elements you have to have a significant advance in technology you know at least to X and maybe on the order of 10x to really have a margin of safety to be able to do it and so there has to be some sort of step function change in the technology and the reason for that is right big companies are very good at incremental change and so for start-up to exist kind of by definition it has to be a big enough leap in technology that's gonna kind of take all the bail the big incumbents by surprise so significant advance in technology and those just don't come along you know often enough and then you need a big market right because if you have a big advance in technology for a small market you'll get a small result and so you need a big market to be able to build a big company behind that and then of course you need the team and you need you know not just the founder or the founding team you also need the ability of the founder and the founding team to bring together a team around them right Peter Peter teal says one of the most are the most interesting things he said is it's everybody focus is unlike how's the startup gonna hire like a second engineer and the more relevant question is how's this going to hire its 20th engineer right and then its 200th engineer and then at some point is mm engineer right because you have to build a critical mass of a team to be able to execute against a big idea and so is the team able to kind of convey the kind of vision required to be able to pull people in and then be able to build the kind of team spirit required to execute against a big goal and so it's just it's like three really big challenges and it's it is you know like it's just it's really hard to line those up that make sense okay I'm gonna see if I can get this out of you order any investments in the last 10 years that you look back and say I wish I had invested in that company we wish we had investment that you had yeah yeah yeah there are okay do you mind sharing or telling us more about this so so I describe the sort of 2,000 to 230 thing there are basically great two forms and mistakes then in our business right there's the the false positive in the false negative the false negatives are kind of the mist or the false positives of the mistakes you make all the time right it's the investment it's the company investing that doesn't succeed and you know those are you know those are frustrating but sort of the statistical nature of a venture capital is that like the Silicon Valley venture capital basically is you know basically half the companies aren't gonna succeed that's kind of the long-run statistical distribution and so you know that you're gonna basically have false positives about half the time if you're doing your job right right and the reason for that is if you're not taking enough risk where you're gonna get some false positives you're also not taking enough risk to invest or the ones that are gonna work right so it's sort of part of the thing the saving grace on the ones that don't work right that you invest and they don't work the saving grace is you you can't you can only lose one X right and it like that's horrible like it does still go to zero so that does still really suck but you know it's it's calf losses then there's the false negatives which is you know the company walks in the door you don't invest and then for the next 20 years you're reading headlines The Wall Street Journal about how well the company is doing like every morning you wake up and you just like you're just being like mentally tortured by how well this company is doing and it's funny because they're right what you've done is this opportunity cost you've missed out on basically the unbounded upside right and so the sort of theoretical opportunity loss is essentially infinite and yeah and every day it's the company goes public and every day the stock price goes up and you just it just gets harder and harder and harder it's a screwdriver in the stomach and so that does get more and more painful and at this point I hope I've run out the clock so I don't have to dance yeah I don't think I'm gonna get a name out of you okay okay let's so let's move on to the promise of AI so you know you have a really unique vantage point I mean you've seen the movie you know we started Netscape you know if you've been around what do you think what is the impact that AI will have on various industries and war maybe some of the killer apps that will emerge in this field yeah so there's a big there's a big debate I think happening in our industry right now and new this conference is right at the heart of it which is okay is AI in incremental advance right is AI a feature that we're gonna add all the existing products and technologies right and so you know what you know at one point speakers were dumb and now they're smart at one point whatever whatever app you have you know it didn't have any I know it has a little AI and right by the way it's it's like every pitch we say you know every sort of funny when these new technologies arrive because you see it you experience it in the venture pitches which it's like you know the entrepreneur will come in and they'll talk about you know they'll have six attributes of the thing that they're building it'll be you know I don't know big data boom boom boom and Bo's stuff scalable secure and then bullet point number six will be like AI and it's like if I can I see the version of these slides from two weeks ago that had five bullet points you know where that six one just kind of got added right before the pitch right and so that's and by the way most advances in technology are incremental right most most of them are just features that you know that you add in and so maybe I ai is one of those things maybe AI just kind of get sprinkled on everything and everything gets a little bit better or maybe it's a profound change right maybe it's what we would we would referred to it's an architectural change right it's sort of a new foundational technology whereas we get it to really work we can all of a sudden rethink things kinda from the ground up right how systems should work from the ground up and so you know instead of computers having screens maybe it's entirely voice right maybe business applications instead of being you know spreadsheets and rows and columns maybe it just gives us the answer right and so if it's if it's that big of an advance then it's an architecture change and then in our industry if it's an architecture change that leads to ultimately an almost complete turnover right of products because because basically products that have to get reinvented from scratch and then over time it leads to an almost complete turn of turnover of companies right most companies they get built on one architecture don't survive an architecture change and so those are the really big transitions right and so what are those transitions been right