← The ChatGPT Experiment - Simplifying AI For Curious Beginners
The ChatGPT Experiment - Simplifying AI For Curious Beginners
Ep 96 - Real Example: Coaching a Team to Use ChatGPT for Scalable Research
00:30:10
Loading summary
Transcript12 lines
- [00:01]
Family Member
Hey, y'.
- [00:01]
All.
- [00:01]
As a growing family, my husband and I love game night. Especially when it's Wayfair edition.
- [00:06]
Game Host
Let's do it. You gotta name as many Wayfair furniture and decor categories as you can. Ready?
- [00:11]
Family Member
Go.
- [00:12]
Sofas, bar stools, beds, ottomans, outdoor seating, bookshelves, kitchen tables, garden sheds, uh, mid century modern lamps.
- [00:19]
Time. Nice.
- [00:21]
Game Host
You got nine out of a lot. Not too bad. Keep practicing by visiting Wayfair.com which you can shop every style for every home.
- [00:28]
Kerry Weston
Wayfair.
- [00:29]
Family Member
Every style, every home. Hey, it's Kerry. Hey. In this episode, I'm gonna walk you step by step through the conversation I had with ChatGPT to help someone I'm coaching solve a complex repetitive problem by building a project inside ChatGPT. If that sounds interesting to you, then stick with me and I'll see you on the other side of the music. Hey there. Welcome to the ChatGPT experiment. This is the PODC designed to help you better understand how to use tools like ChatGPT to get a nugget of value for your personal or professional needs. My name's Kerry Weston. I'm your host. Glad you're here. If you're a first time listener, welcome. If you're a repeat listener, welcome back. And perhaps you've heard me say before that the best way to find value inside a tool like ChatGPT is not to look at it from the perspective of I have to learn everything about it and I have to learn the software. The really the best way to get comfortable and to start seeing value is to find just one thing. The ability for you to have a conversation with ChatGPT is unlike software in general. It's the ability of having a two way exchange. And what I share in workshops and what I share in the people that I work with and coach is just, just dig in and share with it something that takes your time, something that is definable and scalable. If you can explain it and if it can be defined, then more than likely ChatGPT can help. So for this episode, I thought I would just walk you through my thought process and the step by steps that I took here with ChatGPT to help someone that I'm coaching get value on a specific task that takes up their time repetitively. And one of the things I want you to think about as you're listening to this is how you can apply what I'm sharing with you to a task, to an issue, to a repetitive scenario that you may have in your day. Because you're probably and most likely not going to have anything close to what I'm about to explain in terms of the literal details. But more than likely there is something that you do over and over again that you spend too much time doing. And perhaps you haven't thought about how to use a tool like ChatGPT to make that easier and faster and repeatable and scalable. So that's really the reason that I'm sharing this with you today, because I want you to see how I approached it, what resources I used. And I'm finding that when I do this, even the more experienced folks that are using ChatGPT see something that they are not doing or aren't doing thoroughly enough, or they see something from a new angle. So if you have spent some time with ChatGPT, you may find something here of value. If you are a beginner or getting to become more comfortable with it, there's certainly going to be things here to think about as well. So I think there's something in this exercise for everyone. And walking through the specifics I think is the easiest way for me to show what I mean when I say find that thing and talk to it. So I'm going to repeat for you the four part framework that I always keep in the back of my mind. So when we're communicating to ChatGPT, you've heard me say this before, if you've listened to the show AI. For me, I define the, the, the letters AI as the amazing intern. And so I always think about ChatGPT in terms of having an intern, a person, an assistant sitting next to me that I have to communicate to. And I would never communicate to that person. Like I talk to Google, go get me this. Best pizza places, right? I want to give it a framework. I would communicate to that intern, to that employee, to that assistant. Four things specifically. And that's the framework that I approach my ChatGPT work through. And these are, number one, what are we doing? Number two, why are we doing it? Number three, what does success look like? And number four, I always ask, do you have questions for me to help you do your best work? Let me repeat that four part framework. As I communicate with ChatGPT, I want to share what we're doing. Number two, I want to give it a why. I want it to understand the reason that I'm doing it. Now, what happens when I give you the why or I give ChatGPT the why? And step three is what does success look like when I give them those two pieces, I trigger something. I open a reasoning and a logic component of ChatGPT that goes far beyond the task. Okay, so here's what I mean by that. What are we doing and why? When I tell it why, it gives it context. When I say what does success looks like, it actually starts thinking about how to do its work to achieve what it is that I'm looking to do. And the why gives it context to help give a bigger perspective. When I close with do you have any questions for me? As step four, it gives it permission to have a conversation, a dialogue with me so that it can pull out the details that perhaps I didn't share with it. Okay. And one of the resources I