Battery Ventures’ Danel Dayan: Is Generative AI Truly Reshaping Business?

foreign speaks we're talking about Venture investing in the data and artificial intelligence sectors and what trends you're driving those emerging sectors to discuss that I'm joined by someone who's right in the thick of the action with me is Danelle Dan principal at battery Ventures Danelle very good to have with us today James thanks for having me excited to be here you know just to clarify you are a venture investor so so startups come to view and they want to make their dreams come true you may or may not choose to invest in them you must get a lot of pitches is that an accurate description of your of your regular job yeah I mean and it's most simplistic form yes uh I like to think of it as building relationships with Founders and entrepreneurs and companies um and you know if there's an opportunity to partner up with them through their journey I think we're very fortunate to be part of that but in most cases 90 of the time we're we're building relationships here are you and you're typically giving advice to those Founders I assume as you go or is it more like here's the here's the support you know do with as you will no I think advice is a big part of it um we we think of we think of sort of like the investment life cycle across you know building the relationships partnering with them and then joining their boards and being part of that uh that that small group of people that they could find in um for things that are very strategic to the business as well as things that are very personal to the teams that are uh that are around the table yep all right so I know you focus on early stage investments in the data and AI sector obviously there's a lot going on there uh you could probably take about four hours to answer this but in a in a more or less nutshell what trends do you see driving the markets as much to adopt AI yeah happy to jump into it um so maybe just for a little bit of context battery itself battery Ventures itself is a stage agnostic Venture growth fund you know we've been around for 40 years we've been investing in Enterprise software and infrastructure through that period that includes data developer tools machine learning um as well as security uh and so you know as we think about where we are in the cycle that touches data and Nai um I think we're we're very early in the adoption of AI becoming mainstream but we've been investing in this space from the very beginning um with the early data Revolution you know we were investors in data bricks we were investors in data IQ and predicating that thesis was that data was going to be a big component of Enterprises moving forward and I think the way that we're seeing it today is data's manifested itself this year around generative AI how do we take data that a lot of Enterprises have been able to consume store analyze over the last 10 years and actually start to expose that to to their customers um two users and then start to actually create new artifacts from it and I think that's the shift in where we are relative to where we have been but I think this is something that a lot of funds are ourselves included have been investing over the period essentially you know building the foundation um for to be able to support these types of use cases um now that said what I think is really interesting is the building blocks for what these applicate like AI type applications ultimately look like are being built right now um and that means you know everything from the database architecture to how things are cached to ensuring how prompts um and requests into models are going to be consistent secure govern all of that is sort of like being laid right now in terms of the foundation for these AI type applications and so you know when you ask around what are some of the trends I think the biggest one right now is what is truly the architectural decision that teams are making around adopting AI within their within their company um and we're excited about what are some of those foundational layers whether that be Vector databases whether that be you know model Monitoring Solutions that are ultimately going to enable Enterprises to uh to get models into production to get AI applications into production you know essentially you would talk about the you know data in particular in that I think uh you know before the whole AI craze came on board with Chachi BT's debut last year the big topic was data analytics and it feels like we've kind of forgotten about data analytics and it's the irony of the artificial intelligence obviously goes nowhere without mass Storehouse of data data repositories and I would say as you say it there's really an infrastructure going on there AI is not just a fairy dust a pixie dust that we sprinkle on things it needs a big heavy infrastructure yeah exactly um you know I think the last the last 10 years of investing in in data infrastructure was very much analytical driven um but again like the thesis was very much around the there was this exponential growth of data happening um across you know users across devices across Enterprises and it was ultimately how do we actually take this these large sets of data and make them actionable for uh for consumers you know historically that was very much analytical driven that was very much probabilistic driven right and you know these were just very sophisticated probabilistic models that spat out you know your certain distributions for what decisions should be where we're getting now which I think is very interesting is we're able to harness that data expose it in a way that we typically that we historically haven't been able to do and what I mean by that is you know analytical use cases are very much internal facing you're looking at bi dashboards for your Finance team you're looking at bi dashboards for your sales team and as we layered on these Transformer models generative AI models what that allows us to do is actually take that data and generate New pieces of content new artifacts whether that be you know voice images text content and those things can be consumed by external parties whether that's customers users what have you um and so I think we're moving in the way we sort of see it as we're transitioning from data being a very internal um resource that is consumed internally to something that can now be exposed uh and the types of applications that we're just starting to see are what's really interesting and I think we're just scratching the surface on that you know I'm not sure I totally understand the idea of the data being exposed could you elaborate on that a little bit yeah so it initially the way you know the way I sort of is these models allow you to you know take proprietary data uh of Enterprises um of your of your company um you know and then generate you know some artifact or derivative content from that um right and you know historically those would have to be done manually but now the the internal data uh that's sort of the foundation for those models right that's the source of Truth for that generated content um and I think so what it's it's sort of a a much cleaner pipe from data to actual uh to app actual application in a way that I don't think we've seen in the past um what I mean by that is you know the the the generated content is actually a derivative of the internal data that's now being consumed by someone else and in most cases the applications that we're seeing is that those those are external facing um so take for an