The Future of Trusted AI with Marc Benioff and Sam Altman | ASK MORE OF AI with Clara Shih

so one person with a really good idea and a deep understanding of what a customer needs is going to be able to execute on that with what would have taken complex many many person teams before and so giving giving someone the ability to do more uh I think is going to lead to a big shift welcome to ask more of AI podcast at the intersection of AI and business I'm Clara shy CEO of AI at Salesforce this week we're coming to you from dreamforce the world's largest AI event where we just heard an amazing conversation between Salesforce CEO and founder mark benioff an open AI CEO and co-founder Sam Altman Sam covers everything from what he's learned and how he's been inspired by his favorite AI movie her as well as what's been surprising to him about the development of AI take a listen [Music] Sam welcome to dreamforce thanks for having me oh it's my pleasure and uh Sam you know you've been traveling all over the world I haven't really seen you since you got back I saw you just before you left I know how excited you were to go to all these different countries and to give them the prophecy of artificial intelligence The Good the Bad the Ugly what was your biggest surprise making all these trips the the level of enthusiasm hopefulness excitement around the world uh of course balance with wanting to make sure we successfully addressed the potential downsides uh it was really extraordinary to see and you know it's a little bit different in different places but I sort of thought well maybe this is still just mostly a tech Silicon Valley phenomenon and to see what people everywhere around the world were doing with the technology and how they'd incorporate into their lives and how interested uh and hopeful they were was very cool which country really stood out to you as potentially a great leader of artificial intelligence that was another positive surprise the the quality of work happening everywhere um was was really something I think the United States will be the greatest leader in artificial intelligence we have uh we're very blessed to have so many things in our favor but this will be a truly Global effort and people will contribute all together all around the world did you meet somebody who really offered you an incredible Insight or inspired you or somebody that you said you know this really changes my mind on this aspect of AI was there someone out there that you uh were impacted by I mean a lot of like you know really super impressive world leaders had a lot of good advice for me and for the company and that was helpful but the biggest takeaway uh was just talking to our our users around the world and out of all of that and out of what we we heard that they uh they wanted that they had problems with with the technology the features we were missing uh that has really fed into what we'll be launching uh in a couple of months and uh that was probably the that was definitely the most meaningful feedback we got Indonesia gave you the golden Visa was that inspired by your love of Eat Pray Love um a lot of people a lot of people that I met with on the trip were like oh you need to like come here more we'll give you a golden Visa so I got this little collection in my passport now which is awesome and unexpected I thought that was a great book you've got a sparkly Visa yeah passport is a shining bright uh it's passports they're very beat up after the trip I gotta fell in the water a few times but that's okay you swim in with your passport not intentionally but it did happen once um let me ask you you've uh you know you've you've thought about AI probably more than most people on Earth and you've been really having the opportunity to interact with so many great AI leaders not just all over the world but here in the headquarters of artificial intelligence San Francisco um you know what what's been your biggest surprise in the last you know seven or eight years of open AI you know here we are it's 2023 you know we're in September your GPT 4 is coming up on one year you know what what has been the big surprise over the last eight years gpd4 has only been out for six months which is just a you know a good reminder about how fast things have been happening the biggest surprise is just that it's all working um like when you start off on a scientific Endeavor uh you hope it'll work you let you you let yourself dream that it'll work you kind of have to have that the conviction and the optimism but you know when we got together uh at the very beginning of 2016 and said all right we're going to build artificial general intelligence uh that's great but then then like then you like meet cold hard reality and we had a lot of stuff to figure out and figure out any new science is always hard and in this case for a bunch of reasons it was particularly hard we had conviction in a path uh laid out by our my co-founder our chief scientist Ilya um but having that conviction and then actually getting it to work you know this was I think something that very much consensus wisdom in the world was this was not going to work and it's just through the effort of a lot of enormously talented people that it did but that's probably the biggest surprise when did you know that this was going to be a success and that things were going in an exceedingly right direction um sometime a little bit after gpt2 which would have been you know 2019 at some point now as you kind of see gpd4 maybe as kind of a North star going forward you know what's the next big step for openai to kind of get to where you want to get it to uh two things one on this current Hill that we're climbing with the technology of the GPT series we're going to keep making it better we'll make it more reliable more robust um more multimodal better at reasoning all of these things and we also want to make sure that it is really useful to people in all the ways that it's transforming things keep happening uh so you know