Fireside Chat: Pioneering the AI Revolution
Gen AI Summit 2023 - Open or Closed
HubSpot: Yamini Rangan, CEO
Zoom: Eric Yuan, CEO & Co-founder
Fellows Fund: Alex Ren, Founding Partner (Moderator)
SUMMARY KEYWORDS
ai, customers, companies, model, data, features, leverage, hubspot, technology, question, talk, product, work, approach, large, perspective, terms, great, regulate, huge
SPEAKERS
Yamini Rangan, Alex Ren, Eric Yuan
Alex Ren 00:00
They are random and archaea and we do share one thing in common. We are all first generation immigrants to the United States. And let's do some aerobic plus folk. We all have humble beginnings. And remember, I tried a nine times. Reserve and finally came here and found the loan which benefits, you know billions of people theory and after dynamic and younger stories equally inspiring. And I remember she came to us and quantify it. Alright, as a five as obviously a few $100 and network restore was in a restaurant. Yeah, eventually he became a leader in top tech comm with companies like SAP. org, the Dropbox and eventually, HubSpot. Leaders lifetime are truly our role models, and they embody the spirit of entrepreneurship in the blood. So that's, that's the time into our conversations. And we're going to also leave some time for all these to ask some questions for them. Okay, let me still Sunday with my first question. Sure. Okay, so say the outbreak of what generates 10 debriefer AI relevant. So what's your personal feeling? After the lightning your passion again, to work outside?
Yamini Rangan 01:44
I think I can get started here at nine times. I don't even know what happened nine times. This is a story that you have to catch. But it is so fantastic to be here and see that energy and this is the entrepreneurial energy. And it's just fascinating times as such a great time to be here and to answer your question. As Absolutely. I think I've like replaced every one of my Netflix binge watching with an agenda AI podcast watching and there's so much that is happening right now in the industry that feels like things are moving at the speed of light where every week there are 10 announcements and 10 view, you know, kind of big, big progress within the field and so it absolutely is I started my career. You mentioned I did my undergrad in engineering in India. I came here for masters and one of my first subjects here as an engineer was neural networks. And back in this early 90s neural networks were just kind of getting started and to see the level of progress in terms of what neural networks are able to do, which is really the foundation for any of the large language models. It's just fascinating. So great to be here and great to be in technology at this time.
Eric Yuan 03:06
So as you saw when you ask a question as a student, my feeling is I feel very old. So the reason why go back to the early days of internet, right so I spend a lot of time trying to stand for internet works and familiar Internet Protocol, HTML and a lot of Arizona I feel like I can catch on. Now you look at what's going on even generally with the R word right Aragvi all the research papers so many lives have been involved in so many different companies. Also a lot of equities as well, I cannot remember and so that's why I have to wear it. But anyway, so daddy's innovations in the AI industry in particular for gender is as we were happening. That's a huge opportunity. Unfortunately, unfortunately, I'm not the CEO of zoom, otherwise, called very likely. I will start to come to Zambia as the first person to share it, but it will get a benchmark but often it's an ad hoc, right. How do we make sure that everybody by the accuracy, calibration, right response time, or computing consumption, a lot of errors can either be the huge opportunities given to you look at the type of technology stack right how automated a large number one was beautiful by a very valuable model that transformed right how to acquire in a big generally like to all this important use cases huge opportunity. That's the best part for any of you who have no background or interest in EDI have also in generally like as soon as specified to start a company you're already young, so I cannot do that.
Alex Ren 04:52
I took a loan with a never be both. So then, yeah, so after I've showed the Chalkidiki Did you receive a custom cause, you know, actually for more AI integration in your products, so that AI become a top priority for your respective companies since then.
Eric Yuan 05:14
I think before that it's up to charity being formed in California like we already have made investments in a little bit of all the basic features my lack of lack of watching Animal Rights, only the the noise suppression and also like, gesture, you know, the hand gesture recognition, all those features as well. Remember that we are already right. And also this reason why we highly invested the AI resources now over the past several years without help of our two of our most important acquisitions. When is the case in other major so in both of them, they look something like that and inside of that we will be in a copper jet with the iron ore. So we're also going to be bringing that down quite often every time I talk to the customers number of questions recently they asked him how you guys leverage AI to improve your product stress. The second question always, mostly obviously, this is very important. I think everybody needs to have the blue slumbers you're gonna learn today you know that the zoo, we ask every product managers and engineers, how you can leverage AI to improve this feature and that feature all the product roadmap, you know, we're also going to have AI tech science. Well, again, this is more like a very busy stuff for every company recently.
