EP 192: How VC Firms Can Get Ahead with LLMs

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

With just about every company that's raising money trying to sprinkle AI into their offering, how can venture capital firms make sense of it all? Well, by using AI, by using large language models, right? I'm super excited for today's show. We're going to be talking about that and more today on Everyday AI. Welcome. My name's Jordan. I'm your host. And Everyday AI, it's for you, it's for all of us. It's so we can all learn what's going on in the world of generative AI across various sectors and how we can use that information and leverage it to grow our companies and, and to grow our careers. So we are going to be talking today about how you can use large language models in venture capital. So whether you work in VC and you're or maybe you're a startup founder trying to raise funds, today's show is definitely for you. All right, so before we get into that, as we always do, we're going to get into the AI news. So as a reminder, go to your everyday AI.com sign up for the free daily newsletter. We're going to be recapping our conversation today as well as more information on these news stories. And I tell people it's like a free gen AI university on there with now close to 200 backlogs of great in depth podcasts from industry leaders. You can go read every single newsletter. So make sure to go check that out. All right, let's start with Google. All right, always things things always start with Google. So Google in the news today? Well, they've released some new gen AI features into their Chrome browser. So Google Chrome is introducing three new experimental AI features to improve user experience and efficiency when browsing the web. So Chrome now has a new experience experimental feature called Tab Organizer that automatically suggests and create tab groups based on your open tabs. That'll be great, right? Hey, I have 92 tabs open. Use AI to tell me what's what. Also now users can create their own custom themes using a generative AI model, making it easier to personalize your browser. And then the last one. Next month Chrome will launch an AI powered feature to help users write more confidently on the web. So bringing you know, aspects of the bard, you know bard inside of the Google Chrome browser. Also in related Google news. So Google has ended its contract with Appen, a data company involved in training its model for AI tools for Bard Search and other products. So this decision was made as part of Google's efforts to evaluate and adjust its supplier partnerships or efficiency. Or maybe, I don't know, Google got too much, you know, kickback that it wasn't doing well enough, that Bard wasn't performing well enough. Who knows. All right, next, Sam Altman says the AI industry should look in a different direction to power AI nuclear. All right, so OpenAI CEO Sam Altman discussed the future of AI and its growing energy demands at the World Economic Forum, emphasizing the need for a break, a breakthrough in energy innovation. Altman believes AI requires a breakthrough in energy and has invested in fusion fusion projects for this purpose. He also downplayed concerns about AI's impact on politics and the job market and the controversy surrounding OpenAI's use of training data. All right, last but not least, SAP's shares hit an all time high after announcing a restructure and a push toward AI. So sapse, the German software firm, announced plans to restructure roles for 8,000 employees while also investing 2 billion euros toward AI driven business areas. The company also expects to end 2024 with a similar headcount and sees Genai as a game changer for its business. So shares jumped about 7% after the restructure announcement. Also, it was reported that SAP has been using yes Chat GPT, the same tool that all of us use. Yes, a giant like SAP has been using it since the early days and they do plan to roll out it the open AI integration into its own products this year. All right, so I talked about this in my bold 2024 prediction so that this is going to be the year of all companies either laying off or quote unquote restructuring, investing heavily in AI. So SAP, one of the first big companies to make a splash in that department in 2024. All right, that's enough of the news. If you want more, like I said, go to your everydayai.com Sign up for the free daily newsletter. But for everyone joining us live here, I'm excited. So you know Tara and Brian and Woozy, some of our regulars and Raul, thanks for joining us. Let me know what questions do you have about venture capital, about how you can use AI and large language models. So luckily you're not just going to be hearing from me on this one. I have an expert guest today that can walk us through it. So I'm extremely excited to bring on to the show. There we go. We have Roger Thornton who is the co founder of Ballistic Ventures. Roger, thank you so much for joining the show.

Thank you for having me, Jordan. And congratulations on the great success of the show.

Thank you. Thank you. I appreciate that. So tell us, tell us a little bit about what Ballistic Ventures is and what you all do.

Sure. Ballistic Adventure. Ballistic Ventures is an adventure, and it was born out of a passion and determination of a group of US partners that are trying to save the world. We are focused exclusively on, on funding new cyber security companies. We're early stage investors, so we're typically the first institutional money into these companies. Oftentimes the entrepreneur will be working with us before they've even really form the idea. We've been in business now for about two and a half years. My partners, Ted Schlein, Jake Sade, Barmac Mafta and Kevin Mandia are all experts in different areas of security. When you engage the firm, you engage all five of us, and we roll up our sleeves and help you build companies. Several of us have been cybersecurity company founders. I was part of creating two companies, Fortify Software, Alien Vault, and now I'm helping another generation do the same.

