ai.bythebay.io: AI VC Panel

Let's start with the introductions. We'll just do kind of a two, three minute, up to three minute introductions so you guys can tell us a little bit about your firm or we have two folks from DCVC what are your superpowers inside your firm and perhaps how you make this interesting AI investment decisions. So I'll start with myself. I'm Alexey Krabarov. I'm not a VC. I'm the founder of organizer of By the Bay conferences. We do have majority of speakers coming from startups which are VC funded so we organically learn kind of second tier neural network how the hidden layer is making these decisions and basically that's it. Hi everyone

My name is Jake Bomenberg. I'm a partner at Excel. My background is in data. I came from Cloudera and Splunk and at Excel I look after a lot of our enterprise and AI investments and for us you know we're a global fund. We write checks from $100,000 to $100 million certainly focusing more on seed and series A but a lot of our thoughts around AI come from a natural evolution of our investment theses over the years. We first sort of got interested in in the big data world as a result of investments in social networks like Facebook and saw them sort of democratizing a lot of their back end data infrastructure systems for the rest of the world. Of course they were only democratizing that infrastructure layer. They weren't open sourcing their app

So the rest of the world needed help. They needed these last mile data driven software solutions. So we invested a lot around databases, data prep, BI, those sorts of things. But as we move further up the stack, you know, it's not all companies have the skills and superpowers that people in the bay have. And so the opportunity around AI for us is how do we build verticalized software that often you know starts around a particular data mode, a unique set of proprietary data that you get more and more of over time and leverage that with advanced analytics where appropriate. It's hard for me unfortunately to meet any startup ever at this point that doesn't throw the word AI in. And so like despite being at an AI conference like the you know the question I always ask is like do you really need it? Is there an underlying value to your software if this AI doesn't work? And I think we're at a point in time where a lot of very simple things can bring us a long way with the right data. The notion that we'll have a durably differentiable algorithm in the limit seems highly, highly unlikely

Google open source, you know, TensorFlow recently we can talk about the reasons why. I'm reasonably confident they're not going to open source their search relevancy data. So that's a statement by Google in terms of what's most important. And so you know I'd love for the panel to maybe explore a little bit about what's most important. I'd love for the panel to maybe explore a little bit about like, you know, do we do this problem really require a 10 layer neural net or can we get by in year one and year two with something not as fancy and gradually add to it over time. My name is Adam Kell. I'm a partner at Comet Labs. So at Comet we focus on AI and robotics technologies that are transforming traditional industries, specifically ones like agriculture, logistics, healthcare, retail and transportation

At the risk of sounding like I think probably a lot of us share a lot of the same thesis. So I'll give you two things that make us a little bit different from most other VCs that you may have met. One is that, you know, by focusing on basically all of our investments are in AI and robotics. So we spend a lot of time thinking about those core technologies. We spend a lot of time talking to PhDs and looking at the research industry that's happening, trying to forecast what technologies are coming down the line and how that will matter to different industries. And then to the flip side of that is we spend a lot of time looking at the traditional industries who will in most cases end up being the customers of a lot of these solutions. And so those are those are two key things that we're focusing on at Comet Labs. Cool

Thanks, Adam. My name is Jean. I'm at Data Collective. And we and I really am interested in basically how we can use AI to address typically heavy assets industries, things with one foot in the physical world, one foot in the digital world. And how we can use it to make those assets more efficient, how we can use proprietary data and novel algorithms to address CapEx and OpEx for multi-billion dollar industries. Data Collective more broadly, we are basically always interested in hard, defensible technologies that are addressing multi-billion dollar problems with a self-reinforcing advantage. Do mostly seed and series A investing as well. And we really are able to look across all kinds of sectors, including ones in Bradford's talk that I think were at the bottom of what the investments were

