Prompt or Code - You Choose! Data Science with DataLab (Part 2)

Hi there, Data Scamps and Data Champs, this is Richie. So today is part two of our series on new AI assisted workflows for data analysis and data science. So if you missed yesterday's session, don't worry, you'll still be able to follow along with this without any difficulty. But I would recommend that you check out the recording for yesterday's session when you get the chance. Now, if you register for the session, so we're just going to put through a registration link. If you register, then we're going to send the recordings out to you by email. So the short version of what we covered yesterday is that there are two useful AI based workflows. So you can either use an LLM to help you write code, so getting it to write little jumpsuit code, and then also check for any errors you make in your own code, or you can simply use the LLM as a chatbot and have it write all the code for you and then you go back and check it. So yesterday we looked at the start of the analytics flow, so that was covering exploratory data analysis, and today we were looking at using it for data storytelling. Our guest is Joe Franklin. He's the head of curriculum for data literacy and essentials here at Data Camp. So that means he's designed much of data, he's designed much of Data Camp's conceptual curriculum, and he's become an expert in teaching people how to do data analysis without writing any code. So over to you, Joe. Please take it away.

Yes, thank you very much for the introduction. And yeah, data literacy and data storytelling are topics near and dear to my heart. So I really love talking about all of that. And to again give the little fun intro slide, Joe Franklin. I am the head of our data literacy curriculum as well as our business intelligence curriculum here at Data Camp. And it really is an interesting time for data literacy and AI literacy because we're seeing so many new tools come out on the market and so many ways that you can really access the power of data without needing to be an expert on how to do data, how to do all the various analytical efforts. And we want to make sure that our offerings from datacamp are consistent with that. And one of the biggest things that we hear from so many different people is that they really want to get more involved in data storytelling because it is something that we've championed quite a bit here at Data Camp as well as it's championed in a lot of different areas. And it's because data storytelling is super important, it's super powerful. And it really helps us cut through the noise that we're seeing in the analytical sphere. And for those of you who've attended one of our data storytelling webinars before, maybe taken some of our curriculum, these ideas won't be new to you. But for those of you who are first hearing about data storytelling now, there are three main areas that I like to talk about for data storytelling, and we're going to see how AI can really help us enhance the data storytelling workflow and inspire action, ease understanding, and really deliver that lasting impact. Because when, when you share a dashboard with someone, when you just share some basic analysis or something, you want this story to live on afterwards. And that's where data storytelling and AI enhanced workflows can really help bridge that gap between just, oh, I did this, here's a number versus oh, this had a major impact on us and.

We.

Need to take action as a result of it. And in the past, the barrier to generating data stories was actually pretty high. One of my largest bits of wisdom that I used to share whenever I talked about data storytelling is not every insight needs a data story. And the main reason I would say that is there was it was somewhat difficult to generate them. You had to either know specific things in Excel, specific things within Python, or various tools. And it took a lot of work to take a data, to take just raw data all the way to store it. Well, that's not really the case anymore, and we'll see that as we go through our data lab presentation later of just how interesting and powerful data storytelling can be and how easy it's been. So for a little bit more background, the core of data storytelling is visuals, narrative and context. We want to share out our information and we want to share it with a specific audience so that they understand what they're being faced with and they can really digest it in a quick and easy way. I can go on for hours about data storytelling, and in fact, I have many, many different times. This isn't a presentation solely about that. So I will gladly direct you towards our, towards our expansive data storytelling curriculum that we have here at datacamp, as well as the webinars that we've done in the past all around it. But it really starts with understanding your data better and becoming mindful of, of the data that surrounds us. And there's. What we're seeing is with the rise of generative AI and with the just the democratization of data, there's so much more access to data than there's ever been before. There are so many Source systems that are turning on where, hey, I can get access to this customer feed or this sales bit of information or this manufacturing line that used to be siloed away and we just had to guess about it. But with that comes just an overwhelming amount of noise and just challenge to really unlock that extra bit of information. There's actually a great quote from the author of Storytelling With Data, which is probably one of my favorite books. I will tell you. I carry it with me all the time for doing all of these various presentations. You can see that it's pretty smudged up on it of that having all the information in the world at our fingertips doesn't make it easier to communicate. It actually makes it harder. The more information you're dealing with, the more difficult it is to filter. And this quote was, it came out a few years ago before the Generative AI boom. And now I think we can think about it a little differently. For those of you that attended the webinar yesterday, you probably saw quite a bit of our Data Lab offering and I'm going to feature it again more heavily here. But Generative AI has reversed that paradigm a little bit, where having access to huge amounts of data is actually giving us the ability to filter it even quicker than we were before. And that's where our newest and kind of really interesting offering, Data Lab, which is a feature of datacamp, really comes in to enhance your workflow with AI. We have two main areas and we're going to look at both of them throughout this presentation today. The first one is the AI chat, which is where we can analyze our data entirely without writing any code. And that's going to be the one that I really want to focus on the most today as well as there's a fully featured notebook behind the scenes that you can then access and build out for sharing. And you can access the AI within it to make tweaks, to make recommendations. Richie at the beginning of this talked about how the, how the AI tool is really great for authoring code and really great for, for editing things, but actually you can do even more than that. And we'll, we'll talk about that in our presentation of where we can use it on text, just raw text cells within our workbook, where we can refactor our presentations and refactor the context of, for different audiences. If we go back to the definition that we presented earlier of data storytelling was using narratives, visuals and context for a specific audience. LLMs give us the ability to write one report and then if we're sharing it With a business line manager, we.

Can.