have for this particular project and the project here that with the group that I'm working with, it's an advocacy group for an industry trade association and they are looking out for regulatory and legislative issues of concern to this particular industry. And one of the tasks that takes up a lot of time, as you can imagine, is keeping on top of what's changing in certain issues across the 50 states. And it's been kind of a manual search. And they have some go to resources and they do their best to stay on top of everything, but it does take time. This is something that eats into their day. And one of the questions I had is, do you use projects at all inside ChatGPT? This was a group that had tried using ChatGPT, but they found that it was inconsistent in how it was working and bringing back details. So they weren't finding success. I think that's an important way to say they, they have been trying, they haven't been finding success. So they weren't digging more into it, they were going back to doing it in a manual way. And much like perhaps many of you, this was a video call. So I was on a video call, so I was recording it. And I record it so I can pay more attention, so I can be more in the moment and have details when I'm done that I can review. But the other side benefit of having a video recording is that it has a transcript. So when I record a meeting, I have the ability to of finding a transcript. A transcript is just a written version of the words that were used in that call and in that call, because I knew it was being recorded. And you can do this on your own, by the way, just talking into a phone, right? Opening up an audio file in your phone and recording will give you a transcript. But this happened to be a phone call or a video call, so I was able to have dialogue. And so I was asking questions on purpose during this call so I could get an understanding of what the issue is. And I'm just going to go through my framework. I was literally following my framework. I was asking them to explain to me what are they doing. I was asking them to explain to me why are we doing it. And I was asking them to explain to me, if we could solve this for you, if we could figure out a way that ChatGPT could be helpful for you, what would success look like? What would that mean to you? And so I was able to capture that in the meeting notes. Okay. And because they were able to tell me that as part of this exercise that they do on a regular basis, they check external resources, they check state resources, and by resources, I mean publicly accessible websites. So they were checking state resources, they were checking industry websites, magazines and news places. Okay? So I gave them homework, and that was in the transcript as well, but I gave them homework and I said, if you can define for me what you check on a regular basis, give me the actual URLs that you are going to, that would be great. And if you can explain to me what an actual report. So if we're able to do this to paint the bigger picture of what success looks like, what would be helpful to you? Like if we could automate something and have ChatGPT help and it puts something on a, on a regular schedule, a repetitive basis in front of you, what would be helpful? How would you like that to look? And so as a follow up, I received an email that gave me the URLs that they are constantly checking and the description of what success would look like, you know, what information they're getting and what success would look like. And so I took the transcript and the email and I started a conversation with ChatGPT. And I'm going to pause here by saying my goal was to create resources so that I could build a ChatGPT project. And if you haven't gotten into projects, there's an episode on projects here in the archives. But what I was, what I wanted to do was create a ChatGPT project that would give them the ability of hyper focusing on the task and the outcome so that it could help them. And then I wanted to, when we figure out if it's working, layer on top of that a repetitive scheduled task so it could do it automatically. And in this case it would be once a week. Okay, so let me say that one more time. I wanted to set up a mission purposed project inside ChatGPT that knew all of the things what we're doing, why we're doing it, what success looks like, and then layer on top of that a task that would do it automatically. If we could prove it to be successful once, then we could just simply have it set up to do it every seven days. Okay, so how I went about doing that, I give you that background to tell you this. I said to ChatGPT, I have a big task for you. This is literally what I typed in. I'm just gonna read it in front of you. Okay? I have a big task and I need you to serve as my strategic advisor to help me build out a step by step plan we can develop to achieve a repetitive goal. Okay. Notice what I did there. I didn't ask it to go do something. I'm setting the conditions. I'm saying I have a task and I need you to serve as my partner. And we are going to create a step by step plan to achieve a repetitive goal. I am using in this case, a regular conversation in ChatGPT to build out instructions for ChatGPT. Okay? Because I have found that nothing understands ChatGPT like ChatGPT itself. Okay, so that's my first sentence. Then I went on. First, I'm going to attach a call transcript where the problem goal is being discussed with a group in search of how to use ChatGPT to make a repetitive and important task automated. Okay. Second, in the transcript you can see that I asked Mike to send me an example I could play with to explore solutions before the next call. Mike did that and his example is pasted below. Okay, so understand where I'm at now. Then I'm sharing