AI's Defining Moment: A Conversation with Marissa Mayer

This is a familiar stage for you, right? I want to get into Dazzle AI, but I feel like you've been in Silicon Valley for basically since Web 1.0. Here we are heading into Web 4.0. I feel like I should sit back and say, Marissa, what do you make of all of it right now? Well, I think that technology is fundamentally about innovation and invention, and that really comes down to building. And I have to say, there's never been a more exciting time to build. I have to say, at my little startup, I'm having a ball. Just on Monday, Meta announced Wonder. Yesterday, Google made Argon available. We have a harness, an AI harness that can put any model into it. And so there's probably another model announced this morning. When you look at how fast everything's moving, how fantastically you can actually make progress on your own application and really deliver great value for users, it's just an incredibly fun time to be experimenting and playing and building things and so I think it's it's all been building towards this moment. So before we get to Dazzle AI there's been a lot of sort of debate within Silicon Valley with regard to the frontier labs the AI models how are you feeling about that what's your comfort level because you're not out there on X just sort of you know yelling at each other. I think that yeah X is definitely an interesting place right now, but I think that anyone who says they know how this is going to play out, we have a new form of intelligence, and anyone who says they know how this is going to play out is either lying or arrogant or both. So you have to take this with a grain of salt. I feel like I have to do name, I just want to say, Altman, arrogant, well, so we don't know how this is going to play out, so we have to acknowledge that there is a possibility that things won't go well, right, that there is a possibility of that, but in the broader context, I think it's well less than 50%. In fact, even some of the doomers would say it's well less than 50%. And what that really means is the odds are in our favor. The odds are in our favor that really amazing things will happen in terms of science, in terms of creativity, in terms of really freeing people up to spend their time and live their lives the way they want to. And I think that's really exciting. So I'm very much an optimist. I do acknowledge, I think we have to acknowledge and address the possible negative outcomes here. But I think that we get so focused on that that we forget to focus on what's actually possible and what's going to be unlocked through all of it. So you're addressing the positive, that small part of negative. I'm just curious, again, because I think of you as somebody who's been in a perch as a leader. You know the technologies. There is debate right now about how do we protect against that downside. Do you have your own thoughts since this is a platform to share it? Well, I think that some of the things that are being proposed, regulation, collaboration across the labs, really thinking about where we go and how the technologies are used, I think is really incredibly important. I think the right steps seem to be happening there. So Dazzle AI, you were in stealth. Tell us about that and where you fit into the ecosystem. Sure. So Dazzle AI is a personal AI assistant, to your point of that kind of long perch I realized that my honors thesis in college was actually building up a voice assisted travel planner so I've actually been thinking about assistance for a really long time in the background and sometimes in the foreground of my career so what dazzle is is a personal AI assistant that learns from your camera roll on your phone so there's 25 billion photos taken today in the world 98% of them on mobile phones and with 96% of them will be ones you regret. No, I'm just kidding. Maybe, but I'll get to the interesting stats in a minute. But one of the things is that what we do is we really take your camera roll and then we understand your taste. So it gives us the context of you. We understand your lifestyle. We understand your preferences. Dazzle for me even understands the brands that I like and the brands that I prefer. And we couple that with agents, both human and AI, that can help get things done for you. So it's really can do things like create a to-do list and actually have a to-do list that gets itself done. Interestingly, one of the stats, yes, because 96%, I don't actually think it's that high. No, no, that's my life, go ahead. But my previous startup, we were looking at photo sharing, and one of the interesting things. Sunshine? Sunshine, we were looking at it and we were trying to understand, we were building an AI model that could understand when people are likely to share a photo. And that doesn't mean post it publicly, but it might mean sending it to a loved one, sending it to a friend, and what we realized was, About 60% of your camera roll is share-worthy after you de-duplicate the times you take it multiple times to get it right. So that begs the question of what's in the other 40%? And in the other 40% is all the stuff we need to get done. And I have to say that we built Dazzle for all demographics, but it's particularly good for women, and it's particularly good for busy moms because it's that appointment card that you snap a picture of as you leave the doctor's office. It's the time you take a picture of cleats to remember to size them up. Last week, I was browsing social. There was a little program on about summer programs that help prepare you well for college, and I screenshotted it. And Dazzle, basically, you have an AI agent