Yamini Rangan 06:31
No, I completely agree with that. I would say that there's two parts to question and I think one is the our customers asking and what's the conversation you're having with customers? And then second, has this become like a primary priority within the company? And the first part of the question, it's interesting as as I talk to customers, they typically fall into two categories. The first one is, come on, where are you? What's your roadmap, and why haven't I gotten done more features? And, you know, how can I leverage it every day? These are kind of super early adopters. They are already experimenting with it. They're looking for ways to improve their own effectiveness and productivity. And they're all over this. I think the second one is very interesting. And we're learning but we don't quite know exactly how to apply it beyond the hype of kind of going into, you know, another lab and asking questions and so they're kind of waiting. And by the way, we serve small medium businesses, right. We serve companies with typically between one and 2000 employees. These are the folks that have aI experts, LLM experts, model experts within the company, so they're kind of waiting for us and they're interested, intrigued. More than average shepherd of AI. And interestingly, there is actually a whole group of customers that are like, Okay, this sounds good, but I just don't have the time to go to the iPad and learn about the 20 models that you know, Professor Jung just talked about, right? I can't, I can't process all that. They're more of the late adopter. So I do think that technology is developing at a much faster rate than customer adoption of technology. Now, having said that, from a HubSpot perspective, last year, we knew that this was a trend. You know, we've had like an AI group, but this year, we really focused our whole organization on gender bi and leveraging the use cases there. So for your second question has the chip that company morality Absolutely, yes, I think you can talk to our product managers normalise in bright right now they're thinking about their roadmap. They are looking at what we can do with genuine AI that can drive effectiveness for our customers. And so absolutely focused much more in terms of leveraging Jamie,
Alex Ren 09:04
thank you very much. And let's talk about those specific use cases. Right? So how to zoom or passport use the day and I will say impact on your customers.
Yamini Rangan 09:15
I definitely want to hear because I'm looking for the next set of you know, features from zoom. So from a HubSpot perspective, I just mentioned this the customer segment that we serve. Small Medium Businesses don't have aI experts do not large language model experts and we've always really felt like we can take sophisticated technology and really democratize that for SMB. So huge relevance and the way simplistically, we think about it is that we can help our customers generate content. We can help our customers generate insights, we can help our customers think about how they can generate code faster. So there's just like fairly big implications. And you know, from a product strategy perspective, we're just building AI into the foundation. of HubSpot. It's not one where it's choice to show up in one product. It's like foundational at the platform level. And we're looking at how we can leverage m&m technologies and I'd say that the first you know two big areas that we put our feature in watch the first one is called chat bot. And that is natural language interface. And it's very simple. If you think about the CRM, you know, today you're actually a marketing expert. You have to be a sales expert you need to like understand or skip through and be able to get the value of CRM and chatbot sits on top of large language models connects to our API's and provides a natural language interface so you can ask it, you know, what was the best traffic how to come? back last year? Can you create that report and send it to the other one day you can just interface with software which is going to drive a level of engagement? We're pretty excited. In the last two months, we've had more than 40,000 users that are you know, adopting and engaging with them giving us feedback and iterating on it. And that's great. And then the second set of features, again, out of the stage that the cloud is content ideation and creation now, I think it's going to become table stakes, but the ability to be able to look at you know, kind of content that you can put out as a marketeer or summarize the content as a sales person or service person is just exceptional. Again, early days, we're super excited about what we can do in terms of driving value for our customers.
Eric Yuan 11:54
So we already have quite a few new JBI features in beta and reversal and use features in awareness versus ego Samsonite. Opera media is all right, and they can pretend to be new software. And also let's see, if you are late to the meeting. We also can gender we will have to start the software and make sure that you don't miss anything. And also we also have a new IQ test. Which is on Prime Minister slack and the team's files the problem beyond without a feature. I urge you to try that feature whatsoever help is also free. also connected to Jenny AI to help you compose a chat as well. Last year we also announced the Hebrew calendar icon leverage and er, universal Minamata whiteboard, those integrated a Gemini as well. I will conduct a separate virtual agent. I see a lot of features, you know, either in G or in the beta. Again, there's a lot of stuff.
Alex Ren 12:49
After you think those features are they and less questions about actually today's topic is open source or closed source. And I know you guys are working with some large rapid models like the Wii or astropay. And so will you also explore some open source models like Atari and for others?
