Snowflake’s Torsten Grabs on AI and Democratizing Data

foreign speaks we're talking about democratizing data which is a core Trend in artificial intelligence and data analytics these days and we'll look at how this affects application development in particular to discuss that I'm joined by Torsten grabs senior director of product management at snowflake tours and happy Friday to you and thanks for joining us today absolutely uh Thanks James and happy Friday to you as well good question it's almost so big it can answer it I want to ask you about what what trends are driving data and AI these days and I think the the problem with that question is that there's about 18 different Trends and they're all going every which way so like the answer to that question alone could be like four or four hours but I'd say if there's just one or two trends that you see being as close to the market as you are one or two Trends driving data analytics and AI these days what are the most compelling say one or or two as you see it yeah so maybe the the the first one that comes to mind is about seeing more of the data science work uh nml work actually getting productionized and rolled out to the end user so that really the the business eventually benefits from all the work that is being done by Machine learning data science teams and there is still a lot of um a lot of challenges for data science practitioners to actually see their work through all the way to the end um to to getting product productized in in an organization so that is that is a is a big challenge uh still and we can talk about why that is and what needs to happen in order to to overcome those challenges I think the other big Trend that I see and it's somewhat related to that is also the progress that is currently being made in the industry with uh with generative Ai and large language models and uh I I I I call them as related because they also have a great potential to make AI more accessible for non-technical or less technical users uh in in an organization so those those two themes are uh maybe broadly speaking the ones that that I see and that also keep me up at night uh and um and are are challenging me to uh to also help our our users and customers yeah I mean generative AI I mean first of all you get 5000 bonus points for saying the phrase because I think every conversation these days has got to use the phrase generative AI if if not like what's the point because it's just so hot but I also really resonate with what you're saying in terms of keeping you up at night I don't know if I mean I think the industry is thrilled even the consumer world is thrilled but everyone's kind of thrilled and a little scared at the same time uh like wow what's gonna happen what jobs are gonna be affected like there's all that going on again that's a huge question but can you can you give a sense of like generative AI threader promise um I think it's a hugely promising space um I would also say we are still very early on in the process of embracing it and understanding all of its implications so I would expect that there are a lot a lot of things are still still uh playing out um as as we're making progress on the the Innovations there um one thing uh that comes to mind is just a huge potential that I see to make computer use more broadly uh much easier for Less technical users and that may also help with the aspect that we just touched on for data scientists and machine a learning practitioners um it may make the the consumption of machine learning technology much much much easier um because you have a more interactive way to uh to get guided through day-to-day user experiences that otherwise would have required a much more technical team to help you you out with right and you could ask those those questions a large language model based agent and that agent might eventually be able to completely independently guide you through a particular business problem that that you that you have to to solve right and that's that's I think where a lot of the potential sits is to empower more users to be really productive with data with computer technology now having said this I think there is still a number of open questions that we have to solve in the industry and a lot of them tie back to uh some of the social implications of what's going to happen but also technical aspects around security and compliance right now if you are in an organization that has sensitive data um I'm I would caution you to um to be very very careful which pieces of your data you actually want to send into a large language model that's hosted by someone else as a service in the in the cloud right you you may have privacy obligations that are not necessarily Satisfied by uh the llm agent or by uh the generative AI system that that you're interacting with and I think that's something that the the industry just has to Grapple with and solve which will happen over time but right now we are in the early phases and uh I think people should should try out and see the benefits but be careful and cautious about like what pieces of data are they sending where right okay so you're saying that that a Company employee could possibly you know provide a sensitive piece of data to a generator AI platform who knows what exactly is going to happen with that data yeah a clear caution no doubt yes um all right the idea of democratizing data which you've sort of talked about in in a larger sense I mean we we know it means you know not just the c-suite not just the data scientists have access to the data but actually all the staffers the sales folks you know all the managers really the democratizing of data is is one of I think a big Mega Trend running through artificial through AI these days um democratizing data can you talk about that a little bit what that means to you and and how you see that taking shape yeah so one big aspect that I obviously see there is is the data Cloud which makes it so much easier for everyone every organization to get access to more data and also more relevant data for their specific business problems um and that means your own first party data that's easier to to work with but also third-party data that you can consume from a Marketplace or you know from business partners that you want to collaborate over right so that just from a technology perspective is a a much easier problem right now where in the past you had to deal with like sending files over FTP