Build AI-powered games with Gemini and Flutter

[Music] hello hope you all are having a good Google Cloud next uh I'm really excited to talk to you today we're going to be talking about building AI powered games with flutter and Google Gemini um and if you're not familiar Gemini is Google's awesome llm and so we've been having a lot of fun with it um my name is Joanna may I am the engineering of one of the engineering leads at very good Ventures I'm here with my boss Jorge Koka who's head of engineering and uh we're going to be showing off some of the cool stuff that we've been up to um so we're going to end up covering a couple of things we're going to be adding AI to games we're going to talk about how you can personalize the game with AI uh we're going to be talking about how you can uh maximize your success by leveraging building apps for all different platforms uh with flutter and dart and then we're going to be talking about the flame game engine uh for flutter uh which a lot of my colleagues have been working on which is really fun to talk about um and then we're going to talk about adding Gemini to the app to your game and then we're going to discover uh The Prompt engineering full stack with dart and where to go next um when I talk about AI specifically um if you've done anything with games you know for fun or you worked in the games industry you know that like game AI is its own thing you know you've got your uh hierarchical State machines you've got your behavior trees we're not talking about that we're talking about neural network based AIS and what you can do with that by adding those into games so uh just a little bit of a distinction I just kind of want to point that out um AI is good at a number of things um specifically I kind of think of it in a couple of different categories you have um the recognition and classification we've been using this stuff for years U object recognition uh speech recognition image classification um I also think of AI as being really good at doing things that sort of we human struggle to do in terms of inductive reasoning so if you have a large set of data um as a human it's really painful to kind of comb through that and sort of figure out you know um how to like what the big picture is what the the story is or what what's happening inside all that information and so AIS are really good because they've seen so much stuff um they can just sort of infer all that information so they're great at summarization they're great at cinnamon analysis or Anam detection um and also they're getting really good at making things so um we're seeing them produce all kinds of great code Pros poetry Etc uh images music and uh even video now um if you're building you know a game you're you're already often leveraging uh AI anyway at the hardware level with AI upscaling and stuff that the graphics card manufacturers are working on uh which is really fun and then we also uh will probably start seeing and maybe somebody's already doing this uh we'll probably start seeing people augment procedural Generation by adding AI as a second pass or um finding a way to marry with the typical procedural algorithms that we've been using all along um and then finally you know if you're doing mixed reality you're going to want potentially to be able to pick up an object in your room and for that to be recognized and then as we start seeing more and more um cons as we start seeing more consumer focused uh neural network processors on hardware and phones and laptops we're going to be able to start leveraging local models to produce uh on the fly with zero latency and I believe uh Google actually has been working on a local model version of Gemini if I remember correctly um and so we want to talk specifically about using llms in games because that's what we've been playing with and uh it's really nice because as humans for the next I don't know a couple hundred thousand years we're probably going to be limited with our tin fingers so you can only add so many buttons to a game controller before it's not that useful and so we've kind of it's kind of settled out it's kind of evened out and it's just been refinements for the last few years um and so to that extent if you need inputs that aren't you know twitch based and need to be fired right away you can start using AI to create voice interfaces and you can use llms to sort of translate text that someone's speaking that you're using AI speech recognition for and you can turn that into um a sequence of structured data with the llm and sort of turn those into commands for the game on the Fly and the llms are good enough to give you those structured responses now so it's really fun um you can also use llms just as uh like the way you would talk to a person um or a DM you can have um the llm act as a sort of um like a a guiding force or personality that then um takes the information in your game uh evaluates what's happening um using its awesome like inductive skills and then uh you can pass that into your traditional game structures and filter that through all down all through the way through your game components and various services and then you can make it essentially act as a highlevel guiding force for the narrative in your game um and so that's going to lead you into like creating these adaptive narratives um especially if you're building stuff like with large open worlds you can really add a lot of personality to just everything in the game um and even create this huge narrative Arc by using an llm to sort of create a specific sequence of events that makes sense in your game story um and then lastly it's really interesting because you can use an llm to sort of summarize game activity because they're good at processing all that information if you just keep a log of everything that's happening in the game you can have the llm then provide a summary at the end of the day or what however your game works and sort of say oh yeah this is what happened but it can put a narrative Twist on it and put a spin on it because it's like a human Storyteller in a lot of ways and so that's kind of what I think a lot of this is headed and I think with local models we'll start seeing more and more stuff start taking advantage of this and so we actually built something so if you went by the very good Ventures Booth you probably saw um that we have a little game demo running and so we built a little um mystery Adventure puzzle escape room type of game and uh we built that um as a sort of like top down uh 2D 2D uh type of game where you walk around and uh if you find one of the puzzles in