Rework the introduction, rework the paragraphs just right so that we're giving them exactly what they want, versus if we're targeting a high level executive, we give them a much briefer, much briefer summary. And it's really, really amazing how all of that comes together and seamlessly integrates into whatever you want to connect with Data Labs. So if you've got a CSV, if you've got a Google sheet Snowflake, we can connect to it and we can bring it together and really show off the power. But if you don't already have all those data sets laying around, we also have a lot of sample data sets, so you can figure out what makes sense for you and how you want to apply it and where Data Lab's power is. For this webinar, we will be using a sample data set, which is a sample data set that came from that book that I referenced earlier, Storytelling with Data. And it is a data set on the ticket volume for an IT organization. So what, what's happening in this is a, a help desk is getting in new tickets, they're processing them, they're, they're sending out whatever assistance that they need. And we want to figure out what's going on with our data set. And the scenario is that two team members left the team in early May and, and we want to figure out, well, how, how big of an impact was it and how can we share that impact with our stakeholders? There is one important note as you are prompting along with me, AI systems are a bit unpredictable. The same prompt may lead to different results. And actually, that's kind of a cool thing because I really find it interesting when I ask the model, hey, tell me a story about this data set. And it goes and gives me something different because that's a way of injecting extra, extra sets of eyes and extra sets of reviewers. We need to always make sure that whatever we're working with, we are able to back it up ourselves. And we're not just blindly testing them, blindly trusting the models, because at the end, we still need to be the humans in the loop. And for those of you who are frequent flyers to our webinars, you might also be having a small sense of deja vu. In January of this year, I actually featured this same data set in a storytelling webinar where we used primarily ChatGPT and the data assistant within ChatGPT. And the main reason I'm doing this is so that we can show how much better and easier it is to. To do within Data Lab instead of just going directly to raw GPT. GPT is great for a lot of things, but the kind of the extra special sauce that has been built within the Data Lab environment really helps take what would be a large kind of unwieldy LLM at times and make it very targeted for our workflows. So let's hit the highlights. What are we aiming to do here? What are we, what are we going to do? And then let's get into it. So again, we want to focus on three key parts. Our narratives, our visualizations and our context. And we could actually use GPT for all three of those parts and use Data Lab within that to make it even better. And once we can really target those, we, we have a few other areas to to build on and this is really our game plan and we're going to, we're going to aim to do. The first thing we want to do is we're going to summarize it and provide, provide our data set with some context, figure out what's important in it, what's not important in it. We'll do some basic visualizations. Data Labs actually very good about suggesting visuals very early, early on in the process. And it's one of my kind of favorite features within Data Lab is how useful it is for just giving various recommendations and going and giving cool visuals. Then we want to refine our visuals and I'm going to borrow again from the Storytelling with Data book here as well as our curriculum. And there's three key areas we're going to use for refining our visuals. First one going to declutter them. There's a lot of the standard outputs aren't very useful for telling quick, easily digestible bits of information. Then we're going to focus the audience attention. We want to use selective bolding, like for example in line. We want to really draw them in on specific elements. And then the last part is we want to think like a designer. Is there any specific colors we want to use? Is there any bits of extra information we want to highlight? And then finally we're going to take all this together to tell a great story. The last point in on there about refining our visuals, which is the thinking like a designer one. This is the one that is going to be the most subjective. So it's going to be the one that I'm going to spend probably the least amount of time on today. And the main reason for that is when we think like a designer, a lot of it is based on the norms and understandings of your particular organization or your particular audience. Some audiences really love crazy colors, other ones want a grayscale background. I would love if we had some universal set of design principles to go from, but the main part of this is really just thinking and being more aware of the visuals that we're creating. So let's jump into Data Lab and get started with it. So there should be a link in the chat for a to get a copy of the workbook and to be able to work along with it. And when we go into Data Lab, there is a. You're presented with a blank slate and it's legitimately asking, how can I help you today? The first thing we want to do and if you've used the copy link, you'll have the data set already preloaded. So the first thing we want to do is we want to tell Data Lab, hey, let's select the data set. So it's called Ticket Volume xlsx. So it's an Excel file and we want to ask tell me a story about our data set. This is a pretty open ended prompt. It's not something I would usually do right away. And I'm getting a note that I need to text size up. One moment please. Hopefully that looks quite a bit better for you. So when we ask it to tell us a story about our data set, it goes and summarizes it pretty well. Our data set is just a simple one broken out month by month, January through through December of ticket volume received, ticket volume processed. It's not telling us a lot of information. So let's, let's give Data Lab the additional context and let's see where Data Lab begins to take us with it. So we can say in May, two team members resigned. Was there an impact on the amount of tickets we were able to process as a result of their departure? Notice I am talking in very conversational English. I am not using hey, make me a graph about blank. I'm not asking for very specific things and the reason I'm able to do that and it pulls back to that kind of secret sauce that I was mentioning is the power behind Data Lab where it's doing a lot of that prompt engineering for us and it's doing that refinement for us automatically where we don't need to sit here and think about it. And immediately when I asked, hey, was there an impact? Instead of trying to do some crazy analysis and do do whatever, Data Lab just suggested a visual and it's actually very, it's a very useful visual for us to start, and it selected the right visual and that's, that's really powerful and something that we don't see very often with a lot of our AI tools, where it has given us a line chart and it shows, okay, starting in May, our tickets received, our tickets processed, seem to be pretty close. After that, tickets received, which is our top line, started to start to exceed tickets processed. And then it just kept going. The difference kept getting larger and larger and larger. And Data Lab also automatically came up with that solution a little bit within its own bit of context. And that's really, really cool. It's that something like that would have taken me as just a general analyst, maybe five, 10 minutes. It's not a crazy amount of work to get there, but we got there in seconds instead of minutes. And we can now take it a bit further. So some other places we want to do is. So when a ticket is received and then it's not processed, that goes into our backlog because we haven't been able to close it out. So it just sits there and it keeps growing. Well, let's calculate what that backlog is so we can ask Data Lab to do that. So when a ticket is received and not processed. And one thing you'll notice, I am a terrible speller. It's just something that I'm always bad at it Data Lab doesn't mind. You'll see I spelled the word received wrong and it actually, I'm going to send this through with it incorrect. And it will get it. Well, hopefully it will get the, the right answer and it will get us to the right place. And that's, that's one of the really, really cool things about large language models and about so many other things that we've been working with, is you don't need to know exactly what to do. You just need to know that the general idea and then guide it towards that right answer. So when a ticket is received and not processed, it gets added to our backlog. How big did the backlog grow? You'll notice as it's analyzing, it's writing this code in the background and it's just creating this analysis and immediately instead of just giving us a hard number, because I asked it to, just how big did it grow? Data Lab took it and went, okay, I'm not just going to give you a number. Instead I'm going to show that we were able to stick with the backlog pretty consistently and then it just exploded where we were able to hold on for about three or so months and then In July, it just kept going up and up and up because a singular number isn't enough for us to sit here and go, ah, this is, it's just. If I told you the backlog was 210, which, oh, it was 201, you'd be like, oh, okay, that's not, that's not a problem. But this graph right here shows us that the, the backlog is just growing and it's, it's growing pretty consistently bigger. The team's not able to keep up. And instead of just giving us, hey, here's a number or here's a graph, Data Lab's giving us both