Big Technology Podcast - Google's Bard Director On His Vision For The Bot — With Jack Krawczyk

Episode Date: May 24, 2023

Jack Krawczyk is the director of Bard at Google. He joins Big Technology Podcast to discuss his vision for the chatbot, where it's heading, and whether this technology will be as impactful as many imagine. Join us for a fun look into one of the tech industry's most important generative AI initiatives. -- Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice. For weekly updates on the show, sign up for the pod newsletter on LinkedIn: https://www.linkedin.com/newsletters/6901970121829801984/ Questions? Feedback? Write to: [e-mail]

LinkedIn Presents. Welcome to Big Technology Podcast, a show for cool-headed, nuanced conversation of the tech world and beyond. Jack Kravchik is our guest today. He's the senior director of Bard at Google. And he's coming on to give us a little bit of insight into what Bard is supposed to be for Google, how it competes with search and where it's going. I wanted to bring Jack on to get a sense as to how the company is thinking about this product, how important it is to the company,

where it sees it going strategically and, frankly, how we should be thinking about this technology as it starts to be the subject of conversation and action almost everywhere in the tech world. Okay, my conversation with Jack Krafjik coming up right after this. Hey, Jack, welcome to the show. Thanks for having me, Alex. Great to see you again. And, you know, I was looking back at when we first met, and it's starting to get to the point where I've started to know some of the folks in the tech industry for like a solid 10 years.

And we're right about there. We met when you initially worked at Pandora, or we initially met when you worked at Pandora. Then after you left, WeWork. And I just took a look at how long you've been at Google for like three years at this point. I mean, man, time flies. It does. Second journey at Google. I think I'm coming up on seven cumulative years.

and it's still a magical place to work. What brought you back? You know, when I left Google in 2011, I was on this journey of exploration of what might start up life be like. And I had the great fortune to have some amazing experiences, like the one at Pandora that you mentioned, working on a company called United Masters, and the WeWork experience was quite amazing in many regards,

in many regards of that word. There's been some good books about what went on there. And as I was exploring the various chapters of my career, I would say to my wife, oh, this next one's going to be my next Google. It's going to be my next Google. And she would look at me and say, what does that mean? I'm like, I don't know, like a place where people value logic and reason over ego. And I'm very happy to report that, you know, from the first day that I stepped into Google 16 years ago to today,

that cultural norm of respecting each other is still there every single day and it's just it continues to be a magical place to work. It's cool and it's interesting that you phrased it as like the next Google because you're now working on what some see as the next Google and that is barred, right? So I'm kind of curious just to begin with. You're working on this every day and we're going to get into the product intricacies of course, but you're working on this every day And there's a range of thought between, you know, this is going to change the world, this type of generative AI, and especially chat, it's going to change the world. And there are others who say, well, you know, it's, it might be a passing fad. Obviously, Google thinks it's more than that, given that you're working on it.

The company doesn't typically take on projects that it believes are just going to be here today, gone tomorrow. But what's your view in terms of the transformative power of this technology? I think we are still at the very early. stages. I knew you were going to say early days. So no early days excuses, early stages. But okay, but your perspective on where it's going to go, like what is your, you know, obviously we don't know yet, but where do you heart of hearts think it's going? When I look to the core of what's making this moment so magical is you're watching people take their imagination and let it blossom and flourish.

And what inspires us working on Bard as this experiment that's a collaborator for your ideas, we watch people explore this journey of trying to discover something, taking an abstract idea and turning it into something that is more concrete, and it allows them to explore more ideas in less time to really supercharge their imagination. It's magical.

I spoke to a filmmaker in the early days of Bard. And it's funny to say the early days, we've been out for two months. There it is again. I'm going to start ringing the bell. Hey, there's still a lot to learn. And that's why we launched it as an experiment because it's going to change frequently. But I speak to this filmmaker who says, I'm working on this short film for YouTube. And I'm trying to generate perspective of the other character.

And I don't know of another tool that's existed before where it's like, hey, I'm not in my mid-50s going through family transformation, but what are some things that people might explore as they're going through that life transition? And it's kind of fascinating to see this idea that people have in an exploration and trying to use a generative AI tool like Bard to help them build up that. empathy for what that character might be going through. Interesting. So I think that like when people think about the opportunity here, you know, the question is whether this is going to be something. I mean, these are, this is a, that's a great use case that you gave. But I think people when they think about a transformative product, right?

