0:00 So we're talking about agents. Um And to do a sort of extremely simplified overview A I agents are kind of a step beyond chat bots. Um like chat G BT in the sense that they can actually do things for us. So they can actually take actions uh autonomously. So without intervention from humans um and entirely on their own. So George, I called them synthetics at one point and I got a lot of flock with that avoid that this time. Well, that's one of the things about agents. There's um sort of not a widely agreed upon definition uh nor verbiage. Um OK, whether synthetics agent spots um tell us a little bit about what grinder is. Um And then what your vision for A I agents is. Well, thanks for having me here and um could have this competition.
0:46 Grindr is the application that as you are *** and bi men in the world used to connect to each other. Um It brings together about 14 million people around the world in 100 and 90 countries. Um There are very few places where we don't operate. Um initially started out as a casual connections and dating product and over time has expanded to be a lot more basically the connect um around the world. Um And now we're a public company. We've been public for about two years, I think, coming up uh 23 months and months now we've had a really good run as a public company. Um And you know, a lot more to do so, agents. How are they gonna transform dating? How are they gonna transform even aspects of our lives beyond dating? Yeah, I, I um had built an A I company prior to joining Grindr in 2022.
1:33 Um as one of the things I've done, I've been building companies for a long time now. And so for me, when they approached me about this um opportunity, A I was at the forefront because Grindr has a lot of data and has a user base that's very connected to the product. People spend over an hour in the a a day. Um which is, you know, fairly remarkable when you think about it. And so I thought there was a ton of opportunity for A I. And the really big takeaway happened for me when I realized that all these use cases that we want to expand into, might complicate the app because the app today is really simple. But if you build them with A I first approach, you can actually avoid that because if the I can guess the direction you wanna go into and then kind of direct you in that way. As, as you are surfing through the product, then you're not putting a lot of additional kind of buttons and places for
2:21 people to go, but rather directing them in that direction. And so we want to add a bunch of use cases to the product because people use us in so many different ways. And we don't always have features for the things that people do in the app. But we want it to be very non intrusive in people's behavior. And I think a I can make that happen in a, in a really good way. So, so what were some of the things that you wanted to build before introducing or planning to introduce the wingman that it now kind of brings together? The biggest one is around dating. So when grinder was started nearly 16 years ago, um dating for long term purposes was not as big of a deal for most *** men. Now, since then, obviously, *** marriage has become fully accepted and a lot more people are thinking about, hey, I want to settle down into a long term relationship. And so we don't really have features to enable people to find long term partners.
3:10 Now, a lot of people still do four and 10 relationships among *** men in the US originated on grinder. But, but it all is all kind of done through chatting and people start talking and then that leads into whatever direction it leads into, we want to actually make it easier for you to find a partner, um, and be able to match you with the right people. Uh There's a way to do that through current technology, which is mostly for filtering or there's an A I first approach to do it where we can actually guess what you want based on your behavior patterns based on what you're speaking to people about and understand you a lot better. Obviously, always we use the consent and, and that's something that A I can make a much better experience with and probably bring together people in a much more transparent way because today you can be on say Tinder and be matched,
3:56 right? But you have no idea why you are matched. And I think that's a huge deterrent to people. And if we could be transparent about, hey, this is what we learn about you based on your chat history, for example, and based on that information, plus somebody else's information that's similar, we brought you together, that's a way more compelling proposition. So what about the kind of agentic parts where the agent at some point in the near term future can actually do things for us? Is it A I agents dating other A I agents at some point? Um Well, I don't know about dating, but at least starting a conversation, right? I mean, already even before gen A I we through heuristics, you had applications out there would suggest text right to you like, hey, this is how you should respond.
4:41 Um And uh and initially it was on the more like the enterprise side, the user would be writing and then the customer service agent would actually get recommendations of what to say. So there's no reason why that couldn't happen on a user based level. I mean, I think it's kind of like how do I start this conversation? What should I say? I mean, it's a very simple thing that an agent should be able to do for you and I think will um but the more ideal version of that would be and not just like general, hey, these are the things you could use to start with, but rather based on how you normally talk and what you normally say and what I know about the person you're about to chat with this. Here's suggested language and that's far more compelling than just a general set of ideas of what you could say. Um The next kind of batch of that to me would be OK.