with it what success looks like. Right? We're going to achieve this repetitive goal. Now I'm giving it more background. I'm going to give you the call transcript. I'm going to give you the answer that I received from the homework. Okay? And then I went on to say, again, this is the same conversation. I would like you to do the following. And I used a one, two, three step process here so it knows that these are individual steps. First, I want you to review the transcript so you can get a clear idea of the issue we were addressing and the goals they have for using ChatGPT. Okay? So the transcript I mentioned to you in the video call, I specifically asked questions because I knew I was going to use the transcript in this way. So I wanted them to explain to me what the issues were. Right. And the goals that they have. Number two, I want you to review the follow up information that Mike provided. So he gave me the homework, right? Information to the homework. So I'm attaching this. Okay, and then number three, here is the outcome. I'd then like you to engage in a dialogue with me so that we can map out the best path to take to have CHAT GPT assist me with this important task. Then I asked it, does this make sense to you? Do you have any questions for me to help you do your best work? Okay, so that literally was the prompt that I gave it. And along with that, I attached the call transcript and I pasted in the email. Okay? And it came back and it shared with me. Now I want to pause there for a second because you notice that my action step here, number three on what I'd like you to do is I didn't say give me the output or fix the problem or go do it, did I? I said I want to engage with you in a dialogue so we can map out the best path to have ChatGPT help with this. Okay, Very important step there. Because I want ChatGPT to think with me so that we can end up with a solution that will help ChatGPT do its best work. So I'm skipping the step of saying, go fix this problem. And I want ChatGPT in this case to be my problem solving partner. So when I say engage in a dialogue, it's going to do just that. Instead of giving me the answers, it's going to work with me. Okay? So it did that. It came back and it says, here's how I understand the situation. It gave me information about what problem the group is trying to solve. Okay? And then it says, right now it looks like they are visiting many inconsistent sources. They're researching the same topics repeatedly, they're manually filtering what matters and what doesn't, and they're re explaining things. Now, I call this the busy middle. In my work with ChatGPT, I say that every project or most projects that we have with a computer are in three phases. There's the ideation planning phase, there's the busy middle, and then there's the editing and perfecting phase. And I would argue that most of us spend far too much time in the busy middle. And that's where our time and our mental bandwidth gets exhausted. And far too little time in the editing and perfecting finishing phase. Right? We good enough and we just want to be done with it. And our value is really in that third phase. It's really in the first and the third phase. Right? What are we doing in planning? And then the ideation Perfecting, that's where our experience and our value gets put in. But our time gets sucked into that Busy. And so ChatGPT knows that I have that perspective. And it came back and it said that it looks like they're spending way too much time in the busy middle. Okay? And then it says, I think what you want me to do is you want ChatGPT to act like a research assistant. I think you want ChatGPT to. To check specific trusted sources. I like. You want us to check statuses that change over time. You want me to flag what's new and what's changed, provide plain English implications for their industry. Right? So it's telling me, I'm thinking that this is what you're looking for me to do based on the conversations that you've given me, the transcript and the outcome and all that. And then it went through and it told me a little bit more about what it can see in the transcripts, right? And it says, I think if we were going to approach this from a chatgpt point of view, this is what we need to do. We need to describe and design a job Description for an AI assistant. This is ChatGPT giving me the answer again, I've asked it to be a dialogue partner with me, right? Then it says, I think we need to define what good output looks like. We need to decide where the assistant is allowed to look for these sources. We need to decide how often we need to decide how it reports back, right? And then it says, I do have a question, right? If you had to choose one primary outcome, right? Do you want a recurring summarized tool or do you want a deeper analysis? And so what's happening now is, is it's thinking with me. So it hasn't rushed to judgment, it hasn't rushed to an answer. It hasn't started doing something. It's dialoguing with me and it's thinking, okay? And when I told it what my answer was, you know, I want that repetitive summary, it said, okay, I think this is what we need to do. We needed to define the job. We need to create a focused project. Because I mentioned that in my transcript, you know, that project will eventually have instructions and materials to go with it and then lock it down so it knows what it can and cannot do and how often, right? And be flexible. And the thing that I wanted here is I wanted it to work with me so that it knew where I was going and how I wanted to get there. But I wanted it in, in ChatGPT's words, because the Project I'm going to build is going to be for ChatGPT and I want it to understand itself, okay? Because oftentimes we will either miss data or we'll give it