where it's showing that the, it's immediately building that story for us right away. The analysis reveals that the resignation of two team members in May had a noticeable impact on the number of tickets processed, leading to a significant drop. And then the backlog increased massively. So let's, we've got a little bit of our early stage visuals, we have a little bit of our early stage context and we definitely have that narrative already starting to be built. And I do want us to be aware and keep, keep in mind of the thing that I mentioned at the beginning of. We still need to check that this is doing what we say that, say that it's doing. It's great that Data Lab is so powerful for this, but we want to be able to verify it ourselves, especially because if we are, if we are just general data analysts or business users, this report might get sent to a data scientist or someone that's, that's very, very adept at technical parts. So we don't want to just send them a long chat message. So instead we'll go to the code view. So up in the top right, you'll see the ability to switch between different environments. So we can go and we can look at the code view where we can see that it is, how it's extracting those, how it's extracting the backlog and it's using actual best, like good practices for developing out the code that it's providing. And if I was to send this to a data scientist, they would be able to verify that, yes, this is how it's being used and we want to keep it that that methodology is correct and that's super powerful because now we can take some of the more advanced analytics and move it down to a different target, but still have it be auditable and have it be something that we can trust and verify later. Let's go back to our chat so that we can keep interacting with it by going back up into the top right here and switching to our AI chat area. And we're going to continue on with our data story personally. And this is a bit of a, this is a bit of a dealer's choice kind of situation. It's up to us and it's up to whoever's telling the story to figure out which visual they want to use for their, for, for their background. I actually find that our first visual that we referenced up here to be significantly more compelling. It's something that we can work with quite a bit more than the visual that you see down below, which is just kind of line goes up, chart. And it's mainly because whenever I've been dealing with sharing out these, sharing out these kinds of insights, it's been something that's been. You want something that looks a little bit complex on the surface, but actually isn't that complex. And if I was to try to data storytell this backlog growth over time, I think whoever we shared it with would just go, well, yeah, of course the backlog grows over time versus if we used the chart above, we can develop it out and say, ah, yes, our backlog grows over time. And we want to, we want to show kind of the, the battle that went on of in the very beginning we were able to keep up with the tickets and then, oh, August came, the backlog opened up and then after that we weren't able to keep up with it at all. So the team valiantly fought on with it versus, Yep, line goes up. So that's part of the, part of the data storytelling toolkit where you'll figure out what resonates with your audience. And in this case, I think we should be using that first visual that we worked with. So let's, let's build off of that one. So I'm going to ask Data Lab to graph. So let's graph tickets received versus tickets processed. And again, I still can't spell the word received. And we'll see how it generates it. And you'll notice that tickets received versus tickets process. It's back to our original visual and we want to start building this up. I mentioned the three main areas that we wanted to focus on. The first one was we want to declutter. So when we look at this visual, what elements can we removed? How can we make it better? The second part is we want to focus the audience attention. And then the third part is we wanted to think like a designer. So for the decluttering aspect, when I Look at this. I immediately see a few things that jump out. To me, the main one is the grid lines. I don't know why, but grid lines, they just add a lot of extra load to our visual. So let's ask Data Lab to remove the grid lines. And we're not giving it a ton of information here. We're not saying with the previous graph in mind, please remove the grid lines, we're just saying remove the grid lines and it gives us a smoother visual that's easier to just parse right away. And some of you might be looking at this going, oh, but I like using the grid lines. I like being able to cross reference things for standard charts and standard visualizations. Grid lines are super helpful, but we're thinking from a data storytelling context. And in a data storytelling context, I don't want people to sit here and have to go across and look up and check different numbers. I just want them to be able to feel what's going on with it and know that, oh, the backlog has exploded between these two where this data points 156 and this data points 202. And I'm also highlighting a pretty cool feature of Data Lab here as well, where these graphs are interactive while you're building them, because the package behind the scenes in Python that it's using generates interactive graphs. So as you're doing this, you get that fully interactive experience, you get all of the power, all of, of Python with none of the requirements to code with it. And I really, really like it. So what other things can we declutter on? This. And this is actually a case of where the test runs of this are different than the actual cases of it where when I did this in a few different test runs, Data Lab automatically added the little points on each, on each measurement along the way. And it actually hasn't here. And I find that to be really, really interesting. So for the interest of completeness, I'm going to add them so that we can remove them. Just that you can see how better it is in this current state versus how we had it before. So I'm going to ask it add dots at each observation and we're going to see how well this generates it. And this is something that I believe when you would do like your standard Excel output or your standard like report in a lot of different reporting tools, you'd get one with all these little points and all these little bits of information and it to some people that is useful, to a lot of us it's just extra Visual noise. So I'm going to ask, ask Data Lab to remove them. So, so remove the dots and it will go and just remove them. And the thing that I find so interesting. And if for any of you that did attend our ChatGPT version of data storytelling or decide to go back and watch it so that you can compare and contrast it, I would really recommend that you go back and watch it. Of just the speed that we're dealing with here, when I would prompt something in ChatGPT or other AI tools, I would have to wait a significant amount of time and I have to fill a lot to make it seem useful. Right now it's answers at the speed of typing. If we wanted to do it with no preamble of me talking and going through it, we could have done this entire thing in about two minutes. So let's continue on with it. What other things can we remove? Well, actually we're pretty good with what we have right now. Let's start to focus the audience attention. And focusing the audience attention is actually kind of counterintuitive for what we. For the decluttering aspect. Because what I want to do here now is I want to highlight that and show on our graph why you start to see the difference. So I want to add a horizontal line. Horizontal line at May and add an explanation that two employees resigned in May. And we'll see how, how Data Lab handles this and what information it adds onto it. And this is yet another case of where we have some pretty interesting results. It added the horizontal line as I asked for it, but it chose it to be a red dotted line, which is a little strange. And we'll refine that a bit more when we think like a designer as well as it put the legend for that line straight through the middle of it. Not the best place for it. So let's ask it to move it. So move the legend for the horizontal line to the right. We just want to move it to the right so it's not just top dead center of it. This is something that if I had to do it, it's actually, it didn't actually get it exactly right. So we can dig into the back to the code itself. But I was going to say if I had to do it in code, it would probably take me a few, few minutes. Check a help document. Cause it's not something I do all the time. But let's go and look at that code portion because even though you don't have to code, we still do have the ability to check it out. And, and really refine what we're doing with it. So we'll go, we'll look at the code and we'll see that, okay, so it is the, the text equals two employees resigned show, arrow equals true and the arrow equals one. So we're adding an annotation there and we haven't given it any, any direction about where to go or how to move it. So what I'm going to do now, and hopefully this doesn't break everything. And if it does, that's the wonder of live presentations. What I'm going to do now is I'm going to use the AI feature within our coding portion, not the AI chat, to try to give it a more focused nudge. So I'm going to say move the annotation to be right. We're going to say to the right of the line. And what the AI assistant does is it's going to go and parse the whole code and it's going to sit there and it's going to go, oh, and an interesting thing when we did this, it decided. So it has made the changes, but it hasn't actually run them. This is a case of where we need to make the, we need to run it. So it's currently pointing off on June and not in the right spot. Well, we can play