the escape room up pops up a little quiz uh question and so if you get the question right then you can find the next puzzle and keep going and if you get enough of the questions right you win and if you get too many wrong you lose and so it's pretty simple but the twist that we added to this is that we took we essentially took uh your information at the start of the game you know your job title and your company and we sort of used that to create a prompt for the AI for Google Gemini to customize the kind of questions that you would see in the game and so we actually do that as you know as soon as you start we ask the the AI to generate the questions and then we sort of just populate the game world and uh you just play as normal um not knowing any better and so uh I'd like to show you what that looks like here U so we've already filled out the information it kind of went by we'll show it again in a second um and so there it is you know loading and figuring talking to the AI and you know doing all that in the background and so we're just moving around the uh escape room here and we're trying to find that and so it's asked us a question and uh looks like we got it right because nothing bad happened um and so we're going to go find the next uh puzzle question and it looks like we're having a little you know we're thinking about it we're trying to figure out what the right answer for this one is and we asked for a hint which is also AI generated as well and uh we got that right so we're going to go look for the last uh puzzle question and it's over by the printer and so we got that right as well and you can see that it shows us everything that happened in the game um which is these screens here so at the start of the game you've got the form where you put in your information so we can say hi to you later um and then you know you add where you work and your interest and things like that and optionally sign up for our newsletter um and then at the end of the game we show you uh your stats which isn't shown here and then we show you the AI prompt that we generated uh or used to generate your custom experience um and so you can come check that out it's really fun and we had a lot of fun building it and so if you know anything about very good Ventures and what we like to do we've been building with dart and flutter which is you know Google's awesome crossplatform app development framework and we've been doing that for years and we absolutely love it um and so I want to talk to you a little bit about flutter and kind of you know Express I guess what it can do for you and um you know whether or not you know it you're using flutter there's a lot of very popular apps that are built with flutter um you're seeing some of them here and um it's uh it's really unique because Google had awesome people working on it and so they kind of had the foresight to sort of um I guess look back in time and say hey what worked with all these development tools and what didn't and they were able to create what in my opinion is some of the best developer tools that have ever been created it's so fun to build stuff with flutter um and it's fast it's architected like a game engine um so everything's rendered and it can compile down to Native for platforms that need that as well as do Justus in time and so it's just fast it works it's smooth it gives you great out of thee boox support for like the material widgets as well as the lookalike widgets for our other platform friends and um it's seamlessly integrated across Google stack for um for generative AI like this with Google's generative AI Dart package um you can easily run it with Firebase and Google Cloud um and deploy it and it's absolutely um just like super streamlined and uh it's just given a lot of love and care by Google and so we love using it um it's easy to learn um it has you know a lot of experts that have created a lot of great packages which we've gotten to work with and many of us you know work really hard to kind of augment that experience with secondary tooling um and it gives you uh the ability to do things that you might really struggle to do if you were building apps or games with Native tools alone and so for that reason we picked you know flutter to build this and uh it allowed us to this demo more or less in our free time for a couple of weeks tops um and so it was just really fun it was really fast to make this um and I I mean I can't I cannot sing flutters Praises enough it is truly magical so flutter is taking care of the UI for our game and uh it's taking care of you know collecting information and showing the widgets and then you know right below that we have the game screen itself which is done with flame um you know the game engine four flutter and then we're using Google Gemini to add our AI power capabilities to sort of generate your experience and customize that and then lastly we're using dart on the backend to collect your uh information and the stats and all that stuff and put that on the cloud with Google Cloud so it's been really fun we're using Dart across the whole stat because Dart is the programming language that works with flutter and so we can just share everything across all of our code and it just works and it's magic um one of the awesome things about flutter um is that when you're working on flutter it will hot reload and so you'll be writing code and the screen will just change as you edit your code and it uh you know many have tried to do this and it doesn't always work in every framework or every stack and the flutter people pulled it off um and they did an excellent job with hot reload and so that's what makes one of the things that makes working with flutter just so so dang fun um flutter has at this point flutter is can run pretty much on anything and uh you know it's gotten support from Microsoft and others uh contributing platform support so if you've got a device it can probably run flutter uh we're using it from everything from like embedded Linux iOS Android uh desktop uh you name it essentially it'll run it um and the beautiful thing about uh embracing Dart as well for your backend servers is that you can share your models with your front end which will run again on all those devices and you don't have to worry about the uh the conversation where you talk to your backend folks and say hey um is this API model that's going to translate this Json is it going to actually be correct or is it going to break everything and you don't even have to have that conversation anymore you can just share everything um so that kind of covers the the how we structure