example so uh Adobe right you know you have you're you're a customer of adobe you have uh a thousand images that you're working with um historically you would need designers in order to create or generate new images or you need photographers to actually upload those into your library and then you can do some for your post processing on that now you can actually rely on that Corpus of data that you have generate you know New pieces of of content with that and those can be added to your library that then get exposed to your customers in some way um and I so it's it's it's a more elegant way I think of uh building these types of applications that we haven't been able to see in the past gotcha okay so before we leave the foundational piece I want to touch on the idea of large language models which you've talked about models it seems like a lot of companies are grappling with that core idea that do we need to build our own large language model can we you know access one and exchange in a Marketplace somewhere or you know do we have enough for it you know our competitors have a larger large language model I mean what is your sense of like the evolution of the large language model is a key piece in this old picture yeah so you know one framework that we use to evaluate this is whether you require what whether Enterprises are going to require to bring data to the model or model to the data um our view is that models are going to come closer to the to the data meaning Enterprises are actually going to start to to consume more models themselves and build more models themselves and I think we're seeing that with the explosion of all the open source models that have actually come over the last SEO quarter or so um it's it's easier to ingest models to build to build them within your own uh within your own Enterprise and guardrails uh an environment and then use them for a specific use case that you actually need versus pushing everything to a large context window model like open AI or entropic for example um now I think the answer probably lies somewhere in in the middle there will be use cases where it makes sense to actually use some of these you know foundational models like an open AI for example and that could be something as you know as simple as handling customer support or text generation for you know a specific use case that might not be necessarily ingesting proprietary data but for things that are very specific to an Enterprise you know gong for example in our portfolio that uses machine learning and and and and uh large language models to help generate uh feedback on sales calls that's something that's core IP to them um and they've in-house a lot of the expertise around that including the model component piece ah and so we think there's going to be sort of a variance and gradient for how how adoption happens but I have a view that you know models are coming to the data and that means you know being able to build very specific models that um that hit a specific use case so I think we'll actually see a proliferation of models there will obviously be the large language the the foundational ones that the open AIS and then Tropics have and they'll be use cases for that but I think it'll also be very easy to consume you know models you know within your own Enterprise and on top of your own proprietary data interesting all right well then the big question that I think many people are wondering is uh it is generative AI truly radically reshaping business as it appears to be or is it in fact merely another forward moving technology is this just another uh you know cell phone slash internet introduction or is this is this really reshaping everything history of application and infrastructure Evolution you know sometimes you need these types of applications to push technology forward uh and to push Innovation at the infrastructure level forward and I think that's what we're seeing today AI gender bi applications are stressing the existing infrastructure around CPU capacity memory processing in a way that we haven't ski in you know over the last 10 years and that's requiring us to rethink what is the data the structure look like you know what are the chips that are required to support these types of applications how do we increase the efficiency the throughput the ability to consume these these levels of data within our Enterprise in a way that we haven't been able to do before um and I think that's you know pushing the hardware forward and in a sense you know pushing technology forward in many ways so that's one sort of aspect that I think is very interesting that we're seeing with this uh with the rise of generative AI applications you know at the same time the like I said we're it's the first time in 10 years that I think we're moving from analytical uh data workloads and applications to one that are more you know generative in nature and external facing in nature right and so that's also impacting you know the application side of things you're creating new applications that we're still just scratching the surface on um and so you know in long-winded way of saying that I think both things are true we're seeing or reshaping um a re-platforming a rebuilding of the infrastructure layer and the core technology layer to support these types of of workloads and at the same time that's enabling new types of applications to be built in parallel interesting okay I see this sort of a two sides to that answer um interesting and this may be slightly off topic but let's see if we can go there in that he's like that I'm surprised at the amount of fear that I've that has bubbled up about artificial intelligence say before chat GPT there was a sense of the sunny optimism about it seems like the last several months there's there's a brooding sense like hmm this is going to ruin us somehow I'm not sure that exactly impacts what you do as an investor but Mr do you have a personal check on that question yeah I like I I I think um like I think there's there's pros and cons to all sorts of of of these Technologies um the Reliance of generative AI as a medium for for writing for Content creation you know I think could have impact to society whether that's on the educational level um on the critical thinking level those sort of things um I don't I haven't seen the applications move to that point where uh where where we have to be fully reliant on them or we are fully reliant on them I do think that place now where um you know there's still a lot of human input that is required to actually have the desired output that that I think all of us are typically looking for when um when when doing sort of these these use cases um so yeah I think maybe there's there's a world where these models get as sophisticated as as people believe that they will be um but I think part of what we're we're building now and I think part of what we're seeing be built is you know do we have the right guard rails in place so that we believe that the the content that's being generated from these large language models are safe are secure um are auditable um you know our our uh have have have the right sources associated with it um and I think people are are building with with that in mind and I think that's a good place to be the other aspect I'll say is when you think about where the adoption has come from very quickly um it's actually come from incumbents those with a lot at stake and play right it's I think the way that they're building is a lot more accessible as I'm sorry when you were saying