around with it a bit more, but we'll go back into the chat and, and then try to do some more fine tuning of it as we, as we get into that thinking like a designer phase. So let's go back into the chat and, and just say, what other things do we want to do to focus the audience's attention? And when I was looking at this, one thing you noticed me do quite a bit is I was mousing over the tickets process in August and I was mousing over the tickets processed and tickets received in August and I was being like, oh, the difference here is pretty big. And for some, for some of audiences it might be useful for us to add the numbers to the data points themselves. And I know you're going to say, but Joe, didn't we go and remove the grid lines because you didn't want people to look at numbers. And that's where data storytelling is a bunch of interesting trade offs here or there. My goal is to focus the audience's attention on the differences and to really bring, bring them in on that. And a great way to do that is to put the numbers in their face with our new interestingly placed line. So what I'm going to ask Data Lab to do is add values to the data points that occur after May, because I just want to add values to those data points and see what they look like. And. And it goes. And it does it where it's saying, okay, so now instead of having to mouse over August, I can just immediately see, okay, August is 156 processed and 202 received. September is 160 received and 126 processed. Where we're starting to highlight the fact that we're showing these numbers is to get into that idea of. Here is the difference in. In our output and here's the difference in what we're. What we're messing with. So we have a graph here. It's not the best and it's not the best visual for it, but we focused the audience attention on just showing, okay, we're tracking pretty good. So close. We're not even going to show the numbers. And then once we get to that inflection point where we had two employees resign, we're going to then move to. We're going to then start showing off the numbers themselves. So you can see that in the beginning in June, we were pretty close. Like, we were within nine tickets. And then July we were. We were, sorry, 11 tickets in June and then July we were within nine tickets. And then we just. Everything went crazy in August and we haven't been able to keep up ever since. But the graph needs a little bit more work, so let's start thinking like a designer. And one of the easiest things to do, and it's going to sound a little silly, is let's increase the line weights a little bit. Let's make this. Let's make the graph look a little bit more presented. So we're going to ask Data Lab to do that. Let's increase the line weights. And what I mean here by increasing the line weights is I want a thicker line and I'm using a bit of specific terminology here, and Data Lab gets it 100%, but we also could give it more general instructions and we're going to see what happens here. So make the lines even thicker. We're just going to see where it goes and hopefully it's going to do it. I don't think thicker lines are going to help us very much here, but let's see what. Let's see if Data Lab can handle it and how it. And how it goes after that. It's taking a little bit of time to think here, so hopefully I didn't blow it up. But that is the nature of a live Demo versus a shortened one. So we're going to go over to the code area and see where it's going. Okay. So in the code you can actually see it. It's. It's writing, right? And writing. And then we can go back over to our chat and we'll see. Ah, yes, even thicker. So, yes, these are some super thick lines. I wouldn't recommend going this wide, but it's more of a personal preference thing. And the think like a designer phase because. And we can actually see visually the difference between them by scrolling up. And those lines look pretty nice. These ones look a little anemic. And then these ones are a bit too thick. So we're just gonna say actually, actually, never mind, not that thick. And we're. Instead of sitting here and trying to engineer some crazy prompts, come up with a lot of information. And to all of this extra work, we're just talking like a human would to datalab and it's giving us information and it's giving us a little bit more. So these lines are back to our kind of in between our original thicknesses. I actually don't mind it, so I'm going to leave it as is. But we need to continue to think like a designer a little bit more. And one of the interesting ones here is the numbers that we've added to these data points. They don't look the best because it's just randomly anchoring them and saying, okay, July is what it should be on top of it. So let's ask Data Lab to move a few of them around. And I'm not going to spend a ton of time on this part where we're going to ask for ask to move. Okay, let's move July up, July received up and. And July process down instead of just going to focus on September, because September intersects with that line in exactly the same way both in received and in processed. So I'm going to say move the data points to the right for September, where we're just aiming to move it just a little bit to show that we can. And then I'm not going to poke the beast and try to sit here and do each and every little bit so it doesn't again, look exactly perfect. But we can spend a lot of time tweaking. And tweaking is easy instead of being something that I would have to sit here and either look up help articles. If I was in Excel, I'd have to check reference documents if I was in Python, and I'd have to really figure out exactly how to do it. There is one other element that I mentioned here, and it's something that we want to bring in is this idea of, of the workbook. Because right now we have this long kind of chat history, and if I was to share this with someone, it's a little strange to share, hey, here's the chat history I had with the, with the LLM I was working with. So instead I want to go over to our. Go over to the coding portion and go over to the notebook functionality. So when it says code up here, don't be as afraid of it, because as the name implies, you only need a code if you want to. Instead, what we're going to do when we go in there is we're going to round out this data story and we're going to spend a bit more time leveraging the functions of datalab to really build something cool. We're going to go into this coding section and what you'll see here is at the top, this is the coding cell that we've been doing a lot of the manipulation work all of today, and it keeps all of that information. And you'll notice some cool things in here. And I find it very fun. Is the. When we asked it to shift for September, what it did was it said, okay, if row of month equals September, shift it by 20. Well, we don't have to program, but. But we could change it to, let's just shift it by 10 and then run that instead. And hey, that looks pretty good. And you'll notice when we had that, the interesting challenge with our annotation up here, we don't have that same functionality here, and we don't have that same X shift in that ability. But we might be able to change it because it says y shift equals 10. Well, maybe we want to change that to y shift equals 50 and just do some, do some tweaking with it and. Whoa, no, that's definitely not. That's definitely not right, because that's not changing the thing that I thought it was going to change. And we, we do have that ability to sit here and iterate on it. So I did a keyboard shortcut there, which I shouldn't have done, where I did control Z to undo that change and that changed the Y shift back to 10. And then we can run it again by pressing the run button. And you'll see this is. This all looks nice and happy, but we're not here to code, we're here to talk. So I'm going to hide the code. And what we want to do now is let's add a new cell. And when you add a cell, you have quite a few different options. You have text, Python, SQL and then also charts. We're going to add a text cell. Because what I want to do is throughout our whole AI chat, it was telling us, hey, the analysis shows that there's. The number of backlogs remain stable. However, starting in May, significant drop. That's great. It's actually so great, I'm just going to steal it. So I'm going to right click, I'm going to copy it, I'm going to bring it on over to our file and I'm going to put it in here. And what this is going to allow us to do is when we, when we take this report and we take this story, we can just share it with someone using our little sharing button here. And when they see it, they'll see our report with the visual, which if you are sharing this out, please spend a little bit more time thinking like a designer, like I said, and figure out how to move this one just right and also fix those data points. And they'll also see that that extra bit of narrative here, which is our, our first little, the little paragraph that we stole from the chat. But this narrative isn't the best. It's not something that we want to, that we want to just present as is. Instead, let's spend a little bit of extra time refining it because. And I will take you back to our definition that we used at the very, very beginning of our presentation. Data storytelling is combining narrative, visuals and context for a specific audience. So we want to take these kind of general feedback and create and direct it more towards a specific audience in mind. And Data Lab, actually, the AI features in it can save you quite a bit of time on it. So what I'm going to do is a lot of people think, hey, I can only run AI on code cells, I can only run AI on SQL cells. Well, you can actually run AI on any cell. So if we click our little AI button over here, we can say rewrite this for a business executive. Because if I take us back to our original outcome here is our goal was to help justify