the app where we put everything and so for the game itself we leverage Flame which is flutters uh which is an open- source project created to allow you to build games of flutter and um I have some friends that work on it and so I like to talk to them about it and kind of see what they're doing and they're they're working on really cool stuff they're adding like 3D support they're doing uh absolutely awesome stuff with it um and it's growing really fast and so you can see here uh we've been building the uh Google IO flip application that we made and uh flame lets you do a lot of stuff it's just it's really you can do really pretty beautiful stuff with it the same way you can do stuff with flutter um and it's easy to use and uh you know we've also used it for the the pinball application and then we've got here a screenshot of asteroids running on an Android watch which was built with flame and so on and so on um one of the neat things about flame is that it's open source so a lot of people have contributed to it it's built on top of flutter which again runs everywhere so flame automatically sort of just gets this I can run anywhere out of the box uh experience for you which is really nice and just kind of consistent with the whole theme of flutter and it gives you what you want when you need it um it's got all the basic physics it's got your game loop it's got the event system essentially everything you need to create um successful games with flutter flame is either already has it or they're building it right now um when you're building games um if you haven't done much of this uh essentially you know you want to break your game down into like a series of components and so you might have a player component or an enemy or um in our case a little puzzle that's on the map somewhere and so everything is just sort of tokenized um is what they call that and so flame very easily let you declare oh this is a component these are the things that are unique about it um this is how you know this is where it loads this is where it processes input it's just very simple it's very straightforward and it follows along with all the conventions that you would expect from flutter and uh with games it's really simple when it there's really two things you need to know about game components and that's when um when they're added or the game starts up you're going to handle the loading for that component any graphics or Media or assets that it needs and then you're going to handle the update cycle uh where every frame that's called and you have a chance to kind of look at how much time is pass and sort of run your simulation um or pass it off to your other systems in your game that actually handle your actual game logic um and so that's kind of all there is to that and so at this point I would love to turn it over to my boss Jorge who's going to talk to you about how we actually added and integrated Google Gemini with our game to make it the experience that you saw thank you thank you Joanna um so it what's very excited when we decided to do this demo because you know we get to be here with all of you we're like what ideas do we have to share with all of you hey let's build a flatter application we love flatter but what's like like where like the conversation is going these days AI so we were like let's build a game that uses AI we were so excited but then it got the moment to build it and I'm not an AI expert I've been building user interfaces all my life and all of the sudden I realized oh Lord AI sounds really really really complex I had a moment of panic but then you know we decided to take step by step and figure out how can we complete this project and we decided to use gini the first thing about Gemini is that it has full support first party support on for Dart and flatter um this is an screenshot from puff. Dev the package manager that Dart uses like Google is already working on a first party SDK for G that it's called Google generative AI all you need is to get an API key that you're going to get from the Google AI Studio it's like okay I can get an API key I don't need yet like a PhD in math that was my biggest scare while building this project because llms algorithms all that stuff sounds really really really scary but I'm like okay first step completed and then what we realize is like huh the majority of the complexity is abstracted away from me as a developer that has never used AI I've been building user interfaces I feel very comfortable with with flatter the widget system building animations but everything that happens on the back end you know it like kind of like goes beyond my level of expertise so what we did was to build an abstract away this gini client and gini provides this generative model where you can pass the type of model that you want to use this case we're using G Pro but if you wanted to use something higher more powerful but also more expensive you can change the model that you're using or when it's available you can use G mini right for local models I was like okay I can do that we passed the API key now we can start using Gemini it turns out that that entire game mechanic that Johanna showed previously can be done with seven lines of code and out of those seven lines of code three are boiler plate right so let's see what's going on here we have an public API that it's called load game challenges that's going to create the puzzles that we see during the game and we're going to pass as a variable a prompt right based on that we're going to build this content variable that we have on line three and it's an array because one of the powerful things about Gemini is that you actually can build conversations so an array is a great model to model a conversation in this case we're not having a conversation so we just pass one item that it's going to be the text of the prompt and then we call that generative model gem variable instance that uh the SDK was providing us and we just simply call generate content that's an asynchronous operation so the way as synchronous code Works in that is by using Ayn of weight very very simple you don't have to worry about threats or isolates or anything like that and then once that response is ready you just save it in a response pass it to Jason we're going to see how we do that in this particular case and that's it all of a sudden we have the model the information that is going to power the puzzles of this game ready and I was like what a second when I started this project I was terrified because I thought that it was going to be really really complex I had no idea about data training and