Data Storytelling with ChatGPT | Share Your Insights Effectively

Today we will be talking about Data storytelling with ChatGPT. I, I've got some slides that we'll use to kick this off and then we'll get into the, the prompt along and we'll see what we can do together with ChatGPT and really unlocking the value of store data storytelling. I have my normal introduction slide for myself. Like, like, like I was introduced. I'm the head of data literacy curriculum here. I asked ChatGPT earlier today to give me a, a bit of a different title because it's a bit of a mouthful and it came up with Data Whisperer Extraordinaire. I think that it's a pretty fitting title and I think I'm gonna start using it more often. I really enjoy telling people all about the wonders of data and the wonders of storytelling with data because you can do a little bit with just doing the analysis, but getting people to really jump on board and buy into what you're doing with data, that that's where it, that's where the, the magic happens. And that's what we're going to talk about today and we're going to talk about that through the lens of ChatGPT. I, I know the early kind of prompt was where you, where you call it in from, where you joining in from? I'm personally located on the east coast of the United States in Pennsylvania. It's a little snowy here. I saw someone was calling in from Mars apparently. So that, that's a pretty interesting place to come from. And hopefully we can all get some value out of this together. So let's dive into it. What is data storytelling? I know it sounds a bit silly. All of you registered for this session, so you should know what data storytelling is. But it's actually one of those things that isn't a solid consensus out there. There's so many different ways to approach data storytelling and approach what it means to tell good stories. I put up four definitions here from the various experts in the field and they all really try to unlock different things and different bits of knowledge all around data storytelling. The definition that I really like to grab onto and I really use quite often does come from storytelling with data. But it's so interesting how, how many different people just hear the phrase, oh, I need to tell stories with data, but they don't really know why or they don't know how to do it. And that's what we're going to get into a bit here today. But at its core, data storytelling really is the combination of three key elements. Visualizations narrative and context. And the goal of data storytelling is to just share whatever insights you have with a specific audience. And that's what we really want to aim to do. And ChatGPT has unlocked the ability to do this in a way that really was out of reach for a lot of us. I know when I first started doing data storytelling on first really started getting into it, one of the biggest challenges was you had to sit there and manually customize so many different charts, so many different visuals, and it just was extremely cumbersome. And GPT and generative AIs has really allowed us to get back to talking to our data and talking with our data. And that's how we can make data storytelling more accessible for a much wider audience. But why is data storytelling so important? I'd be remiss if I didn't talk about this. It's really for three key areas. The first part is data storytelling helps inspire action. We're living in, in a world that is suffering from analytics overload. Every single day, you probably look at 5, 10, maybe 20 dashboards, depending on what organization you work in, or even if you check where a Fitbit or look at any bit of data all around you. I know sitting right here, just off screen, I've got my phone, a doc, and it's telling me what the weather is when, when lunch is just all of this information that's coming out here. And numbers, they just get kind of lost in, in the din. And we can break through that with storytelling. And I, I like to bring up an interesting example here, and it comes from Stanford University where they had, they had college students go out and share, share stories, share data, share just a presentation with each other. And when one group used data storytelling practices, the, the way that the data made the audience feel was remembered about 63% of the time. When another group used just standard presentation tactics, not data storytelling, people could only remember the individual numbers about 5% of the time. People forget numbers, but they remember how the story made them feel. And that's a very, very powerful part of what we're aiming to do here. And the goal of data storytelling is to really ease that, understand, ease understanding and bring people into your story without just leaving them behind. And at the core, telling good stories starts with understanding your data better and becoming more mindful of the data that surrounds you. There are so many different opportunities for us to capture data, ingest data, play around with that data, and it's getting easier every day. But we need to do it in a meaningful and measured way. So what about storytelling in the generative AI era? I bring back this definition that I used at the very beginning or this little quote around data storytelling from the very beginning. But because this quote was originally From I believe 2017 or 2019, I forget the exact date. But it really understates the issue that we're facing right now, especially in the generative AI era. So what it describes here is when you have all the information in the world, it's so much harder to communicate because you don't know what's important, you don't know what's useful and you don't know how to, to how to react on it and how to really make changes and make decisions on it, and even adding generative AI to it. And you're going to see this during our prompt along today, assuming everything goes as planned as we go through it, we're still going to be battling with that kind of information overload and we need to figure out effective techniques to grab onto our data and get us to where we want to be. So for some reason, I guess I had a slight animation. So let's explore the definition a little bit that I put out there. And then we're going to use this definition to get into our prompt along because I can present at you for probably hours about this, but you're here to actually see us do these things. There's three key elements and we're going to have GPT help us through each one of these. The first one is the narrative. So the narrative is the central message that the story is trying to guide the audience to understand. And basically when we go out there and create a data story, we're trying to give the numbers and give our data a voice. And what we're not trying to do here, and it's actually an