Introduction to Multiple GPTs in One Demo

I'm actually going to do a demo on how you can use multiple GPTs in one. It was inspired by a previous talk or demo that we had three months ago now. Was it three months ago? Three months ago, yeah.

We had some technical difficulties at the time. But it was a really great demo, and it really got me to think.

Audience Engagement and Background

So basically, how many of you have built GPTs on OpenAI already? Not many. That's going to be interesting.

Understanding the Concept of a GPT

So the concept of a GPT is fairly simple. It's a custom, or it's a pre-personalized version, I'd say, a pre-configured version of Chan GPT. It just has a few pre-configured instructions that it has every single time that you execute it. So I can give you, I'll give you a direct idea here.

Demonstrating the Mindstone ICP GPT

So I've got a GPT here that I set up, which is called the Mindstone ICP, so Ideal Customer Persona GPT. And I gave it some information on what Mindstone's ideal customer persona is. Very simple.

So here you can see the context that this GPT has. They're taking on the persona of a COO in a professional services firm, which is basically our ideal customer persona.

They've been hearing about AI. They mostly heard about productivity gains and automation and less about how it helps you improve the quality of your work, which is one of the things we often get into. afraid to move slowly, be left behind, keen to explore options, want solutions to be real and have impact, and focused on business outcomes very specifically.

The Development Process of GPTs

Now, to be clear, all of the GPTs I'll show you today are not particularly developed. I actually set all of these up this morning in order to be able to do this particular demo.

When you actually use these in production, you can get way more detailed on what is your actual customer persona. You can add some documents to it as well. So you can just upload your marketing material for your little customer persona and so on.

Chaining GPTs for Enhanced Functionality

It basically means that ChatGPT will take on the persona of this particular configuration, in this case, the CEO of a professional services firm. Now, where it becomes interesting is that you can chain these together in ChatGPT.

Exploring Different Customer Personas

So I have set up customer personas here, one, sorry, different GPTs, one which will help me with a brainstorm. Then I've got two, Types of customers. We've got the reluctant customer.

In our case, so what we do is we do AI training for non-technical staff. And 1we tend to end up with two types of people that go through the program that we have, which is the people that are really keen to understand how to use this technology and they're really leaning in, and the people that are being told by their boss, yes, you have to go through this program. Two very different profiles.

Creating a CEO Persona GPT

1So I have one set up, which is a reluctant learner, and the other one that is a keen learner. I then have a CEO persona, which basically has the context, and I'll show you that, has more of the context of the business. So here I gave it the context of what Mindstorm really is all about, what we do, how we do it, and then also I gave it information on the longer term business goals. So basically we believe that right now AI, practical AI training is the only skill that people are going to be willing to spend time on and spend money on.

But in two, three, four years from now, it'll be something that's assumed when you'd go into a job. So from a strategic perspective, as we're building the company, we need to make sure that we're not just building for the next two or three years, but we're actually also building a company that will focus on the future most valuable skills as we go through. So again, this is like the CEO persona that takes all of those things into account.

Introducing Diversity in GPTs

And then I have a devil's advocate, GPT. which the only role is to just take whatever was funded and construct the best possible argument you can find to try and deconstruct it. And then last but not least, I have a decision log GPT, whose only purpose is to look at a log of a conversation previously and provide a decision log at the end of it. Now, what is interesting is when you start to combine all of these together.

Live Demonstration

Live demo. you cheated before, means that things might go wrong. So we're going to see what happens here.

Using the Product Brainstorm Assistant

So I'm going to start with the product brainstorm assistant, which is the GPT that I showed before. So I'm going to say here, one of the problems we're seeing, this is not my keyboard. OK, there we go.

Actually, the biggest problem we see with users not understanding or not getting the productivity gains from AI is that they don't see how examples in job X, like marketing, apply to their specific role. say, operations. That's one of the biggest problems that we see at the moment.

For some reason, people know that you can use generative AI to go and produce some nice blog posts and maybe some social media posts. But many people, probably actually many in this room, would not quite see, OK, well, but how do I use it if I'm an HR professional? Biggest thing we have.