guess what I didn't have to do any of that like G and already does that for us right so one of the things that we notice at very good Venture is like hey it is really that simple to use G the answer is yes so all of a sudden people that don't have expertise in AI can actually the most power features of AI using Gemini and that's how we build this game and to me that was fantastic news because that's going to allow us for us that we build experiences to build more power powerful more custom experience for end customers right at this point though I want to talk about prompt engineering because that's kind of like what makes this game a little bit tricky but as you can see it's not that tricky just you need to find what's the right prompt to get the results that you were looking for so the first thing that we need to understand is like what do we want to achieve with the prompt that we're going to create right it's like hey this game what we want is like for every person that plays this game we want a unique experience that only the player will have so if I work at very good Ventures and I do quality assurance I'm going to get one game that only that person will get but if you work let's say at BMW and you build like the mobile apps for BMW and you say like I work in the vehicle industry you're going to get a type of escape room a type of questions that it's only particular to that person so every time is a unique game that no one else in the world will ever play so that's the first thing that we want to do a custom experience the second thing that we want to do is actually tell gem is like hey you're going to be the game logic you're going to be the brains of our escape room be ready for it the third thing that we want to do is like at the end of the day we need to par this information in a model right in a format that we want because if every time that response that we get follows a different format it's going to be really hard for us to determine where the puzzles are going to be on the screen or even how to par that information we also want to minimize latency right you're in escape room there's a countdown of 2 minutes so every second counts if you want to win the game so what we're going to do is like to generate that puzzle up front and avoid making calls to Gemini while we're playing the game and finally we wanted to make it fun and challenging we took this as a learning experience we wanted to push ourselves and share something with you on a stage that we've never done before that's the best way for learning um so the first thing that we realize is like we don't need to test actually our prompts in the game that would be time consuming that would be kind of like a stupid decision right so what we did instead is like follow through that fast iteration on Google AI studio right that's like Google's interface for interacting with GM and they're going to give you the same results that your SDK will give you but I don't need to be kind of like following the game and like finding the challenges every time I can just test all that out on the AIA studio and Achieve fastest results the first prompt that we tried was very very simple create three different challenges to be solved inside escape room where the main thing is technology that gets the job done actually that's going to give you three different challenges but those challenges could be anything AB CD type of questions it could be like a text based question right so what's the challenge with this prompt it's that it's very open-ended and it's not really customized to the player it doesn't know anything about the player so that is an area that we want optimize later on the second problem that we have with this prompt is that the response that we're going to get is not format to be computer friendly it's not going to give us adjacent output so we want to tell Gemini like hey give us a Json model that we can par and use in our game and finally those puzzles are not presented in a way where we know if they are solved properly or not it might give us like a question but it might not be telling us what the answer is and that way we need that information to determine if we win or lose the game so what we did was to start iterating on our prompt strategy so the first thing that we did in here is to improve and add more detail to this prompt right so what we added was to increase the text and now it says my aim is and then we enter in there like the player full name that we collected up front and I am the whatever title and whatever company that it was provided up right so all of the sudden if I say I'm hor Koka and I'm the head of engineering at very good Ventures G and I has a little bit of information about me so I can start making that game experience unique personalized and then we continue telling Jem and I you are the game logic of a digital escape room this game is set in an office environment some people said that it gives a little bit of severance Vibes great we wanted to be a little bit like fun and challenging and finally we said your mission is to create three custom questions that that I must get right to win the game please provide three questions about and then we pass in there the areas of interest that we collected up front or about the company where I work and the typical duties of my role turns out that playing this game I found about a few things that I have or like I'm supposed to be doing at work that I wasn't doing so it's even helping me to be a better professional that said this prompt still presents a few challenges in the results that it's going to give me and we're going to see those results in a second great news the challenges not relate to the input provided they're customized I'm going to get questions that are particular just to me as Joka but like if someone else is playing that game Johanna J do whoever they're going to get a different gaming experience that said the response is still going to be very ended and not like Json formatted and the P the puzzles presented they're still not going to give us if what's the solution or the right answer so this is one example of a solution that we got from Gemini at this point Gemini knows that I'm the head of engineering at bgb and that I can work on mobile development augmented reality artificial intelligence whatnot right so it gave me the first question about mobile development which of the following is a common cross platform mobile development framework that said it just gave me that and it didn't tell me what's the right answer like I can answer a few of those but I don't have infinite knowledge