Things like physical security, things like agriculture, and try to see how you can work together between technologists and people that really know what's going on in those industries, those subject matter experts, to create value there. Hi, I'm David Beyer. I started out as a co-founder of a company called Chart.io, which is a cloud business and telegrams company. Switched to a bunch of angel investing, including companies that span this domain, like Zymergen and Cruise Automation, stuff that I share with DCDC and others. I'm now partner at Amplify Partners, which is a $125 million fund focused exclusively, very much like the folks here. We're focused on machine learning applications. The core difference is we focus very exclusively on seed and A, technical founders, building products for the enterprise. And I think you're not going to find much disagreement among the panel on the areas of our focus

I think the best way to describe myself in terms of the spaces, I'm on an unending quest to find people from very boring industries who don't live in California, who can sort of raise their hand and say, hey, I'm a domain expert. I understand this space incredibly well. I understand the nuances that people in Silicon Valley just don't get as a function of their lack of experience. And I have an idea, but I don't have someone on the technical side who can even help me think about how to conceptualize this. And when you can match those two together, it can be fantastic. It's very, very hard to find just because as a matter of geography, you have to kind of step outside of the Bay Area to find those people. And that's kind of, that's what I see actually is the biggest challenge. It's not so much the technical matter

It's matching those two sets of skills together. Thank you. Hi, my name is Benjamin Mlevy. I'm a co-founder of Bootstrap Labs. We're an early stage VC firm. We focus exclusively on applied AI leveraging big data. I think you've heard the theme all across. Thanks for the punt on international

I mean, interestingly, when we started Bootstrap Labs, we said, look, innovation is everywhere. People have existing talent, technologies, and smart everywhere around the world. The ability to scale your companies, though, is not equally distributed. So we see about 1,000 companies from 60 countries every year. A lot of the things we invest in are here. Even the things we discover outside of the sector system, we're trying to bring here, maintain our new back home if you want. So again, we are agnostic from where the talent is from and innovation is from. We see interesting trends, actually, by comparing what's happening here, what's happening outside

We only invest at seed stage. If you want capital, I think everybody here on this panel can probably invest in your company. We rely enormously on our community. We've built an enormous community around us of experts. We're hosting events on a regular basis to bring these people together and doing a lot of extra activities with founders. We think the combination of a domain knowledge expertise with people that have a new vision of changing things. AI is the new electricity. It's going to touch every single industry and change ways

You can discover how much is disruptive or not. But the bottom line is that there's some really cool new tech that's going to be permeating across the board, and you can start changing the way we've been doing things for the last 20 years and every software written. So that's what we get very excited about. Thank you. all know me. I just did the talk right before this. It's Merlin over there. I'm sitting next to Merlin

Superpower. What's my superpower? Yeah. We can't talk about that on stage. Oops. Sorry, folks. So, the last thing is something I think a lot of AI VCs I found kind of working on. So, we had the Bay Area AI meetup recently. So, you know, this is just one of the things within the community

Mostly we do about 50 meetups a year across seven different groups. So, Bay Area AI is our AI meetup. And so, we had Mike McCormick from Comet visiting. And we did this impromptu AI VC AMA. Ask me everything. Anything. And so, you know, after seeing our group and filling the questions, Mike said, you know, I think Alexei is legit. So, which I took, you know, I was relieved, right? And so, we want to work more with you guys

So, and basically he said in this AMA that the majority of work Comet is doing is kind of sifting through all the people with AI things in their pitches. And figuring out what is genuine AI versus, you know, all the buzzwords. And because Comet focuses on the AI area specifically. So, I would ask you, is this something, you know, you do most of the time? And how do you distinguish between legitimate AI investment versus all the hype AI investments? In any order? I think you first look at the business. Sorry, I will need to give you this mic. Sorry. Hi. I think you need to first focus on the business, as Bradford said