we're now deep into the face of the Enterprise really adopting this technology and getting the systems to be very secure very highly trusted handle data appropriately uh not hallucinate or at least not when you don't want them to hallucinate um that's that's leading to enormous transformation for a lot of companies and we want to keep doing that and then the other thing I'm going to do is figure out the remaining research from this Paradigm we're in right now to something that we could all truthfully call AGI what's been the most complex part of dealing with the hallucination problem well there's a lot of technical challenges and then we could talk about those but the one of the sort of non-obvious things is that a lot of the value from these systems comes is heavily related to the fact that they do hallucinate you know if you just want to look something up in a database we already have good stuff for that but the fact that these AI systems can come up with new ideas can be creative uh that's a lot of the power now you want them to be creative when you want and factual when you want and that's what we're working on but you know if you if you just sort of do the naive thing and say never say anything that you're not 100 sure about you can get a model to do that but it won't have the magic that people like so much if you do it the naive way when you think about your own company and how that kind of reflects back to you I know how focused you are in trust and how focused you are on kind of the deep ethical backbone of open AI what are the core values of open AI going forward what what things in priority do you think are most critical for you to focus on we want to have the smartest most capable most customizable models out there because we so believe in terms of what this will enable for Humanity and that the good will be orders of magnitude more than than the bad but we also want to make sure that our models are aligned and that Enterprises can trust us with data and that we're very clear about our policies around that um and that we're keeping not only the data secure but also all of the uh all of the learnings that a model has after ingesting a company's or a person's data and so keeping the model very secure itself and that balance of continuing to push on the capabilities but also making sure that um from a safety and privacy perspective we are keeping Pace with the capability uh we want to continue to lead on both of those how are you managing between alignment and intelligence so this is very common question right like are we doing enough on alignment relative to capabilities are uh you know should we be slowing down work on capabilities to work more on alignment and I think the frame of it is nonsensical uh it really is the same thing like that What's Happening Here is we have this amazing new piece of the human Tech tree called Deep learning and that approach is solving lots of problems and that same thing which is helping us make these very capable systems is how we're going to align them to human values and human intent and when we make something that is considered an alignment breakthrough rlhf which is how we take the base model and get it to follow the instructions that a human user has that's an alignment technique but it leads from the model it leads to the model going from not very usable at all to extremely usable and so in some very important sense that's a capabilities thing a capabilities gain as much of an alignment gain same thing with interpretability we can look into what these networks are doing that does for sure help with safety in all kinds of ways but it also helps with capability and so there's there's more one-dimensionality to the progress than people think and we think about it like a whole system we have to make a system that is capable and aligned it's not that we have to make a capable system and then separately go figure out how to align it that's definitely been a surprise to me to see how you've LED through rhlf and all these different kinds of Next Generation alignment models got them to work uh invented I think many of these techniques as well that were not really part of the a constant and kind of continuous AI narrative even as I would say recently is the late 1990s we didn't and then into the I would say uh first part of the 21st century I think when I read some of the textbooks that are like even in 2015 2019 the models that you've invented aren't even mentioned are you surprised at how much core research you're doing no we thought we were going to have to do a lot um and we call ourselves a research lab and we very much are you know most this sort of standard way that tech companies work is you start a product company at some point when you're like generating enough profit you're like I'm going to start a little research lab on the side and you know the challenge there is they don't usually go very well we started as a research lab and then we started a little product effort on the side and the challenges we have to make that go very well yeah what's been the scariest thing that you've seen in the lab can I finish one more thing on that geez no I just had a really good question come on I'll get there um our in terms of how we make progress and are we surprised uh not really we are empiricists we know we're going to be surprised we know that reality has a different way of playing out than you think it's going to when you sit there and make your beautiful theories in the Ivory Tower so uh we're never surprised and we just try to meet reality where it is and follow the technology where it can go um and I think that more than any other well there's one more thing which was were extremely rigorous and careful combined with a lot of team spirit and that combination I think is like hard to get a whole company to be rigorous about every part of a system that has to come together but those are two of the things that I think are special about openai um scariest thing we've seen in the lab