hiring, backfilling the two positions that were, that were full, that backfilling the two positions that opened up back in May because it's having an impact on the business. But we haven't quantified that impact and we haven't put that in our context here. We're just saying that the team's reduced capacity is leading to Issues. Well, we want to rewrite this for a business executive and highlight the brand risk associated with not being able to, to quickly close tickets, causing our customers to churn. So what I've said here is I want to rewrite this brief little summary to highlight the, the brand risk or the, the operational risk that comes from not being able to close out the table tickets quickly. And it's going to cause our customers to be more likely to churn, which means to unsubscribe from us and to potentially go subscribe somewhere else or, or just some, some challenge that we're facing here. And when we do that and we hit enter, you'll notice our AI assistant goes and attempts to, attempts to make the changes and it'll actually show the, the difference between the two areas. So red was what it was before, green is what it is afterwards that you can go and you can audit and figure out, oh, is the changes something that's actually useful for me or are the changes something that I want to say, no, I don't like this. Let me reject it. Because all throughout all of this, the data lab and the whole product is learning. So if you're using this product and you're, you're saying, hey, this answer stinks, you can give it a thumbs down. I'm not going to do it because I don't want to mess up our data here. And you can tell it, hey, I don't like this answer. Instead I want it's, it's not useful for us. We're going to accept it. And when I accept it, you'll notice something interesting happened. It gave us a pretty, it gave us a much better, much better summary. It put in action required where it says to mitigate the risk, it's imperative that we address the capacity shortfall immediately. That's like, that's a pretty strong wording here. It then goes on to suggest other ways to do it because maybe we want to hire additional staff, maybe optimize current processes, leverage automation. And this is where I take it back to that original idea of we still need to be that human in the loop. If I am a manager here, and I know that we've already done as much process optimization as we can, or we've done a lot of automation and other things, I need to go and proofread this and edit it and change it because GPT and Data Lab and all of the LLMs in the world will only do as much as you tell them and they only give you kind of generic answers for it and it's up to us to really tune it but and also a really cool thing that this is doing and I'm not sure how many of you have played around in Data Lab's text functionalities is Data Lab leverages something called Markdown. And what Markdown does and you'll see the formatting up here is it goes and it's the way of formatting things like you wouldn't Word. If you want to bold something, you want to make a, make a title of some case and right now you can see that it added these little asterisks and when I hit this view button you're going to notice that it has bolded those, those things where it's, it is shown a little bit more information and it's given us just that little extra bit of special sauce on top of it. And we can, we can do that even more. We can say okay, add more Markdown to make the sections clearer and easier. I can, I need to spell the word easier. Easier to parse where I just go and tell Data Lab, let's, let's, let's lean harder onto the Markdown functionality. And for those of you that again are well versed in Markdown, what you'll notice here is it's decided to swap out. Instead of using Bolding, it's now using headers. So I'm going to accept it and then I'm going to view it and I'm going, oh, this does look a bit better. So we're using that same idea that we used on visualization within our text and we're doing it entirely through Data Lab's built in functionality. Because if you didn't know Markdown or you didn't know how to format this or regenerate something, it would be challenging. You'd have to go out there and read a bunch of things and try to figure out a bunch of information on it. But instead we can just use the AI within the tool to be able to take us straight from raw data all the way to being able to share a report that when someone looks at it, they just get it the visual right away and then the analysis with the brand risk and customer churns and then the actions required. This is a good start of a data story. If we were going to send this out and really highlight it, I would definitely emphasize us spending quite a bit more time on the acting like a designer phase. Like I mentioned earlier of we'll tweak this a bit more, figure out how to really get datalab to move that over, but start to finish, we could have done this much, much quicker than I ever could have within a traditional programming tool. And just anecdotally, Data Lab is something that I use every day within my role here at Data Camp, generating curriculum. It is remarkably useful at bridging the gaps in your understanding and also teaching you new ways to do things. And that's kind of crazy to think about this idea of learning by just interacting with an LLM, but it's, it's super useful because you'll. You noticed that when I was going and looking at the code behind the scenes, I will tell you, and I'm going to review the reveal the code cell again here by hitting a little show code. So I changed the code area, I hit show code. I will tell you, I have never in my life seen this line with Y Shift or anything. As someone who has done a little bit of programming, I have never, ever, ever done that before. But since Data Lab was able to generate it, I could then go and play around with it, change a number, see what happens. I hit run and it didn't do the thing that I wanted it to do. But I've now I've learned, oh, that's what that does. And it's fun for tinkering all the way to finished reports. And that's really why I love Data Labs so much and why I want to highlight it as a great tool to add to your data storytelling repertoire. Because you can take something all the way from raw data, which is what happened way back in the beginning of our AI chat here. So I'll scroll all the way up to the top here, all the way to raw data, to a finished data story with a little bit of extra tweaking with minimal amount of coding and none. If you really want to spend enough time and you only code exactly when you want to, and you can become a better programmer, a better analyst, all of that by just leveraging what's already within Data Lab. And I want to spend a little bit of time here at the end to highlight kind of that, that bit that I mentioned about how different Data Lab is from using just traditional LLM tools like ChatGPT and other ones. And I'm actually going to bring up the analyst that I use, the conversation that I used from our, from the previous webinar and just show the difference in the chat histories that we came across. So we've seen this one, and I'm going to bring this one over for us, which is our, which is the, the data, the storytelling session that we had before. When I uploaded the ticket volume data set I asked it to tell me a story about my data data set and GPT. It went on a very long winded story. So even though the, even though we had a great even though this was called Data Analyst and we were in that custom GPT, it didn't actually do a lot of that analyzing that we would expected it to do. When we were within Data Lab it immediately when I asked it to tell me a story about our data set we which frankly is a very weird prompt to throw at it. It just gave me that right information and immediately took me into doing that analysis versus the more open ended just it literally wrote a story about our about our data set and then it took quite a lot of battling for us to get to a graph. GPT's first version of a graph really wasn't that strong. It had a lot of weird elements that it put in and I should size this one up as well. It really wasn't that strong. It had a lot of extra visual clutter. It threw in so much extra effort versus Data Lab's precision made implement for doing analysis. So while you might have used GPT for a lot of data analysis, I really would recommend give Data Lab a try. It's so much more useful and can really get you where you want to go. So much easier than previous tools and previous implementations of tools because it's that that synergy of the prompt engineering behind the scenes that makes it just a joy to use. So I'm going to go back into the presentation. I've got like one slide real quick and then we'll do some some Q and A portions. So where do we go from here? I've all I've mentioned quite a lot of of of places that you can go if you wanted to go and and play around with Data Lab more very much recommend it. It was a great tool. The AI really makes interacting with it super super useful and it's something I frankly can't work without at this point. As well as if you're really interested in more of that data storytelling aspect, I do recommend our data storytelling skill track. While I referenced the book Storytelling with Data quite often today, we do have a dedicated curriculum that expands on these ideas and takes them even further. So you can go all the way from communicating data insights straight into case studies at the very very end with two very very useful hands on case studies. So I do recommend you checking out our curriculum quite a bit because it's something I'm Very proud of. You'll see my name on the first two there as well as just play around with a lot of the AI content that we have to offer. So with that I will turn it over for what questions can we answer for you? I've spoken quite a bit and I would love to hear more from all of you.

