OpenAI’s First Investor On Why Robotics Will Have Its AI Moment Next

0:01 let me ask you what was it like being the first investor in open AI um you know were were you just putting money on on 21 let it ride you know did you have Insight that you thought it would be big is it what you expected and what can you tell us uh as an audience uh as you've seen that kind of evolve um that we should know that you only you know well the thing to realize is you have to

0:27 have conviction about what you're doing these things especially looking into the future as Yogi Bara said uh is are very hard to predict uh but you have to have conviction there's a directional uh Trend going on that's important I've long believed in AI as a directional Trend wrote about it in 2012 uh thought it'd be important and hence when the open AI opportunity came about which was we made the decision in

0:58 2018 to invest uh twice the amount I'd ever invested in any other company um as a first investment and that was because I had conviction about what AI could do I had no idea what the timing would be or when the Breakthrough moments would come but directionally I felt there was so much talent going into AI it'd be really really important venod will we be capable of having free AI doctors for every person

1:29 AI tutor for every child um uh 7 to 11 you know that store 7 to 11 no no 247 no question about it you know this is not even a question anymore with the capabilities of AI uh we have to go through the full process of FDA approval of an app as a primary care physician which sounds crazy to people but almost certainly can happen if the a FDA lets

2:00 it happen no question if we do a oneon-one comparison of 10,000 encounters uh between humans handling it and an AI handling it the AI will do substantially better especially the more diverse the community the most diverse input cases that will absolutely be true venod will we have a billion by pedal and other robots freeing humans from the servitude of undesirable jobs in the next 20 years or is that not

2:32 going to happen uh I I believe uh robotics is coming up with a chpt moment here pretty soon in robotics in the next four to five years uh could be as soon as two or three years uh where robots aren't programmed anymore they are learning systems and I think that'll be the seminal change in robotics uh and they understand the physical world they understand physics um real world Dynamics all that kind of

3:03 stuff uh and because of that I'm I'm actually pretty optimistic that many of the things that have looked very stunted and terrible because we are humans are trying to program them and those are like rule-based AI systems of old uh we'll see this rapid change um so I'm I'm pretty optimistic we'll see that happen and then you nobody has to work at a GM assembly line for eight hours a day for 40

3:35 years that's that's not a job that's servitude so Veno did you ever use Punch Cards to program I did s on Punch Cards I started my programming career in India on Punch Cards so in in 32 kilobit machines yeah you you could probably guess where I'm going with this um will there be a billion plus programmers all programming not in punch cards but in human language dramatically increasing the scope of computers will computers adapt to humans

4:07 not humans adapt to computers like the Palm Pilot they couldn't get the Newton to work to read your handwriting so Jeff said hey let's just have people adapt are we gonna have a a Tipping Point where we flip back um I absolutely think so those are two entirely different concepts uh but let me explain them just a a little bit uh almost certainly uh people will be able to program in

4:37 natural language um and and so many more applications essentially programming will also be near free and many more people will be able to do it hence the use of computers will expand dramatically you might actually do use one software because so easy to do almost your disp aable or throwable packaging for your shampoo bottle you throw it away after use um I

5:09 I think Computing will become more like a utility uh more like electricity in the background you you most people most of the time won't be like sitting in front of computers uh hopefully things will happen for them uh I'm pretty optimistic about that the other major change we will see is uh you know so far we've had to learn how to insert a row row in Excel or use the menu to do something or God forbid

5:39 the macro programming language in Excel or learning to use sap uh computers will learn humans instead of humans having to learn computers I think that'll be another major Trend and you know devices like the rabbit you can talk to and they do things that you need done and they understand it break it down into the various tasks that might be needed uh I'm pretty excited about that so computers will become a lot lot easier to use vode um do you um do you get out

6:14 to movies do you like entertainment I love entertainment and design and I know where you're going yeah so my my NE yeah so my my next question is um how will AI play a role in entertainment and design like some people say uh you know the Beatles won't happen and then this this next wave of AI because uh you know you don't need a group you can use AI to do it will the human collaborate with the AI and and be

6:43 creative and what's the future of entertainment design is that a dead end opportunity or is that just getting started it's just getting started uh five years ago when when I first talked I give a talk in Toronto give a talk about AI generating music it was met with real skepticism even derion not only do we will it happen but also that we don't

7:14 need it uh I think this notion some techno haters have of uh some human elements uh will will will sort of disappear I think uh I think AI alone or AI Plus in conjunction with humans will have much more creativity we'll see much more what what sometimes is called soul in the music um in uh we

7:47 also see much more diversity of entertainment of the not only music but everything else and so I think broadly we will see much more content near free much more available much more personalized to eat you might get music for the mood you want to be in just before you give a talk or just before you at an athletic event you might get very personalized music for your brain uh to you so um now I don't want to say

8:18 the the celebrity fan relationship's going to change you know T TOA Swift will still be there and swifties will still be there but that doesn't take away so this AI music doesn't take away from that relationship and that kind of performance or entertainment um next question is just a one-word uh uh answer are you a Swifty no okay all right internet agents access what what comes to mind on those

8:50 words you know um I think most use of the internet will be by agents it won't be by humans uh 247 there'll be billions and maybe tens of billions of Agents running around multiple ones for each of us doing specialized things for us uh almost if each of us had multiple assistants to do things for us so most

9:22 trained in that access in my view will be by agents not by human beings that might have some implications on how content on the internet needs to change um so venod uh I'm curious open versus closed uh what are your thoughts on that and um you've been uh are you passionate about one way or the other and has your opinion uh changed over the years on that you know so uh my opinion hasn't