So I'm going to give the brainstorm assistant this particular problem. Now, the brainstorm assistant is set up to look at a problem and provide me with three potential solutions.

Sorry, first it's going to ask me a bunch of questions and then it's going to give me the solutions. So at this point, when I tried this out earlier this morning, it asked me a whole bunch of extra questions, but here it must think that it has enough already.

So could you clarify if the challenge primarily due to lack of relatable examples, or is it also about the difficulty in translating generous AI applications to their specific tasks? So it's mostly examples I'm going to say. Often this problem gets solved when you sit down for two minutes and you go through their workflow and you give them an example on how it could work in their day-to-day and suddenly that whole thing disappears.

In other words, do users need more tailored examples or do they need help in general concepts? Okay, I'm going to say examples.

Okay, interesting.

So it's coming up with three different approaches. Role-specific content tracks. Develop dedicated content tracks for various roles within the company. Or customized learning paths. Allow users to customize their learning paths by selecting their job roles and specific tasks that they perform. Peer-led... Peer-led role-based workshops, so organized peer-led workshops where employees form similar roles, from similar roles within company or across different companies, share their AI use cases. Okay, so these are three examples it comes up with.

I happen to not quite like all of them. that much, but that's not the purpose of this particular demo. You can go and dive into whatever the thing is. Also, because I set these up pretty quickly, I might have added more context to the brainstorm GPT so it knows more about what mindset is about in the first place, and it might have produced different ones.

Feedback from the Forced User Persona

So the corporate forced user persona. OK, so now it is running through based on the persona of the forced user. So this is a user that has been asked to go through our program.

What do you think of these solutions? And these are three solutions that we're thinking about adding to our product. What do you think of these?

So it goes through usefulness, appeal for the first case, the second case, and the third case.

Insights from the Eager User Persona

Then what I'm going to do is MindStone corporate eager user persona. What do you think of these? So now I'm getting more feedback from two different angles. Now again, the more information I would add to these personas, the more accurate the feedback that would come back on the idea that I'm trying to develop.

So it's giving me more data here. In order to go, so I'll just kind of go through what it comes back with here. So this is the eager user persona.

So I'm looking at here, approach one. So role-specific content. Clearly define... It's actually figuring out what that would mean. And then evaluation.

So this approach is highly useful as it directly relates to the user's daily task, making the learning process more relevant and practical. So indeed, role-specific content tracks. You can see how it would work, but it's a ton of work if we tried to go and do it.

What is the... The personalized nature of the post makes it appealing. And then here, usefulness on the customized learning paths. High level of personalization ensures content is relevant and practical for each user.

Customizable nature is highly appealing as it offers unique and user-centric learning experience. And you can see how it's rated it as well. So eight, nine, and then here, seven and eight respectively.

Let's see actually what the other did here. So this is one of the things that I... Okay, this is actually different. So here you get... It's written themselves the same?

Same scores. Yeah, same scores. That's what I had. So this is one of the things I would still have to kick out.

Considering the CEO's Perspective

So this is actually a direct specificity of some of these language models, which is that once you have a thread that goes through, you have to be really clear that it doesn't just predict the same scores that came from the conversation that happened previously, because it's all about predicting the next token. So you get stuck in some of these loops. So I was actually thinking for this demo, I was going to remove initially the rating because you end up with this type of behavior, but that you can do yourself.

What I would do from here, I'm not now going to say the, I'm going to take the ICP. I'm going to take the ICP. What do you think about this? So now, I'm going to look not at the user, but what does this actually look like for my buyer, the person who actually decides to buy this solution in the first place?

And so here, they would say, OK, we're looking at scenario two there. OK, so scenario one. directly addresses the need for relevance and applicability by providing examples for each role, helps users see how AI can be applied day to day, and so on. So slightly different approaches as you go through.