there's been a few mainframe to mini-computer mini computer to Unix UNIX to PC PC to smartphone you know like those have been big ones client-server at a web actually client-server the web was a probably a bigger change web the mobile is was actually not as big a change in the way you know that is because most of the big web companies actually became mobile companies right it you didn't tend to have her plate like you know many example as an example you didn't have a new mobile search company just placing Google as a as the low search company just because mobile arrived so even mobile was not was not even like a complete architecture change in a lot of areas our sense is that AI definitely has the potential to be one of these changes and that sort of our approach we're assuming that it does and we're actively looking and funding companies that are thinking about a complete reinvention of how things work I mean some of the best examples are now you know incredibly obvious but we just as an example we have this with this drone company called sky do that's complete complete autonomous drone so it's it's like the drones that you buy you know DJI drones or whatever you buy it's like a flying camera but it doesn't even ship with a controller like it doesn't even have a controller because it's entirely it flies itself it you you just tag it to yourself or a person and then it just it basically as a flying camera operates and follow-me mode and it's just incredibly - in terms of its level its level of autonomy and so that's the kind of thing we're all of it like assuming assuming we're right about that and you can see the product side if you look up sky deal and YouTube you can see this in action and so if that's the case then the previous generation had drones like literally are all obsolete today right because there's now a fundamentally different way of doing things and I and and so our sense is there's a real possibility that this will happen across lots of areas I see I see I thinking so basically it's the 10x difference so that's what your connection even bigger it's almost like it's almost like binary it's like we couldn't believe there's a set a whole set of things we just simply could not do and now there's a set of things all of a sudden we can do and the it's like the the inventors and the architects of the previous generation of products hadn't even envisioned that this would ever be possible so phase shift basically makes us yeah okay so think about that when you're pitching okay all right by the way it's yesterday I should be the first bullet point exactly so you know one of the major things we've actually heard today is about the skills gap and also yesterday a lot of companies are struggling to find data scientists that are skilled people that know they've develops data engineering data science and you know a lot of them say you know it's really about the data the data part is the hard part I'm curious when you talk to enterprises I mean you must talk to a lot of these big enterprises what are the data challenges that they mentioned to you that they have what are their struggles yeah so I think the big the big one is skills and cultural and we can come back to the data but like this talk about skills and cultural stuff cuz I think it's it's the thing that dominates us so I think there's a really good thing that's happening and there's a really kind of scary thing that's happening so so the good thing that's happening is as follows so if you had asked me three years ago actually how we because we were got excited about AI you know back 2012 2013 2014 when this all started to change for the first couple years we were fairly disconcerted by the whole thing and the reason is because the view at that point was AI is like a black art like it's this like hyper specialized thing right in fact it was a field that had gotten you know quite discredited for a long time and so they were they're actually not that many AI PhDs coming out of the top research programs you know basically for the preceding 15 years it's just a limited number of people who were trained in the area and then the big incumbents you know the googles and Amazons kind of figured this out early and then they figured out how valuable it was and so they started hiring all the PhDs four gigantic amounts of money right and Google like has had it has like an eternal spreadsheet basically showing like if you have you know an AI PhD applied to this part of how ad targeting works we can incrementally generate another hundred million dollars of profit a year just through algorithm changes and so we'll pay this AI expert whatever ten million dollars a year it's you know it's easy and hope he doesn't discover the spreadsheet because then we have to we have to pay him even more and so basically in the first two or three years the the incumbents basically hired many of the experts and of course for for us for in venture that's really bad because it's like okay well then who's going to start the companies or who's going to go to work for the startups make this stuff work and so we were discouraged out of the gate we were wondering whether there would be AI startups that were viable just because of the talent restriction and of course the same question that had to be asked of how existing companies and other industries right how our car company is possibly going to adapt how our aerospace company is going to adapt or medical technology company is going to adapt if they can't hire the people I think it's been dramatic in the last three years how much the field has really opened up and expanded and you know obviously this conference is a great example of that and it's sort of this all of a sudden it's like the rush of enthusiasm right couple of the the real-world results has led to number one just a massive increase in the number of people interested in the field a massive increase in the number of students at university learning about this stuff and then also a massive kind of re-education happening inside the industry right and a lot of you know many many programmers today are we training themselves at AI in fact one of the companies we've back to Udacity it's actually funded by they're created by an AI pioneer out of Stanford and actually it's a it has online courses to basically help programmers retrain out AI and so we see kind of a dramatic retraining taking place and then also this rise of open-source right the right you know tensorflow and all these other technologies and then the rise of the cloud AI offerings from the big cloud providers have really opened up the field and so the you know the number of people who are going to be able to basically do AI from a technical standpoint is expanding very fast so that's I think the reason for optimism that a lot of companies are gonna be able to adapt the scary thing or difficult thing is you know this really shines a spotlight