Eric Yuan 13:12
I think you know, our approach is we look at everything from India's perspective because an inquest connect him prior to that and we spend a lot of time talking to customers to try to understand you know, their demand, right, what kind of issues fishing so we came up with a federated AI approach. So meaning, you know, not only do we have our own last name in the model, and also impress others as well. You know, we announced the collaboration with open AI back to the inverse net, and also recently invested in solving images to create a large number of other companies. At the same time it also is open source as well. And in particular, you know, Adam, customer, some large enterprise customer, they told us, Hey, we all have our own last metamodel you don't want this annual model you have and you support them, you also support them. In the we also can you know either who is integrated with our customers, you know, the large number model as well is SB that's a random AI approach. Nobody open source or closed source or causal model of our own or that we asked for that, you know, we just sent around the customer demand. That's why we're the flexible approach.
Yamini Rangan 14:21
I completely agree I think you know, the question really from from our perspective, we are Maathai LLM friendly, right, so, we're working with open AI right now, but I think what is happening is that open source is catching up really, really quickly. And I think it's all a question of matter of weeks in terms of the improvement in catch up. And so I think, again, very much like you, Eric and zoom, we approach it from our customers perspective, and I think our customers will demand that we are multi LLM. And it's pretty exciting to see the kind of improvements that are happening based on not months or years. So
Alex Ren 15:09
you know, you're really out to train a good model. So you need the five qualities or so I'd love to learn more about the data strategy. Like you have some privacy, they'll be able to do the model or find some of the moderates. What's your strategy?
Yamini Rangan 15:25
I can start I think it's a really good question. So right now we're working to get in a couple of like, large language models. And I think about it as data out strategy versus data instruction. What I mean by that is, data out is if our customers you know, sitting on top of hotspot or wearing large language models, we provide them to control they can ask you to opt out, and it's pretty easy for us to say none of their data prompts can actually chain large language models. So control is really in the hands of our customers in terms of whether their data is allowed to be able to train large language models or not. So I think that's like number one. I think in terms of the data in strategy data within you know, HubSpot, again, we've always focused on customers and giving them the ability to figure out you know, what they can share and what they cannot share, even this age, because really important in terms of transparency. I think Professor Jung talked a little bit about transparency. It's really providing clarity to our customers of what we're leveraging their data for, how we're leveraging it, and providing the clarity for what they control and what they want control. And I think that's how we brought the project. Now there is you know, troves of data, it's a question of how do we leverage it for data in essence data, marrying a very good answer a second question, the second column.
Alex Ren 17:09
Okay. So, you guys know So recently, there was a congressional hearing, you know, we I see some automatic and there is some like, AI regression that we needed and what are the federal risks associated with general AI? And the lie what is the what could be the implication for enterprise software industry? And what will make this
Eric Yuan 17:33
increased AI regulations? I think you're speaking so I think the difficulty is all right. So you know, we should all family the less the less developed myself the long toss I think that no person has no teeth. I will say you know, Arabic comedy right. How to oversee you know, take the you know, generally I've heard versus meeting the Godfather, pushy I you know, very very safe. And so, you know, otherwise, you know, we're likely looking to lose their trust, right, because you cannot view the customer data at your home here, right. You have to make sure for the customer, the end user security, and a trust and a central purchasing body alibis whereby people with the commander's right that's why I had assumed almost every journey and features. Nobody thought we're not allowed to travel. All right, we have to let a customer of the other of us if not take that approach. Even if I have a great you know, error model, guess what the customer may not trust you when the customer lose your trust, my God, nothing new work. This is this is very important. That means, you know, philosophy. So
Yamini Rangan 18:41
I completely agree with that. I mean, let's just take a moment. If you have not heard the congressional hearings and the whole conversation, please do it. Because what was just incredible listening to that conversation. I'm not political. So I'm not making a political statement. For the first time bipartisan. Everybody wants to solve this problem together. Second thing that struck me was that you have an industry leaders like Sam going in and say, Hey, regulate us. Like, you know, I started my career in telco. We didn't call you to the regulators and say regulate us. And I think the fact that we're having these conversations and the fact that we've learned something from the social media model, and we're kind of leaning in is just incredible. Then the question becomes, what do you want to regulate? And what are the policies there for us? To guide and it really comes down to transparency, and customer trust. I think the question then becomes, we want to make a game well, I want to know whether I'm interacting with a human being or an artificial intelligence why? I want to know how my data is being used. I want to have a little bit of control of where my data goes, if I do decide to opt into something. And so I think it was just great to see the kinds of conversations they're having. And I think as an industry, we should be part of, you know, really working with the educational institutions but also with the government and policy makers to make this something that is value and ditto for humanity. So I'm, I'm very diverse, but yeah, we're
