ChatGPT Code Interpreter vs Tableau | ChatGPT, Tableau GPT & LLMs

hey it's Tim here over the weekend the chat GPT open AI the company that runs chat gbt added the code interpreter for chat jpt this is a separate model from the gpt4 that we've been using or GPC 3.5 if you're using the free version It's a separate model that it's able to do a mathematical data analysis because it's using python behind the scenes in this video I put it head to head with tablet as ever let's get stuck in okay so here we are we're in gpt4 and if you're wondering how to enable this you have to go into your settings and there's a little beta section there when you click on that you're able to click on the toggle to enable it and then it appears up here on the top right hand side I've actually been using plugins already the plugins have been around for a while but code interpreter is a new feature and that's available here and you have to choose one of the models if you use plugins for example you're not able to use the code interpreter so these are the plugins that I've been playing around with I found them extremely useful for various things that I do but I I can't use those plugins if I'm trying to use code interpreter at some point I'm hoping chat GPT enables the capability to use plugins and the code interpreter because then you could do some really crazy things okay so I'm going to put my face up here now what I want to do is see if I can do the basic analysis on this dashboard this is the default dashboard that ships with tablet and so what I want to do is try and see how close can we get to some of these numbers okay and uh is the analysis that it's going to do actually up to Snuff is it going to work is it reliable there's already a couple of glaring issues with this whole process for example the only way that I'm able to be able to do this is to upload my data to open Ai and for most people that's already a big No-No you shouldn't be doing this work with work data with any data that you don't trust I'm doing it with Superstore because it's sample data this is the demo data that comes with Tableau and so we're happy to use that so what I always like to do is to discover the features that chat jpt uses by simply asking hey can I do this thing so hi can you analyze a data for me okay and this is actually one of the big things that's been in tweet threads and I'm always like listen this tool is not going to take over your job right like these tools are driven by intent and they don't have the autonomy to come up with that intent without being given some sort of instruction so as long as these Souls don't sort of just go off and decide to do things which by the way if they did the world would be panicking because then they could go off and decide and I take over things that people people people people are you know concerned about but nonetheless here we are so this is what it can do what is the nature of your data what are the key variables are there any specific questions you're trying to answer also if you have the data ready can you upload it here and we can get started so I absolutely do have the data set I actually have it on my desktop so we'll go ahead we'll bring it in you'll see me drag it in here and I'll just put it inside of the text box and I'll say okay here is the data okay let's just have that conversation and you'll see that it's working now this is the crazy thing um when it says it's working it just says hey look finish working and if you actually expand this box it's going to type out a bunch of things is going to tell me that it understands what's in the data and this is already pretty impressive I mean I saw this and I was like wow this is this is pretty interesting and so what we can do is we can actually open this up and you can see what it does it wrote The Python for this so import pandas as PD so it basically gives it an alias and then it loads the data using that Alias so PD read the Excel this is the file and it's essentially mounting that uh this file that we've uploaded and it's able to then you know load up the the headers and that's what it's basically using and then it's this is this is what's crazy it's actually dumb some initial analysis it's actually gone to see like okay what can we get here here's the customer information and so it's gone got customer ID customer name customer segment it's gone and got the order information and it's got that kind of nicely grouped up it's got some country region specific information subcategory but ultimately all it's doing is just listing the fields there's nothing sort of you know crazy about this it's just I think this is just the order the data comes in so so far so good we know it has the data now we can start to ask it the basic questions so the basic question I'm going to ask and I've checked that this is the right version of Tableau the right file because each version of Tableau comes with its own version of Superstore I've checked it so we should be getting exact numbers so I'm just going to ask it what was the total sales in this entire data set so let's go ahead and do that we'll look past the first test what was the total sales in this entire data set so I'm giving it context and I'm giving it sort of a purpose so let's go ahead we've given an intent let's see what it comes up with there we go now nearest I don't know if that's what we wanted I had it in pounds and there was nothing in the file suggesting that it was Euros but 2938 and 89. uh 2938 and 89.06 oh god let's is it correct to be doing that I mean I'm gonna assume it it's correct if I just go into the sheet um you probably want to