All right, super. Thank you so much, Joe. We are running quite close to time. I should have known when I asked Joe to talk about storytelling, the session was gonna run long. So yeah, we got tons of questions from you and for anyone else in the audience you've got questions for Joe, then please do ask them in the chat now. So we're going to run a bit over time. If you've got a jump, then please do catch the Q and A session on the recording. In the meantime, before we go to audience Q and A, I just want to talk about some of the upcoming events. So tomorrow we've got a session on building high performance data and engineering teams. So anyone who's a manager, actually I think the advice is gonna be good for if you're like a, if you're running a data science team or a data analysis team as well, that's good for you on Thursday. Next week we've got another code along on Tuesday. So it's Data Camp's chief technology officer. He's gonna talk about how to use BigQuery. So BigQuery's a large cloud database. We actually use it here at Data Camp. That's my good session. And then if you want to get a job as a data engineer, we've got a session next Wednesday, so please do sign up for those. All right, let's get to some of the audience questions. So Christiane asks, are there any built in tutorials or resources within Data Lab to help beginners get started? Yeah, Joe, I don't know if you want to share your screen again and just talk about or just show the built in data sets that are available.

Let me share.

Yeah. One of the tricky parts is deciding, well, you know, where do we get a data set to get started analyzing stuff. So on the left hand side of Data Lab there's navigation bar and then there are some sample databases to use and some sample data sets. There are also some template workspaces as well with just some example analyses in there just to help you get started. Sometimes there's a thing authors have called the tyranny of the blank page, where he's not sure how to get started. And the same thing can be true with data analysis Data Science as well. Just that blank notebook can be a bit scary sometimes. So we do provide some resources so you can sort of get started.

