Introduction

Today, I am going to try, in the next 15 minutes, I'm going to build an email bot that's going to use ChatGPT to live answer your emails. So every one of you that has a phone, hopefully you have network here. You will be able to send in your questions, and it will be able to answer. And there's gonna be a small step, which I have faked, because basically we're gonna go from a very simple one to a more advanced one, and I'll walk you through what the differences are between the two, and the advanced one I have prepared in advance.

Audience Engagement

So, how many of you have already heard of, well, not heard of Zapier, everyone has. I mean, unless you were sleeping the last 10 minutes. How many of you have used Zapier before? OK, interesting, good start.

Technical Difficulties

Let's see if my internet will work. The internet's going to be the limiting factor here, I see. It's actually really going to be the limiting factor.

In order to counter that, I am actually going to try and hotspot instead. Yeah.

Apologies for the slight delay. It'll accelerate the rest of the demo.

Oh, come on. Don't think this has ever failed on me yet. Let's stop with that.

One more time. I'm gonna try, oh wait, I'm gonna see if I can do Okay, there we go, good.

Thought there that was gonna be a problem. And now my hotspot doesn't quite take it either. Looks like internet is just not, like live demos without internet is a hard thing to do.

Okay, well, we have something here. We'll see how long it resists.

Demonstration Begins

Okay, so here we go. I'm gonna start, I'm gonna jump. I'm going to jump right in.

Creating a Zap with AI Helper

So in the left top corner here, you have a Create button on Zapier. You can create a Zap, it's called. And they rolled out about two months ago, I think, maybe three months ago at this point, an AI create function, which is a little widget. It's not that useful, but it's still impressive, which is interesting.

1So here, to start, I'm just going to say the automation that I want to build. And so when I get an email in my Gmail account, send it to ChatGPT, and send an email back from my Gmail account. Very simple. So I literally just told it the automation that I'm trying to build here.

And on the left-hand side, you can see that it's working. So Copilot, hang tight. That's a new thing that they did a month ago, a little less, a few weeks ago. And it got it wrong.

So here you can see it got the trigger right, which is when an email hits my Gmail. I can actually add this. Then what I want is I want a ChatGPT account, but I never asked it to summarize text. I don't know where it got that. I'm still going to add that.

And then the fifth step is send an email, again, back in ChatGPT. But the way that it's done this, in this case, it's actually created too many intermediary steps. So this was not really that useful. But many times, it does get this decently right.

Configuring the Automation Steps

In this case, it didn't. So I'm just going to remove these steps in the middle. And I'm going to configure the individual steps.

I'm going to gloss over some of this a little bit. So I'm going to make it from demos at mindstone.com. It is going to look only at emails that go into my inbox. Important aspect, because the first time I did this demo, I forgot to do that.

And so it looked at emails in my sent box. And I got 100 people to send me emails. And then it replied. And then it looked at the sent box. And by the end of the demo, I had sent more than 10,000 emails.

And I was locked out of my Gmail account, which was not what I was looking for. So I'm testing it, which basically gives me a test result. It's looking at my Gmail account now. It takes the first result it can find. And it uses a test.

I'm now in step two. Step two, as I said, it was actually, it selected the wrong action. I was not trying to summarize something with ChatGPT, but what I do want to do is I want to send it to an assistant.

Introducing OpenAI's Assistants

How many of you have ever used assistance from OpenAI? OK, very few.

How many of you have used GPTs in the ChatGPT interface? OK, many more.

So assistants are the API equivalent of GPTs in the interface. I don't know why they named them differently, but they did. This is the most technical this will go.

So you can go to platform.openai.com and you can create an account. You get free credits for that as well. Basically the assistants are created here just like you would create a GPT.

Let me look at here, this one. So This is what the GPT looks like for me.

So these are the instructions that I gave the GPT. GPT is like a customized version of ChatGPT, or not a customized, it is a pre-populated context version of ChatGPT. So it has a certain set of contexts that you gave it ahead of time, and it'll always execute based on that context.

And here, this is the context I'm giving it, which is imagine you're an experienced salesperson, answer questions in a way that make you more likely to buy Mindstone's platform and solutions. The process is read an email that comes in, consult your knowledge documents. So I actually uploaded documents in relation to Mindstone's offering.

Compose the best possible answer to my query based on your goal, and then think about a thoughtful, non-obvious question that expands my thinking in a way that makes me more likely to buy Mindsled's platform and solutions. Respond to the query by writing an email reply using the template. This is an important step because otherwise every email really feels like it's written by ChatGPT.

By forcing it through a template, you will dramatically change the way that this comes out. So that's the assistant I'm going to use.

Setting up the Assistant in Zapier

I'm going to go here back to Zapier, my conversation with an assistant. I'm going to choose account, open AI, continue. And now, so as you saw in the process, I told it to look at the email, the thread that was going to send it, right?

So here I'm going to select the body of the, or I'm going to say, one sec, yeah. This is the thread. I'm actually going to do two things. I'm going to give it both the subject and the email, the body of the email.