honestly nothing is nothing super scary yet we know it'll come we won't be surprised when it does but at the current model capability levels nothing that scary or you're newer in India I was listening to one of your interviews and I was really taken with something that you said about intelligence that you said that as you look at these large language models and specifically how they're coming up with their answers that you think it speaks actually or is a reflection back on human intelligence and what maybe what human intelligence means what were you trying to say I don't remember exactly but probably that intelligence is somehow an emergent property of matter to a degree that we probably don't contemplate enough and that it can happen it can happen with electrical signals flowing through neurons that are reconnecting in certain ways it can happen with electricity flowing through silicon um but it's it's something about the ability to recognize patterns in data it's something about the ability to hallucinate that to create to come up with novel ideas and have a feedback loop to test those uh and as we study these systems which are easier to study than the human brain for sure you know there's no way we're gonna without doing a lot of collateral damage too there's no way we're going to go figure out what every neuron in your brain is doing but we can look at every neuron in gpt4 and look at every connection let's go a little bit step deeper because I think when you talked about that in India I was really surprised you're talking about intelligence as this emergent property of matter do you feel that that's what's happening then in the lab that you're seeing this kind of wake up that you're starting to see the kind of intelligence emerge from the software that we have inside the people in this Auditorium um well I don't think we're seeing anything wake up but I do think we are seeing intelligence emerge from a very complicated computer program and how does that go forward I think the current GPT Paradigm we know how to keep improving and we can make some predictions about we can predict with confidence it's going to get more capable but exactly how is a little bit hard like when you know why a new capability emerges at this scale and not that one we don't yet understand that as scientifically as we do about saying it's going to perform like this on this benchmark um but I suspect I'm pretty sure that there are major new ideas to still discover and that if we assume that this sort of GPT Paradigm of the world is the only thing that's going to happen we're going to be unprepared for very major new things that do happen like what kind of things the one that I would be tempted to say is the most important is the ability to reason gpt4 can reason a little bit in some cases but not in the way that we we mostly would use that term when we have models that can discover new scientific knowledge at a phenomenal rate you know if we let ourselves imagine a year where we make as much scientific progress as a civilization as we did in the previous decade or even the previous century and think about what that would do to quality of life what's possible uh that's that's pretty transformative I thought it's really interesting you know thinking about intelligence as an emergent property of matter um and that feel like there is this unremarkable but also highly correlated connection between like working with these chat Bots but also thinking there is a some human aspect of it is it coming mostly through the training or do you think it's actually inherently there the human aspect of talking to a chatbot I think it's coming mostly through the training um we're really training this in some way to like be all of humanity like that's that's what it trains on is the output of you know a huge fraction of humanity and what would be the next step well we can talk about the obvious ones and the ones that are going to be more speculative so the obvious ones are just the models are going to get dramatically more capable they'll be dramatically more customizable and dramatically more reliable um and and really I think that that those you know the model itself uh that is the fundamental enabler of everything else um we will continue to build the features around the model uh like we're talking about earlier for Enterprise class usage and we'll continue to build consumer products like chat GPT to make it easier for people to just start playing around but in the same way that the internet and then mobile just kind of seeped everywhere that's going to happen with intelligence and you know right now people talk about being an AI company and there was like a time after the iPhone App Store launched where people talked about being a mobile company but no software company says they're a mobile company now because it would be like Unthinkable to not have a mobile app and it'll be Unthinkable not to have intelligence integrated into every product and service it'll just be an expected obvious thing um and you know companies will have their AI agents that can go off and do things and customers can interact with and all sorts of other things that we're seeing happen right now um but this will be a big shift in terms of how we interact with the world and with technology when you think of dramatically more capable models what is a dramatically more capable model can you try to describe it to us um I think it'll happen in all sorts of different ways but one example A lot of people use chat gbt to help them program and you know maybe they say hey it writes 25 of my code and then it'll go up to 50 and then 75 and then 87 and a half and then 90 and then 95. and at some point it's not just doing more code but it's letting you do things you just couldn't do before there's uh you know I'm a big believer that quantitative shifts lead to qualitative shifts at some point and so if you have better tools if you can operate at a higher level of abstraction if you can keep more of the the