[Music] w [Music] oh hello and welcome to live coding happy hour this is the service now developer show that brings you an unscripted demo we are going to roll up our sleeves check we're going to we are going to take our knowledge of what we know on said subject and do our best to achieve a goal and you get to come along for the journey make comments you hear the thought process you see us skin our knees and bump our foreheads a couple of times if that's what it takes but we're going to do our best let's start out with some introductions beginning with that man hi everyone My Name is Earl duate I am a senior developer advocate here at service now uh prior to that I was in the higher ed space doing development for several customers at the University of California um one thing that we forget to mention also is oh I guess you might talk about this later but we're in the middle of our Washington DC content so uh if you are interested in seeing what other things we're going to be covering for this season of Washington DC uh make sure to go to Dev link.net DC uh to find out everything that is there to that we're doing live streams podcasts blogs and everything uh but yeah that's who I am I'm Earl and then I'll pass it to Victor yes thank you Earl Thank You Chuck and uh I'm a outbound product manager for um you know the topic we're going to cover today uh I have a Utah t-shirt the Washington one is not in the mail yet but I'm sure um you know once that comes in uh you know I'll I'll probably be wearing that one more often as well wow they Martin called it out so fast he called it before he even started talking about it you know same same country you know just different area slightly different area easier climb the easier climb to the White House or the capital building than it is up to uh anyway another time my name is Chuck tamasi I am a also a service now senior developer Advocate and I've been with the company since 2010 I was a customer for a few years before that and a long long history back into the early 1980s in Tech and computers which we will share another time let's go around and do our drinks because this is live coding happy hour and the tradition was started back in 2015 2016 whatever it was where we get together and share a beverage while we do our work so what is your beverage of choice Earl so uh in spirit of the St Patrick's Day this Sunday I have a Guinness a drought timely Choice and then Victor yeah for me on the west coast not quite a drinking hour yet although it is coming but so for now I do have a lroy black Ras Rasberry spelled r Azz so cheers and I've got something that's kind of green so I get to a lame excuse I've got a margarita that's my drink of choice this afternoon so with the beverages out of the way we're ready to embark on our journey and Earl is here to drive Victor is here for moral support and I'm just the color commentary so Earl what are we doing today sure um it's funny because uh when we do generative AI stuff on life coding happy hour these streams actually do fairly well and a lot of people tune in and watch it um I think we hit search engine optimization pretty well um we're one of the ones that are making the content and having a lot of fun with it and so I want to try to keep that spirit in so I'm going to show off some of the stuff that we that are in that's in Washington DC specifically uh with that pertains to generative Ai and then I thought we were going to uh tackle an idea I had and just have fun with Gen with Gen and start making some silly stuff like we usually do on this episode just to show how fun it is to incorporate gen into our service now instances and kind of like unlocking that kind of potential and that's that's one of the things that our team uh Chuck myself Lauren and panov we're going to be constantly doing over the next several um months and stuff like that is showing just like hey it's fun to put these tools inside of our instances and so that and just showing you off that stuff to use one of the things that we want to be doing yeah so if um YouTube can keep an eye on the active chat and let me know if I need to pause or answer a question or anything and if you're in the chat and you do have a question just let let us know and we'll mark is letting us know that it was July 22nd 2016 for the first lo I should know I was there that one's lost Co messed with my brain on time um my first live coding happy hour that I appeared on it was sometime in like 2018 or 19 I think it was pretty early actually um yeah that was I didn't know it went far that that far back actually okay so I'm in a uh a fairly fresh instance um what I'm going to try to do right now is try to show off the virtual agent designer generative AI stuff so um they added uh the ability to do input generative AI uh inside a VA designer so going to try to tackle that I haven't touched VA designer since I think the 2021 hackathon so it's been a minute and so in the spirit of Life coding happy hour it's going to be an interesting dive in because we're really G to be trying to uh struggle through this if I hit my head against a wall changed in the designer between Quebec release I suppose and end now when I clicked into VA designer I was like it was the Gandalf me meme when I was like I have no recol I don't have any memory of this place so it's a little rough um but yes okay so before we jump into this instance um and you all always know if it's my instance because you see this little uh dancing service now logo but um the two things that I have already done is I've turned on plugins and I've already turned on the now assist panel and I'll show you where the p panel is but just FYI I've already turned on um now assist for uh itm and the virtual agent plugin so those two things are installed but pretty much else than that it's a fairly fresh instance because I've had to ZB booot so many of my instances recently