I'm going to choose the assistant, which is the AI competency assistant. Everything else I can keep the same. I can test the step.

So now it's actually testing. It's taking the email that it took in the test from the first step. It's sending it through to the assistant and getting the answer back.

Refining the Automation Outputs

Okay, and then this is the, it's not technical at all, but this is the most finicky step, which is that basically you get so much information back. Zapier tries to cut everything up in a way that you can easily use it in next steps. So you get so much information, you have to now search for what is the field you're actually interested in. And in this case, I'm just interested in the response from Tag2BT, that's it.

I'm not interested in anything else around it. And that is called the field called full response. So here you can see what ChatGPT is actually answering.

Great question about the potential risks associated with investing in . Interesting, OK. That was an email that it took from the demo account. And it's answering the question here. I'm going to hit Continue. And then I want to send an email back.

Crafting the Email Response

So the whole idea was I create an email that an email bot is actually able to reply to your questions, right? So the email back is going to be two. I'm going to select the from email. And it's going to come from the demos account.

Actually, I'm going to call this... I think I called the assistant Ava. You can see here, actually, the template says Ava, MindStone AI sales director. So there we go. So I'm going to say this email comes from Ava. Ava from MindStone.

And the subject is going to be re... subject line of the email that came in. And the body is very simply going to be the response I got back... So the full response. And what I'm going to do is I'm actually going to include the email I got in the first place so that when I respond, it looks like a thread, like an actual conversation.

That's all I need to do. I'm not going to test this step because if I test this step, it actually sends an email out at this point. So I'm just going to skip the test. I'm going to hit publish.

Live Testing the Email Bot

OK. And now, whoop. If you, let me check that it's actually published. Yes, it is published. If you send an email, to demos at mindset.com, have a try.

Ask about AI mastery. So high level, so as you're composing your email, what we do is we go into companies and we help all the non-engineers within the company understand how to use AI in their jobs to either improve the quality of their work or to help them automate parts of it. So if you have any questions around that topic, we call it AI Competency, which is the program that we run, send it to the bot, demos at mindsum.com, and you will see that it starts answering them, your questions, in real time.

So as I'm doing that, you can see here Zap runs. I'm gonna just refresh. It takes a few seconds to go back and forth. So it's waiting. Obviously, it has to wait for your emails coming in.

Now, I was going to say, the other thing, and actually, I will wait here first for a second. I hope that some of you are sending emails, because a live demo without those emails is going to be a little bit harder. Okay, it is all well configured. Maybe it's Zapier that's taking a little bit more time this time around. There we go.

Okay, the first one came in and got answered. So, okay, so the first one, and if I refresh, there'll probably be more than after. So you can see here there's an email that came in. Let's see the data here. Where is the actual content of the email? How many atoms in the ocean?

Very good question. We will see what ChachiBT answered. So I'm going to look at the email, though, and I could go into step two and see what ChatGPT did directly, but I'm actually going to look at the data that got sent out from the last step, so the email that went out. Thanks for reaching out for all queries about how the Mindsome platform can support AI trainings. I'll be glad to assist.

Have you thought about... Interesting. Okay, so it's not answering the question, and I just want to point your attention to one thing, which I did do. which is part of the assistance instructions, is you only ever respond to queries about, ah, this is the competency. Interesting. Mindstorms AI Mastery, or AI Competency Program. So this is one of the things I put in the actual assistant instructions, that it's supposed to restrict itself to the actual topic at hand.

And in this case, you can see that it has not actually answered the question. But because the template that I gave it asked it to also answer a question, it's still asking a question to the user. Now, you can think this is a good way of doing it, a bad way of doing it. The important thing is it actually did execute on the instructions that I gave it.

And I see that quite a few more of you have actually since sent an email. Can you confirm that you actually got a reply back at this point? Yeah. DAVID J. Perfect.

Question about the formatting. So on the phone, the formatting is a little off. Is there a way to do that? Yeah, so I put plain body. You can actually do HTML content if you want to. You can make it as sophisticated as you want. And to that point, this is the next step. I'm going to switch off this particular step.

But I prepared another one here, which is the crowdfunding email final bot demo. But what I did here is I created another bot.

Introducing the Advanced Email Bot

Actually, if you got here, how many of you are aware that we're currently crowdfunding? But so right now, we're actually crowdfunding. We have a whole bunch of stuff, and I'll get into that in a second.

And this is what is really, this is what most people forget is that the first example I gave you is very simple. It does all the things that everyone always says, which is ChatGPT just creates text, just answers a particular thing. And this is the first example I gave you. Here, I have a much more complicated flow.

1And I say complicated, it took me about an hour, hour and 10 minutes or so to put this one together. But what happens here is I get an email. I have one assistant whose entire purpose is to tell me if it thinks that based on the data that it has, it can confidently answer the question. And the only thing it does is it gives me a confidence rating back from zero to 100.

I have another assistant whose entire purpose is to figure out is the question that is coming into the email related to AI training or related to our crowdfund and how people can get involved there. I then split it up into different paths. If the bot thinks, if it has a confidence rating of 90 or above, I just let it answer the question.