big picture problem in your mind at one point you can just do dramatically more and that's difficult to say exactly what that's going to be like but if you you know we could like reach back into history for an example and if you think about the kinds of problems that a programmer let him or herself dream about when they were working with Punch Cards versus what you can think about with a high level language of today versus what you can think about where you can just say in natural language this is what I want a program to do and I know how about this actually that's a really interesting idea that I just saw emerge here how about this um and the the cycle time in the the the sort of iterative feedback loop is so different than it is today that will change what a single programmer is capable of that will change what a single person running a one-person company is capable of and I'm tremendously excited to see I remember in one specific dinner that we were having I was describing all the different things that I wanted to do with Salesforce some of them that we just showed at the dreamforce keynote and then other things and other things and then I just kind of shook my head and I just said I just don't know how I'm going to hire all the people do do this work and then you turned to me and went Mark you idiot you're not going to have to do any of that the computer is going to do it all for you is that what you're saying not not quite um I'm smart that I'm an idiot or the what I'm saying is that the the capab the the the amplification of one individual's capabilities so one person with a really good idea and a deep understanding of what a customer need needs is going to be able to execute on that with what would have taken a complex many many person teams before and that ability to uh give people these tools and let things happen with less resistance with less friction faster easier um it's very easy to kill a good idea you know it just takes like one person to be a little bit less than supportive they don't even have to say all right no like creative ideas are very fragile things and so giving giving someone the ability to do more I think is going to lead to a big shift on a bigger message you know I think a lot of families try to encourage their kids to go and become computer programmers and take coding classes and coding camps and all of these kind of things that do you think those kids are going to come back and really be quite upset at their parents I remember my grandma trying to tell me in like second grade or something that was really important I paid attention and learned cursive and like class I don't even think they still teach it probably they don't but I was like I have very bad handwriting and that was a real struggle for me anyway and you know I like cheerfully went through it I knew at the time that this was like no way this was gonna be important because I was already like typing on a computer um but I'm not upset about it [Applause] when you watch all these kind of unusual movies about artificial intelligence you know Minority Report or Hal or her or uh war games or all these various things you know there's been so many of them you know which one is your favorite I like her which part of her do you look at and go this is never going to happen well I mean a ton of it that I think that's not the fair question because and looking back at sci-fi it's always easy and you just fill in that last time that part that you know which part is going to happen which part do you think will never yeah yeah so so but I I think it's like unfair to dunk on Old sci-fi for all the parts that got wrong and we can we can cover that but like that's always true what's amazing is that people get it right at all and the number of things that I think her got right that were not obvious at the time like the whole interaction model with how humans are gonna use an AI this idea that it is going to be this like conversational language interface that was incredibly prophetic and certainly more than a little bit inspired us so it's you know it's not just like a prophecy it's like an influenced shot or whatever um but I think this idea that we all have like a personalized agent trying to help us um and we talk to it like we talked to you know Chad gbt um that was actually not what most movies I mean most movies thought if we interacted with an AI at all it was gonna be like robots shooting us or something um but you think that that will be happening soon I do not I do I realize it like is a compelling movie but no I don't I think you know I'd it'd be great for Hollywood to have some new tropes um but yeah I think her got something deeply right on the interface and that is no small feat one of the hobbies that you have and we've talked about it is you think about if you actually had to prepare for a moment in life when you had to kind of escape and kind of protect yourself um where did you find that interest in that that kind of idea I mean I think like a lot of little boys I thought Boy Scouts was really cool and I like outdoor stuff and survivalism and I have no delusions that any of that is going to protect us or me or you or anyone if AI goes wrong and I sort of think it's like silly that other people assume that's what it's for it was just like it's like a Boyhood hobby that's stuck so you're not escaping into one of these Indonesian Islands now that you have a golden visa and that's your new um the speed of light is really fast I think the AI can get there much faster than I can get there um you know you have such an interesting background in life and the first time we ever had a meeting uh it was because you came to me and you had some political interest and we were at my home and we were talking about your political interest and I was really it was many years ago and you're very very young um you know where did that political Interest come and where did it go and is it still in there um I love California I would like to spend my whole life