so everything's really fresh uh yeah so a good experience um so when I was talking about analysist panel for your analysis stuff for the generative AI things to show up in a lot of places um particularly what I'm going to be showing off today too is you need to make sure the panel is turned on and now that I've already turned on the panel so it doesn't appear here but if you haven't if this is the first time you're jumping into a now assist capability you'll see this you'll see a little box over here saying you need to turn on the panel um and that makes it show up across the platform for a bunch of different things like you'll see this little twinkle for analysis up here uh that wasn't there until I pressed the wait was it there I can't remember but I had to turn the panel on so that was one of the steps if you're going through VA um what's it called Victor the the guide is it virtual yeah virtual agent console or gu um if you if you're wanting to go through the virtual agent console to do your your chat Bots then you need to make sure you turn on your skills for that so to show where that is real quick I'm in the analysis console you can go to features under platform you'll see virtual agent and going into there you'll see that uh there's three available skills I've already turned I again I turned these on already so um they would appear up here to say you can you can turn these on and then it has descriptions for each of them but I've already turned those on and I um was I haven't deactivated them so that th those are there um if you're going through Straight Up through virtual agent designer like I'm about to do uh this step I you can skip the turning on the skills because uh now as this panel turns it on for VA designer straight up straightforward so I'm going to go ahead and go to um virtual agent designer and real quick while that loads Earl I you said you mention you had to turn on the panel to access the designer that's something that was probably there left left in code for Washington early availability you will you will not need to turn on the panel in the future in order to use VA it's it's it is actually kind of two separate things there so that's why sure um you VA is completely separate to to be able to see this part to change it over to the now assist for um llm assistant part that's what you need the panel for I think right in in early availability but not ga oh okay so you can even skip that stuff thanks this is why we love when PM sitting on our live streams because they can correct um our fiddling around uh so if you didn't know already virtual agent if you're use one of the main things that you can use for topic creation and stuff like that is uh natural language so nlus um but with analysis you can change it over to llm and it's a lot faster I mean everything that I did so far this time around was faster like turning on all the plugins was so much faster and I think that we did that in Utah or Vancouver but just everything seems so much faster uh but yeah so okay I got to stop talking and I'm just going to start making stuff um okay so I'm creating a new topic uh what was the last who was the last thing this is the lasto topic he llo is the last person I saw in the chat so I'm gonna put that uh what do you want this topic to do okay so girl make that bigger oh yeah yeah yeah everyone drink because we had someone had to remind me to is that part of the game it's not life coding happy hour unless someone forgets to zoom in okay um so at this part uh what do you want the topic to do and it's actually this is actually building the llm the the access to the now llm part of it so we need to actually you know tell us what it wants it to do um what's uh what would what could we be gathering information for what's a example oh um this will uh what was the example here LEL says in shots every time you hear his name llo llo llo no um help users see other PTO okay so help users determine what drink they should have for live cing happy hour uh for a technical live stream I don't know exactly the content that would be that sounds that sounds good um yeah this is basically top to help with topic Discovery as we call it right like how how do users get to this topic just by typing in their query right right so in nlu used to we used to use utterances and intents and entities no longer now all you have to do is just describe what this topic does and this is a a great description you have here if I were to like I don't know enhance it be like oh what drink such as soda or beer they should have for a technical live stream yeah exactly um you know the more objects and stuff that you can put in there um the more the Live Language model will be able to like int it and understand you know what this topic is for wow no more utterances and intent nope nor entities yeah and if you're used to gen um like it's often you hear sentiment analysis and stuff like that when when it's figuring out if this is the topic that matches the query or whatever the person's typing in the virtual agent it the you'll see in the uh it's like it's sure of this being the topic or it's it's like a confidence level so it's similar to a sentiment analysis but it's taking their their their chat and saying based off of the llm and this description that this is probably the one they're looking for so I'm going to go ahead and create this is like the mo the most context for every step along the way for anything in live coding after hour that we've ever had uh okay so I'm in VA designer so um this is the view that we're all used to and then uh you'll see these little the little twinkle stars that that appear on all these things that means that this is uh llm enabled um and and uh what that means is if you start collecting uh info from the person and I'm just going to add one of each of these and I'll go back and name them in a second oop uh we'll do bullan and a