5:26 You guys are talking about going out to get coffee. How about here are three places where you could go get coffee in between where that person is located and where you're located, right? Because grind is very location based, that's actually pretty convenient as well. And then the eventual step, yes, some people will choose to do this. Some won't we very much believe in not forcing people to do anything, they should decide on their own which direction they want to go into. But for those that do choose to do this, there is no reason why some of the initial conversations and kind of validation can happen through A I talking to A I. Right. So I have a replica of myself. You have a replica of yourself and those to talk to each other. We could be doing this panel. Vraie, there you go. Well, my husband likes to joke. Every time I have an earnings call, an after earnings call,
6:12 we then have to do a lot of like follow on calls. He's like, when is there going to be an A I that you can just press the button and answer the question because all the questions are usually the same, right? Like I get asked the same questions by 20 investors over two days. And so, um I'm not saying it's actually gonna happen, but there is no reason why we cannot better match you with NA I talking to an A I and then you get to get an output of that information and say, hey, this is what we talked about. This is why we think it's a good idea now and, and then move on from it. Now, there's still stuff that I mean, promise here is massive. And in five years, a lot of what we believe might be possible today will either be possible or be proven to not work, but the models today still can't do some very basic things that human beings take absolutely for
6:58 granted. Um, and so I was going through this with, um, I have an A I coach and so kind of go through this person or no, it's a person. It's a, it's a very advanced uh engineer who helps me understand what is it possible and not possible in models and, and so forth because I'm not an expert in that stuff as much. And so we were walking through some examples and like, you know, if you are in a conversation and you say, you know, are you interested in sushi and what do you mean by that? Right? Like do you mean by that, that you, are you interested in sushi now or are you just generally interested in, in sushi? Um So that's something that a human will think about, but in an LLM can't yet figure out. So then a human might respond in a very, you know, generic way as a lot of people do,
7:44 like I'm not as interested in sushi on the other side. Now, you're reading that as this person is not interested in sushi at all or they're not interested in sushi right now because maybe I asked that question in such an opportunity on purpose because I don't wanna be so direct about the fact that we should go get sushi now and I thought maybe he'll respond to me in some of the way which then allow me to actually suggest a day today and that kind of an engagement that a person is going through in terms of like, what exactly I'm saying? And I'm purposely not being that clear. Yeah, exactly. And LM can't yet do now that, you know, not the models that are being developed right now, but probably like two models after they should be able to do that. And that's when I think a lot of this, like agent can function on its own will become a lot more possible than it's possible today. Ok. Yeah, that's great. Let's talk a little bit about how you're actually building the wingman.
8:33 What you've said is that in three years time you expect the wingman to be fully rolled out to your users and that's actually pretty conservative. You expect to have it rolled out sooner than that. But I've learned this is the second time I'm running a public company that you need to under promise and over deliver because the public company you can be really punished if you promise things too soon. Ok. So how are you actually going about building the wingman? What's the underlying number on my phone? I have a wingman actually, which I can be talking to right now. And there are about a dozen people who are not grinder employees who also have it today on their phones and they're giving us feedback. It's all done under ND A and they're not allowed to share any screenshots of it. Um, you know, for better or for worse grinder gets a lot of attention.