interesting thing, is we're not trying to remain purely objective. Our goal here is to guide the business or guide some organization or something towards rewards a outcome. So we want to make sure that we, that we support that with our data and we, we lay out why it's so useful to make whatever decision we're aiming on aiming to make. The second part is our visualizations. Visualizations are the key to a great data story. We're going to take a, take a crack at making kind of a basic visualization for this. I would definitely say we want to do a little bit more fine tuning towards the end, but visualizations help us digest the information and share it much, much, much more quickly. And then finally this last part is context. Context provides a broader picture and tells us how this story ties into the rest of the world. This is actually the area that GPT struggles the most with because you'll see as we bring in our data sets and just upload them and throw them at at GPT it doesn't know anything about our data because it's our data. And we need to either tell it that that context or give it or just add it ourselves without any inference from GPT. So let's talk about our data set a little bit here. So our data set is a sample data set on ticket volume for an IT organization. It actually comes from the book Storytelling with Data. I really recommend this book for anyone. I actually had it last me Lean. I have it myself right here. It's a fantastic book all around how to tell great things with data and I really really enjoy and recommend people check out the book just so that you can get a much deeper understanding of how to tell great data stories. So the scenario that we're going to be working on is we are a manager investigating the impact of two employees leaving the team earlier in the year. Our primary goal is to just understand that impact and craft an effective data story all around it. And we want to see well what happened. Was there any large uptick in number of tickets received or is there any uptick in the number of tickets processed? And that's really kind of the gist of it. Where going to leave it a bit open ended and we're going to jump into our kind of prompt along portion. So I'm going to, I'm going to pause for a brief, brief moment just to give people a second to a second to get the data set. It's. There was a link shared with it as well as it's within the slides as well so you can grab it. And in the meantime I'm going to cut over to our GPT instance so that we can start talking about how to, how to really play around with it a bit. I saw a question in the chat of where can we get the book? I personally have got it on Amazon locally in, in the United States but it is available all around the world. There actually is a new edition of it that was put out there so and I've seen that question. So the book is Storytelling with Data and it is from Cold Master naftlec. I'm butchering the name but if you go to our attribution it's actually the information for it is available at the very, very bottom for it. So it's storytelling With Data A Visualization is a fantastic book and I really, really recommend working through it. So hopefully a lot of you have the data set and you're ready to prompt along with us. For today's demonstration, I'm actually going to be using one of ChatGPT custom GPTs. And to get to that, if it's not already in your little dropdown box up here, to get to it, you're going to select on the Explore GPTs option and then go down to the ones that were made by ChatGPT. And we're going to be using the data analyst GPT. For some of you, this might require a premium subscription or an enterprise subscription. It varies based on regions and if you don't have access to that, just feel free to follow along and you can take these techniques and bring them over to the normal world of GPT as well. So I'm opening up our Data Analyst GPT and it, it actually, it gives us a very familiar and, and kind of easy interface. For any of you who have worked with Chat GPT, it's, it has that same look and feel. But the main difference here is we can go and we can drop files into it. So I'm going to take the, the, the data set that was provided, the ticket volume data set, I'm just going to drop it right in here and I've uploaded it. Well, I give it the opportunity, then I'm just going to hit send message. So we've uploaded the data set. It's going to say, hey, great, what can I assist you with? Our goal today is to make a data story. Let's ask GPT, Tell me a story about the data. A story about our data set. GPT is going to begin to do its machinations. It's going to look through the data set and try to find us interesting things. And it's, I know this prompt, for many of you that are seasoned in prompt engineering probably seems a bit rudimentary and that's kind of by, by default here. And the goal is to just show how even a basic query such as this one, tell me a story about the data set. It can give us a lot of information. But like we said earlier and wow, GPT has given us a lot of stuff. It's, it just kind of runs with it. And this really ties back to that quote that I mentioned from the author of Storytelling with Data of it. It's just giving us too much. It's just going to keep going over and over because I told it, hey, tell me a story. About this data, but I didn't tell it what a story, what kind of story I wanted, or how I wanted to look at this data. The interesting part of this is it's actually getting it somewhat right and we're going to come back to the story at the beginning and really try to dissect it. I mean at the end, not at the beginning and try to dissect it because there's a, it's finding the insights that we're looking to find, but it's hiding them within all of this kind of long winded the, the year end crescendo. Well, it's, we've done a few test runs of this, as you can see over here, and this is the longest response it's given me for the beginning, but it's actually moving pretty quickly, so I'm happy about that. So it's giving us a huge amount of information and frankly not much of it is useful for the central tenet of data storytelling, which is where we want to share all that information quickly and easily. Because if I sent this as an email to my boss, it would just immediately be set to set to the deletions or just marked as spam or something because this is just, this is wild. So instead let's, let's find out what we have in our data set because that's actually an interesting thing. When I uploaded it, I didn't ask any information about the data set. I just sent it to it and said, hey, what do you have? So let's go back down to the bottom here. I want to ask, describe the attributes within our data set. We just want to know what's in there. Because we actually