9:53 changed a lot you know sun was the first really open source company NFS was what of the first major product you think of even before Linux uh so uh that's uh pretty exciting uh you know we invested in gitlab so I've been a big fan of Open Source now in the case of state-ofthe-art models I feel like open source shouldn't be prohibited it's like open sourcing the Manhattan Project because we have an

10:25 adversary in China and I believe the AI races so important whichever country whether the West wins it uh or liberal value countries build it more generally because I consider uh many many countries around the world liberal values companies or China wi it will have huge implications for the planet and I hope we win it and and hence we shouldn't be open source saying state of the art models everything else is fine in terms

10:58 of AI what you most excited about and and what what keeps you up at night and you're most worried about you know the the hard thing in AI is to predict what capability will show up uh GPT 5 hopefully is coming soon what will GPT 6 be like when will GPT uh X start train training X+ one uh those are a set of things to worry about it impacts the rate of change

11:30 um I wonder if once we have ai programmers like Devon are we going to see really really interesting progress but at some point these AI systems will be designing other AI systems and then you have um real issues there um that's on the on the model side I'm also pretty optimistic about something most people don't believe that within the next two two years or three years we'll start to

12:02 see uh more more interesting alternative approaches that add substantially to the llm model is the right way to say um so you might see other approaches like problemistic programming there's a group at MIT very interested in that you might see symbolic logic uh I'm pretty excited about a company we have called symbolica that's doing something called category theory in fact it's funny nobody's heard

12:33 of it but I've run into two or three startups doing category Theory including one at MIT um so these are esteric areas there they're sort of the what open I might have looked like uh in 2018 um you know so the theory is there the attention paper was out and then you didn't see its implications so I think they will all all be additive to uh current models uh so that's sort of the model side I think will continue to

13:05 evolve pretty substantially um then we have the application side and we've sort of talked much about it you know uh there'll be a lot of co-pilot type applications almost um Everybody uh should be able to use a co-pilot for most things they do why a physician would ever have to use epic I don't know they shouldn't need to uh an AI will do that for them uh but areas are and

13:38 there'll be lots of new consumer applications how shopping is done uh we've talked about music or games will all be impacted there'll be regular Enterprise but the area I'm most excited about is when you replace human expertise uh we've talked about near-free doctors we've talked about near free tutors uh I think we'll have near oncologists Structural Engineers uh you name it will so expertise we'll have geologist so we

14:10 can discover a lot more mineral resources that's an area I'm very interested in uh so that that that site will also go through a lot of innovation in fact I think a lot of media will be reinvented too we have a book publisher that uses AI in the can they're 10 acts better than the best book publishers in in picking out hits that might be million dooll wins which is a good level for a book so venod um what advice do

14:42 you have for college students uh who are thinking of dropping out or maybe finishing on areas to focus on is software engineering Dead uh and what skills are needed I don't think it's fair to make any assumptions um I think the best software Engineers will become 10x or maybe 100x Engineers well they'll still be needed but there'll be a lot more capable a powerful Amplified by AI

15:12 um simple programmers will become better programmers what I would say in general uh to young people is be as diverse as possible in the things you learn don't plan on learning your career now but do the courses or that help you learn how to learn so you can adapt rapidly the one thing for sure we will have to keep changing what we do nothing we learned today in class will be relevant 10 years

15:43 from now uh so mostly what you have to do is learn to learn and be able to change Fields rapidly did you know that different advice for PhD students that's a very different thing yeah um let me ask this did you know that there was a board kind of Shakespeare drama at open AI were were you aware of that no I didn't hear about it yeah so you know this is an intimate conversation you and me here do you want to say anything

16:11 weigh in on that that no one knows what really happened or went on well the only comment I would make in general is uh is the broader issue is one of how you handle risk AI systems do have risk but we have lots of risk as Society we uh an asteroid could hit the planet too that's a real risk another really big risk is another pandemic so Humanity faces a

16:42 whole basket of risks and so you learn to manage them you learn to pay attention you don't go overboard on one thing especially you know EA was a Jonestown like religion it's just completely irrational it pretended to be it's completely irrational uh and so um ignoring every other risk in humanity ignoring all the benefits of humanity um is just

17:14 silly um how do you like how do you use AI like when you wake up in the morning you know does it brush your teeth you know are you an AI right now do you use AIS to um you know how are you using AI and how has that changed from a week ago a month ago uh five years ago so before I go I would say to you I think we have to take a long view of AI even AI

17:43 applications and the silly you know you have to realize the press needs headlines so they have to make headlines and they have to show how silly X or Y is or why this hand that mid Journey produc has the six finger is not five by the way nature does that sometimes too um but we have to take the Long View all of these systems are so new version 10 of them will be mature or somewhere near

18:14 what we might consider mature and so we have to be patient and say they they're good at some things they're not good at others and what we have to do is use them enough T the flye data flywheel and what what doesn't work so we can fix it in every generation every year will be better than the last and that's what I'm optimistic about the rate of growth will become exponential systems that evolve to be better are almost always better

18:43 than engineered systems that are precise in one one dimension and once AI gets to be really Learning Systems which is it isn't quite today I think we'll see even pasted progress yes um so my last question to you uh venod is um are you having fun I'm having a blast I mean how can it not be fun working on all the things we've just talked about and I work on those and a lot of other fun things like

19:15 fusion and mark five flight and all that kind of stuff well thank you very much yeah ladies and gentlemen you've been venod coad venod [Applause] kosla e