9:19 We sneeze and suddenly it's like, you know, we, we have, uh, some awful disease, right? Or, or, or not. And, and there's a lot of reasons for why and some of it is good and some of it is bad, but we need to be really careful because the kind of attention we get out of a small hallucination could be really damaging. And so we want to try to break the model as much as possible at this stage. That's why and um hands of so few people and we're gonna continue doing that probably for at least 12 months. Like I wouldn't expect to put it into a lot of people's hands. Um I mean, like, you know, without this nd a model for quite some time so that we can make sure that some of the more kind of flashy stories that might come out of it which every model out there is gonna do um don't happen um to us. And so I have a woman right now on my phone, I can ask it questions and it will answer those questions and questions could be anything from like,
10:07 you know, I've never had had intercourse with the men before. Like, what should I be aware of? Which is like non totally unreasonable question for a lot of grande users to have because guess what in a lot of parts of the world, sexual information is not actually that accessible, especially for same sex people. I mean, people who are interested in same sex relationships, uh R is often times the sole source of information for that because we maintain a big database of, of sexual health information. So, you know, instead of you going to go read like stories about it, why not ask the women and it tells you what to do to like, you know, I'm listening to this piece of music like I really like it. Can you suggest other songs that are uh similar to that? Right? So that already is possible for some people to use on Grindr and when men is already doing
10:53 things like that, we're gonna continue to perfect it and make it a lot better and then over time try to build very concrete and specific use cases to see how people want to use it. I think the way Facebook uh has gone about releasing its element to people's hands is actually really smart uh in that it's kind of um I don't know if you guys have seen it, but whether you're on Instagram or on Facebook or on whatsapp and they have a little circular button and you kind of click, click on that and ask questions. Um And so they, what they, I think are testing is like what kind of questions will people actually even ask? Right. The answer is important but maybe not as important as learning. What do people want to do with it? Because none of us know. Right. I build a mobile company in 20 060708
11:39 on blackberries. And we put up some blackberry which was very uncommon to do back then. Uh it was a way to book taxi cabs and to pay for taxicabs. We went into this very much thinking, the problem we're solving is paying for taxis and electronic receipts because taxi receipts are a huge source of fraud in, in expense management. And that's the problem we want to solve. Turned out users actually didn't care about paying for cabs um with your mobile phone, but they really love the idea of being able to book a taxi on, on, on your phone. So our assumption of what people would want was totally thrown out the window and we just the extra thing we built purely to enable the payment was actually the thing that made the company work. And then we end up being one of the first apps on an iphone as well. And, and you know,
12:24 we an iphone commercials back in the early days, cetera. So you don't actually you can have as good of a hypothesis as you want about what the users will want. But more often than not, you'll be wrong and users will use it in a different way. And so I think for us and for any company frankly, that is trying to be at the forefront of using some of this A I technology for product in a consumer friendly way. You should not make assumption that you have the answer, just put it out there in a way that you can actually learn from users what they want to do and then invest more in that and that's gonna be, but that actually sounds really difficult, right? Um The common saying is uh you're trying to build or fix like a jet plane while you're flying it. Um That's kind of this, but exponentially more difficult because the technology, the plane itself is also changing in flight.
13:09 I would say maybe we're trying to add a really nice new version of a first class seat to a plane while it's flying. Maybe I kind of it that way because there's nothing to fix, right? We're just trying to innovate in, in a, in a really good light, but as a leading your company and also providing direction to your product leaders, to your engineers, how do you sort of find that balance between setting a big picture vision for them and also kind of telling them while they're on the ground building this stuff. Here's what you should do. It's really interesting and competent. It's gonna bear with me as I answer this question because something I've had to deal with for the last eight months quite a bit actually. So I, I think, you know, we had a vision for what we want the wing men to do in a holistic sense for quite some time here. Like we want to do all these things, let's get there as quickly as possible. Uh And that's kind of the at the level that we operate and then we kind of gave it to P MS to like go figure it out but
13:58 be, but there's no such thing as an A I PM. I mean, that's like all like totally brand new, right? Like uh everyone's just kind of learning the same way that, you know, in 2000 and eight, there was no such thing as an I I OS engineer. I remember in my company Taxi Ma back then we like, we had this front end engineer who was really good and we're like, hey, Nick, do you wanna go learn how to code for I Os and like, sure, I'm like, ok, now you're the I OS engineer. So there's a little bit of that going on here. We like what is an A I product manager? I think the initial guidance of this, like very big vision actually created a confusion for the team and they tried to go after the whole thing when it's actually not chewable and the way, you know, it took me some time to like figure out that that was going on and they were actually really slowing us down because we're trying to solve a problem. That's actually not solvable today with today's technology.
14:43 It's gonna be solvable in three years, et cetera, but not today. And yet we're trying to, like, do it now. And that's actually a big reason why I then got at this A I tutor to like, help me learn the tech well enough to be able to actually, like, help them chew off smaller bites and understand what might be a way to get from here to there in reasonable steps that actually can give people results sooner. Um And so I do think that in this sense, more direction from the top is probably more necessary than not because there is so much newness here. And ultimately, you know, whether it's a CEO or a founder or CTO or whatever, whose vision is being implemented probably needs to be more deeply involved in the product than you might be on a, on a regular basis. So that's something I I've been doing.