on which companies are really set up to run top-flight software engineering efforts right and which companies are going to have top-flight software engineering cultures and you know say what you will about you know the issue is the Silicon Valley companies like Silicon Valley companies know we know that we live or die like on the basis of our technical technical cultures and our ability to attract and retain top end software engineers you know there's a lot of companies out in the rest of the world that it's like well the important people are you know the business people or the marketing people or whatever and like then we have this IT department and the IT departments kind of kept in the back and they're not really invited out of the back they kind of stay back there you know they dress different like they're you know they're funny habits and we don't really want to send that much time with them and they're they're not top-flight members they're not they're not important influential members of the organization in the way that you would need if technology all of a sudden that's become the most important thing and I think there's a lot of companies the car companies are great example this but a lot of the incumbent car companies have an opportunity and a challenge to figure out can they basically make software engineering as important right in their company as Hardware engineering has been historically and I think honestly it's an open question I mean Tesla hires way more engineers than the other ones so yeah make sense it's a culture of change and a process change it needs happen I want to switch gears a little bit I know we were almost out of time but you know you started that scape you've been a big proponent of open source you know being behind a lot of these things in the past how do you think open source business models around that will evolve especially now that we have cloud and cloud vendors and it's being hosted in the cloud what are your thoughts on that yeah so it's really fascinating so open source you know this has been a big week for open source for those of you who follow these things Microsoft you know buying it up is one of those amazing events that you know ten years ago it would have been like saying Donald Trump will be president you know just like thank you this thing flat out implausible and yet here we are and so you know it's just like I mean that in the sense of like it's just amazing how far Microsoft has come as a company right it's sort of Microsoft 10 years ago viewed open-source literally like the two words that they kept using were cancer and communism which is kind of amazing if you think about it because it's like isn't cancer bad enough without also being communist like of all the forms of cancer to get Communist cancer is the worst so so to go you know and obviously Sasha and that the team at Microsoft is there have you know gigantic credit for having navigated that company to a point where they've not only fully embraced open-source like they're doing these amazing things on development of open-source you know they built some of best development tools you know with github they now have the best the best the best of the best code environment and so you know that's just like a great example of a company that's evolved you know from basically you know butt-flap hate all the way through to love and embrace and like really doing amazing things in the field and so I think that's a great specific instance and then just more generally I mean the scope for this way open source I've been involved in open source so long at this point I started working at open source when literally you could work on open source richard stallman this is actually true richard stallman who you know is kind of the Godfather of his he calls it free software but the origin of a lot of this he literally was he believed in open source so thoroughly that in the early 90s you could actually remote log into his UNIX account at MIT anybody could username RMS password RMS and you could be Richard Stallman and like and it was because egalitarianism right his sense was basically people are good people want to contribute people want to build things and if somebody wants to log in as Richard Stallman and work on his code while he's asleep fair enough go for it I used to it you know middle of the night there I am and so that that went from kind of that extreme and idea at that point to something that's now this mainstream right the world we live in today all of our smartphones like every smartphone that everybody here in this audience is running fundamentally on open source right some derivative derivative of UNIX and so the open source snowball I think is now rolling down the hill in a really amazing way and as I said I think it's it's become absolutely central to AI like it's amazing how I went from being a black box technology to something that's now usable usable on open source and then I also think like open source and cloud are like a fantastic fit because right open source you know makes makes technologies very widely available and then cloud makes open source like an open source derive technologies much easier to deploy at scale right and so to be able to have kind of a continuous kind of thread where you can learn about these things and program these things but then also deploy them at kind of you know supercomputer level scale in a practical way is a really big deal and so I think it's you know there's never been a more exciting time for software and I think that's a big reason why also that's great to hear I mean especially Frederick's we've got an open source bet on the cloud on AI so I have a personal question towards then so I want to close with this when ben horowitz came to you and he said you know i have this these are some crazy guys out of berkeley they have a you know this thing called spark you know I won't invest in them what questions did you ask and what did you think whatever anything you did to pressure test or stress test well the problem is weed hurts have really sketchy things about the CEO so you know what happened normally we do a full battery background checks but then said in this case we should probably just not even run those but though in all seriousness know that you talk about 10x advanced like the the nature of the 10x advanced that these guys had done and this old community had done was I you know was we thought just obviously obvious and underrated by their by the rest of ecosystem at the time and so that that's the best kind of investment I see also well I'm glad that into those background checks take some okay great thanks everybody [Applause]

Channel: Databricks

Views: 7,189

Rating: 4.8666668 out of 5

Keywords: ai, ml, artificial intelligence, machine learning, databricks, apache spark, spark + ai summit

Id: bI3rAJcTbtI

Length: 20min 48sec (1248 seconds)

Published: Wed Aug 29 2018