Before we even really dig under the hood at the AI, we look first, is this something that people want? Or are you telling them that they want something that they absolutely don't? And after you kind of get past that hurdle, then you look at the AI. You kind of look at, is this something that some smart grad students could spin up on TensorFlow? Do they have production engineers there that can actually make this into a real product? I mean, you can build your model, but if it's insecure, if it's not scalable, if it's going to have to be deployed in this massive 12-month implementation, you're never going to get any customer traction. And looking at the entire technology around the AI is often equally as important as looking at the model itself, in my opinion. Yeah, everybody else is fine. Sure, I can jump in a little bit on that, too. I think Gene points out something very accurate, which is basically that AI may not be the right solution for a lot of problems. I think the thing, when we talk about buzzwords, I think the thing that we think of is people who initially had a business model, and then they basically changed their business model based on the tides of funding to be able to shoehorn AI into that. I think that's the thing that is not the most exciting thing to see, for sure

But I think the important thing Gene already pointed out is that it's not, and Bradford talked about in the talk previous, but it's not like a grad student who has one hammer and just running around and trying to find that nail. The confluence of factors that determines whether or not that type of technology can actually work is a lot deeper than just the technology is now available. So I'd simply say that most AI startups are not going to fail on the basis of their AI. I don't know if that's an actually controversial statement. The number of startups that require fundamental basic scientific research advancing the field of AI that aren't, you know, like one of four companies, three of which are in the Bay Area and one of which is in China, you know, it's just not a lot. And so like, why do people fail? They fail because they fail to identify a problem that actually needs solving. And one of the most peculiar things I meet that happens time and time again when I meet with AI and just very technical teams in general is they come and all they want to do is show me the prototype. They want to show me the technology

And if you think about the process of building a startup, what are you trying to do at every step of the way? Reduce risk. And if you have, you know, a PhD from XYZ University, PhD from Stanford, PhD from Berkeley, and you know, you work for the right professor, you can walk in the door and I'm probably going to believe that you can build what you say you're going to believe with advanced analytics. And so if you have an hour in the day, what should you be doing? Should you go build some prototype of something that I'm already going to give you credit for? Or can you actually de-risk to go to market? Can you produce a human being on the other side that's going to say like, I want to buy this and this is how I can consume it? And guess what? Like, you know, you know, I'm not stupid, but like I might not have all the technical skills you have. I might not, you know, understand SQL. I might have a very unique set of domain experience and like the UI and interface I need to get value at, right? You can't just sell to the smart people on the planet. I would love to be able to do that. And a lot of people try to build products for this really smart, sophisticated data people. There's just not that many

And we can talk about how much education is or isn't going to happen. There's always going to be a lot more, you know, less sophisticated data analytics people. And so the solution that really worked for them to solve a problem, that's what's going to make or break the average AI driven startup. Fundamentally, I just don't think that AI being involved changes the dynamics of going to market or the value that you're selling. It's just you're selling something that can be material or substantially different than anything that existed before. And to the point of hype, truly there are a lot of problems and businesses that can be created just by applying logistic progression. No, there. I did not

Thank you. Just by applying... Is that good? Micah, micah, micah. Here. Basically, I'm just saying that the presence of AI, quote unquote, doesn't fundamentally change the dynamics of how you say take something to market. It doesn't change the notion of packaging software so that a user can use it. doesn't really change the value proposition except in so far as it lets you do something substantially different. But again, that's still packaged in the normal way of doing business

A lot of problems can be... Is this good? A lot of problems can be solved with logistic regression. And you can probably build businesses that claim to be AI that wrap logistic regression and go after industries that have seen very little work in that area. So it's not... I don't really care about the AI per se. I care that you're solving a real problem and that it also is a second dependency. The industry you're going after, you have a credible sort of plausible approach to aggregating data from those customers and potentially hopefully learning over time. Because if you don't, it is hard to build something that will be sustainable and really last

So... And then just for the fun of it, I tell you, there's so much AI startups out there. We just have an AI powered eight ball. We shake it and we say yes or no, invest or not. I mean, joke aside, I think to your point, we're not going to change the way you build a successful company. The premises of huge markets, passion into the space, changing the way you're going to market or the technology using... You know, we look at the world and say, okay, no, we finally have the data. You know, AI is not new 50 years into the making