that's sitting on your camera roll, came back and said, do you need me to put together a list of the requirements and deadlines for these five programs for Macalester? Because Dazzle even knows that Macalester's my child is older and is going to be going to college sooner. What's the business model for this? Right now, Dazzle is free, and I will say we're invitation only, but we're extending invites to anyone who's at the conference, so please try it out and tell me what you think. But we think in the long run, it will either be a subscription model, where basically for part of your everyday life, and we're starting to see with our early testers and users, they start using Dazzle for half an hour, hours a day, which is really incredible to see. so it's a subscription or on a per transaction basis so so far we've seen cases like dazzle has helped people to replace the roof on their house you know it's helped to fix the garage door it helps find that hard to find bag that you snapped a picture of that you've always wanted you know all of those types of things so in addition to helping manage your calendar and some of these other tasks there are transactions that dazzle drives so we may apply a transaction fee as well and I think that there's it's possible that we will have subscription transaction fee and I suspect we'll always have a few tier let me ask you about it's nice to see you here on stage and we've talked about sunshine in the past and obviously if I go back how many times were you on the Fortune most powerful women list several years in a row right and I don't know I do know I was really honored when I was back at Google I was I think the youngest you were the youngest you were that no no Sarah London but you know that was you know you'll be some person there was when i think about the voices in Silicon Valley that are out there i'm always struck that you know you you Cheryl Sandberg there was a real vibrancy of of hearing from women with regard to the tech conversation it seems to have some ways gone back what are you kind of seeing and hearing and what role do you want to play really to be part of this debate? Sure. I think that when I look at Dazzle AI launched two weeks ago today, and so we, and as I said, we're an invitation only, but when I look at who's signing up and who's asking for invites, I will say we're happy for all the users, but they're predominantly male, which really surprises me because I really think that, yes, I get that's often the case with early adopters, but this product is so good for women. AI is so good for women. I really think we need to- AI is so good for women. I think AI, I think that in terms of helping you manage lots of different tasks really being very busy you know taking some of these more mundane work streams away from you my friend Brit Morin wrote this great piece that says like AI isn't coming for your job it's coming for all the stuff you didn't want to do anyway because nobody wants to wait on hold no one wants to spend 15 minutes checking every site trying to figure out if this is available in this color and this size and if we can actually offload that to agents it frees up everyone but you know for for women who often are in in the role of doing that coordination or or doing the shopping and the buying i think that it is something that could be really really good overall and i think it's important for us to be part of this building the infrastructure layer of ai is complete will it continue to improve it will but we're at a phase now where we're building out the application layer the context layer how does this actually contextualize into people's lives and i think this is a place where women can add a lot and actually i would argue in many ways are sometimes better skilled at that how do you take this ai and or this technology and apply it to a practical everyday problem so we're going to take questions so we have mic runners here we do so the paddle so if you have a question just um raise your hand So as you're thinking of your question, we go to some, I want to ask as a leader, so you're now leading this company, you were leading Yahoo. Do you feel we all mature and grow? Do you feel you lead differently now? Are you a different, give a different perspective in terms of the perch you occupy? Well, I think the scale is very, very different, right? Yeah, the scale. Versus, you know, we're like, you know, a dozen or so people. I will say with AI agents, it feels more like a hundred people because they're all programming at like five or 10 X. But I think that you have perspective, yes, both because the situation is different and the order of scale is different, but also just, you know, experience over time, different things that you've learned in terms of what's worked, what motivates. And I think that being able to paint a picture of what it is we're trying to achieve, how this is going to, you know, be useful for people, the anecdotes start to add up and become data. but I think that it is I think early in my career I was very wedded to data and I say now I was and I remember because you and I talked about this you know obviously famously with regard to how little maternity leave you took and stuff and I know your kids are much older now I mean how are you feeling about just balance work and just the conversations that we were having when you just had kids to where you are now Oh I think it's I think it's interesting I think that again context matters and I think a lot of people missed in things like my maternity leave was the context of the company I was running Yahoo couldn't afford to have a CEO who was out for three or six months