date time um what it's going to what it looks like it's basically before you would have to say like oh what question you want to ask and you type out the question and um and you have to word it and uh figure out how you want to uh engage with the user how friendly or whatever all that kind of stuff but now you just have to say what you're what the thing you're Gathering from them and then the llm will generate all the text to engage the user on how to collect that information um so let's just start building it and that will make sense when we start testing it in a second so the input collector is um I'll just call it input um no what's the next who's the last person that talked again par uh he's in the chat and what is the reason this the llm is collecting inputs from the user uh we I always like to look at the example to get the user a temporary building access B to get to get enough information to recommend a drink to the user all right so that's what the instruction is for the llm once for Gathering all this information and then if I go back through all the different inputs that I put um my mind just completely blinked out I'm so sleepy okay I'm catching up my um I to my toddler I had to wake up several times throughout the night to help her so I'm like Dazed and Confused right now I'm I'm not just drinking Co um beer but I'm also drinking coffee at the same time to bounce it out so it's an Irish coffee essentially yeah essentially the universe is going to implode in your head um oh what would be a text input for gathering information um what is the topic today uh drinks right no oh no but like we're we're looking for it to recommend a drink aren't we yeah so what would we ask what's a text thing that we would ask somebody like uh what do you usually look for in drinks ah okay okay you're GNA feed the input of that to the llm yeah drink preference tell us the detail I need their shirt CHS I need to know what they usually prefer in their drink choices so see it we're not tying out the question we're typing out what we need and then we're going to let the the the L llm the generative AI to determine what that is uh choices um uh drink types I need to know if they have a specific uh drink type of preference so we'll do like beer wine um nonalcoholic uh soda and we'll throw in other what Boolean what's a good Boolean good bullan uh that's where you could have gone alcoholic or not but um true are you over 18 do you mean 21 21 sorry I'm an old man it used to be 18 when I uh sure are you over the age of 21 and if and if not we can you know we can ask you you know what what what sodas what what what types of milk would do you like to drink chocolate milk uh date and time input oh what would this one be for um what day are you available to drink or what or is the show I don't know yeah I wonder what it'll do with that what what when you need to know when when the uh episode live coding happy hour is well just so to clarify this is for generating the text that will go into the input um collection uh this doesn't actually generate the answer this is that would be another step but we're just focusing on this one uh for right now the input collection uh I'm just going to add a random a last step of confir like a confirm message here okay thanks we'll let you know whatever okay so um so from here I'm going to test it this is I've not gone this far in testing so we'll see if it works okay so what we some I said life coding happy hour so let me try to just uh oh okay this is live Cody this is good this is fine uh click restart say like I need a I need a drink recommendation I need a drink choice for my live coding happy hour or something like that okay okay still not working um so this is I was wondering what will happen here because I'm used to the nlu tab for this part yeah so that that shows that will show things when it catches it if you go to prompt Discovery just trying to REM remember what it goes so it's got that and it actually if if you expand that window vertically a bit um yeah it it actually picks up the topic you see it says llo topic so it picks up the topic I don't know why it doesn't run the topic itself for llo topic I need a drink for for technical live stream okay so it's not working so I know I heard I've been told there's another way to try to force it and so I'm going to try to do it here and I'll back out if it doesn't work but um let me try show me okay it's not there either okay so this is where I was told that like I if I want to just test the input thing for the stream I should back out and try it here because I know it works s so I'm going to go back to VA designer on the on the selection page oh hold on let me um let me save that for I didn't publish it let me publish it maybe well do you think it doesn't have to be published to test it right right I was goingon to say that would be counterintuitive I'll just try real quick uh hedr says the topic is inactive was it before you published it it was inactive but when you test that sh should it shouldn't matter it shouldn't matter yeah okay okay okay so we're GNA we're going to do it from this view so I went to back to VA designer the selector and I said we're doing this through now assist I'm going to say test active topics which will open the same window and um I think the error still happens here and I need to figure out what's going on there I'm again I've done nothing to configure virtual agent in this instance so I'm not sure if that's one of the reasons why um but I know uh if I go to show me everything here here's llo topic um and because I'm not really testing for the utterance part of it I wanted to see what it would generate right here it is so okay so um remember I didn't type what it should say for the question but it's asking me hey what what do you usually prefer in your drink choices and I will say um uh I'm not much of a pops Sky sorry you're giving