know well how do I know it's correct let's go into the sheet and let's just right click format we'll just go ahead and say measure values we'll just choose a number number custom will add two decimal places okay actually that's not going to work because it's the sales value sorry that I need to actually adjust here so we'll go currency custom to decimal places and it's a little bit small so we need to make it larger 0.062 so it was it was actually correct 0.06 that's what it's got there two decimal places okay we trust it it works it's done the total pretty interesting okay that's pretty basic I'm going to assume it can do the other things let's go ahead and ask it the profit ratio now the profit ratio is more complex because what it has to do is it has to go and do the calculation for profit it's only got the base data the the profit ratio is not in this data so has to go and figure out the margin then it has to aggregate everything and come to this percentage so let's go ahead and just check those fills so you can see here there is no profit there's a profit value but there's no profit ratio so let's go ahead and say ask it um what was the profit ratio across the entire data set as a percentage and being specific about percentage because um I think it does help it sort of get things better and actually that's how you'd so that's how the numbers have actually shown so let's go ahead and see how long this takes and we'll look at the working as well a 12.69 absolutely nailed it so let's 12.69 let's let's just go back into this I I want to see if this is being rounded up and I actually don't know about it so let's go ahead let's go into the profit ratio this is a percentage is 12.69 that is so good I don't know why that makes me happy but this is crazy because previously this was not the case Okay so this means that for every one Euro it still thinks it's euros in cells approximately um um this amount this amount of profit which is which is perfect so okay it's done the basic maths um this discount this discount is the average discount across the data set so let's go ahead and ask it that um what was the average discount across the entire data set so this is a difficult one because the context isn't always clear when you do averages you have to like assume something and in this case I think it's done the right thing um it's gone across the entire data set and I think it's just aggregated it all down as Tableau would have done 10.31 again I think this is going to be correct um I think we can confidently say this is pretty easy for this thing to do and if we say this is a percentage 10.31 is the exact answer the tablet has so at least at face value this top level here easy stuff for um chat jbt I'm not even going to bother with the others because those are just sums it's going to get it right okay now we get a little bit harder okay now I want to try and see if I can get some of the information that I'm getting in these charts because these are sort of more complex questions okay and I don't want to just do that I want to go in here and pick a filter so we'll go ahead select Central so we'll narrow this down to Central and we'll give it a Time range let's go ahead and say that we want to go from the uh first of this day and what I'll do is I'll copy these dates into chat GPT when I do the query and on top of that I want to know the value for Germany okay so Germany Bavaria here and I think it's um well this is interesting so this is this the oh no these this is a region so this is a I need to go in and be specific about what exactly this is this is a region yes so Germany Province Bavaria is a province in this particular concept so State Province so what I'm gonna need to ask chat GPT and I need to kind of be specific here because I think it really does want you to be this specific because I need to ask it in Bavaria for the central for um the specific date and time what was the total sales okay so this is going to be a little bit hard but let's let's let's do one thing actually let's go in here and um if I if I just duplicate this as a cross tab it will give me uh this field and I can just go out here and I can can I copy this I can edit the Alias I cannot spell so I want to grab this and just say Okay cool so go back to the overview and then I say okay what was the total sales in central region uh paste that Bavaria in Germany between the dates let's go ahead this is a little bit of a tricky one so I hope I get this right between the dates this one and let's get this out this is boring to watch I'm sure but hopefully you're all still with me here and that now am I missing anything total sales for Jamaica bavari Germany I've got no other and I don't want to click my face I don't want anything else everything is good okay we can just go ahead here and circuit was a total sales in central region Bavarian geometry between these two dates okay here we go Moment of Truth What's it gonna do I've given it two dates I've given it all the other contexts that is like an interaction that's why I think this is a more complex question there's a lot to get right in there and here we go now I see here this is interesting the total sales in central region specifically in Bavaro Germany between those two dates was 24 000 Euros now that doesn't look right because did I did I give it the right data oh no it's it's got it right 24 651. Jesus I'm sorry if anyone's religious I apologize I'll try and keep it clean here that is that is nuts wow okay that is that just blew my mind away I I cannot believe it got that right let's look at the working I mean I don't know python so I