I then have a second split of the path, which is if it's in relation to training, I have a bot that's trained specifically on the data in relation to training. When I say bot, in this case, every time I'm meaning assistants, so the ones that I showed you. I can actually show you here. You might have seen it before.

the crowdfund mastery qualifier this is the one that tells me is it training or is it crowdfund i've got the crowdfund expert assistant which answers questions on the crowdfund i've got the crowdfund mastery confidence which gives me the confidence zero to a hundred and then i have the ai competency assistant which is the one that answers questions about the ai training so i'll go back here if the confidence is over 90 then if it's about training, then I send it to the training assistant and the training assistant answers. If it's about crowdfund, the crowdfund assistant answers.

If the confidence level is between 70 and 89, then what I do is almost the same thing. The only thing I did here is that in the CC field, I add myself. This is to make sure that if it thinks it can answer the question, but it's not quite sure, I want to make sure that I can keep an eye on the answers that are going out so that I can correct it, so that I can make sure that I can correct the record if something wrong has gone out. But otherwise, I do the exact same thing.

I have the AI competency one that goes out, and I have the crowdfund answer that goes out. If it's under 70, the only thing it does is it sends out an email, hey, first name, thanks for your email. I'm afraid I don't know how to answer your question, so I'm looping in Josh to help.

I don't look at the ChatGPT answer. I just send the question through to myself. So if you all now send an email, either on the crowdfund that we're currently doing or about how our AI competency program works, you should have different scenarios that now evolve.

Closing Remarks and Crowdfunding Pitch

And as you do that, a little bit of context on the crowdfund. So I mean, we're actually going, for those of you that are interested, so in the next week or so, we're going to be closing it because we're already in overfunding and we already got more than what we thought. So if you are interested, there are two things that are interesting here.

One is we got here through the community, like literally. So this is not for the money. We have a whole bunch of professional capital that's in the business and we can raise more professionally. The whole idea was to open it up to everyone else.

The second bit is that all the investors get a discount on AI Mastery. So if you're interested in actually using, joining one of the cohorts to help understand how to use this in your job, then becoming an investor gets you a discount on that.

Yeah, and in that time, I hope, I didn't see too many people actually send an email. I do want to demo the actual result that goes on here because you can highlight the path that goes through. So I hope. Okay, there are a few. Good, good.

So now I can show you here a question that came in, and you can see the path that gets highlighted. So the email came in. It's going into path two. So confidence level was between 70 and 89.

So I'm interested now what was the actual question that came in. I wonder... Let me just go into the question and come here. Okay. Inquirer.

I would like to know how I can hire your services for my company. Do you offer special packages training? Okay, interesting. Interesting that it thought... I wonder what the confidence level is that comes out here.

So I'm going to look at that. The confidence level here that it thought it had to answer this question was 70. I would have expected this to be slightly higher, because I think this is covered in most of the documentation that we have. But let's look at the answer that actually comes out.

So the email that came out is thanks for reaching out about Mindstone's AI training platform. We offer both structured packages and tailored training solutions fit for the specific needs of your company. Our AI competency program is divided into three core levels, starter, chaser, pace setter, and so on.

Additionally, we provide highly customizable learning environment, and so on. Have you thought, and then the question, have you thought about how adopting this structured, flexible approach to AI training could rapidly enhance your team's productivity and job satisfaction? Interesting. Thereby contributing to your company's growth.

The whole idea was it ends with a question that helps the user kind of go one step further. I thought this was pretty good. So one, it actually got the answer correctly, even though it thought the confidence level was 70. It has looped me in.

So if I look at my emails now, I should have received Yep, I actually received that same email in my inbox as well. And that's it. That's the end of the demo.

This is just to show that, very simply, there was no code involved in any of what I did here. There was literally just a series of steps and a little bit of reasoning.

It did take me, out of the hour and 10 minutes that I spent to set this up, I would say 30 minutes was to think about, what do I actually want to build? What do I want the flow to be? And that is the work that you have to do. But actually putting in the flow is extremely easy.

And I think too many people are not yet understanding the power of everything that's about to come. I am in the process of building an entire sales agent using exactly this.

And just look me in when I can be useful. But if questions can be answered where I am not useful, why would I be answering them? And this is going to be the norm as you go through because everyone can start to build these themselves.

I often talk about we're going to an era of the compounding productivity, personal productivity platform, which is that each one of us in our respective jobs, you can go and do your job and you can be extremely good at your specific skills. but it's going to be really hard to start competing with someone who has built up many of the routine tasks, who can spend twice as much time on the tasks that they actually need to spend time on, and they removed all of the tasks they don't have to spend time on.

There was a survey, it was Microsoft, One week ago came out with a big report on the use of AI in the workforce. 70% of people in a position of hiring today would favor a candidate that is AI literate and junior than a candidate that is senior without knowing how to leverage AI. 70% of hiring managers. Interesting stat, right?

So all of that to say, I hope this was somewhat educational, and come and talk to me if you're interested about the crowdfund afterwards.