Alex Ren 20:41
gonna see how it works and personnel for the US, both public, to public public companies, and all the requirements, cost reduction and investment in AI. So this question about the movie, how they seem to be able to do
Eric Yuan 21:04
that on many fronts right here is going to change everything. I mean, from our perspective, no gotta look at it from a product processing people expect from products that we just talked about that you'll have to increase the icon level we add more and more weight to the prototype product experience. As an eternity we look at a process by our rapid AI to improve the process workflow by fully mutual process over an automated and also that AI to feed our what are those holes, right, you know, for success, also very important. And I've asked Well, these days in the heartland AI program, Katie is very important obviously aren't better farmers reason why I normally look for that side, you have the risk AI on many fronts, because it's going to change areas what we do so
Yamini Rangan 21:47
yeah, I you know, I'm of the mindset that AI is not going to replace humans, but humans that leverage AI, we're calling to replace humans that don't leverage AI. And we're in the business of helping people leverage technology and democratize technology. And so I do think that you know, when you think about, like production data or some rote, odd tasks, repetitive tasks, novice tasks that will get on it, but that also means that the value that humans can add is actually going to increase, not decrease. And so we look at, you know, the same, you know, product and equal but how do we help a marketeer become even more effective, not just generating volumes of content, but high value content? How do we help salespeople and even more effective in connecting with customers, and how do we do the same thing so I think that's the lens with which we are approaching
Eric Yuan 22:51
the argument that a leader would have definitely will I want to add and replace you. With that all the work of setting the standard to be academic, Dr. Jason dump, I tend to be helping patients so throughout the whole five days a week down the road, if a guy gets more and more popular gets for three days or four days. So that's why I think it should replace most of you.
Alex Ren 23:16
I like that I want more vacations. Yeah, so many other. They're talking about sort of the mid audience of founders to investors having the distribution, schedule over modules and data, how hard is most of the labs to compete?
Eric Yuan 23:53
I think that again, as a mission variable, huge opportunity, right? And if you have if you're entrepreneur, if you're gonna start a company, I would say is the best opportunity I think, don't always assume you know, you know, they lost confidence they can get even faster but we're gonna open guy, no one ever was on five years ago. They were very smart today. And so we could say that no, that's not started. Before. Nowadays. Political confidence, right? Funny thing is recently a friend of mine, Mike. He's not from high tech. You asked me hey, have you ever tried next up ultimately, I think Oklahoma is a different company, the startup company right now like an email develop those cool technologies not that developed by us. They the Congress is 1000 times outside of the years, a startup company I think a huge opportunity because they are very big. We move very fast right now Innovation Suite and much better. That's reason why we subsidiary is coming back, open your eyes. So like agenda, AI is the best time for startups, because you're ready.
Yamini Rangan 24:54
I would completely second that. I think. Some people didn't see advantages. And then there are some stars advantages, right? You think about these events, the data and what you do with the data is the ability to get feedback from your customers, because one of the reasons why we were opening I got to millions of users in a very short period of time is it put it in front of users so they got the feedback and therefore they're iterating much faster than others. And so I think, you know, incumbents have the advantage of the human feedback loop as well as the ability to leverage large data. On the flip side. This is a time where processes that have power and knowledge workers should be imagined. The you know, you keep up like every process that a knowledge worker does. The lines are completely blurred in terms of what is AI driven, what is AI assistant, and what is human driven and if you're a startup, you're an entrepreneur. You know, just do that. Like you got to start from a clean sheet of paper and reimagine what work means to be a Heineken hour and so, a great time to be more established, you know, technology better as well as a startup and I think we're gonna have a great deal of fun over the next decade. It's
Alex Ren 26:35
totally up to you too. And so as a investor, I think I always use this analogy of like Martinez being the 9095 96 in New York offices in the new space, and today opened my eyes as for me, something opened this whole space, but there's huge opportunity for startups and other companies to apply AI to different ways because unless I agree with that,
Eric Yuan 27:05
so So I guess the implication of your question is read about their work as a good idea on startup company. So it's on his eyes, they have you know, fellow spark investing guys, okay.