I love it. And so tell us a little bit about how the industry as a whole is changing. Right, so cybersecurity. Right. So we'll get into the VC side. But you know, I'm also curious because, you know, artificial intelligence has been used in the cybersecurity space for decades, but how specifically have the more recent advancements in generative AI and large language models, how has that changed even what's happening at, you know, these cybersecurity companies?

Yeah. So big picture, we can put it into a couple buckets. I think I've got to start with, and I heard one of your previous podcasts talk about this. The, the, the people who break things and steal things and spy on us, they use these technologies too. Right. So, so the, the malware is getting more virulent. The attacks, a great example. A lot of sophisticated attacks are based on fraud. And to defraud you, it's basically a con game. Right. So the more you trust me, the more I can relate to you, the more, the more you might fall victim to that. An attacker who had only a conversational understanding of English and maybe no understanding about your field, it's not going to be able to go very far in terms of defrauding you. All of a sudden, that person speaks perfectly and maybe you're a scientist working on the next big breakthrough for your company, and this person communicating with you clearly understands that science. They don't need to. They need to have access to a language model that does. And now they can communicate with you like a trusted source. Deep fakes. We should talk about that a little bit. So the adversary uses technology just like we do and they've got a big boost just like we do with AI. And then for the cyber security industry, it's, it's really interesting. You know, the first one, you might think about it, but some of the most important systems being built today are based on AI and a lot of our security controls weren't. And so there's new security controls that need to be put in place. As little as 2%, a poisoning of the data that's training your model can, can completely destroy it. So you've got to make sure your data didn't get poisoned. These things give answers that are really critical. You know, forecast the next six quarters of, of financial projections. You CFO might love to see that, but you might not want the intern looking at that.

Right.

And so access control is a really big thing. The other big bucket is, you know, probably, probably the largest one. If there was something we were doing before, chances are we can bring AI to it and make it more efficient. In some cases not. But the part, I guarantee you my job is to see new startups. Whatever you're doing, whatever your business is, there's a young innovative startup that's going to try to disrupt you with an AI approach. So you might want to beat them to that.

Yeah, yeah, that's, that's a great point. And you know, even more recently, Roger, how, how has that changed things, you know, even in, in your space, working with these cybersecurity companies? Because, you know, my thought is you probably have to have, you know, decade, decade plus of experience to, you know, be a major player. Has AI changed that? Has, has it been able to give these, these young startups, you know, an edge, maybe even if they're just not operating the way that everyone else has been doing for, for decade or decades. Does AI give companies an advantage, you know, startups an advantage to compete in a different way?

Yeah, great question, Jordan. You're spot on. So every so often the technology industry has these, you know, think of them as a great leap forward kind of Cambrian moment. The Internet was that way, the cloud was that way. AI is clearly one of those. And, and there's two really important components. I've, I've got new tools to, to bring better solutions and a might, but, but frankly, probably the bigger one. The customers want to talk to people that are bringing these new solutions. So, so you could be a, a big vendor and you got thousands and thousands of customers and they love you. And off you might have been on a 10 year run where they're not going to talk to anybody else because they're happy with you, but now they're watching your podcast, they're reading, they're seeing that this is a huge impact. They don't want to be left behind. So that little company that never had a chance of getting that audience with your customer has it right now and they're taking advantage of it. And our industry is a great early indicator. You and I talked before the show. It, it's probably 80%, maybe as high as 90 of the companies that are inbound that have some AI element to it. And so, you know, to my peers in the venture capital industry, you're gonna have to, you're gonna have to understand AI. Not the deep technical stuff but, but, but the business implications and how to vet whether something you may win or not.

Yeah, and how, specifically how have large language models changed that vetting process? Right, because I can imagine not just for, you know, like you all at ballistic, but VCs across the industry. It's hard, right? It's, it's, it's hard because AI has allowed so many new startups to pop up so quickly and maybe even these new emerging technologies that so many people don't even fully understand. So how can large language models actually be used to help in that process to vet and better understand these companies?