from one organization uh to another that is no longer necessary right so that opens up uh just a more collaborative way to interact across cross-company value chains with shared data Assets Now having said this obviously they're again I already alluded that in the context of generative AI there are certain privacy obligations that organizations have and you will have to satisfy those in the data Cloud as well which is why you want to make sure that your permissions and entitlements and things like that are well defined even in the context of collaborating over data in the data cloud and there are ways to to do that in a meaningful way through for instance data clean rooms where you have very clear guarantees around uh sensitive data and what can be seen versus what cannot be seen by different organizations that come together in a data clean room um in addition to that I think the other key change that's happening is in terms of programming languages that are becoming popular python is becoming increasingly popular also for uh let's say analysts and advanced analysts and that's going to help a broader community in businesses and organizations to become really productive with uh with data there's a lot of packages in the open source Community around python that are primarily dealing with a data analytics and data processing and that now will become available for a much uh broader user population um and the interesting aspects then are also how can we help um developers and advanced analysts become more productive with python and how can we then also help those people to shape their their python programs in a way so that even a non-technical end user can leverage them in a business and uh they are a key step that we're looking at is uh is streamlit which has the ability to essentially wrap around a piece of python code that could be for example a machine learning model that a data scientist has implemented in Python and now you want to share this with with business users um obviously you cannot just simply share a Jupiter container with a bunch of python code in a notebook that wouldn't be approachable for let's say a marketing manager or a line manager in a large organization but if you wrap that in a streamlined application with clear user interfaces as visualizations that are approachable for the end user are with the ability to provide input and interact with an application in a meaningful way that creates now an opening for these data scientists or machine learning practitioners to operate in Python but then wrap their results through a streamlined application that makes it so much more approachable for for an end user in an organization so we see that as a big opportunity on that first challenge that I hinted at which is how can we make sure that more work from data science and machine learning practitioners get successfully productized in an organization and having a meaningful way for end users to uh to interact with the result of your machine learning practice through something like streamlit is really crucial to accomplish that goal make sure I understand about streamlit streamlit versus containers same idea similar idea could you explain that please oh no they're they're very different I mean the container gives you a great way to to to package a runtime environment and code in it and it's it's a very popular environment for ML practitioners to just do do their day-to-day work but still in the in the end the result of what you're doing in the container is a piece of python code and that is then what you would lift into a streamlined application in order to make it make it accessible for for an end user if that makes sense all right well let's talk about the snowflake product offering what is snowflakes snow park how does that help people yeah it helps people in in multiple ways so snow park uh first of all is infrastructure that allows us to host uh programming language runtimes natively on snowflake compute and um that infrastructure has allowed us to First launch a Java virtual machines as a uh as a co-processor running on uh snowflake compute natively and that same infrastructure is what we are also Now using to support python natively on on snowflake um compute and uh that's kind of the the server side extensibility part of snow park if you want in addition to that uh the server side capabilities are complemented by an SDK on the client side that provides developers with a familiar abstraction um mostly I'd say a data frame concept that has become very popular through pandas or or spark which allows them to in a more declarative way Express the data processing that they would like to do or over the data set that sits in Snowflake that becomes a very convenient way to write very powerful data transformation data processing applications in Python but also in in Java and Spark and we take literally the the programs that you've written uh in the client side through those apis and then render them as a distributed parallel execution plan that runs on the server-side pieces of of snow park uh other other snowflake product highlights what's going on with snowflake these days that we should know about yeah there are there are a couple of uh a couple of interesting pieces so first of all um snowpark for python has been uh NGA for for some time but um we we keep adding additional capabilities um to that and we're obviously looking at our upcoming Summit conference this summer in in June and in addition to to that uh uh something that's actually very relevant in this this context of uh productizing machine learning and data science is uh our native applications framework you can think of it as a way to package uh code and data of a snowflake application in a meaningful way so that you can put it up on the marketplace as a snowflake application that then a snowflake customer can consume and install into their snowflake account and that also is is a great way now for machine learning practitioners to take a machine learning model that they might have implemented in snow park with python that they have uh that they have created a streamlined UI for if they package all of that through a native application they then have the ability to put that on the marketplace and literally every snowflake customer if they they want it could consume that native application from the marketplace and install it into their account so that has the potential now kind of thinking uh