it a challenge got it based on your previous response you mentioned that you're not much of a hop Sky now I have a question about drink types for you do you have a specific drink type of preference you can choose from beer wine nonalcoholic soda or other so you see it's kind of like um it's trying to make it conversational and it's making it so that it incorporates previous choices it's trying to just be a little bit more friendly and it's and this is not and I didn't do anything to tell it to do that all I just said was what I need so it's creating that text from that so that's kind of what this feature is in Washington DC for before you go before you do anything don't don't click anything type in Chardon oh I don't how is that how you spell your yeah I think that is oh okay so does it show me the um where's the it should invariable drink types it doesn't upload update the thing I was hoping that like because it's um it's large language model powered much like obviously for bullying like you don't have to type true or anything you can just type yes yeah whatever and people will generally know that's true like I was hoping that the llm would go through the synonym and know sh8 to be wine but I guess in this case you need to be more exact but anyway that was just a a test I didn't know what was gonna happen that's good uh got it I see that you can legally drink alcohol by what would you like to have by when would you like to have this drink uh I we have another lch by for next week so we'll put 22 oh okay now it updates oh interesting interesting uh K thanks we'll let you know um age is true uh I like that type java. lang. string so it it inputed a new type even though that didn't exist in the choice list yeah interesting yeah it should have just maybe errored out but I don't know or I would or intuitively know to put as other or under wine I mean other on maybe but yeah that was the drink type but it wasn't your recommendation wasn't it supposed to come back with a recommendation no no this is just for input collection generating the response is completely different from this and I just wanted to show off what's new in Washington DC okay so let me I want I'm kind of interested to see uh what you're talking about for the when it was asking for the Boolean I'm just going to say yep and see if it puts it as true um I prefer drinking on the show floor at knowledge in the he which is a true story and I got in trouble for who was busting you in Amsterdam for drinking on the show [Laughter] floor it's because the it's the the venue was foring the drinks so us drinking our own drinks was not anyway um uh let me just type in whatever and then let's get to that bullan Les says pinky swear yep I pinky swear got it here's another question all right well let's see if it properly classifies it as true true wait it said what uh it said it says false actually oh false Oh weird interesting no more pinky swearing to an llm okay let's try not to fool it I'm just gonna put reasonable answers yeah I'll just say yeah yeah yeah yeah and it worked yeah [Music] affirmative Oh I thought I broke it for a hot what oh it it really I did break [Laughter] it as long as it's not poison I'm [Laughter] game okay now I'm going to say yeah and hopefully that sets it to true yeah AG is false J AI peeps you got some splaining to do oh well huh I guess that's another time another time we'll talk about that but uh at least the actual generating of the prompt is uh accurate good yeah yeah okay so um I'm going to back out of that that's the Washington DC thing I wanted to show off uh for the rest of this episode I'm going to I want to shift gears so uh shift gears I wanted to build something um on this app instead of just demonstrating a feature um so I have I want to make um an app and let me jump into I already created the scope for it um it's it's a completely empty scope which means it's empty so I need to make the table H okay let me just oh the product teams are going to get mad at me for jumping into the old Studio but I'm going to do it the audience won't complain I'm so used to it okay I'm I'm okay so my idea was we have all these live streams um and I want to create summaries or blogs based off of the live stream uh but I don't want to rewatch the whole thing and so I had the idea of oh and also I used to when I wanted to do this I would grab the um transcript after pulling downloading the live stream video putting it into a Premiere making that generate the transcript and then I would go through that transcript and um summarize it and stff like that but one thing I noticed was that YouTube you know how it has those autogenerated captions over um they appear after some time after the stream ends uh you can just download those and so I was like Oh's let's just take those captions and then uh throw it through geni to make it um create a summary so we're going to try to do that so uh lch did I already make it no I didn't make lch summaries create wait am I creating a new no hold on I already I already created the scope I need to create the table for it my bad okay I have to zoom out to see the button permission granted yeah this uh non next experience UI is giving me flashbacks I suppose this you knowt ah all these lists and forms um a place for everything and everything in its place so one real quick the reason why I'm putting this at full utf8 is because um the transcripts are huge uh so I want to make sure it can fit inside of these fields uh but for the blog itself it shouldn't be that long so I'll just put 4000 pretty much anything over 4,000 won't get truncated yeah I mean if you leave it I think it's 1,000 or 4,000 is where it stops truncating I think it's 4,000 I think you can up it to 10,000 in properties I think that's what it is interesting okay so I have the table and then so what the goal will be is uh hold on let me go back into the regular view so