You know, the question is can enable humans to do things that they just couldn't do before in meaningful ways. And a lot of this stuff you could, for instance, like accomplish by things. or, you know, the greats might have had it in an innately anyway, being able to empathize in that way. Or you might have been able to find these solutions by searching. So I'm curious, like, again, at the heart of this product, do you think this is just going to be sort of a cool new way to do things that we already did or a brand new way to do things that we never could? It's a fascinating way to put it of, like, projecting where it's going.

to go is a very hard experience because when you think of a new technology and a new capability, you don't know the directions that it's going to go. And I've studied a variety of history of technology development for a long time. Like take refrigeration, for example. We had ice boxes for a really long time and then we created artificial ice boxes. We call them refrigerators. And at the time, it felt like it's a natural transition to go from an box, so now I can plug this thing into a wall and it starts to refrigerate. Did anyone at that time that was trying to keep their food cold think it would enable couples struggling with fertility issues to have children, like the enablement of IVF through the ability to extract eggs

and freeze those eggs? You don't know the compounding effects that technology will have. Part of what gets me excited about something like Bard, is this idea of more ideas and less time. Like, our ability as humans to place meaning on things is profound. And it does compound. And so to have a tool that allows you to more seamlessly scratch that itch of your curiosity and see where it takes you, you're compressing the time. And when you compress that time, magical things can happen. And so, yes, I believe the opportunity is profound. I mean, we look at some of the things that AI has enabled. You know, you look at Google in 2020, we announced AI's capability to do early detection of breast cancer, and we published it in Nature magazine. And these opportunities

are just so profound. And what we're really rooted in is thinking through how can you take that idea of being that filmmaker, exploring that, and you can create a film that you may have never created. Right. Back in the day, by the way, when the refrigerator came out, if there was a Twitter, I'm sure you would have VCs doing threads about how that stuff would be good. No, I'm just kidding.

No, but it's fun, right? We're still allowed to imagine it. I think part of what I want to do over the course of our time together is sort of dream a little bit about where this could go. Yeah, I think that's what's so exciting about this moment. you know, the technology has been under development for many years, and again, compounding over neural models into transformers into where we are today. And, you know, we're now talking about multimodal models in the future. And the pace of change is remarkable. I mean,

it's just absolutely remarkable. And part of the thing that makes this so fun is the ability to dream about what if and what could be. Right. And I think, like, you made a good point. about what announcements Google's made in 2020, looking at something like computer vision, but the big shift here has been the generative stuff. And of course, like, I don't think anyone's going to deny the fact that AI is going to be everywhere. It already is, right? But the path that the generative stuff is going to take is quite interesting. So let's talk a little bit about the innovator's dilemma, because the perception from the outside, at least has been that Google's been stuck a little bit with an innovator's dilemma, which is that it has this great

product search and these type of bots can threaten search a little bit. So there was a CNBC article that had some elite comments from you at a Google All Hands that I think is like pretty interesting when it comes to like thinking about how this this balances search and whatever else it might have. So you say to Google employees barred and chat GPT are large language models, not knowledge models. They are great at generating human sounding text, they are not good at ensuring their text is fact-based. Why do we think the first big application should be search, which at its heart is about finding true information? And then you say, I just want to be very clear, Bard is not search. And then you say, we can't stop users from trying

to use it like search. And CNBC contextualizing this said that you mentioned that Google is still catering people to people who want to use it for search. And you did build a search button. inside the bot. You can Google things now. So I think that, and this is great. This is like awesome to be able to talk with you about because like we can actually go a little bit deeper into these comments. From an outsider's perspective, the way that someone could read that is, well, why would, it seems like a lot of effort to go through to talk about how this isn't search. And like kind of the only place that that would happen is a company whose bread and butter is search. Because otherwise people wouldn't really be worried about it. So I am curious.

how yeah how you think about the balance between the two and you know there's obvious reading through the lines it does seem like there's some worry inside google that this could displace search at least in part so what do you say to that i won't comment on uh meetings that we have uh internally but what i can't say of part of what you're scratching on is this question of um what is barred and and why did you launch it as a standalone experiment? And that was a very intentional decision because of the known limitations of large language models, their ability to hallucinate,

paired with a massive excitement and enthusiasm for their ability to be collaborative, for their ability to sit down with a musician who's thinking about writing their next song and they're stuck on the chorus because they can't figure out that next line in the story that they're trying to tell, and they just need to bounce ideas up with something. This notion of being a collaborator is the root of what gets us excited of part of this experience. Now, as you collaborate, you do want to understand some topics more or less.