You know, I'm going to start with, you know, Dr. Heal thyself. I am sure there's some brand new venture capital firm out there that's going to base all their decisions and strategies on, you know, what the Oracle tells them to do. I'm not that concerned about that because I am one of those technical experts and kind of know what it can do. Well, or not. Now what, what can it do? Un, Unquestioned. Our job requires an insane amount. It's probably similar to your job. We've got to learn a ton of new stuff very quickly and we've got to be able to communicate things very concisely and completely. If you're in that job and you're not using chat GPTA or Bard or one of the other LLMs to help you. You know, you could take a massive. The SEC has new directives on companies reporting security breaches. You could read the entire thing or you can push it in the LLM and have it summarize it for you and the ability to be able to move four times faster is really important. Likewise, and this is, I'm going to make a plea to everyone out there. You can use the LLMs to make your communications more verbose. They'll pass the thump test, right? You can also use them to make them more concise. So I rough things out with LLMs, I tear them apart, I work at them. The very last thing I do is I send it back and say, can you make this more concise and easy to read? And they do that wonderfully. If we all do that, we're all going to save ourselves a little time.

If you're listening on the podcast, I am just smiling so hard because, yes, it seems like everyone's doing the opposite, right? Everyone's, you know, using large language models to, hey, make these 10 bullet points into, you know, 10 paragraphs. But yes, can we do more of what Roger is saying and using LLMs to just make your writing more concise. Like, we're, we're all tired of reading, you know, super long, you know, pieces of text that don't really mean anything. So, like, Roger, I'm even curious specifically, how are you using personally, you know, but, but in your role, how are you even using large language models, right? Because I'm sure there's so many companies throwing stuff at you saying, hey, invest millions in this. Look at this shiny new thing that we got. You know, we're fighting cyber, cyber criminals this way and that way. How are you even using large language models in your role? And what has for you personally been the biggest gain or the biggest way that you've won back your own time?

So I'm, I'm going to bookend this. One of the very first things I did with the large language model is, is entertained myself and got an understanding of how they work. I, I, I was a CTO for much of my career management person, interviewed a lot of people. And, and when I, when I got to the more senior roles, I'm interviewing people the team wants to hire, right? So, so, so I had to be very concise. I had this one interview question, question, and if you interviewed with me, you probably got this question. I put a blank piece of paper down and a pen and I would say to you, write a program for me. These are for technical. Obviously didn't do this with the accountants. That could have been. Now with ChatGPT, I couldn't. They do fine. But I say write a program. Here's your spec. It should be something that's interesting to you. Interesting to me and take you about 10 to 15 minutes and I'd come back and someday you could do a podcast on the range of answers there first thing I I really ever asked chat GPT was that it it wrote a program that generated math puzzles and it did a lot faster in 10 minutes that would have been about the 80th percentile of answers and of course it I could type it in compiled so that that's at the beginning and the most recent thing. I also just have to tell you because I think this is the way it's affecting people's lives. I've got a son he's a third grader give my background math is real easy. I've always loved history. And then there's English grammar and it's hard and so you know I I went upstairs at 8 o' clock is 8 o' clock an adverb changes when you now you're a professional you know that it's a adverbial noun and and I I've chat GPT I could give my son English grammar feedback like the same level I can math now what I'm teaching him is the questions right Remember the most important thing it life was never about the answers. It's about the questions asking the right questions and prompt engineering or what have you. Okay I'll end with what I do at work because I think those personal uses more interesting frankly. The.

The.

Anyone in venture capital knows the bane of venture capital is we we we have to write investment memos and it and it's not that we don't like doing it. They're they're arduous tasks and they're very thoughtful pieces of documents and and and should that company be a multi billion dollar outcome someday everybody's going to read it and you're going to be judged on infinite failure. You're going to read it and judge it. So a lot of care goes into these it it you can do them in 20% of the time it used to take to get them done that time crush down the quality of the output. Remember the last thing to ask is make it more concise. You've got 80% of that time left over to go meet that new company or go help somebody in the portfolio and it's it's that simple. It's just the stuff that we need to learn the stuff that we need to produce is better more concise faster.

Are you still running in circles trying to figure out how to actually grow your business with AI Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, or you can partner with us too, just like some of the biggest companies in the world do. Go to your everyday AI.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen AI.