living in California I was concerned then I'm concerned now about what's happening with the state um I think everyone has some civic duty to think about what they can do uh so I was like you know lightly looking into it but I don't think I'd be a good politician at all and I think I have something else to work on that's really important to me um but I think you'd be very good I don't think so that's really nice of you to say um I don't I wouldn't sound like a super self aware I had some interest around that I had like a desire in seeing the state and the country at this point too get better but I think we've all got to figure out the way we can contribute I think the way I can contribute is scientific and technological progress um but I do think we need to get more people to jump into politics well you're a very young man and you know when you're an even much younger man you did have a very big political interest and that was before you started open Ai and now you have you've obviously now come out on a global stage as clear to people you're the Visionary capabilities that you have and Leadership capabilities that you have tomorrow you're going to go back to Washington DC you've already been out there quite a few times tomorrow you're going to interacting with I think mostly with the Senate so as you meet with maybe just talk about the United States for a second and dealing with the white house or the Congress or the Senate and and trying to explain to them what's happening here in San Francisco and the technology Revolution that you're leading you know what what is your message to them and what has been your biggest surprise dealing with them pleasant surprises for the most part I think our leaders are taking this issue extremely seriously they understand the need to balance the potential upsides and the potential downsides but you know Nuance is required here and I did not have I did not go in with particularly high hopes about the balance of that about that Nuance being held appropriately and it really it really is um I don't know exactly this is all going to play out but I know that people seem very genuine in caring and wanting to do something wanting to get it right and the message I think our role is just to try to explain as best as we can realizing that we don't have all the answers and we might be wrong technology takes weird turns sometimes what we think is going to happen um where the control points can be uh where the dangers might lie what it takes to make sure the benefits are broadly distributed and then let the leaders decide what to do when you have these discussions with these political leaders which part of it has been the most shocking or surprising to them the first time that they see uh you know gbt4 they're usually does not usually exceeds their expectations I'd say it's the technology demo and realizing that you know this thing that people have been talking about for a long time is here has that also been true on the international leaders as well uh yes although by the time of that trip most people had by then already played around with it to some degree like when we did that trip it was sort of like Chachi petite already had its news cycle if you would have one goal with the US government between now and let's say next year when the elections start and potentially there'll be changes and in administrations and so forth what would be the goal do you think for the next 12 months from a political perspective I think getting a framework in place that can help us deal both with short-term challenges and the longer term ones uh even if it's imperfect starting with something now would be great it's going to take people a while to learn I think solving this legislatively is actually quite difficult given the rate of change uh you know I think an agency a new agency would probably make the most sense um but getting something going so that even if it's just focused on insight and not oversight so that the government can build up the muscle here I think would be great so are you suggesting the CIA should change their letters into the Cai uh I mean I can imagine worse agencies of less competent people to hand it over to but I really do think a new one would be appropriate um what part of all of this are they just getting really wrong this is a very human thing I mean I don't just think it's government 's getting this wrong I think most of us get it wrong to different degrees too uh you show people an exponential curve of Technology and they believe you that it's been exponential but they don't believe you it's going to keep going exponentially I believe it's about to level off and that happens in all sorts of subtle and not subtle ways and it's a very difficult it's a very difficult bias to overcome particularly when if you accept it it means that you have to confront such radical change in all parts of life and so what are you doing to help them understand that um I mean a bunch of things one is I believe in the power of repetition here another is to just keep showing up every year and say hey that thing you thought was impossible or that thing you said oh well I don't really have to worry about this that your last Model couldn't do it try our new one see what happens the last few times I've arrived in the United States I noticed that the Customs also has now a camera and sometimes they don't even ask for my passport they just put up the camera in front of me and the AI is able to know it's me we know in other countries especially certain Asian countries that there's a lot of that type of Technology already in place do you think we're moving into a surveillance economy do you think that this technology will accelerate the move into greater levels of surveillance I do one of the things that I struggle with is I don't see a world where if AI is as powerful as we think and people can do significant harm with it I don't see a world where we don't have less surveillance and I don't think that's a good thing but I