Cowboy too who do you mean in this case well the microsofts of the world the Googles the Amazons adobe's they've been very quick at actually releasing generative AI type applications through their existing platform and through their distribution channels and it's interesting to see because one I think that requires a level of robustness and sophistication and security um that you typically wouldn't see if it started in a more Grassroots effort or it started only specifically you know targeting startups or smaller companies that maybe didn't have the resources to think about those um those implications from day one and I think the fact that we're seeing some of these larger companies adopt it lean into it means I think it will be built with more reliability in mind more safety in mind okay well then certainly there are there are some winners and losers as this new technology gains ground there always is um who in your view gains and who loses ground is as generative AI gains adoption yeah so I I think you know touching on the last the observation of this cycle is just how fast some of these incumbents or larger companies have actually moved into the space um that being the microsofts of the world adobe's Notions figures for example uh and I think that's a strong validation that these types of applications are real that the demand for them is is real um and the benefits you know whether that be economically uh or you know from the types of of productivity games you know users can get are also real um and I I you know you rarely see that Dynamic happen with New Market Cycles but I think that is a strong signal of you know where the potential winners can be or where the potential value is going to accrue um you know I don't think that I think that the the the loss ratio of um adopting AI is actually quite low I think there's a lot of benefit said another way right I think if you introduce AI into your application um the the actual floor of uh or lost potential for example is a lot lower than I think people perceive meaning if it if the AI application produces a bad result on a chat you know a user will just stop using the chat right and you can improve on that experience until you get it right so I actually think the benefits of of this are a lot higher and a lot more uncapped than I think people are actually realizing um which is exciting to see that these these big companies are able to move this quickly and exciting to see how much uh excitement there is from the early stage ecosystem as well okay uh well let's look at the future I think that is really really the the big question the future of generative Ai and data in the Enterprise I think one of the problems with forecasting is that there is a certain exponential nature with AI it can it can learn unlike other Technologies you can learn by itself so it does create an element of unpredictability still I'll ask you what what do you see in the future in terms of you know guiding your own Investments yeah the big thing we're looking for is honestly Enterprise adoption um and and you know it's kind of interesting to to say that after the conversation just we had but the adoption of where we are today is still very much sort of development and and test in the development and testing phase um and I think we're just scratching the surface on what are the requirements needed for Enterprises large Enterprises you know think the G's of the world um Fox telecoms what have you uh to actually introduce this type of technology and expose it as part of you know their Suite of offerings um and the things that we're looking for from an investment perspective is are you actually able to enable Enterprises to on-ramp more models into production more AI applications into production in a meaningful way and we're just still so very early in in understanding the requirements around that um and so that's really what I'm I'm looking for uh I think the tip of the iceberg will be once Enterprises really start to adopt this uh in a meaningful way we see a little bit of that with our investment in a rise which which monitors models in production and a big qualifier for them is you know how's how many models are you actually running um and does that sort of you know is there enough breadth there that you require a monitoring solution for that and so we think that's going to be a leading indicator for you know how this category and space uh ultimately uh evolves you know for we're not here for talking Enterprise it's one of the issues is that they know it's going to be expensive and they sort of have no choice they need to invest in it but it's it's it is expensive it's not just like adding another Microsoft module to your Cloud you know platform you already have AWS it's a significant amount of money so they're wondering hmm we need to get on board it's a lot of money is it going to pay itself back I don't wouldn't ask you to answer that question today but it seems like it is a question that the companies are thinking about these days yeah absolutely the way I think we're seeing pricing uh uh arpu average revenue per user uplift if you will um you know for example notion May charge ten dollars per user for their premium offering and then charge you know five dollars per user to for their AI assistant um it's a 50 rpu on top of their existing pricing model uh so you know seeing some of that validation come through in terms of pricing and business models is is an indication of you know the value that can be captured um Microsoft has you know obviously pioneered this as well by introducing some of the AI features through their office 360 products um but that's sort of like you know one aspect to to think about it um the other the other way I would think about it is the the investment that's already been made across the data infrastructure uh and data ecosystem so databricks for example is really enabling companies to expose and operationalize machine learning and artificial intelligence in a meaningful way and in a production grade way um and I think a lot of the investment has already been made by companies and Enterprises in building that foundation around data bricks and Snowflake and you know your your cloud data warehouse um and now they're allowing you to add a catalog on top of that to manage your models to manage the data assets that are going to be feeding into those models manage the databases like a vector database that actually help retrieve specific information from your proprietary data through your model so all of these I think Investments have been actually made over the last couple of years the question is actually how much is it going to cost to operationalize all of this in meaningful way and I think that's what we're still getting to grips of there is a component of training and inference and I think that cost curve is coming down very quickly and that comes back to my prior point of you know we are rethinking what is the hardware what is the compute infrastructure layer required to support these things and I think it's just a matter of time before we get to to a place where it will be fairly cost effective all right well that that is optimition I can be sure uh you know managers will be glad to hear that Daniel I think you said it it's a lot of good stuff I learned a ton uh thank you so much for sharing your Insight and please come back and talk with us again sometime Thanks James is gonna be on foreign

This transcript was generated automatically from the video's captions and may contain errors.