people don't get mad at me okay so I have uh this field I'm going to put the input in here and I want it to generate the blog up there and I should have done some preconfiguration on this instance but I didn't so we're going to walk through setting up the geni controller real quick um give me one second I got to go find the documentation for this oh that's the whole screen my bad do you need my help I can walk you through the the setup for the um no I I found it it's fine so I have the doc uh it's this one I was looking for sorry okay so there's a table um and if you go this is in the documentation so you'll find this table under setting up um uh a provider for generative Ai and in there I'm going to use generic prompting which is in flow designer uh and under config I get to pick what model I want to use and I have a key uh for live coding happy hour that I'll use for this one so I'm going to turn on chat GPT 4 for this change scope first oh I usually jump into the record and then just press edit okay yeah and then I go up to here click default update it and there you go okay so now that it knows to use that one when I use generic prompting inside the platform and then now I have to create a credential uh for the Shrubbery that's a think I'm just gonna take off my screen real quick all right and the the capability you turn out was that for open AI or Azure open AI I I didn't catch that I'm pretty sure I C I I'll recheck okay okay so um I got my key and I'm I pressed this related link create new credential uh and then it's I inputed my API key nice uh and it okay cool it did it and so let me backtrack back to one extend oh failed to create a new connection oh seems like one already exists did I already do it maybe oh man I'm like completely lost with what I've already worked on well hey if it's there then it's then it's then it's there and hopefully it's there we'll see if it works in a second um one extend okay let me make sure I put on turn on the right one yeah opening a cool yeah looks good all right so that's it that's all I had to do to set up that part so I'm going to jump into flow designer and uh attempt to make this flow oh one thing I am worried about is it failing because the transcript might be too big for the API that's one we'll see if it works what this show is about yeah if if not then I have an idea of how to strip some of that um you you log a defect to Sam mman to increase the token count for input it's already using a higher one so the API itself gets a higher token count and then Chad bt4 has a higher token count so hopefully it's not a problem yeah um I probably should have hit subflow whatever it's live going out who's the last person in the chat Andreas Andreas I hope people know that if they're watching this show it's not all about best practices always it's about having fun getting things done showing us actually working through some of the errors that we run into uh oh edit flow that's different when did that when that showed up in Washington oh Washington okay cool no accidentally deleting steps while looking at something that's nice are are uh you're not I see you haven't did have you got a chance to test out flow recommendations or flow generator I haven't tested it yet is this where it would show up that's a Washington feature as well you basically in natural language type out what kind of flow you want you know generate it generate an email when something happens and it'll build out the skeleton of the flow and then you just kind of fill in like all the other details so that's that's analysist uh product coming out soon too and I just noticed that I got automatically loaded into workflow Studio I didn't even think about it yeah the the module that goes to flow designer now goes to workflow Studio it redirects okay so I picked the generative AI prompt for generic prompt and I'm going to give it the input field from my table oh but first I want to say generate a Blog from this transcript of a service now live stream technical live stream ah so this is where we would want to make it little verbose so if if we have time I will flesh this out to actually give it like some of my old blogs that I've written and say stuff like make it feel like this blog and it'll try to model it off of that so if have time I'll keep adding more features to it but uh I mean that's that's really it and then updating the record you just made me think of someplace I want to use this yeah you want to talk about it while I typed some stuff uh social media posts so when we do a podcast or something we can throw in a a transcript and have it generate the social media post because where you post matters on the length the content the hashtags the at mentions all that stuff so um you know we could automate that quite a bit um uh just a shout out to one our colleague or our teammate uh panov who just created a short talking about that it Autos saves on Flow designer now um I was up here like where's the save budon and uh if you make oh plug please like And subscribe to the channel in this video it helps a lot I guess but um if you are subscribed to the channel you might have seen panov's uh YouTube short on this but he talks about he showed off that it Autos saves and flow designer now I'm sitting there like how do I save okay it's already there uh I'm just gonna test it live because it's live coding happy hour we do it live yep uh summaries okay so I'm uh oh and to show you what I mean by um the transcript I already so I have our YouTube studio for our Channel and I go to content let me Zoom go to I think I took that picture just a few minutes ago yeah right ja hands uh and then you just go to one of your shows so I think it'll be funny if we let's try to do the last year's or the last episode we did for generative AI okay which is this one is it this one yeah I think Victor you were on this one too yeah August the good old days and