didn't, didn't even do any exploratory data analysis. Instead we just threw it all in there and went with it. And we can ask GPT to describe the attributes of our data set. And when we looked at our data set, it's. Well, if you look at our data set, it's actually a pretty simple data set behind the scenes. But this process works for huge data sets, massive ones. If you have hundreds and hundreds of columns, if you have rows, if you have many different places, you can still go and ask GPT, hey, take, tell me about it. It's just trying to. GPT tries to give it some meaning and says this. The data set appears to be designed to track and compare monthly volume of customer service tickets received and processed throughout the year. This data can be crucial for understanding customer service workload efficiency and potential backlogs and ticket processing. So let's key in on that and if you remember from our, from our definitions, and I will go go back to this. The goal here is three main parts. We want a narrative, we want visuals, and we want context. So let's, let's start on that first one, the visualizations element, and start to see what we can do and how we can assess it better. So let's ask GPT to plot the difference between tickets received and tickets processed, just to see if there's any interesting insights that come out. So create a plot by month for tickets received and processed. And luckily my Grammarly saved me there because I spelled the word received wrong. But it's actually an interesting thing and I really do love this about GPT is you don't need to have perfect English, you don't need to have perfect anything, because if you screw it up, if you, if you say create a plute or something like that, GPT is going to go and find that little bit of understanding and really make it seem like you're talking to a friend instead of just talking to a computer. And this actually, this graph right here shows us something very useful. And unfortunately it slid off the screen. So I'll slide this back up. We just asked it for a plot of tickets received versus tickets processed. And we see something start to happen here of our data set. The lines are tracking together and then they start to deviate. But the graph itself, it's not that good. And I actually want to bring us back to our presentation area to just talk real quick about the idea of what is an effective visual. Because like I said, visuals are the key part to so much of this. And the goal of it should be to allow the audience to easily inspect the data set and find any trends. It should accurately and honestly reflect the story's insights and should focus the data store. And our goals here is because we want to pick the right tool for the job where we don't want to be plotting something. We don't want to use a pie chart if the data set doesn't lend itself to it. We want to keep it simple. And the last part is we want to eliminate visual clutter. So with those three things in mind, let's go back to ChatGPT and we're going to see what does it look like and how can we make this graph better and is it the right graph for the job? So the first part is let's figure out if it's the right graph for the job. And we can do that by just asking GPT show me other ways of graphing this data. So show me other ways of graphing this data. And we're going to see and because I scrolled up it, it won't show we're going to see different ways. And it actually suggests a few different parts possible options. We have the bar chart, the stacked bar chart, the area chart. It actually did suggest the pie chart, but it does say that is a less conventional option. And then finally the heat map. This from my tests earlier. This is going to take a little while to go. So while it's spinning, I just want to show an interesting idea and an interesting thing within ChatGPT because while we've been just prompting this directly for some of you that are that know programming and know how to do the data analysis behind the scenes, GPT is actually showing its work. If you notice this little thing right here, the View analysis button, it actually shows us the code that is going behind the scenes. And, and if you wanted to, you could take this and put it into your Python environment of choice. I would recommend datacamps workspace. And you can replicate the graphs that it is doing. You can replicate everything that it's doing behind the scenes from ingesting the data and creating those plots. And you can learn from it and refine your technical skills while you're also working on your more conceptual and soft skills. So let's see if GPT is ready to give us some answers. And it looks like it's giving us some good things. The first one is a stacked bar chart. And one thing that you'll notice is GPT is not the best at making graphs. This first one here, it's using two different shades of green for some reason, and it's really kind of difficult to see the difference between one versus the other. It's just kind of frankly strange. The next one, it gets even weirder because it's fully stacked everything instead of just showing kind of the difference between them, it's showing, oh, here's tickets received and then putting tickets process just right on top. It's not very useful for what we're aiming for. And I still, of the ones we have seen, I'm still liking the line chart that we saw earlier the most. But let's keep going. What else did it, what other options did it give us? It actually only ended up visualizing those two, even though it said, hey, here's a bunch of different options up there. And that's one of the more frustrating parts sometimes of doing this is we can only guide GPT so much at times. And that's why I like calling out the ability for us to see the analysis and see that and edit the code because you can blend these two skills together. For most of us, going directly prompted is the way to go. But for some of us that really know how to tweak things and adjust things, that's when we can find that little extra bits. So let's, let's go back to our, our line chart and we'll start kind of tweaking that one and applying the principles of data storytelling to it. So we're going to ask GPT, what did we. Did we call this line chart anything? Nope. We just asked it to create a plot by month. So I'm just going to ask it to do that again just so that it's, it's easier. So let's create a line plot by month, tickets seed and tickets processed. And the reason I'm doing this is just because I don't want anything to break and hopefully this doesn't break by doing it this way. Because I could ask GPT in a conversational way, hey, let's change the axis or let's get rid of the charts. And most likely it would get it right. But the wonders of live, not television, but live webcasts. I'm going to