And so, of course, people are going to ask questions about, deeper understanding of a topic and as you see that you know through using the product there's a disclaimer bard may not get things right it was a very intentional decision to put that google it button there because we want to guide people down the path of knowing what's right hallucination is a challenge of this technology that remains unsolved and you know we're certainly working toward improving it getting it better every every day but we also want acknowledge there are multiple tools for people on their knowledge journeys and it's part of what inspires us at google i mean you heard sundar talk about it recently we have 15 products with over

half a billion users we have six products with over two billion users one of our core values is to respect the user and really listen and understand what they what they're looking for and what they need and that's part of what you see and in some of the design decisions that we made uh in the way that you see BARD manifest today. Now, I'm kind of curious, like, you know, the idea of not being able to stop people from using it for search, you know, if there is this wide consumer behavior where people say, hey, I'd prefer to use these things, whether it's BART or ChachyPT or Bing, to search versus the traditional search box, does Google kind of have to go along with them and enable that, even if it sort of falls into some of those worries that you're talking about?

because the other option, effectively, is just kind of sitting by the side and hoping the old way it prevails. One of the things that is amazing about this notion of respecting the user is you respect that they have multiple ways to get information. You look at YouTube, for example. YouTube is one of the largest search destinations in the world.

And when do you search for a video on YouTube versus Google search? There are a variety of inputs, a variety of ways. the key is just making the journey seamless for the person based on the context that they're coming in. Recently, I was looking up, what are the hours of a local business near me? And to be honest, I can't tell you when I use Google Maps to do that search and when I use Google search, but it's about making it seamless. And so, again, we're two months into Bard.

We're still learning how people want to use this technology. We're iterating on it. We announced a lot of the things that are coming out. We're launching images in responses because what we're finding is part of the benefit of launching direct access to a large language model, rather than building a large language model into an existing service, is you find these new ways that people are exploring and using it. Interesting. So actually, you know, it is an experiment, which is true. But I'm also curious, like, what type of weight Google's putting behind it.

Can you share anything to give us, like, a sense as to, like, the magnitude of this project inside Google? How many people are working on it or anything of that nature? I won't comment on the size of the team, but I will comment on the energy of the team, which is, honestly, it's infectious. I come into work every single day. I drive 90 minutes to get to the office. And I listen. Are you still in the East Bay? I'm in the North Bay. Oh, okay. And, you know, it's a great time to listen to the big technology podcast while you're, while you're on that day. I love that. Thank you. Got to get that plug. Yeah.

No, but it's like I am excited to wake up every single day, well, three days a week and make that drive into the office and see the collaborative energy that's coming off. Like, you're standing next to a whiteboard. You're watching people build and riff and grow on these ideas and learn. We're at this pace right now, working on this product where, yes, we're very enthusiastic and hopefully you see it through our experiment updates at the pace at which we're able to generate these new features that are coming out, these new experiences in Bard. But it's really about the speed of insight that this cadence is doing. And that's the thing that's infectious. Like going back to that Google value of respect the user, the number one way you respect the user is listen to them. Like, listen to the person that's using this technology. Listen to the person that's telling you, the thing that gets me so excited about this technology is it makes the journey more enjoyable.

It's not the destination. I think like one of the things that we saw when, you know, the emergence of this generative technology came out, I would hear things like, oh, this thing's amazing. It's going to make this travel itinerary for me. And like, bing, bam, boom, like two minutes. I've got my trip to, I've got my trip to the Jersey Shore planned. Tell you go somewhere that it's hallucinated. No, it's, well, you know, certainly hallucination is a problem that we continue to work on. But, like, you listen to how people are using it and what inspires them about building a travel itinerary.