15:30 It's I think something that our chief product officer is doing as well, kind of try to be a little bit more directive and, uh, you know, I, I think kind of clear about what result that we want to achieve. So by the same token, how are you also thinking about the risks? So what kind of keeps you up at night when you're thinking about the ways in which the wing man could go awry or uh could hallucinate or tell users. Um you know, when they're seeking information, they might not know sexual information, um it could lead them in the wrong direction. Um There's also questions about liability. So are you responsible in that instance, uh Data privacy is incredibly important to your users uh the risk of a cybersecurity breach. Um and also prompt engineering, the same sort of like prompts injections that can be applied to LL MS now could
16:15 also be applied to agents. I I could go on, but how are you thinking about all of this? It, it's really complicated and I think everything in life is a measured risk, right? Like you have to measure risk with potential reward and how much risk are you willing to take? You know, I'm probably more on a higher risk tolerance levels than most people. I mean, I um I've started three companies and, you know, taken them to something and, and most people wouldn't even choose to do one. And for me, like it's totally normal to, to, to do this. Um So my response is, is pretty high. That's actually a big reason why I think big enterprises usually get kind of challenged by start ups is because they're not willing to take the risk. So the biggest risk we face actually is to do nothing. The same way that there were many products for people to meet each other prior to
17:02 grinder existing, there were web-based products only. Um And then the iphone came out and there was a platform shift happening and those companies chose not to move to the mobile platform gave grindr a huge opening while those companies don't exist anymore more. And grindr has, you know, the market share that they gain or gain from, from them. Um Because, you know, *** men were online way before anybody else was in, in, in practice. Um You know, that's the biggest risk whether we don't actually innovate and then someone comes out and does something that is a I first and suddenly we are then trying to compete with them. So that would be, I think actually the single biggest risk but in getting to, from where we are now to where though obviously privacy is, is critical and um we are extremely privacy first. Uh We actually get challenged a lot about like, what,
17:47 why don't you ask for ID verification? Why don't you ask for photo from a person that's a face verified for? Well, the reason is because a lot of people still don't have the luxury to be as open about who they are as we would like them to be. And we and the world that we want them to live in. And that's not just true in, in countries where it's illegal to be *** or maybe not illegal but very difficult to be ***. It's actually true even here in the United States, 30% of the users are actively choosing to be discreet, right? And so we need to be extremely careful with their privacy information because it's not a matter of like, oh suddenly there is a breach, it's actually a matter of like, is there a life potentially at risk, et cetera? And so privacy is really critical and we are very privacy forward. We just released a new privacy policy recently which users have to actively
18:37 accept that allows us to use data inside the app to train the model so far, we've not done that and, and we'll only do that with, with user consent. Um And, and ultimately, a lot of what we'll be doing over the I will be based on user choice. It will not be, hey, here's a product, you must use it this way, right? It will be, this is available to you. If you'd like to use it, you can, but you don't have to if you don't want to. Um And because that, that is really critical. Um secondly, uh I think uh I think it's really important to chew off where you can actually kind of eat in any given moment. And so don't try to do something that's too much kind of do it in chunks and see what the results are with it on all fronts and then do more. So that's gonna be our approach.
19:22 That's partly why we're being so careful and who gets access to this technology from us um quickly, which, you know, probably slowed us down a little bit. But I think it is a better outcome um eventually at the end. Uh But, but again, I think for me, our users deserve to have a product that's not just good for among the other products that they use for something like this within the community, but actually totally deserve to have the very best product you can have in the world. And that's the kind of company I want us to be, right, like we are the second *** company to be public uh ever and the other one is not public anymore. So the only *** company mean like a company that was the first plan it out back in the kind of first tech tech bubble um period in 2000, 2001.