way too much data and it doesn't need it all. And so the reason I'm having it engage in a dialogue with me, to think out loud, is this is going to lead to me to say, okay, now that we all understand each other and now that we know what the purpose is and what the challenges are and what the next steps are, you know, I can then use ChatGPT to build those resources, okay? Because I've done the hard work. I've done the hard part here. I've said out loud through the transcript, through the homework and through my explanation and setting this conversation up with ChatGPT, as far as I'm concerned, I've done the hard part. I am 90% of the way done by the time I start that prompt. And that's one of the things I want to share with you, is there are things you can bring to ChatGPT that will engage ChatGPT as a problem solving partner that will allow it to take what you've already done and keep moving forward. Okay? I speak with a lot of people that are frustrated because they are constantly correcting in redoing work with ChatGPT. And when I dig in, I will tell you that a far majority of the time, almost always that is because they rushed the conversation to get ChatGPT to do something. Okay? So what I'm doing here is I'm practicing my own philosophy is I'm getting out of the busy middle, I'm letting ChatGPT do the busy middle. I gave it the setup, I gave it the purpose, and now I'm just going to sharpen it, perfect it and edit it so that it's good. But I'm letting the busy middle, I'm letting the hard work being done by ChatGPT because I've already had the conversations and provided the resources, okay? So I share that with you because you know, there's a long back and forth with ChatGPT that I won't go into because there's a lot of minutiae there, but I did have that dialogue. Okay? And the important part here is we've already determined that it's going to be a project. And a project is a set of resources designed for a specific purpose. But it also has memory, unlike a custom GPT that doesn't have memory. A custom GPT is like a tool I pick up the tool. I use the tool, it does exactly what I want it to do. But when I put that tool down, it doesn't know that I've picked it up before. Okay? Think of a ChatGPT project as a tool with a memory so it can remember and build upon itself and have just as sharp a focus as a custom GPT, but. But be there with you to grow and learn with memory. Okay? So that, that's a big deal for me. And I told it as we were going through here that I know that I don't have all of the information because things are going to change. So what I do not want to do in, in the print world, I would say I don't want to laminate anything, right? Because the minute you laminate something, you can't change it, right? So I want to have something that's flexible. I want to have something that can be changed for a couple reasons. We're going to use this tool when we're done, and we're going to find things that we like that it does, and we're going to find things that it's missing, and we're going to find things it does that we don't like. And I want to keep refining that. So I don't want to make this overly complicated to manage once we start using it. And I also know that there are going to be changes. The sources they're using today might change the website or the issue may change. So I want this to be a modular approach. Okay? So for those of you that are looking into projects, here's what I did. I understand that the project has a brain, and the project's brain is in the instructions. So project instructions to me is the brain. And So I told ChatGPT that I wanted this to be modulated. So I want to make sure that I have the freedom and fluidity to change the sources, the issue, and even the criteria that they're using to do their work. Okay? So let's have the brain understand that you are an assistant that's helping us do this task. But the details themselves, I want to be modular. Okay? So it said, great. I think what we want to do then is we want to create the brain so that it knows over and over again what we're doing. But we should create separate files, okay? Separate files that will allow you to upload those files to the ChatGPT project. And if the brain knows that those files exist, then all you have to do is keep those files current. Okay? And as long as the brain knows what Files you have, what they're there for, okay? And how to use them. You can change those reference documents as you wish without changing the project as a whole and without having to dig in a lot of resources. So we ended up with the brain, which was the overall instructions. It was the anchor, if you will, things that won't change, okay? The reason that we're doing it, the association, you know, the goals of the group, that kind of thing. And then we end up with four resources. And one of the four resources was called Sites to Monitor. So in the instructions it says you have a document called Sites to Monitor. And the reason you have these is these are the trusted URLs. Okay? And you may only look at those if the source is not listed. It must be ignored. Okay? This contains approved sources. So as this grows, I simply replace the Sites to Monitor document with the most updated list. And the instruction knows to look at that and only use that list. I don't have to change anything else, just that document. The second document that I've created is called Material Change Criteria, because what I've created here is I don't need every instance of mentioning of the legislation or the regulatory or whatever it is we're after. I don't need to know every time it's mentioned, but I do need to know if something's changing, right? And so I have created a document called Material Change Criteria, and that defines what qualifies