It's that I'm sitting in a, like, I'm sitting in a room that's maybe with some drab lighting and it's raining and it's 40 degrees outside in March and I'm dreaming of where I am going to be at the beach. You want to make that journey fun for people. And you want them to start thinking about the lateral capabilities of bringing these things. Like, one of the things that we don't necessarily have a tool for is, like, I was talking to my four-year-old the other day about the places that he wants to go. And Ratatouille is one of his favorite movies. And he goes, Dad, I want to do a Ratatoui trip. And I was like, what does that even mean? And so I started playing with Bard of like, what would have?

a Ratatouille trip look like to, well, the movie takes place in Paris. I would love to go to Paris, but I'm terrified of the nine-hour time change for my four-year-old. But you start to explore these new ways of reconfiguring information in a way that didn't exist before. And so we've spent so many years making knowledge retrieval more efficient. And people know how to use that. and, you know, getting back to the, what's exciting about this technology, we're building upon knowledge retrieval and now focusing on knowledge creation. Like, how can you start to get these ideas? Like, I don't know how many people have ever searched for build a Ratatouille itinerary for four-year-olds.

Look, I know that this is like an exciting thing. I just want to get back to the, we could go on for this forever. I definitely want to like look at the product stuff, but I definitely want to get back to this magnitude question. So I know you can't, maybe you want to share the size of the audience, but can you talk about, like, do you see, is this a project that Sundar is involved in? I think you see the amount of energy that we talk about. I mean, anything you see on the I.O. stage is something that all of Google is excited about. And there are plenty of things that we're not on the I.O. stage that Google is very excited about as well. And so, of course, there's involvement from across the company of seeing the excitement of what we're building. And I'm really thankful to have partners across the company that are willing to share their ideas.

Hey, have you explored this? Have you considered that? We'll soon be launching tool use. We announced extensions at I.O. There's going to be Google tools and services that are built directly into BARD. And so it's a collaborative effort across the company. Yeah, one of the things I've listened to in a recent interview with Sundar is that he said, this is the most exciting thing Sergey has ever worked on in his life.

Has he been involved in the creation? You know, again, going back to like the first go at Google, I had the pleasure of seeing the founders involved in the creation of products of the company. And Sundar has said, you know, Sergey, Larry, have been involved in the discovery. various discussions of product development for years. And so to have them continue to be involved in Google, they sit on our board. Of course, they have a vested interest in all the bets that Google's making to be successful. You know, this has been interesting because it's happened so fast. Google sort of became famous for being the company that really got this first when it's

kind of interesting what makes you famous. but when Blake Lemoyne, a software engineer at Google, said he thought this large language model that you guys have, Lambda was a person, that sort of like, you know, really caught a lot of people's attention. I spoke with him. And, you know, the thing that I really came away from that conversation was, was, all right, sent to you or not, dang, these models are extremely impressive,

like their ability to hold a conversation. It was unlike any chatbot I'd ever seen before. I'm kind of curious, like, how you felt internally when you, you, I mean, is that something that brought your attention to large language models? And what did you feel when he made that statement? I mean, I think the question of AI sentience is something that's broadly discussed. And frankly, I don't think we're going to get to the bottom of it today. I don't think we're going to get to the bottom of it.

But one of the, I mean, so first, flat out, like, no, these things are not sentient beings. But it touches on, A, the importance of. of the conversation that we need to have in rolling out this technology. Why did we take this foray by making Bard a standalone experiment? Because for the people that are interested in the technology and wanting to try it, we want to give them a destination

to directly engage with the technology and try to understand it because that conversation is how you drive this term that's used across the industry of alignment. How do you make sure that the technology that you're building is responsibly aligning with human values? the things that we expect from this technology.

And one of the things that you learn is you get these very loquacious, convincing large text responses is they are convincing. And so you have to go through the design process of, yes, we as humans, again, like we talk about unlocking people's imagination. Like, imagination is a very powerful tool and people place meaning into things that may not necessarily be there. one of my favorite concepts is this notion of peridolia you ever look at a cloud and you know see a pig in there there's not a pig in there but we put yeah actually it's there's not a cloud in the sky today in california uh but it's one of these things where that is a human phenomenon like our ability to see faces in things is a human phenomenon that's existed for years and as we roll out this technology again uh you know maybe you can do like one of those ding counters like

But just a reminder, like, we're at the early stages of a very long arc of technology. And so we need to engage in these discussions. We need to figure out how do you build technology that's both bold and responsible? And the bold part is give people access to the capabilities, get them to engage in it, listen to that feedback and iterate quickly. And we want that to be a dialogue. And so with a product like Bard, you see something that we launched called experiment updates. And so when we update Bard, we not only tell you what's changed, we're also telling you

why we changed it. And we have this hypothesis when we release the what and why that people would engage in the conversation they have. But one of the things that it's helped us see is, hey, this is part of that path into helping us understand why you might take the next step. Like, why are you launching images? Well, we're launching images because people love the journey. journey of creation. And so we have that, we have that capability and we should include it.