And we're now at a stage where from a world that has a thousand decisions being made by humans, we're going to go down to 20. And that's going to have massive impact, no matter how you look at it. And so you're going to have two types of AI. They're going to AI, they're going to really focus on productivity improvement and taking the jobs away from humans that are actually doing it. And they're going to be AI that are going to be focusing on augmenting what you're able to do today. So I like the second one a bit better, but we're going to see both. And I think we can look at the world in that way. And it says, what kind of vision do we have for the future? Hollywood has been terrible at painting the picture of what AI is going to become

And that's pure AI, true AI, AGI, which we're not investing in. We're far away from for now. So, you know, I'm interested in people that have a sense of design of the user experience, right? Whether it's inside the enterprise or not. Because I think that's really where a lot of the innovation is kind of lacking today. People are lacking creativity around what is the job of tomorrow going to look like. And when software is more autonomous and more acting alongside you in the environments you live in. Thank you. Okay, so if I have to chime in on this one, I have a more boring but at least much simpler answer

So how do you see that the AI startup is legit, right? Was your sort of initial question. So I just look at two simple things. At the end of the day, you want product market fit. So I look at, you know, do I think that these guys can create a product that meets a real market need? Like some of the other guys mentioned that. So that's the first thing I look at. And then I look at what I call team product fit. These people have the right team that can build this product. And that's it

It's just a two-step chain. And, you know, it's pretty simple. Like if somebody is building a computer vision startup, I want them to have actually come from a top tier computer vision research group. That's it, right? It's like if you're telling me you're going to do that and you don't come from that background, then I don't believe you. Period. It's pretty simple. That makes sense. So let's pick up on this

I think it's actually a great kind of connection to the next question. So AI is one of the kind of technologically most involved kind of groups of startups emerging, right? And so I think for over several years, I kind of see there is a dichotomy of VCs. There are leaders in technology like who have a huge staff and technology come from technical founders. And there are followers who kind of co-invest, right? And so how much due diligence should you do in AI? How much can you do? How do you do this? Right? Because there's some of the most involved technologies. How do you vet essentially that team product fit? Well, for me, it's just like I was saying, you know, do they have the background to do it? Do they have the educational background? Do they have a track record of doing such things? You know, we frankly are seeing a lot of LinkedIn profiles that have miraculously been updated, you know, over the past 12 months. And, you know, I think I think that's something to watch out for. Right. It's like, is there a real background here or not? You know, because this stuff is actually quite hard to do correctly

You know, if you and that's, you know, again, like I think that some of the other people said, assuming that you really do need to do this stuff in the first place. And sometimes, you know, companies really don't even need to be doing it. But assuming that they do and there's good business reasons for doing so, then you want to just make sure that they have the right background to do it. And I think that means on the technical and the subject matter side both. Right. If you aren't acknowledging the need for the subject matter expertise and you think you're figuring it all out on the fly, you're not coming with bias BD relationships, you're not coming with a baked in knowledge of like a lot of things that you can put into your product in advance of just getting it from the customers. I think that's a mistake. So I would just look for those those two things, you know, and again, like the track record in the background really has to be there

It can't just be that they are passionate about this or something. It doesn't work for stuff like this. If I can augment on that, just I think fully agree. I think the team product market fit is super important. But every team is imperfect, especially at the early stages. You have two, three, four people. So we always look at it and, you know, what are the core skill set? What is the founder really good at? We like to look at assertive humbleness. I mean, what are your really strong areas and you realize that? And what are your weaknesses and are you willing to listen and acknowledge them in a way that actually will attract much more talented people than you are in those areas? And so that's why we rely also on the big community around us