don't know what I'm going to be able to do here but this is absolutely nuts it's gone and got the data and it's written what looks like fairly simple python here it's gone and basically it's like it's written SQL saying hey this is the information go get this and figure it out and by the way it's been doing this all along so if you're not familiar with python um it's basically been sort of building this up so you could also use this as a really nice way to sort of go backwards and figure things out what is alarming is that and I think this is why companies like Tableau are really you know really worried about this technology is that the threat isn't this this the threat isn't that it can do this this is not the threat the threat is that the interface for How We Do data analysis might be changing right in front of our eyes everything I've done has not been drag and drop here in tablet just imagine just to build out that chart to build everything here would have taken out more time than typing the questions and I'm pretty good at Tableau I can confidently say that it would take me less time to type it than it would to build what's in front of me even if I was racing and trying to do everything really as fast as I can the other thing is that this little box here anyone can use anyone can type into that it doesn't just take me a data analyst it takes anyone who has a basic skill set of asking good questions that is an important skill can use this search box so in my mind the Innovation behind open AI all these AI models is not the models it's actually the interface that allows anyone to use them well that is actually really hard thing to do in open AI have done a lot of work to train these models so this to me is the Innovation Tableau everyone power bi all these products they're racing towards AI you might think they're racing there because you know this is the next big thing no they're racing there because I think they all see this little box as the biggest threat to their product if they can't get their product into this box in the next five years they might as well give up right now because someone out there is doing exactly that and it's actually alarmingly good at being specific about certain things now this is where tools like ask data kind of went down but they were doing something which wasn't as good as some of these models and the key thing to remember here is that openai have had a massive investment in technology to get to the point where they could build the model that is what we're using here they've had lots of money pumped into them by Microsoft and lots of other investors to get to where they are today to be able to then ship this model out to companies the actual model the final model they build is very easy and portable for them to share with companies in deploy the hard thing is training and that takes time people resources and a lot of Hardware but pretty much only Nvidia makes so um this is pretty crazy I I just cannot believe this so we've seen it answer questions let's see what it can do with charts let's go ahead I want to go to a different dashboard just because I'm I'm interested in this okay so let's ask it for a sales by product category broken down by year and month okay so let's do that let's go back to this and say hey we don't like tables but we'll ask for ones hey can you build me a table showing sales by product category category oh category with years as a subheading and months as columns along the top so I've put some information here about how I want to see the data and let's go ahead and see what it does so it's going to have to do the processing fast but then it's going to have to visualize it because I'm asking for a table it could print the table to be Ferris text it maybe doesn't have to but here we go let's have a look category yeah oh this is interesting okay oh my word oh God it's doing it so then the next question is is this number correct six seven five one for furniture 2020 Jam let's go ahead six seven five one six seven five two oh I'll give it that's just a Roundup that is that is remarkable that is I cannot believe it got that right I know it sounds so simple you're probably thinking Tim this isn't that Sensational like no but it is because all I did was type it in and it nailed it it nailed it and I'm comparing against a finished solution I'm literally drawing the gap between you don't miss yourself and typing the question that is that is ultimately what is groundbreaking here and so far it's got everything right there's not been a single mistake in this entire uh pipeline this is pretty crazy now it did what I thought it would do which is it didn't actually do any sort of visualization you can see here it's just literally just done a pivot table and it's it's actually printed the result as I thought it would here and then it's just basically gone and styled it in HTML which is even nicer okay and it says the values in the Euros we know that's wrong I could have corrected that earlier but we didn't bother okay it's past that test nice okay we'll ask for this sales and profit by product names okay this is a this is a this is a difficult one says let's just grab this I'll try and give it less information this time um so can it show us sales and profit by product names oh this is a this is a really this is really difficult this is all about the art of typing the good prompt so wow what I'm what I'm gonna do this what I'm gonna do is I'm gonna put tablet on another screen and I'm going to put chat jpt here and what we're going to do is we're going to type the prompt in front of you so you can see what I'm typing