specifically about machine learning and data science to create a Marketplace for machine learning models where those could become native applications in the marketplace that can be consumed and installed by a much much larger larger audience and the marketplace infrastructure then also gives the providers of these machine learning models the ability to track usage across snowflake customers and then also monetize that essentially bill for the usage of the machine learning models that they see through other snowflake customers on the marketplace so that is that is one uh key key area we already talked about streamlined can I interrupt you first that's interesting is the idea of the marketplace for for machine learning models that does exist or it's being built completely explained that a little bit if you would so I mean the marketplace exists um I'm primarily the focus uh right now is through the data Cloud to provide a Marketplace uh for for data where people collaborate over data okay but the next step for us is to take that Beyond data and apply the same principles to to code to applications and that really depends on the ability to package and publish code in a meaningful way into the marketplace and that is exactly what we're building this native application framework for um it is currently in private preview so it's being used by a smaller number of customers and we're working towards releasing that uh later this this calendar year and then it will be more broadly accessible that's interesting because I think about um pre-built models as being one of the hottest things is going on in machine learning at AI because companies don't want to build themselves they can't build themselves so they really rely heavily on pre-pre-built prefabricated models so be curious to see what happens when that when that Marketplace actually opens up to the general Tech public that's going to be fascinating yeah um well you know I think the big question is that the the future of artificial intelligence and and data analytics on snowflake or or in general I mean what do you see what what's coming on on Snowflake and what what do you see the larger in the industry yeah so I think the the the two trends that we're seeing kind of coming back to the the challenges that I mentioned um uh in the opening the one is really about making making data science nml practitioners uh more effective in an organization seeing more machine learning models getting successfully deployed and delivering meaningful business value in organizations that is going to be an important topic for us going forward and with some of the pieces that we touched on in in our conversation we hope to actually make progress from a snowflake front and help the industry uh on on that front and the other aspect is obviously around uh generative AI large language models and how are they going to shift the Dynamics uh between more technical teams and and less technical teams um there's there's a lot of work still uh to be done for for everyone in the industry we are working towards uh a number of these capabilities um I touched on the native application framework streamlit running natively on snowflake compute um uh interoperability over data through embracing Iceberg as a data standard for for for your data Lakes so those are all areas that we are actively working on and I think the next milestone for us here is our user conference the snowflake Summit in in June and they will have more to tell about everything that's happening on all of these fronts that I just mentioned good I'm glad you you mentioned the conference that's going to be interesting uh the the last question it's about data and and being Cloud agnostic I'm not even sure how to ask a question let me let me fumble through this so I know the snowflake is of course Cloud agnostic which is one of its claims to fame and not all platforms are Cloud agnostic or or proprietary agnostic uh and yet some of them do quite well being not necessarily you know completely open platforms I guess do you see the the future of data and AI being really completely open or or will the this proprietary model still be really strong um I think I think it's a great question so first of all I think on the on the on the data side of things the the the the open model is the one that that that will succeed eventually because otherwise organizations will just find too many hurdles in terms of getting access to a business partner's data that may reside on a different cloud in a different region and at some point this becomes prohibitive and it becomes a competitive disadvantage if you cannot collaborate over over your data so I think on that front the the right pressure the right incentives are in place to continue to open up that that environment for data collaboration now um on some aspects where I think there's a lot of innovation happening there at least for some time I'm I'm convinced that we are actually going to be seeing some proprietary advances so think about like the the large language model space where I think having uh the best large language model in the industry is seen as a competitive advantage and that may put pressures in the system to uh to keep that proprietary um until there is enough demand from customers uh that are saying no I want this uh I want this on my cloud or I want to run this actually close to where my data processing happens and that then over time will create enough demand and enough pressure on the proprietary vendors to open up their ecosystem and make their their capabilities more broadly accessible actually can you say who who you think has the the largest you know language model this time or is that is that a forbidden question um I think it's Gonna Change probably on a monthly or maybe even weekly basis okay um I think right now clearly uh the the major Cloud providers are on the Forefront of the Innovation there and there are a few smaller startups that are also highly competitive in that space so it's it's interesting to see and watch over the next couple of months to see how that that that plays out yeah a lot of activity a lot of people working a lot of hours in that particular area no doubt about it um Tristan you said it that was fascinating I learned a ton thanks so much for for joining us and I hope you come back and talk with us again sometime absolutely thank you so much James appreciate it

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