We want to make sure that that product market fit assumption validation iteration loop, it goes as fast as possible. So I think all areas of expertise is working side by side with founders, helping them figuring out the product market fit, which is what I think is the most critical stage of the life of a company. I think Naval said before you say reserve on you're going to make 9000 of the 10,000 most critical decisions for your business. So I think that's really where it's happening in the tranches and we love to be there. I think, you know, I'd add there's also an element of timing and it comes back to, again, what is the fundamental problem you're trying to solve? If you want to do deep image analysis or find like features and videos or something, okay, like you probably need to have someone with the right skill set on the founding team. There's plenty of other problems that, you know, a lot of data can be collected for a long period of time and then you can add more intelligence later. You know, a great example from our portfolio is Slack. I think if you looked a year ago, probably couldn't find, you know, anyone that I think would have like a deep, deep academic appreciation of this field

If you look now, it's very, very different and they were able to be, you know, quite successful before they hired their first AI person. And so, look, I think everyone on this panel is probably going to be one degree separated from some of the best professors and thought leaders on the planet. And so if we can't personally bet the person, there's always someone right next to us where we can just ask and say, hey, how does this person stack up? So the evaluation of the AI talent, it might like seem like that's a huge part of our diligence process, but like, I mean, that takes five minutes. I think my answer to this one is pretty simple. It's called do the freaking work. You know, just read the papers. You can have the smartest people in the world and if they're not really thinking about how to stay ahead of their business, they haven't really thought several steps ahead strategically on the technology side, you can figure that out pretty quickly. Some of the things that I really often ask is around the data that you need to build a lot of the new AI models, right? Like it's one thing to try and do logistic regression

It's a completely different thing. If you have like a 10 layer neural net, you really need millions of data points. And for one company that we were doing, saying I think one of the most convincing things for me was that they actually showed me the curves of how many images will it take to it for us to get to here, there and what their projections were on the technical side, not just the business side, not just your sort of made up five year projection to get to the accuracy that they need. And the same with other people whose algorithms that they're saying are going to give them that 18 month edge, even if it's not sustainable, you can ask them about all the other techniques in the field. And if they don't have a good response to them, you know that they're not keeping enough tabs on what's going on around them for that to actually be a sustainable advantage at all. I had a hard time with audio today. Just to put one last point on this, just referencing what Jake said, very few of these startups will fail because of technical reasons. I've just, I've personally never seen it

I'm an investor in one quantum computing company, which is the most technically risky thing I've ever touched. Everything else will fail because of the typical reasons startups fail. They don't understand their customer. They don't figure out the channels to go to market. They don't recruit well. So I think the, you can, to some degree overemphasize the importance of the AI in this, in this equation. I really don't think it matters as much as all of the other things combined. I'll ask one more question before we open the floor to the audience and we'll have them probably about five to 10 minutes

So I want kind of an insight from the guys and a problem which is very kind of near and dear to me and I was trying to solve it with a lot of different VCs and I found I think there is a gap in the VC model. Right? So here we have a community conference. And we have, you know, I run seven different ops. I very rarely see VCs in high profile at the community events. Right? So they tend to build communities around them. So we have this kind of walled gardens with compartmentalized communities and access is important. So if you are an Andresen founder or a DCVC founder, you have one kind of community. If you're a comet lab committee member, you have another kind of community

So you have different mentors who gravitate to different VCs and also kind of they can have multiple relationships. So a specific problem I kind of posed to Andresen. So Andresen has a hiring arm and they have actual recruiters who recruit for the portfolio. Right? But these guys don't come to the events much. And my question was, why don't you guys hire for the portfolio? Right? Why don't you raise the profile and, you know, like IBM and Salesforce, why don't you have a brand where you can route folks to your portfolio companies based on your brand? And they basically say, we don't do that. We will let the portfolio companies do that. But then my question immediately, the portfolio companies are stretched. Right? They are, you know, hectically trying to build prototypes

They're not going to sit all day in the conference. Usually they're not going to sit on the lawn of Stanford University with a T-shirt. Right? So I sense from my community side, there's a big gap. So I think that's a big gap. And I think that's a big gap. And I think that's a big gap. And I think that's a big gap. gap

And I think that's a big gap gap. gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap gap