Alex Ren 27:17
reasonably early investor technology that he has faced in the last few months. Yes, because you agree with that. So we have a lot of time to take questions. And obviously, your name and colleague
Eric Yuan 27:45
in the back Donald works.
27:55
out I think it's a rather bizarre question, but the authentication of what people say in conversation via Zoom is an application of efforts are fascinated by I'm the CEO of a company called omniscience we basically build these retrieval augmented generative systems so that all of their data everything you've ever written or seen before, talking about people is behind the larger picture. And not only is this an awfully nice zoom calls to augment my decisions to answer my questions in ways that are similar to what I just said, but that much better detail. And every employee that I have to perform is actually the face of that kind of feature. And I wonder if it's to have to be brought to the fore. So that says,
Eric Yuan 28:40
yes, he's sort of a big factor in where again, you know, you know, because the composition we all have those days of it that you record a meeting again, as a mission right in the meeting software feature will be available very soon. So not only do we get all that but also download resources for planning as well. You can apply your own you know, model you know, to kind of cosmic try and make accommodation, you know, all the data in the open, you know, to those, you know, the meeting hosts like you can find your own model also our own model just for that basically essentially what we have that person real time conversation. Yeah, that's the exact real time and also the posting service for us. You know, like when our making awkward things or with gender that also you repopulating the media. As I said earlier, we also support as well also expose a car for you from your home. You're planning as well. That's fractions.
29:30
Sisters progress from flying ventures. One of the things that keeps me up at night is making the wrong investment. In the world of generic epi looks like there are a lot of opportunities out there are companies placing a bet on a fine shoe foundation tomorrow to upset somebody else. What is your opinion whether these are the three products the risk given that you know, as we said Congress is going to regulate that industry?
Eric Yuan 30:03
I think it's a it's a tough question. You know, I do not have a good answer to that. But uh, however, I will say, you know, given all those, you know, opportunities so I think that you know, it's hard to to bet on one company by technology, right? So no copy has to be opened by the FBI. So he's back to like, maybe years ago right. So he really does have internet so you're gonna have to take an open minded approach. You know us the problem that a lot of accomplice regardless all kinds of challenges or your office, you have to give a lot of wisdom so that's why you know, our approach is always like a clicker open minded approach, adopted all the last nine modules and the focus area is that accompanies us to almost every possible successful companies and stuff. So that kind of my, my wins.
Yamini Rangan 30:51
Your job was exceptionally difficult to get out and I think all of us is when you're doing a big transition and a technology transition like this, I think you have to look at like every layer of the stack, and who is getting disintermediated who's got competitive votes and who's going to like survive and it's, at this point, you can take every layer of the stack from the hardware, to our labs, to you know, software on top of that to like, existing SAS companies and you get like with that, and you can have your own thesis on, you know, who's gonna win who's not going to win and who's got the advantage? And it's, it's pretty hard. I wouldn't say that. You know, you'd asked like, maybe six months ago, you'd have thought that oh, the world is gonna have a few elements. Well, now six months after you will say no, I don't think that there's going to be two other months. There's going to be many, many specialized MLMs in there and I think every layer is going to likely get, you know, really transformed as we figure this out. So, early stages, you have to have hypothesis, you have to like, you know, make a few bets and normal that, you know, to Eric's point your views can, you know, have to like adapt and shift very quickly. Me meaning. at HubSpot. We actually look at it from our perspective, who's got the advantage? What is the advantage and we spent a bunch of time talking to a lot of AI experts you know, from Academy from the industry. And there's not clarity of where we will be three years from now or five years from now. But there is actually pretty good clarity of what you can do right now, which is get a ton of feedback, iterate and really identify use cases that are going to be table stakes that you have to have, but then identify use cases that are going to be specialized based on the data that you actually have. And so that's kind of the approach that we're taking, but it's exciting times, but it's also times of great transition.
Eric Yuan 33:00
So when once you look at it as such in the buyback, netting and environmental analysis, there were so many similarities right? To the party invest is what finally, you know, at that time, Google, you know, Google was bought from them and then if you lose the focus, while you say oh my god, this boy is over, right? I'm not invested in companies. Agreements. I think as soon as you can get back to this to this video, so you have to be patient, open minded, you know, try to always keep a close eye on those systems agree to start complex, right? Otherwise you cannot always do when you should. I agree.
Alex Ren 33:38
With that. Thank you very much for your sharing and let's give a big round of applause.
Yamini Rangan 34:04
Thank you so much out here our source control, insightful discussion. I know I personally