Yeah, and you know, Roger, I'm also wondering about other use cases and I'm sure, you know, some venture capitalists are starting to do this, but it seems that VCs struggle to find the time to adequately address everyone, you know, or every company that's, that's looking to raise money. Mainly because there's so many companies out there looking to raise and they're, you know, sending it to everyone. Right. They're also using AI to, you know, craft more personalized pitches to, you know, 100 VC companies versus five. What, what do you think is going to happen from other professionals in your space and how they can maybe leverage AI to make, you know, kind of AI versions of themselves? Right? Like, are you going to see, you know, VC companies creating a, you know, ballistic ventures GPT and they compile all their information in there and then in the future, you know, come, you know, people looking to raise funds just talk to the GPT and they see if they, they make it through. Right. Like, is that something that might happen in the future with AI and with how easy now it is to, you know, upload all of your knowledge into a large language model and for it to work kind of accurately? Is that, is that something that might happen in the future?

You know, I join, I don't want to say yes, but I'm, I'm certain it's already being done. The fact is ours is a, you know, it's a service industry, right? So, so we, we have customers and they're primarily venture capitalists, I'm sorry, entrepreneurs, and think about anybody you interact with when they take the time to personally Connect with you and meet with you versus somebody that makes you go through anything that's artificial.

It's better service.

Right. We're in service of our portfolio companies and so generative AI can definitely help the, the learning, the, the areas of productivity. But I think ultimately in service, it's, it's, it's about the human connection. The, the other thing without, without diving into the deep technical details. LLMs are amazing, but they're basically retrieving what, what they've been taught. Right. So there's, there's this corpus of information and they're just, you know, what's the next token? What's the next token? And, and so, you know, there's a famous old philosophy book, came across it in college. It is a paper, a conversation with Einstein's brain. And it, it, if you can get your hands on it, go read it. It'll blow your mind. Because basically if you took Einstein's brain and modeled it so that all these synaptical responses from stimuli produce the same output, could, could you speak with Einstein in a way? That's what the LLMs are doing. And you think about the greatest investors of all time. Could you take everything they wrote and feed them into an LLM? Absolutely. Could you take every novel concept and idea that they had going forward and put it in there? I'm burdened with too many technical, too much technical knowledge to think that's possible. Yeah.

So, you know, another thing I'm curious about is what, what next? Right, so like, what are you all working on next, specifically when it comes to how you can better use large language models? Because it seems like once you implement something in your business and, you know, everyone's saying, you know, 2023 was the year of exploration, 2024 is the year of implementation. So after you start to win back some of your time there at Ballistic, what, what's the next iteration for, for you all and, or maybe for the large language models next?

Yeah, I, I'll hit a few. And this by no means is exhaustive. A lot of security can be approved not, not after the fact by protecting things, but by, by building things better. And there's a, a movement in security, a strategy. We call it shift left. So rather than take the broken thing and, and, and put a firewall and other controls around and try to protect it, build it better, build it stronger. And so we, we, our firm focuses in that area. We've got a lot of great startup companies in there and, and AI can play an enormous role. I, I promise you, everybody who programs a computer today is experiencing enormous productivity gains. If you're a security person, you got to be very careful with the intellectual property rights, you know, the, the AI code it generates. Do you own it, what have you. But, but that aside, there's big stuff there. One of the other areas that's really impacted by AI is, is this notion that we call, you know, protecting a free and open democracy. You know, we live in a, a wonderful system where, where everybody has the right to express themselves and share their ideas and, and we all come together and wrestle with each other on that. And that's the way our system works. There's not a lot of controls on it. With generative AI, there now is the ability for somebody with really ill intent to create information that's deceptive and fools you. And I don't want to step into politics, but there's anything my adversary says is fake and so bad. But no, I'm talking about people that aren't in that game and just want to harm the system. And so this so called notion of deep fakes is going to have an extraordinary impact. And I don't know if everybody saw the news cycle yesterday, but the, the President of the United States was doing robocalls in New Hampshire. No, he wasn't. That was a, an amazingly convincing sound of his voice. Ten years ago, the expertise to pull that off was only found in Hollywood movie studios. Now somebody sitting at home, two minutes of somebody's voice, our voices on this podcast could be, could train a model. So then, now someone could sit in a keyboard and have them say whatever you want. And this is going to have profound impacts. And so we're working with a couple of companies creating a couple of companies that can help, help sort that.