take a more conservative approach. So we have our graph back here and we want to start improving it and start making it easier to read and easier to understand. The first thing that I like to do and it's a thing is the grid lines here don't really give us any value. I personally never find myself going over, tracing over and it just adds a lot of visual clutter to, to the display as well as the, the difference between the tickets received and the tickets processed. It's extremely difficult from just looking at it from a colors perspective to know which one's which. So we're going to ask GPT to just make those changes. So remove the grid lines, make the oh and create a. Oh, geez. A better distinction between tickets received and tickets processed. We'll see if this gives, gives us what we're looking for. And I'm trying to avoid being overly prescriptive with GPT of telling it, hey, we, let's make one line red and make one line green and notice it didn't actually remove the grid lines. And this gets back to the part of sometimes these things go a little weird. So let's try one more time remove the. And we'll be a little bit more prescriptive. This Time, remove the X and Y and see if this gets us where we want to go with it. But while that's spinning, I want to bring us back to the idea that we had earlier of our goal here is we are managers and we're trying to understand the impact of two team members leaving our organization. And the additional context here is they left the organization in May and as soon as I say oh, they left the organization in May, you go in and you can look at this chart. Even though it's not the best chart so far, you can start to see that beginning in May, our tickets received and tickets processed begins to deviate. Where the team is able to keep up with the number of tickets pretty closely in June, in July they're also sticking pretty close. It's not the they're not achieving the same 100% clearance rate that they were getting in previous months, but after that it just starts to fall apart where the team is just overwhelmed and they're not able to keep up. And as a manager in this IT organization, my goal is to make our data story really tell my managers that we need to add more headcount efficiencies that we're finding. It's not getting us there. We're falling behind and the backlog is growing more and more every single month. And you can see it clear as day through this, through this graph. So once we make this graph a little better, that's the narrative as well as the context that we want to start bringing into it and we'll start to see what kind of interesting areas GPT allows us to, to build into that. It is still telling us that it is updating it without any grid lines, but they're still there. Thank you GPT. Let's see how else we can try to prompt this one. Let's see, make the grid lines and hopefully for those of you who are trying this at home and I would hope, I would love to going to make my chat pop back up. Hopefully some of you, the grid lines are disappearing so please let me know in the chat if it's working for you versus not working for just me because that'll be an interesting kind of non deterministic way of how GPT does things. So make the grid lines white. We'll see if we can make them go away that way and go from that. But if not, we're just going to leave the grid lines up there and just know that for making this a truly production ready data storytelling interesting we had three people, hey, making the grid lines white Works. So we cheated a bit. I saw Jennifer said that it didn't remove the grid lines for her, but then a different Jennifer said that it did remove the grid lines for them. So it's. It's very interesting how for even two different people, it's these same or relatively same prompts work, but for others, they don't. But luckily, my little trick of say, make the grid lines white actually worked because behind the scenes, GPT was able to do it. So we have it updated where it's a bit less visual clutter. There's less for us to mess with and less for us to really try to understand. One thing that I personally don't like is the utilization of the colors here. Just because we are using the color green. The color green can be misinterpreted as a good thing and can be misinterpreted as a. As just something that's. It just doesn't stick as well as you have. Red, green, colorblind. We could try to tinker with the colors a little bit more, but we're just gonna. Actually, let's. Let's try it. So make the series colors color blind friendly and don't utilize well. Actually, we'll just say make the series colors colorblind friendly just to see what GPT thinks is colorblind friendly. In my previous role in the insurance world, my lead designer and he was the guy that made all of my visuals, made everything everywhere. He always came up with fantastic things. And then, like, two years into working with him, he told me he was colorblind. I was like, that's interesting. I never would have guessed it. And it actually explains why I never have to talk with you about how colors are strange and how things are just different. And it's just a way for us to begin to work with it and understand it. So one thing that we want to start to do is let's make this visual. Let's focus our attention here. When I was giving you the context around it, I said, we want it to that our teammates left in May, so let's add a vertical line in May. So add a vertical line. And this is. Hey, did I spell the word vertical right? It looks like add a vertical line on May and add a note that two team members left organization. Oh, I spelled vertical wrong. I'm being told organization in that month. Let's see if. If me spelling it wrong gets it. GPT still gets it right. And I really, like. I sound very, very silly here of how much I love the fact that I don't need to always get it right because this is actually one of the things that really moved me into the world of conceptual versus moved me into the wonderful world of programming is in programming, when you get something slight syntax wrong, everything breaks. When you do this with GPT, you actually can get it a little wrong. While it might need to, you need to battle a little bit more. It did add this. I don't like that it added in red, but we can start to do some little tweaks with it and start to make adjustments all around it. So let's make the vertical and I'm going to spell it correctly this time. Vertical line. We're going to make it black. So make the vertical line black and change the note to be more descriptive. And what you're seeing here is I'm now using GPT to help us build out the context and try to give us just that little bit more of understanding. So I'm asking it to change the note to be more descriptive and state.