about all the questions in the world so we need to actually continue trading on our prompt to get the right answer so we continue adding more information to the prompt right we started expanding that promt that we already had and we appended the following provide me with the right answer for each question as well as seven other wrong responses per question and hints for those questions please just in case you don't want to piss the robots return your response in Json format with the fields name as first first question second question and third question this is going to be very important so we're starting telling J and I I want Json results but also this is the name of the fields these are the kind of like the property values that I want to see in the adjacent response each one of those questions must contained a fill with the question the answer at least that includes the right answer and the wrong ansers named options and a possible hint all of the sudden Gemini has more context about what I want so I'm being more specific with my Quest and it almost got it right so this time we get a validation response success we get field names that are corrected most of the times we'll see that in a second we get three questions so now we're even closer to have like a game that we can execute every single time and all the fields that I requested they're been included in my Jon response unfortunately sometimes this is not consistent some sometimes I get the Json field as like first question but sometimes they're returned as first puzzle I don't know why Gman is like that it needs a little bit more of support right so what we added was to actually add an example be really concrete and specific yeah it's a large prompt but it doesn't matter what we're getting here with G is like a specificity that's a fun word it's very very important to intera with AI right because that's the only way you can achieve and get the results that you want with the highest level of success so to The Prompt that we had prior to to this one we added examples we actually added an entire Json payload uh in this case we're only showing you like the first example but like we added three questions so we add an example to our prompt that says this is an example of a Json payload that should match what we expect from you Gemini right this is the first question what is the most popular internet browser of the moment give me a hint give me the answer and the list of options so with this one with this format all of a sudden we guarantee that those challenges are now fully customized we guarantee that the response that we get actually matches the Json model that we're expecting to Parts therefore we can present that in the game and the puzzles are presented in a way that we can solve or like we can determine if the player playing the game loss because we know what's the answer we know what's the hint right so to me when we finish this in a real realized like huh the only thing that I have to do with gemani is actually be really good at prompts I don't need to train data I don't need to create algorithms or anything like that in open up a whole set list of options and ideas in my mind that got me really excited that I haven't experienced in many many years actually probably since I started working with flut because all of the sudden I'm not tied to just Android or just iOS I can be a product developer I can deploy the this game on a mobile app on a website on a vehicle on Las Vegas sphere as long as there's a screen I can run this experience in that screen so I'm decoupled from the operating system in the hardware and with AI I'm even decoupled from the traditional constraints of apis I have more flexibility I can be more custom and meet my customers create experiences where they need it right not Everyone likes the same thing and this is very powerful so to summarize the key elements about promp engineering right it's very very very important to give clear and Specific Instructions make sure that you defined the right task that you expect AI to perform make sure that you specify any constraints that you have with the data that you expect out of it and make sure that you define the format of the response if you want to parse like something into adjacent model or things like that right add contextual information as much as you can that's going to actually context generate customization and that's very very important and if you can even include a few short examples right like hey I expect you to do this in this format this is an example that's even better because that's got going to guarantee consistent results that is very very important not when you're having a conversation with AI but when you want to use AI as a data source for your application that said this is not the end CU one of the things that we wanted to do with this game is like hey you're playing with this it's a custom experience you want to save those results we might want to follow up sending you an email with how the game was built we're even doing like a ruffle later uh so we wanted to save this information in a database so we wanted to still build an API just to kind of like interact with a database and we needed a full stack solution right so what happens here at the end is like hey you receed your stats like we want to collect the stats so we can determine who was the fastest winner who like what's the average like winning time or losing time and things like that right in order to do that uh we decided like Hey we're building this experience with dart let's challenge ourselves and use Dart everywhere because I don't want to be switching languages I'm not that proficient with JavaScript or python like I'm very comfortable with that and turns out that like my colleagues at very good Ventures developed this amazing backend framework called Dar frog it's very similar to remix that run in the sense that your API path is going to be file based as a supposed to having to create your own routers and all the stuff so all you need to worry about once you define that file path like how do you want to respond to the request that you're receiving I don't have to leave the tools that I'm comfortable and familiar with I can deploy to Google Cloud run without a problem so it was a huge win because it's just in a matter of minutes I can build that API and interact with the data that I receive so what did we get out of that first of all speed we were building this in our free time we don't get paid for that unfortunately um so we wanted to finish as soon as possible without actually losing quality but guess what that