and I also just found in terms of my own rhythm I was better off kind of keeping my my foot or my toe in it and actually like going from a hundred miles an hour back down to 60 or 80 rather than zero and so um in terms of in terms of work and so you know for me i really try and keep it well well integrated but the context i think is something that really matters um and people have to do what works for them i think that i once had this interesting conversation with Tory Birch where she just said you know in as women and in motherhood often there's just too much judgment and when you you know everyone's going to do this their own way in a way that works for them and there isn't one right way to do this and you shouldn't be criticizing people who have to do it a different way or have to find a different way works for them that's great do we have any questions in the audience i love it so you keep going oh we have i think we have some up here oh there's one right there we can just um if we bring the paddle over to you um i um First of all, Marissa, I turned around and commented to the women sitting next to me, your articulation is beautiful. And it's really clear that articulation is an important skill today, to be able to articulate not only your intelligence, but in a way that it can reach other people. But that's not my question. My question is, in a world where I feel like I lost the data privacy war the first day I touched something, what does Dazzle plan to do with the data that you're collecting because it's really who owns the data, whoever owns the buffalo owns the ranch, right? So whoever owns the data wins the game, yeah, and the trust, right? Yeah so what is your plan for the data that you're going to collect? Sure, I think that when you look at AI assistants, everyone needs context. You see some, some the AI assistants are asking for email, we're asking for photo access by the way, Dazzle works if you don't want to give us photo access or if you only want to give us select photo access, that does in fact, uh, work, um, and I just a couple of observations before I get to the exact question, but the I would say that one of the things I've often said is if I had to be your executive assistant for a week I'd ask for your email. If I had to be your personal assistant for a week, booking travel, buying gifts buying clothes keeping track of kids' schedules, I'd ask for your camera roll because from that I can understand your life so one of our beliefs at Dazzle is that photos are the richest source of personal context available at the lowest friction. That said, they are sensitive. In my view, are they as sensitive as email? They are. There's a lot more in my email but I actually find that we find higher utility information tension in the camera space but on dazzle in particular one you get to decide our level of access so you can pick full select or none and the other philosophy we have is as we're analyzing your camera roll and trying to understand your taste. When in doubt, leave it out. For me, I have 55,000 photos on my camera. You do? Mostly of what? Mostly of my kids, but also travels and other things. But the interesting thing is if we find a photo that's sensitive, right? It has personal identifiable information in it. It might show just too much flesh or skin, right? I don't have those kinds of photos on my camera but if we if we actually raise your hand but interestingly but interestingly if we find anything like that we're like oh we know our users don't want us to have that and the other thing is we don't need it to do a good job so when in doubt leave it out as soon as we detect a photo is sensitive it's one of the first screens we do we immediately discard it and delete it because by the way we don't want it on our servers either and as i said like if i throw out one percent of my photos, I still have 54,000 photos, 54,500 photos to do a great job understanding my life, and so we can easily throw things away if there's any question about whether or not the user wants us to have them, and that's ultimately how we're handling this, and if at some point, if at some point people get uncomfortable, if they delete their account, all of the data gets deleted, and we aren't training AI models on your data, we do have of data but as I said but we've actually filtered that data to make sure that it's not things you don't want us to have I think we have a quick question from Annie up here and before we get to that I almost want to go back to the people who are on these AI models what you must still talk to them I just want to get sort of a quick thoughts we've got Mark Zuckerberg out there drinking Guinness looking very different than when you interacted well give give us your sense as to how the the bros that you grew up with they're all still there here's Marissa Mayer what's your take on on all that I was I think that when I reflect on people leading the frontier labs I think that they fundamentally love to build as as do I and I think that you know if it takes a lot of taste and and sensibility to really and they I think they're quite good at that in terms of what to build and how that technology can be applied the tips you're being very polite i like you're not going to comment on but i but i do but i think that what what is the common thread is is the fact that we love to build obviously there's a gender difference and as i said i i hope that there's more women that are inspired to actually build and grow some of these new technologies because i think it's really important for us to have a voice in how ai develops i think we have a lot to add yeah so very quick question annie