That.

The team is now understaffed. And what we want to, what we want to do here is get GPT to start just building out that context and building out our bit of understanding so that we can make a good visual. And it's going to take a little bit of time. And one thing that I really do kind of caution against as we start to play around in the GPT world is the, is the, the sense to go off and do something else. Because what happens here is every time you send a prompt, you just kind of sit there and stare at it. And I know personally, I found myself when I hit enter, I go and look and go, what's going on? And sometimes GPT is a bit, being a little slow, but a whole part of this data analysis and understanding is really sticking with your analysis and not getting lost somewhere else. And as GPT gets better and as the processing power increases, you'll see that you'll realize that you're kind of floating away from it and, and you're losing your train of consciousness with it. So it's added the little bits to it, but it hasn't added it in the right spot. I mean, it kind of has, but now it's, it's weirdly truncated. So we're just going to say move the note to the right side of the, of the line and make it break into two lines. We'll see if this gets, it gets us where we want to go. And while that's going, I want to take us a bit back into our presentation. Just so that we can talk about other elements of what we're looking to do here. So we're trying, we're to trying working on crafting an effective visual. We're getting kind of there, we have a few more elements to go with it. But the next part is really bringing in that right context. I mentioned it a bit here, mainly because this is a hypothetical situation, we don't have a ton of context to grab onto. But really the goal here is to highlight the impact of the data story on the broader worldview. So the context that we're adding, and hopefully GPT will have it ready by the time I switch back over here, is we want to show that the team dynamic changed in May and that's leading to our outcomes of having less and less capacity and less and less throughput to be able to work on things. And the main thing we want to consider when we're doing this is we want to really know our audience and know what their expectations are. If I'm presenting this to a C suite executive and they don't really want to sit here and be bogged down in the nitty gritty details, then I can just change my context. Instead of saying, hey, two people left. I can say that we're, our brand loyalty is going down because of we're under staff or some other bit of information to really know what they expect and know what to present to them. And we can use GPT to help us adjust that message for whoever we want to target it. And then the two tips that I really like to mention around here is when in doubt over, explain to a small degree because it's always easier to skip over something in a presentation or skip over something in a visual than it is to have to add it after the fact. And then secondly, get a second opinion because the biggest thing around context is making sure that everyone can understand it. And if you're the one that has made it, if you then show this visual to someone else, they won't really, they might not know what you're talking about. And this is one of the biggest things around data storytelling and it's actually something you can do a little bit with ChatGPT and is when we're done and at the end and if there's time, we can take whatever visual we've made and ask GPT to interpret it and feed it back to us and it can act as our second opinion. So I want us to talk a little bit about narrative and it's the kind of the last part for what we're working on here. And really the goal of a narrative is to just give a convincing view of the data set and really drive us towards why whatever findings we are finding are significant. We want to inspire action and we want to make sure that people understand what we're doing and why we're doing what we're doing. Because without it, these are just numbers that someone's going to forget tomorrow. And this is. It's just another bit of information that's being shared with us and just not sticking and not really being useful. Data stories are useful. Data itself is just kind of numbers on a page. Let's go back into data storytelling and see back in the ChatGPT and see what it did. So it kind of got it right. We need to ask it to move our block of text down just a little bit. And that's kind of the interesting part of if we wanted to, we can open up our code block here and we can scroll down to the part of where we want a horizontal alignment equals left color. But we're not going to mess with that. Instead we're just going to ask it to move the annotation because that's what the little thing is. It's our move the annotation down. Yeah, we'll just ask it to move it down. We'll see how far it moves it down and see how much it actually gets us there. Hopefully this gives us what we're looking for, but it really shows what we can do and how we can start to mess around with this. Once this gives us the answer, we're going to change gears a little bit. We want to start to ask GPT to interpret this for different worldviews for different audiences so that we can see how we can start to craft our narrative and craft our. Our bits of knowledge for. For specific audiences to really bring this data storytelling all into one kind of key area in the interest of time. I'm not. The things that I would like to continue to mess around with are for this visual is. I would want to give it a more descriptive title. We're going to leave that as an exercise for the reader as well as I would want to move the little annotation just a little bit off that line. But instead we're going to dive in on this idea of, well, how would this findings be useful for X Persona? And that's something that GPT is really, really useful to do. So what we're going to do is we're going to give GPT a little bit of context. I've been telling you that two team members left the team leaving the team understaffed in May. But we want to ask GPT so write a short paragraph explaining that two