20:07 Um and, and so like, I think for us, we have a little bit of a responsibility to be successful and not just like, ok, successful but exceptionally successful because if we do that, we change people's perceptions about *** people around the world and that's really, really beneficial. And so I want us to be the very best product, not just because I want our users to have their best product because I want that to send a message. Hey, look at what's possible the same way that Martina Navratilova becoming an amazing tennis player in the 19 eighties and 19 nineties force people to change their perceptions of what it was to be *** and helped make it much more acceptable to people. I think Rwanda being really, really successful will force people to think about what it's like to be *** and, and force acceptance in, in a, in a better way. Right.
20:53 And also being, um, the first major dating app to talk about having an A I agent. So you're kind of, um, setting the bar in technology, hopefully everyone's gonna talk about it and then it's gonna do something. So, yeah, that's not why I'm talking about, I'm talking about it because actually I want to get people used to the fact of where we wanna go. One of the reasons to talk about this stuff in advance is so that when it happens, uh users aren't as surprised that, hey, this is the direction that you're going into. But my point is I think we have a responsibility to build a really amazing product that releases the full potential of what we can be for users. And that's a big reason why I think it's important for us to be at the forefront of this. All right. So I think we have time for just one or two questions from the audience here. Well, I have plenty more questions for you, George.
21:38 Um So when we talked about the sort of risks, um you'd said that you can't afford not to make the big bets uh in technology and also that um you're asking user privacy, you're asking users if they want to consent to this upfront, you're being very proactive. Um But still, how do you think about liability? Um Are your investors concerned about what A I could do uh where it might mess up? Um I think the biggest risk on the messing up, at least that I can imagine right now is around hallucinations whether uh and we can have liability protections on that. And, and, and we will obviously, but nevertheless, you know, a I says something that it's not supposed to say or it's not good to say,
22:24 then there's like screenshots about that, then someone tweets about it and then it kind of takes over, I mean, I don't know if people saw but like a parody site during the Republican National Convention had this little story up saying, you know, grinded usage is up 30% or something uh in, in Milwaukee and literally 200 million views of that video took place in three days. So I, I mean, I don't know how much brain marketing would have been, like, how many dollars you'd have to spend to get that kind of brain marketing, but that happens all the time. So I can imagine how a hallucination, like imagery could be really negative. Um And so that we have to be really careful about that and we are, and we put a lot of protections in place. Um to try to manage for that, but it will still break and we just need to be prepared for that. Um And then secondly, you know, it could lead people to do something that's not,
23:10 they're not supposed to do um themselves, right? Because it messaged something. But again, ultimately, users have to be responsible for their own lives. Right? I get into my Tesla at my house every day and literally right out of the garage. I turn on self driving and it drives me to the office and you know, there's one place where I have to intervene and, and um, and that's the only one we intervene in otherwise it drives itself, but I'm taking on that liability and, and that's my choice to do that. You have the alternative to get lost. You're not blaming Elon. No, I'm not blaming you on that. And also if I get into an accident, I'm not blaming Elon. That's on me. On the flip side. You could take the approach that Wayne Wayne Mo has taken, which is like we're gonna make this absolutely perfect before it's in user hands. That's too risky. I don't know if you'll ever kind of get there. Um And frankly whether it's going to be economical or not,
23:56 I think we're almost out of time. So let me just end on a quick one. what are some takeaways for the enterprise? I know you're building for a consumer audience but, and being a company that's sort of at the forefront of talking about agentic technology. Now, what have you learned so far that can be applicable to any company? Well, I think for enterprise, I think the bigger question is on how are you gonna use all this capability to make your teams a lot more productive? And I think, be using these technologies inside your own organization, right? I mean, I think, I think engineers in three years are gonna be on average two x more productive than they are today. Uh And that's gonna have massive implications for our workforce. Uh And, and how we work and I don't think we're grappling with that almost at all because there's uh days
24:41 coming when you actually won't really need a lot of junior engineers. I'm not saying you won't want them, you still won't want them because you cannot have senior engineers without junior engineers because somehow that people have to grow up to be senior engineers. But there's gonna be a period of time when you have a lot of senior engineers and then they could just be supported by a bunch of a I to do what the junior engineer does and that's gonna make companies way more productive and way more profitable and, and how we deal with that. I don't think we're in any way prepared for it.