so oh go ahead no no no yeah so once you get into the um the video go to subtitles and it's under here so you can download that transcript my goodness um and it it puts it in an SRT file so let me um Shrubbery real quick and download it and open it up so what is an SRT it is I don't know what SRT stands for but it's um it's captions with timing it has a oh okay so it is text yeah yeah and here I just loaded it let me show you like oh at two seconds you know this is a tech there it goes yep yeah I've seen that before pretty standard looking stuff yeah so this is it's telling YouTube exactly when to show that thing on the screen or uh it's a standardized file most multiple platforms no SRT yep and so at one um one minute in one second say hi everyone welcome on the screen so that's what it is so I'm going to take this transcript this thing is huge that's 100 it's 100,000 characters I don't it's gonna that's going to break the API all right we'll find out find out find out so I inputed it and um low should be running why did it say this field was updated I just did it all right well just let's check the flow so executions [Laughter] error dang it Sam mman if you're watching this no I'm just kidding um okay so let's just you know let's I'm going to redo it with a much shorter one and see if that works okay let me where's how do I refresh this page the refresh icon on the left of that panel right ah thanks okay so this come completed I think it's that I think gets the tokens okay there it just summarized it fast that was fast yeah well I mean it was only 37 lines worth of text I was going to say it uh yeah it created pretty verbos it got the it got the title right okay so um so I have given this a little bit of a thought so this is what I would be this is what I would do right now and I think we have time to try to do it so I want I'm going to try to remove all the super super Superfluous information from here to try to limit down the token count uh and so how I would do that is Rex so um oh this is going to be rough let me let me go see I did think give this lot I have Rex ready so I'm going to put my example in here okay so I want there could be a new line a number well it could be one or it could be zero or more lines and then um a number which is a digit and that could be several of those and then specifically another oh and another new line that could be multiple of well no that could be multiple of those go watch my our Rex video videos on this channel if you have no idea what's happening um I'm sorry just use chat gbt to tell you what to how to create a a Rex to strip out XYZ that's true that's an interesting use case I haven't thought of yet but I'm trying to flex my reject skills y' okay my yeah this is where the the old adage comes that I I used to have a problem so I thought I'd use use Rex to solve it now I have two problems MH yeah I am hopeless when it comes to Rex that's I rely on sites like this heavily and then you go past it and goes I'm sorry said on the Mac doesn't understand that expression wait I thought Rex was Universal okay there we go I think I I think I'm getting it now there we go perfect okay see everything is selected besides the actual text that's my that's me flexing my R all right gonna go I'm going to do this in a business rule and I'm going to say whenever there's a new record Advanced uh who's last in the chat uh Andreas again uh who before that Edgar Edgar why can't I edit what is happening oh it's Lo it's still loading UI again this is a freshly ZB booted instance so a lot of things aren't cached so when you first load business rule after a new cash business Ru is one of the slowest loading things okay um Edgar yo what's up ad yeah before insert where's the advanced tab yeah where's the tab and check it and check it again oh my goodness okay what's Happ you jumped the gun on that one it really wasn't ready current. input equals current uh current. input. replace two string. replace my Rex Global while while Earl Flex his Rex and and and scripting I would have just installed now assist for create and code assist and in natural language tell me to do all this fancy stuff right now because I can't do any of this off the top of my head we wrote one line of code on live coding happy hour today yeah okay at least it's like an impressive line of code right it is an impressive line of code thank you unless it doesn't work then it's not impressive where's that record uh oh man where did it go see you'd be able to blame it on hallucination it didn't work with code assist but now it's it's all you Earl if you watch this show before blaming it on me is not is not entirely inaccurate so I'm gonna test it with a shorter thing real quick it didn't work pick the executions no it's supposed to strip the stuff oh the block part worked um it did it before insert is that what you said yeah interesting okay let me go to business rules real quick and look at that is that a minus or an equal where current. input Dash is that what you meant to do Dash yeah current. input is your equal to me that's an equals okay my monit oh there you go Zoom or I'll Zoom yeah uh we got a few minutes if you want to debug that I'm going to um I'm going to take a hammer approach I'm I'm going to guess it's not meant to be new line it's supposed to be um white space of some type space so that could be God please work oh now it's too zoomed in okay so let me delete that and insert in state to re-trigger it there it goes there hey there you go hey I don't know what the extra one is there for let's see what it does to the the summarizer um what's that extra oh you know what the extra one is from uh it's this because it goes into double digits I didn't factor that in you needed a plus I really want to get this to work give me a second okay so I need it to be back SL D+ right but start an asterisk is zero or more where do you have where do you