because we're running sure the landscape is changing so quickly how will dazzle evolve as news and dots come out and if you look Two years from now, how many agents do you think a regular consumer will have? Right. Sure. I think that, you know, one of the things that we're seeing is we've been working on Dazzle for about a year. So there's been a nice validation. We were doing iMessage. We needed the context layer. And there's been this huge surge of AI agents, which I think shows that there's a real need there. I think as we evolve, we're looking at how our users are using Dazzle and how do we better support them in those tasks. I think it's likely that over time we will add email and for our super users add more context overall. But I personally believe that a user will have probably more than one agent that they use, but it's probably going to be a small set, right? It's probably less than half a dozen. It's probably more than one. Because I do think we naturally want to contextualize and compartmentalize, you know, overall. I think that just as many people have, or some people have an AEA and a PA, right? Like, having that kind of differentiation or just a separation and compartmentalization in those tasks is helpful. You might want to have someone different looking at financial and economic matters, money matters, rather than some of the personal pieces. So I think it's probably a handful. So, Marissa, before we get to some final thoughts, you know, you've seen the euphoria of Web 2.0, obviously a lot of 1.0 forgive me the zeitgeist has shifted against tech and ai's and what what do you think about that and any thoughts as to what will turn the tide because especially in the U.S. it's quite negative. Well, I feel like it is a pendulum that swings and I think we've all you know, I've been around long enough now that you see the pendulum swing at different moments as I said and as I started, I think there's a huge amount of to be very very excited about but I do think as some of the early models came out and really surprised people in terms of what they could do and how capable they were it caused people to say wait, you know where where does this end and I do think that some of the initial positioning was perhaps too cavalier. What does this do to my job? What does this do to my family? What does this do to my children's future I personally think it can make all three of those things better it will be different right it might be different than what you envision but I do think it can be better and I think we need to stay very focused. I worry that sometimes we lose in the negativity. We lose the fact that, again, I believe in the majority of outcomes they're actually more positive. And I think we should also latch on to that. Both Dazzle AI and AI writ large, whether it's this time next year, we're moving at AI speed, how is the conversation going to shift? How is the company going to look? Well, I think that there's a bit of a bifurcation happening. At the Frontier Labs, as I said, they've become very capable. And there's many, one good analogy I've heard is that the frontier labs are now moving from a master's level understanding to a PhD level understanding. And I think that we're going to continue to push on that because that's how we're going to get all the advances in sciences and some of the breakthroughs. That said, today already, many of the frontier models are shovel ready for everyday applications, which is why I think we're seeing this big moment in assistance so I wouldn't be surprised if it's by the time next year we see a bit of a bifurcation right people who are using the models to really push the frontiers of PhD level and scientific expert level Nobel Prize winning work forward as well as people who are using today's AI and AI will continue to advance on everyday on everyday tasks and I also think that we'll start to see a proliferation of how you apply different models because you know I'm on the board of Walmart you were just saying to them like you know you don't need a PhD level understanding to understand that there's a spill on aisle seven right and so you know so being able to right-size the model to the task at hand can give you a lot of value and also to some of the earlier questions you know about data security and privacy can can allow a compartmentalization that's at times comforting and responsible. Do we need a broader list, whether it's voices? But let me go back to, you started at Google. You know, you were a young engineer, very sort of excited. What is your advice to sort of the next gen right now? Obviously, you've got a startup. How are you feeling about Google? Oh, I'm very bullish on Google. I will say we were blown away. We put Argonne into Dazzle in our staging environment yesterday, and it's tremendous. It's a really great model, and there's a lot of really great models coming out now. So I continue to be very bullish on Google. There's a lot of very talented people there, and I'm very grateful to them because that's where I grew up. And I think for new people, like, get out and use it. I know that some of these tools are scary. I know that life is busy. But I really find that when you actually use it and understand it and think about all the different – we get new use cases at Dazzle every day. Yeah. Right? And so, you know, it's like, wait, I never thought about it being applied to this or that. It's really exciting. but it really means you have to make the time to immerse yourself and try things. Thank you for joining us. I look forward to continuing the conversation. Good luck with Dazzle AI. Please join me in thanking Marissa Mayer. Thanks.