is not something that like Johan and I we care because we were doing in our free time Enterprises care about that you want to do more with this you want to ship fast and ship safe right so that was the perfect solution if you were already investing in flatter Tools in your front end on top of that it's very lightweight there's barely any boiler plate uh it kind of like the memory footprint is very small you can deploy it I mean you can dockerize it and deploy it anywhere right so we wanted to make sure that this is something that you don't have to be an expert in the back end to make sure that you can actually deliver high quality experiences and finally as we said with one single language we can build our front end we can build our game engine we can interact with AI and we can save and like interact with our different apis all with the power of Dart know it looks like that so if you wanted to do like a post call to be like hey I want to send the results of this escape room to a particular email right all we have to do is to create a file in a route SL escape room SL send results slail as a variable and this is the entire code that we need in this case we're using toio to interact and send those emails so we retrieve our toio instance for the middleware and just send that information right forget about like I've I've been doing like Ruby on Rails in the past for many years I don't need to set up controllers routes models anything like that just respond to my API this is very easy to unit test this is very easy to deploy so it's a huge win-win for everyone so where do we go from here right like we've been talking about building these experiences we've been building a game with flatter that we know that we can deploy anywhere as long as there's a screen flatter can run there right and no one will notice the difference like you can put the same application built with Native Technologies and with flatter and I'm telling you no one will able to separate one from the other I'm sure that many of you are actually using multiple flatter applications in your phone already if someone is a staying at the MGM Grand the MGM application is built with flatter hint hint um we've been also using Gemini and we've been using the first part SDK that we got from Google to build that experience to to create like those challenges everywhere we've also been using flame to run the game mechanics right like we need an unload method to run our spr our Sprites we need that uh update API to move the player around the the screen to determine collisions and all the stuff right but it's all part of the same toolkit then we've been using Dar frog to deploy our bucket apis so to me it's not about what we can do today it's about what we can build next right like if we can build this personalized custom experience in just a couple weekends what can you do as a company with the right focus and dedication and the right expertise so the first idea that comes to mind is Extreme levels of customization unique experience experiences for everyone right we just build this game getting your name the place where you work your title and a few areas of Interest that's just like three pieces of data but usually you know your customers better than like what we do with this game right so you can actually offer them the solutions that they need when they need them and that to me is a power that unlocks AI but then if you think about the cost of that like if you have to create those experiences separately for iOS Android your web team embedded systems like if you're a car manufacturer you have to build it for the naviation system if you're like in healthcare you might need to build it for like you're like uh um you know like the tablets that they have in the in the office to check you in and all that stuff that gets expensive and that's where flatter shines one single codebase worry about your brand worry about your product and build those experiences for every single screen you don't have to worry about the cost it's all streamlined it's also going to give you a higher level of focus on the real complex problem it's not the platform provin that you have it's not going to be whether you can deploy this to Android or iOS right like we need to CLE ourselves from that mentality and just focus on the product that we want to build right and that product that is the real complex because you know what's important today for customers might change two years from now you want to make sure that you adapt and react to those changes as fast as possible even better not just react lead you want to innovate and that's the flexibility that flatter in combination with Gem and and just streamlining your toolkit gives you it's not a technology problem anymore it's matter of execution and finally let the data speak be more detailed and throw through the analysis than ever before right like that to is the real power of Gemini and all these llm models we've had huge amounts of data both public both private think about all the information that we have typically on even Confluence or J things like that those are great historic records but we don't do anything with it but AI can Empower that right can unlock the meaning of that data and that to me is super super powerful but again if it cost you three four times what it should cost you to build those experiences there's not going to be there's not going to be any appetite to build that experience and that's precisely where like that combination of flatter Gemini Dart and loocks for everyone thank you so much this is the end of the talk uh if you want to come and play the game actually you can stop by the expo hall on Booth 1160 and we're like running the game for everyone we're also doing like a ruffle for Google pixelbot so it's a great motivation also to play the game and also if you want to learn more about how the game is being built Beyond this session check out our blog series because we're going to be releasing like a blog post with like more like codest s pits and things like that we're even going to open source decode right like at very good Ventures we're are big Believers that the best way to learn is through examples through having things that you can read share try to imitate and then you can build it on your own so we're going to be publishing this as an open source solution ReliOn so please stay tuned it's been a pleasure to talk to you and yeah stop by and say hi Booth 1160 thank [Applause] [Music] you

Channel: Google Cloud Tech

Views: 1,941

Keywords:

Id: _NRDyKiVIa4

Length: 40min 21sec (2421 seconds)

Published: Mon Jul 01 2024