Joe and one other area I would recommend for resources on figuring out how to get started is our competitions area. So competitions leverage our Data Lab backend. So if you're ever interested in learning more about how other people are doing things, competitions are a fantastic resource to learn about it because part of it is a community judging aspect. So you'll see how people interact with datalab and the challenges that they're interacting with every day.

Absolutely. I definitely second take part of those. All right, so question comes from, so what if a manager did this and didn't bother checking, go straight to decision mode.

And that actually is a, a big challenge that we see within, with, with the rise of LLMs is people putting too much emphasis on that the LLM is right and I don't want to verify it. And that's why I really, really like Data Lab over a, over a more like ChatGPT style way of analyzing data because Data Lab is auditable. Like it generates a workbook at the end. So while I would love to be able to tell that manager, hey, you've, you're, you're wrong, you need to go back and check it. You need to check your work and all of that. Once the manager starts circulating that and using tools like Data Lab, we are able to then go, oh, hey, there's an issue here, let's fix it and, and let's, let's take it that next level because while we are increasing the speed of decision making, we're not getting rid of decision making entirely with this tool.

Yeah, I think also worth checking is if it's something you've written yourself and you give it to your manager and your manager goes straight to decision mode, you should probably have checked it yourself before you gave it to the manager. If something the manager's doing on their own, then yeah, maybe there can be a problem. But the more, the bigger the impact of the decision, the more you have to check it's correct beforehand. Like if it's just a kind of almost disposable analysis where it's like very minor consequences, it doesn't matter too much if you get it wrong. And speed of getting to the answer is probably most important thing. If this is like a live or die situation for your entire organization, then yeah, you're going to want to double check everything for you. Make a decision. Oh yeah. So there's a, a thing from Jeff Bezos, the former Amazon CEO about type 1, type 2 decisions about whether they're reversible or not. That's further reading for you. I think that's like a whole separate webinar on decision making, but definitely worth looking up. Okay, so next question from M. How do you go about reducing the velocity and, and how do you adjust this to the level of the user, whether you're beginner or advanced?

So to reduce the kind of how verbose it is and how, how crazy it is, just ask is pretty amazing how well it'll work if you tell it to respond only in a sentence, only in two sentences. I know when GPT and LLMs first kind of got on the scene, one of the first things that people did was ask it to write like a pirate and all that stuff. And it did very, very well. But that same idea and that same kind of engineering of like, let's make sure that we want to guide it applies for targeting audience levels. Targeting, targeting how verbose they are targeting everything. If you want it to be, hey, I only want a one sentence summary here. It'll give you a one sentence summary and then it's up to you to do that extra tweaking on it, to really build it and to take it back to Richie's idea of the tyranny of the blank page. That's where Data Lab and LLMs are so useful is being that blank page breaker and getting something on there so that you can say, no, LLM, you're wrong, I'm better than you at this. And it really helps you get to that next level.

Oh, definitely. I do that a lot when I'm interviewing people, tending to come up with a list of questions beforehand and often ask asking that and be like, no, I hate those questions. But that will give me some inspiration of what I actually want to ask about. Okay, so question for Charles. Can we have web scraping webinars also?

Yes.

So I'm not sure web scraping is a whole webinar in itself, but we can definitely do some sessions where we download data from a web page using web. Actually, this is a great example of use case for LLMs. I can never remember. Beautiful soup syntax, very confusing, but GPT is very good at it. Apart asked what kind of algorithms are used in Data Lab?

Yeah, go on, Joe. So I, I won't, I won't say that we can definitely narrow down a specific one, mainly because the product team is constantly iterating on it. Like I mentioned in, in the, in the webinar itself. When, when you give that Thumbs up, thumbs down. Someone is going to look at that and someone's going to say, oh, this didn't work exactly the way that we wanted to. So maybe we need to bring in a new tool or bring in a new way of processing that data. So any information I give you now could be outdated by tomorrow if a ton of you go thumbs up or thumbs down something. Yeah, actually.