Yeah, it's, gosh, I've talked about that. Yeah. And we talked about the, the, the, the Biden AI robocall deep fake on the show yesterday. Yeah, it's going to be a huge problem. You know, I've been saying since, you know, the first day of the show that this is going to, you know, the 2024 election here in the U.S. it's, it's going to be extremely chaotic. So getting back to though, you know, from the VC angle because, you know, Roger, you, you walked us through so many different ways that, you know, venture capital firms can leverage large language models and generative AI for productivity gains. But maybe at what expense or what are maybe some potential downsides of large language models that, that VCs or even startup founders pitching to VCS should be aware of.

Great. Let me give everybody a real simple framework. And this is important if you're a business person trying to evaluate a project inside your company or maybe an investment. AI is their programs, right? And they're not programmed with, by humans. Humans work on them. They're programmed by data. So if you want to know about an AI, go look at the data. No data, no AI. Bad data. Bad AI, marginal data. Marginal AI, really great proprietary, insightful data, very powerful AI. And you know, thinking about ChatGPT, I think I would also say unbelievably large amounts of data. Very interesting AI. So you kind of focus on those last two and whether you're an entrepreneur trying to build a new product, you've got to ask yourself, you know, what, what data am I going to have that somebody else is not going to have an edge on me? Now a lot of these startups, the first innovation is building the infrastructure that can generate the data. The company we're working on that deals with deep fakes has enormous sophisticated infrastructure to create deep fakes. Some other time, I'll play you all sorts of voices or images, people. And we don't use them criminally, but we use them to train models, cycles. Right, the, yeah, so, so follow the data and, and, and you'll know if, if you've really got something and, and you know, for, for, for venture capital firms, I, I think there's two things you got to be wary of. You got to be very careful that the information given to you by portfolio companies doesn't get shared. Confidentiality and trust is so important in our business. So if you're tempted to take something given to you by someone else and put it in a third party system, I'd be very careful about doing that. And if you have the wherewithal to build your own custom internal ones, that's a different story.

I think that's great advice there. So we've covered a lot here. We've gone into deepfakes and we've talked a little bit about cybersecurity even and we talked about, you know, the pros and the cons of using large language models and how VC firms can, can use those to really save time. But maybe, you know, as we wrap here, Roger, what's, what's one piece of advice that you can give to maybe other, you know, whether it's VC firms or startup founders pitching VC firms, what's the one best piece of advice that you can give them related to using generative AI or using large language models?

I'm Going to take a twist on this and I'm going to give a piece of advice that every single person listening to this can benefit even better. The adversarial use of gen AI as we mentioned, allows the ability to create these deceptions of reality. And we saw it with the President yesterday. It, it, it will eventually happen to you. There will be a family member, loved one or dear friend that calls you and you know, has a situation where you need to move some money tonight, when you go home with your family members, create a family password. And, and, and, and if you're gonna do that, create two of them. The first one is it's really me. The second one is it's me but I'm under duress. And if you get attacked with generative AI, that simple ancient little technique will protect you to make sure it's, it doesn't need to be a complicated password, something you're all going to remember. And, and if I swap back to the, to my friends in venture capital and, and, and startup companies, you know, don't forget, not everything has to be solved with, with AI. One of the most important things I think we're gonna, we're gonna experience over the next few years is, you know, software and technology businesses themselves are, are very complex and there's a lot of art to making them happen. When we go into AI, it's very complex. There's a lot of art to making it happen. You need to bring these two fields together to produce very successful companies. Don't, don't underestimate, you know, how much work there needs to be to, to, to integrate those.

Well, such good advice. Wow. Like even someone that covers AI every day, I took something from there. I didn't even think about the second, you know the second safe word, right?

Yeah.

Like one is it's me and one is I'm into rest. This is so, such good information. Thank you so much, Roger for, for joining the Everyday AI show. We really appreciate it.

Thank you so much Jordan and, and all the best with the show. You're doing a great job. Thank you.

Thank you. And hey, hey. As a reminder, we covered a lot. We're going to be breaking it all down. Today's conversation with Roger Thorton from Ballistic Ventures. So if you missed something or if you just wanted to double down, make sure to read our recap of today's interview in the newsletter. Go to your everyday AI dot com. So thank you for joining us and we hope to see you back tomorrow and every day for more Everyday AI. Thanks. Y' all.

Bye, everyone.

And that's a wrap for today's edition of Everyday. AI, thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.