have talked to a lot of people about this and have not yet heard a great solution I know that you like to just kind of get away and maybe get on a small boat somewhere and so like a sailboat and and uh be with a friend or two and just Sail by yourself and completely disconnect um when do you think it'll be that you're going to take that trip and that sailboat will not be quite so disconnected and we'll know where you are no matter where you are and maybe there'll be some autonomous characteristics to the boat as well well we have Starling now like that's in some sense you know that's a transformative thing I think even more than people realize um I'm okay with that I'll take that trade for sure but I think this the sort of idea that I'm gonna be like off on a sailboat unreachable unless you really want to be like that's that's over when you now look out at your product pipeline for the next couple of years with open AI you mentioned a few things you especially mentioned these remarkable new models you use the example of you know the computer science model and the ability to write any code and maybe become a whole engineering team can you take another industry or another application area and give us an example of something else that you think will be a remarkable achievement in terms of a model moving forward I'll mention a few of my favorites um what what's happening with education is incredibly gratifying to us to watch um and the Ingenuity of teachers of entrepreneurs building edtech companies and also just of students themselves finding new ways to use Chachi BT to learn is quite remarkable we know what a difference it makes to a student to have a great one-on-one tutor it's like two standard deviations better outcome or something but Society before AI just like realistically was not going to be able to afford that and so we didn't even let ourselves think of how we could deliver it and now I think we see a path where with a combination of humans and AI together we can offer this to everybody in the world and if we can deliver on that and again it looks to me like we're going to be able to with the technology that that that will be a transformation for the world that I think would be a huge Triumph in healthcare I think we can imagine a world where not only do these systems go help us cure a lot of diseases but also that the actual Healthcare product we deliver to a person um again this hybrid of a doctor and AI together means we can just offer something far beyond what we think of as even possible today um in Creative work it's actually quite remarkable to see what's happening what an what a visual artist for example can now do with the latest damage generation tools we're all going to get like better art than we've ever had before are you especially surprised to see that level of creativity and art and and how it's connected back to these systems in some sense yes if you had asked me to predict um 10 years ago the order that AI was going to disrupt Industries I would have said physical labor first cognitive labor second and maybe maybe never but certainly last creativity and I hadn't actually thought that hard about it but it's like what all the experts were saying it was what like my college professors said whatever um but it was what everybody thought and for creativity to go first which clearly it has um is an update but also as we started working on it and we kind of saw where the system strengths were and where the weaknesses were um it's a kind of area that can tolerate the flaws in our current systems quite well yesterday we gave a 20 million dollar Grant to our local public schools to help them augment their understanding of AI and we did some seminars and the one in high school that had the most attention was on music and kids were just fascinating with how the AI was able to work with them with music but also wanted to understand all the legal aspects the trademark issues and and and brand issues and everything else associated with music when you look at the music industry overall are you surprised to see the kind of this dramatic change in music and the level of interest in AI music from kids I mean not at all surprised on the interest people love music and it's really important part of life and what people can do with these tools is great um unfortunately I think the music industry has a reputation such that companies just don't want to deal with it and are focused on other areas and I think that's a loss because like we're seeing with visual art I think this will be a tool that amplifies humans not replaces them all right well thank you Sam thank you very much great interview please thank Sam Altman would you [Applause] what a great conversation three takeaways for me first that creativity and hallucinations are two sides of the same coin and just to appreciate that number two even Sam didn't expect AI to be so capable in Creative tasks and so a lot of these Technologies they're going to continue to surprise us last but not least I was struck by how Sam describes himself as an empiricist openai was not the first to develop large language models but because of their pragmatic approach they were able to go to market first with chat GPT that's it for this episode of ask more of AI the podcast at the intersection of AI and business follow us wherever you get your podcasts and follow me on LinkedIn and Twitter or X please [Music]

Channel: Salesforce

Views: 622,579

Keywords: AI, Artificial Intelligence, Salesforce, AI Podcast, Ask More of AI, Data, CRM, customer relationship management, Generative AI, Predictive AI, The Future of AI, salesforce, salesforce crm, Salesforce admins, artificial intelligence, AI trends, AI news, Chat GPT, A.I., future of ai, future of ai technology, salesforce podcast, tech podcast, ai podcast, ai news, artificial intelligence explained, Clara Shih, Marc Benioff, Sam Altman, Dreamforce, crm conference

Id: uRVOeqSSZtQ

Length: 40min 46sec (2446 seconds)

Published: Thu Sep 14 2023