team members left the team in May causing a ticket backlog. The intended audience for this is a hiring actually we'll. We'll say the intended audience for this paragraph is. Is a. An executive in charge of allocating resources for next It's a pretty basic Persona that we're aiming for. So our intended audience is a. Is a. The intended audience for. Is an executive in charge of allocating resources and the thing that I love about GPT is it just helps us break the page. And this is what I really love is it's not the best explanation here but it's something that would have taken me 10, 15, maybe 20 minutes to just get something started. And now we can sit and we can do proofreading on it, we can edit it, we can change the tone. For example, this response here it's a bit too dry. It's a bit too boring. It technically does target the executive in mind but we want to ask it to be make it more of an executive summary and include specific data points from analysis. And what we're going to do is we're with this we're asking GPT to go and actually like pull into the data set and grab things that. That are useful. And what. What we can see is it does start to it actually gets it wrong quite a bit here of where in April said that there are 149 tickets received in 148 process while in May we also have that same difference of one but it has the right idea of where we can start to pull from those little bits of information and start to support it. Interestingly it's giving us way too much information and way too much knowledge here and we want it to be even shorter. So let's ask it for limit your response to three sentences just to give it a very specific targeted limit because this is something that will. That you'll mess around with the more you do Prompting is the. The idea of giving focus and giving a clear guidance to it really becomes more and more useful for how we. How we digest and work with our kind of prompt along strategies. Here it's and one thing and I see a note pop up in chat about how you'd be pretty wary about trusting an AI that is prone to hallucination and I 100% agree with you. I 100% agree with you. And that's why everything that we're doing here. It's a great place to start, but it is not the only way to do things. And that actually brings me kind of round to the kind of summarizing thoughts as well. It gets us where we want to go is GPT is fantastic for getting you started in data storytelling. When we started off here, we just had a bunch of numbers. So let's go back up here to the very, very beginning. We had a bunch of numbers. We asked it to tell us a story and it gave us a pretty wild story about January's fresh start, February, the love of detail, where it was pulling in things. I don't know if that was specific Valentine's Day part, but it started to get the March, the spring surge, April, the ebb and flow May, steady as she goes, but it actually got it right. And that's the crazy part is I never told GPT about, about the fact that two team members left. But if you notice here it starts to call it out in this little story. It starts to say that we're maybe perhaps a sign of summer vacations impacting the workforce. But it wasn't exactly the summer vacations that were doing it. It was the deviation in our team members and them migrating to a different area. And it starts to really show that GPT is a great tool for us to get started and it can actually produce some pretty good visuals. Let me go down to our. To the one that we started to work on. We didn't finish it up, but it can, it can start to make things that if I tried to do something like. Let me get to our last one. If I tried to do something like this in Excel manually, you'd have to sit there and do each individual little bit. I tried to do this in Python. I'd have to first learn Python a lot more than I already do. And it really prevents us from getting the value out of it. And that's how I like to talk about GPT and generative AI in general, is it helps us facilitate a greater understanding. And when we bring that idea to data storytelling, where we want to make visuals, we want to give it context and ask it to restructure that context as well as we want to help it use it to help us craft a narrative, it can really allow us to make some interesting things that would have been out of the reach for a lot of us. So let me go back into our little presentation thing and talk about how would you want to continue your journey? I have left 10 minutes for Q and A, so If there's anything that you, that you, you have burning in your mind, please feel free to get it ready. But I have one area that I like to mention for a lot of this is the our data storytelling skill track. Within Data Camp we have a track all around data storytelling that's useful for getting to learn all about it where, where we go into all of these details quite a bit more. It is a great little skill track that it's very powerful. You'll see that I have actually I was the author for the first two courses and the final two courses I went a little bit deeper in. And then one other area that I also wanted to mention I actually forgot to put a slide up on this is the prompt engineering part. So you notice that we spent a lot of time working on our prompts and kind of tweaking our prompts and I see a question about is this the Data Analyst GPT? Yes, we were using the Data Analyst GPT and we have a course out there on our platform called Understanding Prompt Engineering. And an interesting part all around this is the course is completely free for all users until the end of the month. So you do not need to subscribe in any way whatsoever to Data Camp. Just go on, follow that link that was put in chat by Reese and anyone, any registered user can take it and complete the course in its entirety. I really recommend it. It is a great course that I put a lot of time and effort into with the instructor and it will help you unlock this idea of how to prompt with GPT and how to mess around with it. So with that in mind I will and we'll turn it back to kind of what, what questions can can I answer for you? Is there anything in specific you want to talk about? GPT storytelling?