So Philip, who is the VP of engineering for the datalab team, who did the session yesterday, he did cover a little bit of the architecture of how Data Lab works and how it interacts with GPT in the background. So please do check out the recording from yesterday. It came towards the end of the session. So yeah, somewhere around like 45, 50 minute mark I think. All right, so next question. Question from nilabja. Do I really need this kind of experience to get a job as a data scientist or data analyst?

What should I do for, for this? I, I won't tell you. It's absolutely required to be an expert prompt engineer. But I can tell you it definitely won't hurt because knowing how to work in, in the space of Data Lab or all of the AI tools and just being proficient with AI or whatever's coming next will help you be more resilient as a data analyst or as a data scientist and that will translate into better outcomes for yourself as getting a job as a data scientist. I will tell you, before ChatGPT and everything came on the scene, I didn't know a ton about large language models and other things, but I was able to take my knowledge of how I interacted with it and how I interacted with other, other analysis tools to directly apply it. And that's, that's one of those things that is super. It's not well covered in the data scientist and data analyst world of just how cross functional a lot of the skills that you learn are and AI literacy and data literacy is one that is, is becoming even more important. Cross functionally.

Absolutely. So thinking about like a long term, your career long term, give it five years, everyone's going to be using AI. So this is a skill you can need for the long term. One thing I will say though, so I do a lot of hiking and there's this saying in the hiking community that technology is great until you don't have it. So you go out, yeah, you're in the woods and then you drop your phone in a creek, suddenly you can't use your GPS or you know, you go out of signal range. You need to know how to read a Map. You need to know how to use a compass or ideally have some clue of which trail you're on and where you're going to. So if you don't have access to AI, and a lot of companies are blocking these AI tools just for data security and privacy reasons. So you do need to be able to write the code as well yourself for the occasions when you don't have access to this. All right. Okay. This is a. It's a hard question. It's hard to find a job nowadays. Do you have any suggestions?

It ties into the. The previous statement and it. It's the. The biggest thing that we're seeing in the analyst space is that it's. That it's. You're. You need to become a jack of all trades more and more than you used to be able to, as well as you need to have a little bit of industry knowledge for some areas. And I know that's kind of a bit counterintuitive is when people out there are going and looking for like a data analyst or something, they really want you to be able to attack every part of the life cycle for their specific use cases. So my biggest recommendation is if you're looking to break into an organization that does like, say, healthcare market research, it's a field I used to work in quite a bit. Spend time researching the specific analytics for that area. So you can speak to that because it's so useful and it's so powerful that we're seeing the analytics mature everywhere, that you do need to have some background knowledge on it. And we make sure to address that in our curriculum where you can go and pull up case studies around specific, specific use cases, projects around specific use cases. So you can get exposed to that. So you can be more targeted in your search.

Yeah. So there are two ways that you can demonstrate that you're good enough to get a job. So the first way is having a qualification. So for example, you can take one of the data Camp certifications. We have certifications for data analysts, data scientists, data engineers. And that's giving you that piece of paper to say, I have these skills. Second way is to show off that you can perform like full projects through having a portfolio. And datacamp also provides the portfolio facility. So there's a portfolio feature within your profile. So you can say, okay, this is the work I've done. These are things that I can talk about, like in the job interview. Look, I can do a real project. So certifications and portfolios, they're the two things you need. We regularly run Webinars on both topics. In fact, I think these are both going to be covered in the session next week on how to become a data engineer. So come back next Wednesday for that session. All right, so three more questions. Do quick questions, then we're going to wrap up. So is DataTube free when you're subscribed or is it extra payment? Joe, don't. Whether you want to take this or whether you want me to.

So datalab is, I believe there is a free tier to it which has some limitations on it and then there is a subscription that you can then go to a more unlocked version of it. So I don't know the exact specifics on it. Maybe Richie, you know the. If it's like five prompts or something.

Yeah. So Data Lab, if it's one of these code along sessions, you get it for free. If it is one of the projects of datacamp, projects are powered by Data Lab, so you get those for free. Data Camp competitions I think might be free, but otherwise on the free plan, you get three different Data Camp, sorry, Data Lab workbooks, paid plan, you get unlimited workbooks and you also get four times the computational power. So four times as many processes, four times as much memory. Okay, so one file here. How would Data Lab handle multiple files and their relationships?

That's something where Data Lab doesn't even skip a beat with it. And I will lean on my, my application of it within internal Data Camp processes of within Data Camp. We have Data Lab connected to all of our source databases and I can ask it, hey, how many customers did XYZ thing? And it will find each specific table and link it all together for me in a query and build it out. It doesn't bat an eye when you start to bring in multiple things. And that's, that's on purpose because it's meant to be an enterprise tool that really delivers, delivers all of that information.

Absolutely. Yeah. It just works if you, if you're dealing with like thousands of different files with different formats, that's going to break the LLM side of things. But from Data Lab side, it's. Yeah, it all works. All right, so last question from Mohamed saying, can we use Data Lab for commercial project and how much to pay for it? Yes, you can go to datacamp.com pricing. You'll see how much it's going to cost you there. But yeah, we have tons of business and enterprise users using Data Lab already for this sort of stuff, so yeah, definitely possible. All right, excellent. With that, we are long overtime, so I just want to say thank you again, Joe. That was really brilliant stuff. Yeah. Such a fun like wave to create a data story. So, yeah, a lot of good. A lot of good stuff there. Thank you to Rhys. Moderating. My voice is starting to crack. I'm talking too much. Thank you to everyone who asked a question. Thank you to everyone who showed up today. Hope to see you all again tomorrow or next week.