the as is right here no on that's your white space oh oh okay after backd okay yeah perfect okay thank you Chuck for seeing that one oh okay so that worked okay so now now let's test it for truly go back to my summaries table I'm going to take the whole stream which is 10,000 character 100,000 characters long I'm G to input it here and I know it ends in music always so save it stripped it out into one long trans wow okay there we go I hope it still has enough white spaces to make a sentence and now we wait to see if it worked oh I should check the execution if it failed feel free to take a drink oh yeah refresh refresh that's an old one who's the latest one 1418 error a it didn't work it didn't work 1420 or I didn't I don't have verbose um logging on right now so I would need to change it is it just too long still you know what let's try with a shorter stream um isn't that the point of summarizing unless I ran out of API SP like funding on it you have used how much money do you have in your open AI account you know I only put the minimum at any given moment have it right the blog post as we're talking right now um let's just do what would be a shorter stream uh this Creator toolbox is shorter let's try this one to answer Edgar's um question he did use gbt 4 so I would also have thought that gbt 4's context window would be much would be able to handle 100,000 characters well okay so token count C let's try 10,000 Oh see it's just above the limit oh well cut off the last couple lines okay yeah we'll just do that because I think it's it should work uh just yep you let's just be sure a little bit more and insert in stay you insert and stay and while while that's happen I did insert and stay yeah so while that's happening let me go grab the other stre let me grab this file oh doesn't have this one doesn't have captions uh oh all right let's let's dramatically reduce it now pack it in half in this episode they lost connection halfway through in progress girl can't wait like faster that's not screen flicker people Earl is just it work take a drink celebrate the little things let's actually read it okay in this live stream we talked geni uh can connect to third parties we talked about open AI andure open AI uh yeah yeah yeah looks good looks good it almost reads like the transcript though because they've got first you need to do this once that's set up I mean but that's kind of how that stream was though remember from back we like talked about how to set up this the controller and everything yeah okay okay and you didn't ask for like a four line summary your prompt just make a blog post out of this that was that was another idea though um cuz one of the things I'm not great at is going back and putting a good uh summary on the YouTube description so I'll probably keep this app um here and then in a couple days when it generates the transcripts now I'll run it through it and then I'll say create a YouTube summary for the description for this according to this transcript and I'll put it inside of this YouTube video so if you're watching this uh like one week later after this airs then come uh hopefully you can look at down at the description and it's actually accurate to what we talked about and if you want to influence uh this transcript right now Chuck or Victor go ahead and just say the most random of things wild boar appears a wild boar appears just to ruin the transcript I don't know I would call that a success Chuck is Chuck is doing some prompt injection as we speak with the transcript Purple Monkey Dishwasher yeah you know if we have time uh I can share uh speaking of virtually in designer going back to the first topic I can share kind of what I did for my uh uh for my instance so if I were to share screen real quick here oh yeah sure let's see this one so this was my way less fun virtual topic which is about trouble shooting collaboration apps right like let's say my topic is a troubleshoot or fix issues for a collaboration app such as team slack or Zoom issues may be video audio whatever hey Victor can you zoom in buddy yeah sure oh let me let me zoom in here first there we go there's the description yeah right and then here let me zoom in again let me Zoom that's interesting but zoom out that one you got to do that way yeah I gotta do the other way yep I can zoom in more and here I basically you know same thing you did Earl I just created app uh impa collector and I just asked them hey what apps do you want you know describe the issue and you know I my description was way wordier than yours LM is gonna just kind of sand that out and then whether or not it's a high party that was my bullying example right and when I test this hopefully you see the popup if you don't I could re I can reshare the screen you might need to I think we only see your let me oh it says uh share this tab instead I don't want to do it instead let me switch my screen share to window or desktop sorry uh okay cool let me loading loading loading so I click test again and remember when I said no more utterances inputs or ENT remember when you had to like create an entity for each kind of app right or each kind of catalog item I don't even have I don't even have to do that and I'll still know what I'm talking about so if I say you know I have a sorry uh it's not on our screen oh oh Sho we're still looking at the uh canvas why is uh presenting the oh you know what sorry present screen uh window uh screen not window my bad that's my bad let's try that one more time some Zoom versus streamyard terminology here we're seeing more and more of your computer screen every time you you share that's completely fine there we go we see it now y great okay so you hit test and then yeah so I can I can say I have a video issue with slack right and then like what you what you showed Earl it it knows the it it can find the topic relatively quickly but it'll show more analysis uh