Really?

Any of it?

Awesome. Thank you so much Joe. That was such a great presentation. We have a couple of questions from our audience. First is going to be from EPIC and I would like to know does sample size have any impact on ChatGPT other than statistical implications?

So GPT there are some limitations for how much information you can put up into it. If you want to put millions or billions and billions of rows, it doesn't exactly work all the time, but for the most part it's the power of the cloud. So you can go in and give it some pretty big data sets and allow it to work on it, but you aren't limited for most traditional data sets. But for truly big data, you're not going to be giving it to GPT just yet.

Yeah, we've had a couple of other questions in the chat about like how big a Data set can ChatGPT allow and how much information can we give it? And that was a really helpful and comprehensive answer. Thanks. Our next question is from Hamna. Hamna would like to know how about asking ChatGPT what are the possible ways to analyze the data by suggesting us possible queries, Would that work?

Yeah, and that's actually one of the, one of the kind of the best ways for getting started with using kind of the data analyst GPT or really other ways of getting at it is you could ask it just a totally open ended question of hey, what do you see within this data set? And it will analyze it. But one thing that I do like to caution against is there's a chance that it will find things that don't exist. Whether it's spurious correlations, just relationships that aren't truly there and we still need to have that human in the middle for us to sit there and find what is actually useful in this data set. And does that analysis actually make sense? Because it did suggest a pie chart at one point and I could rant for hours about why pie charts are evil. So GPT, if you asked it to, would do it, but we don't want that.

That makes sense. Yeah, pie charts, I'd love to hear that ran someday. Cool. So our next question is from Asha. Asha wants to know is ChatGPT capable of measuring correlation strength of two measures or a dimensional measure pair? From what they understand, it doesn't work well with math.

So that actually GPT's famous inability to do math was one of the things that really kicked it off. I know if you asked it what's like, what's one plus one? It would frequently, if you told it no, you're wrong, it would frequently tell you one plus one was three or some other crazy numbers. But and this was actually something that we were doing here, the data analyst GPT specifically as one of the custom GPTs available for either enterprise users or people that subscribe to GPT plus, it really, it's gotten much, much better because as we were seeing as we were going through it, G itself was, it wasn't doing the math and it wasn't doing the doing. It wasn't calculating those correlations. It was reaching out to Python or reaching out to different languages to do it instead of trying to do it itself because a large language model is awful at doing it on its own. But Python, fantastic.

That makes sense. Yeah, I'd love to have that argument with someone trying to convince them that their answer is is right or wrong. Mike asks a really interesting question next. So Mike says how would you handle sensitive data that you don't want to share with the model?