One of the things I would actually like to try is if you add a marketing step in between and you have a marketer that basically tries to package these three different solutions in a way that would be appealing to your ICP, and then you layer in the ICP, you can then get to another level. Now, before I actually try and get to a decision,

I'm going to look at the CEO persona. So here we're looking at content development resources. Significant resources are needed to develop high quality content.

And what are the considerations to take into account? Product manager would look at everything that everyone said, figure out what is the right option to go for.

Decision Making with the Product Manager Persona

What would be the next step? So I'm now going to do product manager persona. Which option do we go for?

It's now going through all the pros and cons of the three decision process. And then recommended option, customizable learning paths, step-by-step implementation. It actually goes through what would this mean?

What could you take out of it? Data privacy implements stringent data privacy measures. So this is directly from the feedback of the CEO that this was one of the things that needed to be taken into account.

Challenging the Decision with Devil's Advocate GPT

And now I have gotten to a decision, so I can do two paths.

I'm going to start with devil's advocate, because I want to figure out why was this the wrong decision to make. What's wrong with this decision? So now it's going to point out all of the different issues that relate to the decision that the product manager had taken previously.

Technically complex in cost, data privacy concerns, content coverage and quality, building and maintaining a comprehensive high-quality content library that covers all roles and tasks is challenging. Ensure that AI curated content remains relevant up to date with industry trends.

Actually, I wonder if it... OK, sorry, it's doing this for all of the solutions here. I thought for a second it had gotten confused with which option we chose.

So strongest argument against customizable learning paths. So it's just done the strongest argument against each of them instead of the solution that we had chosen, which is a shame.

Conclusion. you get all of the points that you can go through. And again, there's way too much text to go into a live demo.

But then last but not least, I am now at the end of a decision. I've got all this content that's there and we went through a whole bunch of things.

Normally, if I would actually use this in production, I would obviously challenge each one of these steps. It's like, I think this is true. I don't think that is true.

Like maybe feed a little bit of context for the persona, for the product manager, for the CEO. Like you can really spend not 15 minutes on this, but 45 minutes on this so that you get as much context in there as possible.

Summarizing with the Decision Log GPT

But I'm now at the end of a discussion, and I can do MindStone decision log GPT. Can you provide me with a log of this decision? It's now going through the full context, what has been discussed, what were the pros and cons, how do I summarize this in a way that if I hand over this particular bit to an external party, they should be able to read through and understand everything we just went through from all the arguments that have been presented from these different personas.

Now you can imagine, it even created a signature block. Love that.

The Power of Combined GPTs

That I did not actually build in. You can go way deeper on this, but you can start to see how powerful this is when you start to put all of this together.

These GPTs can be enhanced with all of the collaterals you have internally to the company. So if I wanted to really push this, and I would probably not do this in a live demo because then you end up with sensitive data that goes through.

You'd give it all the important information on your company when it's about the CEO and the strategic context, like are you having investment conversations? What are you thinking long term? Do you have strategic partners? All of that would be part of the CEO persona.

If it's the product persona, you'd probably upload all of your product documentation there. What are all the current features that are already part of your app so that it has all of that context to consider as it's making different choices?

That's an easy one. You can take all your marketing collateral. If you've done your job from a marketing perspective, you have key data on who your personas are, and you can upload that to the GPT. Same for the ideal customer persona, the one you're actually selling to.

And the more data you give, the more feedback you'll get, the more accurate these conversations become. And then the only thing that you do different here is that pushback, is that in order to get to the best bit, you don't just Take whatever comes out. I'm not suggesting you should like I'm going to take this and now put it in production next week. This was way too fast. You push back on a few of the things that you think don't work. But as a thought tool, it's extremely powerful. And you can now get to a point from a from a thinking perspective within one hour where previously you might have needed a full week.

Conclusion and Audience Q&A

So hopefully this was useful and it makes you think about different use cases you can use this for yourselves. And if you have any more questions, I have two more minutes, so I can take two questions.

Yes, Ben. How did you sort of, did you use ChatGPT to give you the personas for each buyer? No, I did that. Yeah, I did that in my head.