Brainstorming with the chatbot

- Let's say you're a team manager who keeps getting asked why projects are late. And you've got some great ideas for a dashboard that could help you pinpoint where the problems are, but you're not a developer and you're not sure how you would even prompt for something like this. So let's vibe code a dashboard to solve this. And in the process, we're going to learn a ton about prompting for tools, best practices, and the process of working with AI. For my prompt I'll type, I'm trying to build a dashboard that explains why some projects keep slipping. I need a realistic sample dataset, project tasks. Can you help me choose fields that show early risk signals, the cost of delays, and simple sentiment or confidence levels that reflect clarity or requirements? Now, sometimes ChatGPT will give you way too much information. In this case, it gave me a list of 15 items. But if it gives you too many items, you can try to rewrite this prompt to make it more precise. And here's my first tip. For something foundational like this list of items, you want to finesse the output to make it better. Don't just take ChatGPT's word for granted. Now, sometimes, ChatGPT will also give you an idea of where it might be going. In this case, it created a couple of different dashboard styles to show me where I could be going. And although I didn't ask it to, it explained a little bit further what each of the items meant. Let's assume that we want to modify this role right here to account for multiple people. One nice thing that you can do with ChatGPT is make a selection and click on Ask ChatGPT. This will quote that item and let you ask a question about it. I'll say, let's make sure we allow for multiple team members. This is too many dates, I'll say, can you simplify this list of dates? And I'll go ahead and modify these ratings so they can go from one to 10, giving me a little bit more depth. I'll say, let's make these go from one to 10. Let's also modify the notes sentiment and see if we can expand these options a little bit. I'll say, let's expand on the amount of options and use fives instead of three. Even if you start with your own dataset, a back and forth conversation with ChatGPT can be useful. It's gonna improve on your assumptions, and will make sure that the tool has the best possible data to build with. Let's go ahead and ask, can you generate a mock dataset of about 200 tasks using these fields? Make the values varied, realistic enough that charts and insights will be meaningful. One of the things you'll learn about ChatGPT is that it's very stingy with resources. So even though I asked it for 200 tasks, it only gave me 10. So let's talk about a few ways of trying to fix that problem. I'll say, let's create a downloadable CSV file with the 200 records. And here's the generated data. It's not going to be particularly creative with things like names in the same way that normal ChatGPT would. And most people don't realize this, but these chat bots have two modes. The first is the one that you're used to when you're talking to it. It's very creative and can brainstorm really varied datasets. The second one though is the code interpreter or the Python developer mode, sometimes called the advanced data analysis tool. In that mode, it's a good programmer, but it's not using the same sort of brain it has when it's being creative. And so you'll see that it makes more programing type decisions on how to approach solutions. You can see here the task names are really not that interesting, so I'm gonna go ahead and change that. I'll say, the task names aren't really interesting. Let's generate 200 interesting tasks for the tool and then add them to the CSV file. If we take a look at this new version, you'll see that it is repeating things quite a bit. So I'm gonna say, the names are very repetitive. Let's generate the names without the code interpreter first. And let's go ahead and send that. That's a much nicer list of names. When you're working with the code interpreter, you can always take a look at the code that it used to generate, and I know, you're not a developer, but I think it would behoove you to learn to read some of the code that it writes. And although some of these things may look like gibberish, they're actually very simple English type languages that are somewhat readable. Take some time and experiment with just looking at the code and see if you can figure out what it's doing. Try reading it and see if you can pick up some of the patterns that ChatGPT has to work with. You can see that when it generates data, it will sometimes create a couple of lists of words and it'll just combine those words together. So often, it's best to know when to use the code interpreter for programming and use the power of ChatGPT for creative generation. This list is a lot better. I'll be honest with you. I'm just as guilty of jumping in and handing the tool some data, and pretty excited when I get a sweet looking dashboard. But without some planning, you're not gonna generate operationally accurate and consistent repeatable results.