So that's, that is the kind of the number one question that comes from every C suite executive everywhere is well, I don't want to upload my data. And that is a very, very key question. And that is something that I want us to be very cognizant about as well is the data that you're uploading you for better or for worse, don't have full visibility of where it goes. There are enterprise plans for some large language models where they will be hosted on prem, which means on premises within your environment. But for the most part I don't think it is a good idea for you to be uploading extremely sensitive data or things that you would be worried about it getting out into the world. The large language models are in their infancy and we want to, we want to make sure that we're not accidentally sharing things that we should be sharing and having them be leaked to competitors or being leaked in any way that we don't want to. There is a place for uploading data to it and I do play around with it, but it's really finding that line, especially what your organization allows, allows and really the value and risk associated.

With each part that makes a lot of sense. Yeah, definitely something to consider when using these large language models. Our next question comes from Special Eddie and special Eddie would like to know what are your thoughts on LLMs and prompt engineering disrupting data scientists and data analyst jobs.

So it's actually an extremely interesting question because when, when ChatGPT originally kind of came on the scene, I know I personally was a bit of a doubter of like I'm not going to be able to do much with it. It's not useful. And then we saw kind of the hype cycle with it of oh, it can replace everyone, it can do everything. It's super great. In our own example that I did four test runs on it didn't do the thing I wanted it to do at the end of the day GPT it is a non deterministic which means it doesn't follow the same path every single time method of working with data. Sure it can help you do your job better and do your job quicker but it most likely isn't going to replace anyone doing it. In fact I actually see a bit of the other way of it helps get people that don't know those Python skills or don't know the base things super well and it gets you familiarized with them and, and then you can use that as a way to build your skills more. I know in the early days of GPT and that sounds silly considering it's only been out for just over a year. One of the things that I used it for was helping me put together a pull request for a package that was written in JavaScript. Something that I've never done ever before. But it helped me walk through it and like it actually helped me learn a little bit about it. But, but at the end of the day I had to be the one that did it. And GPT isn't going to fully replace anyone, but it is going to change the way that we work with it.

Yeah, that's a great way of looking at it and can be super helpful if there's something in a language that you haven't encountered yet. Next question is from Marwa. Marwa wants to know, is Python the default language? Could you request it to use R?

That's actually an interesting question that I don't exactly know the answer to, but I would love to look into that for you. I. I know any time that I had queried did default into into Python, but I've asked it to write R for me before and it was. It's done it. So I have a feeling it probably could, but I'm not sure if that is a base functionality of the, of the specifically data analyst GPT but play around with it. Maybe, maybe it will.

Yeah, super interesting question. Thanks Marwa. Next is from Jasmine. Jasmine says could you start out by giving chat some context in terms of data storytelling or technical terms related to your data set and would that help with better visuals?

Yeah, and that's actually something that is kind of a more advanced topic with the way that we were approaching data data storytelling and really approaching it is we were using custom GPT, the data analyst GPT but if we wanted to we could make our own that would kind of act in that conversational way. So it would ask us for that context, it would ask us for just a little bit more and we could actually have GPT build and just go a little bit deeper on it. And it is a fantastic way to prompt with it and it gets into this idea of prompt engineering of there's so many different ways to approach it that your method, and this is something I encourage, your method is going to be one that you refine through many different iterations. This is just a starting point.

Absolutely. And then let's wrap it up with a question from Asha. So Asha would like to know, do you have any resources on ethics and data storytelling? I'm sure Reese will share a link, but, Joe, do you want to tell us what some of your favorite resources are?

The main resources that I talk about all with data storytelling, it really does go to. I'm going to hold my book up yet again, storytelling with data. And I know my thing's mirrored so you can't actually read it. So storytelling with data has fantastic insights all around, kind of the implications of telling stories and sharing information. Because if you don't really know what you're doing or know what you're sharing, then it's hard to kind of act as the bad guy. With data storytelling, that's something that we really try to get away from, is misleading. With data stories, as well as the data storytelling skill track, that is something that is super useful for you to digest. We get into it in a few different places where we want to talk about how to learn all about data storytelling and use it in ways that make sense. I'll borrow from Spiderman here. With great power comes great responsibility, and data storytelling is a fantastic power.

Awesome. Yeah, absolutely. With great power does indeed come great responsibility. Thank you so much, Joe. Those are all of the questions that we're going to answer today. Before we wrap up, I do want to let you know that we have two more live sessions this week. Tomorrow we're looking at driving value creation with data science in the mining industry. On Thursday, we're running a session on data Camp classrooms. Head over to datacamp.comwebinars or scan the QR code on on screen to register. That's all from us today. Thank you so much, Jo. This has been an amazing session. Thank you to everyone who's prompted along with us and we'll see you again soon. Bye, everyone.