as material change. It also defines what doesn't qualify, okay? And it gives the rules, if you will, as to the thinking and the logic of whether this is valid or not. And that may change as well. And so the external document called Material Change Criteria is our guiding rules, and I will change that. But the instructions again, knows that it relies on this document to know what it's going to report back on. Okay? And then the third one, the third external document that we created was called Time and Look Back Rules, because right now the group wants every seven days, but it may change maybe once a month. It may be every day, whatever it might be once a quarter. And so the Time and Look Back Rules is a document. And by the way, I want to pause here and say, when you do this, know that ChatGPT is really poor at knowing what the date is. So by default, it doesn't really. Not all the time. These aren't global rules. But what I've found is it's really poor with dates. So in that document, I've asked it to first determine what today is. And then our Look Back rule is based on after it's determined what today is, go back a certain amount of time and that's all defined in that document. Okay? And then the last external resource that I have is the output template, because that's what success looks like. That's what I want the report to look like. And again, I know what we want it today, but a month from now we may want something different. And I don't want to put a lot of work into changing that. So I have told the brain the instructions that there's an output document that you should follow each and every time you do a report. And that document defines the required structure and the formatting, it must be followed exactly. And you're not going to add any sections or remove or reorder anything. This is literally what I want. And then when it's time to have a different output, I just change that document and we're off and running. So the reason I'm doing this in a modulated way, and that's really the takeaway I want you to hear today, is setting up a project that has a fixed purpose, a single reason for existing. Giving it the brain and the instructions, giving it the big picture one, and then allowing it to look at its resources, its documents, its external tools for all of the nuts and bolts will give you a very fluid way of managing things as it changes without having to dig in and look. Now, the other thing I want to share with you is the only way Instructions knows what to do and what it has for resource documents is it has to be listed. So if you were to add another document, for instance, for whatever that might be, you should edit the brain to tell it that you've added that document and how to use it right, or what it's for. And if you take a document out, of course you want to delete the instructions, but that's really the only time, unless you have material changes, that's really the only time that you need to modify the instructions is when you're doing something different with the documents, adding or deleting. Right? So again, if you've been using ChatGPT for some time, hopefully there's a nugget here that I've shared with you. Whether it's the way I'm talking to it or the modulated approach, or even projects in and of itself will be helpful to you. If you're new or just getting comfortable with ChatGPT, I wanted you to see how I approach with my own four part framework, even a complicated task like this. And I'm giving ChatGPT the permission to think and work with me rather than just say go do this thing and I hope it works, that is really. I'll repeat this again. When somebody comes to me and says I've tried ChatGPT as an answer to this problem and it's not working, it's typically because they've rushed in and they said here's what I want you to do and they're dissatisfied with what it comes out with. Okay, so a little bit of work, right? Let ChatGPT do the heavy lifting. Engage it in a dialogue. Have it think with you. Have it create resources for ChatGPT, right? Have it do the busy middle. Have it do the heavy lifting and utilize those resources like your call transcripts or emails or whatnot. Get the information in early because the more you do up front, the easier it is to rely on ChatGPT to do the busy middle. Right? I hope that was helpful to somebody. Thank you for being here. If you want more information on what I do or archive on the podcast or some resources, the website for the podcast chat GPT experiment.com you will find. I do training sessions for groups and I do workshops and keynotes as well and one on one training. You'll find all you need for information there@chatgptexperiment.com Reach out to if you've got something that you want to share, please do. I love hearing how you're using tools. I love sharing it and if it's relevant to something I think the group would be interested in, I'd love to share it as well. Maybe even have you on. So that's it for today. And as always, as you use this tool, the biggest, I think attribute that will give you the most success is your own curiosity. So dig in and be curious. Okay, we'll talk soon. Bye bye.
- [30:41]
Kerry Weston
Well, the holidays have come and gone once again, but if you've forgotten to get that special someone in your life a gift, well, Mint Mobile is extending their holiday offer of half off unlimited wireless. So here's the idea. You get it now, you call it an early present for next year. What do you have to lose? Give it a try@mintmobile.com Switch limited time.
- [31:01]
Family Member
50% off regular price for new customers. Upfront payment required $45 for three months, $90 for six month or $180 for 12 month plan taxes and fees. Extra speeds may slow after 50 gigabytes per month when network is busy. See terms.