Yeah. And so if going out there and having this dialogue and working with users is so important for alignment, I'm kind of curious why you think that the first company that came out with a large language model available to the public was open AI and not Google. If that was always a value inside Google to get it out into testing, then why did Bard come after some of these others. We've been testing Lambda inside of Google, dating back to when Sundar announced it on stage at I.O. a few years ago, it's been a tool that we've had inside the company. And part of what you see is the discourse that happens inside of the company that's magical, engaging in this conversation. Are people ready to talk to a thing that sounds convincing in response?

And so we have been iterating for a long time. And what you've seen over the past few months is this signal from the broader populace that we're excited to try this thing. Like, yes, it has its limitations. Yes, it has angles for responsibility that we need to deliver. But like the signal was there. Like going back to our core value of respect the user, people were saying we're ready to try this out knowing its limitations.

And so we rolled it out in a way that hopefully people see we're very clear about the limitations that are known, but the improvement that's happening alongside it as well. Yeah, I have this feeling that, I mean, I think, I'm kind of curious how you think Bards tax up with the competitors. But, I mean, I have this, I think that it's behind. But I also have this feeling that Google has a way of rapidly innovating. and we could see you guys catch up and exceed the others pretty quickly.

And frankly, like, it's cool in some ways to have this competition because as a user, like I love using these things. And, you know, if I could have, for instance, like, oh, I remember like one of my early interactions with Bing was trying to get it to like search my inbox for things or send emails. And I was like, I'll give you my Gmail login and password, just do it. And it wouldn't do it. And like, wow, like if Google could actually get good at that, it would be really a compelling use case to use BART.

So I am kind of curious, like, where you think the effort stacks right now. And, yeah, and whether whether you feel like that theory is somewhat accurate. Well, I've been very pleased to see the response that we've been getting with every new iteration that we have of BART. And one of the things that people had asked when we launched it, that use case that you mentioned is one that is very commonly asked for. And the question that became, like, can you do it in a way that responsibly meets user's needs? And we didn't launch with it for the same reason why we didn't open up Bard in its initial waitlist phase supporting code because we didn't feel like it could give. helpful enough answers. But then again, you listen to what people are asking for.

You see the improvements in helpfulness. You get that insight like, oh, as it does this, it can start to understand how to use tools. And so we start asking the question, like, could it use Google tools? And that's why you heard us announce a few weeks ago now on the I.O. stage two weeks ago, that tool use is coming to Barton. And we see this as this progression. And what I get excited about is people want, people want rapid iteration.

They want to try out new capabilities. And they want to find both utility and appreciate the fun aspect of using these things. I want to make sure that that doesn't get lost in the discourse. Like, these things are fun to use. No doubt. Google's a verb. Is there a risk of moving too slow and having like chat GPT or Bing become a verb? I mean, we talk about being bold and responsible.

And so being responsible means you have to understand and appreciate what are the risks of taking a product to market. And one of the things that we are really building into the bold steps that we're making as we release images, as we bring tool use to the fold, is that you have to do all these things with the user's trust in mind. And so tool use, why are we not launching it today? We're still working through the exact tuning of, hey, what are the right ways to bring these tools to people where they feel like they are in control because that is a core principle from which we're building?

And it's not that far away. But we want to make sure that when we bring something to people that it meets with both their expectations of the good and the expectations of what's actually happening. Awesome. Jack, thanks so much for coming on. Great chatting with you. It was great chatting with you.

Alex. Good to see you. You too. And that'll do it for us here on Big Technology Podcast. Thank you so much, Jack, for joining. Always great to speak with you. Great to catch up. Thank you, Nate Gwaddeny, for handling the audio.

Thank you, LinkedIn, for having me as part of your podcast network. Thanks to all of you, the listeners, for being here. We'll be back Friday with a new episode breaking down the week's news and stay tuned in my conversation with Astro Teller, the captain of Moonshots, and also the CEO of X at Alphabet. It's coming up next week. Thanks again for listening. We'll see you next time on Big Technology Podcast.