say I'll say um can you build a scatter plot I want it to show s and profit by product names with category and sub category breakdowns in rows nested and then uh what is on the top uh what is that what is that official terminology let's just go I think this is customer segment yes correct so we want segment and then segment as column sections okay so this is a slightly more ambiguous thing because I've asked it for a scatter plot but I'm talking about rows and things being nested and then the segment is column sections let's go ahead and hit enter let's see this is this is tough because my prompt could have been bad I don't think it's that bad it probably has typos but let's let's wait and see I'm not going to do the work I'm not going to show the working I always look like to look at the working afterwards oh it stopped so what is going on here I think what it's doing is it's potentially firing up another thing oh wow okay yeah it got busy okay so here we go so uh what did it do it imported a matte plot lib by pi plot so it's imported a library and it's imported Seaborn and then it's commented it out so it has map scatter plot create facets um it's a Seaborn seems to be able to do some of this segmentation say hold on okay and it's it's created it and then it's basically kind of pulled the chart and that's what we've got below here but did it do it correctly so we've got corporate consumer home office yeah okay now what it didn't do nicely is this this thing here so what it did is it kind of said each scatter plot is going to be a subset of the total so it didn't put it didn't put them along the left that is something that it kind of is confusing and I thought this would be confusing but it's not done a bad job of giving me the alternative and I could still see the outliers so I can see the outliers clearly here in many ways it's maybe made it easier for me to do the analysis because in Tableau these would be in one line um just to show you what I'm comparing again so I'm comparing it against the bottom bottom chart here and that is a pretty interesting uh sort of uh take on that isn't exactly what I wanted because I was starting with something that was already you know articulated the trick here is going to be prompt and I think if I did a better prompt it maybe would get closer to the answer um also I'm not sure if the thing I'm asking for is specifically possible you know having a table like structure on one column on one sort of axis and then a a like a section setup on the other and it got some of that but it's not perfect and it goes and explains what it's done and again you've always got the ability to show the working so that that is that is truly remarkable I don't know I don't know what you're using this for if you're using this let me know what you think um there is a couple of issues with it it still can't do all the all the um the things you would expect so if I go ahead and ask it hey um can you plot a map with each data point sized by cells at the level of cities in the data set so I've sort of asked it in a really bad way there's even a typo in there um I'm sorry for the inconvenience but currently in this environment I'm unable to access Internet services which includes mapping services such as Google Maps or any other geolocation service it could be is therefore I'm unable to create a map plot for your data fine however I can guide you to this using python libraries such as geopanders and tholium so it's basically saying hey you can do this on your own environment and by the way this is the code to go and get there's a little bit of geocoding that it does so you see it takes the city column and it uses geolocate using the latid and longitude to basically get the exact location and then it's able to then import volume which takes that information and for each row in the data set is actually able to pass that and then create the map so this is telling you how it would work if it had access to that but the the other crazy thing you've got to remember is that chat GPT is firewalled from the internet everything you get back it's not able to search the internet it can only use its large language model or model that's built into it for everything that it does and in more recent features they've handed this ability to browse the internet with Bing but they actually have to disable that because it was doing things they didn't like and but in bing you can actually get it to go ahead and still do that so in here in charge apt it's completely sort of isolated still and that's how it's working anyway what do you think I think the biggest threat to pretty much everything is this search box this search box what is incredible is um this tool if I didn't expose the working every single time what I'd be seeing is just answers okay and you might be wondering well what is the what is the confidence level of some of this work well I walked away with pretty high confidence the things it couldn't do it told me and things that didn't quite get right were based on the prompt so that's another sort of interesting angle but yeah um if I'm Tableau I can absolutely understand why they're racing to build something like tablet GPT because again if in five years your tab your product doesn't fit in this text box I I'm certain something will come in at your lunch anyway thanks for watching I'll see you in the next one

Channel: Tableau Tim

Views: 7,164

Keywords: tableau, desktop, salesforce, version, tableau prep, tableau online, tableau prep builder, tableautim, chatgpt, chatgpt code

Id: 4P5VCHg-KAU

Length: 24min 32sec (1472 seconds)

Published: Tue Jul 11 2023