pretty soon coming up so yeah that's what I was been trying to get to work on my instance I I couldn't get this this tab to work but you got it to work okay yeah so it says like this is the topic I'm I'm wanting and and it could already matched like the confidence score that you're talking about and some of the variables that I mentioned uh in my input collector it kind of found out here right and it already knows look okay I got a video issue don't need to answer that question skip it it knows I have slacked don't need to ask me that skip it is it a high priority issue or not and I can say you know yeah it is that's awesome so it filled in the variables knowingly oh okay you don't have to like bind or fill in entities anymore like you used to uh with nlu it just llm just knows like okay slack means slack video means screen share or video right that's awesome and then um what I've done here is let me see if I can resize and zoom out a little bit with bullying you can you notice there it said you know it is a high party issue confirm that I I actually said hey decision tree give me a decision if it is High party if Hy partty is was true log me a log me a ticket if Hy party was false send it to some generic you know Chad gbt and then let that answer collab issue for me right and and we we showed this in um in August which was that other live coding video like how to in how to embed controller into uh designer that's basically what what what what this is where I passed that describe issue V uh variable here that it captured into the prompt and the prompt then generates the result which I I spit out here you know we ask gbd4 and came up with this response and then it just goes along this thing it's either that or login issues so you know that's the way you know the the less fun more work oriented topic but one that uh you know the one I used to to show people about the the possibilities of of va8 with uh large language models that's dope that's really cool I now if on that Boolean if you said yes this is a high priority in here are details would it have filled in the next question too so let me try that let's try let's do it again um let me re I'm restarting here this I think you have a a better patch mine is like Day Zero or a day one install so my my VA is a little slower than yours I have a super high I'm going to be extra dramatic High prity issue uh on slack and I'm not going to mention video anymore so it might ask me of hey what's your describe your issue and then you could be like it's a video issue in here or some details yeah or I can it's a uh let's see I think I used free text here so I can just type whatever I want uh let's see H pry it captured that as true so most likely it will skip it sometimes it doesn't you know a large language model it's it's generative so I don't I'm not able to like I'm not always going to like 100% land on the same exact thing every time um but yeah knows it slack describe the issue I have a video screen share issue and let's do this oh I I put it as yes if I said no it's not a high PR issue would have just asked Chad GPT uh it was and then it keeps it it probably go to you know login incident so so yeah that's the uh Power of it can it highlights which uh Branch it's going taking me towards oh what it do it didn't do anything okay well we'll demo that another time I think we need to start rounding up that's that's sted it it worked uh in the in the previous uh try so yeah that's that's my example at least nice nice I I like our generative AI streams with Victor too because we always uh he always ends with a a working thing that shows what it's more capable of so it's really good to have Victor on these things with us like the cooking show this is way it was supposed to turn out to finish fast fast forward and they pull out the already prepared one out of the oven here it is oh thank you so much Victor for showing up you uh you definitely helped put some color on this Earl nice job getting through that and thank you everybody who joined us quick reminder we do have more content as as Earl said Dev link.net DC I don't see the banner up there have to get a new one uh but we do also have Creator con content coming out so if you haven't already looked get signed up this is the developer event at knowledge coming up in May I believe what is it 52 days from now something ridiculous oh my gosh com yeah it's uh not long not long uh if you're in the Washington DC area I invite you to come out for the Washington DC launch event you can find registration information for that uh let one of us know we'll be able to get you information for that it's there's a developer meetup on the 18th and then the event is on the 19th for the Washington launch there should still be room for that don't know yet but uh it's filling up fast so you're in that area the Virginia Maryland DC area we'd love to see you there Lauren McManaman will be there mark Obie will be there I'll be there uh Sarah Gary's coming in so a lot of the community people that familiar with will be on site to show off all the cool new stuff in Washington DC so hope to see you there until next time what did I forget Earl uh raing our beers if we want raing our drinks yes how could I forget it's been too long go ahead Earl uh I really like gness I'll give it a solid four oh it's out of five okay yeah I I'll give this uh black rasat Taz uh a solid four out of as well uh I may I I may prefer tandron Moore personally personal preference but this is pretty pretty spoton I may have added just a wee bit too much tequila to my margarita but I'm not complaining I'm just making an observation it was really good I'm giving this a 4.75 nice oh Technically when they're success we add another 0.25 that's tradition on our show so up we'll be singing karaoke tonight thanks everyone thanks everyone a great stay again for joining us bye everyone [Music] bye [Music] w w