Unlock Your Accounting Potential: Master Prompt Engineering In 6 Levels!

But now we're going to be talking about prompt engineering mastery and the six crucial levels as you develop your skills. So let's first start though with what on earth are we talking about? What do we mean by prompt engineering? What is it and why is it so important? So I'm sure you haven't failed to notice that over the last 12 months there's been an explosion of developments in English AI and in particular large language models, which is a type of AI, arguably the most powerful and certainly most relevant to us in the accounting profession. There are many large language models. The most well known is ChatGPT. So I'm going to focus on ChatGPT today, but almost everything I talk about will be relevant. If you prefer to use Gemini or Claude or any of the other large language models that are available, I'm going to focus on ChatGPT. It's the most common one at the moment. And prompt engineering simply means it's the skill of communicating with an AI and it's a skill that we need to learn. It's extremely important because AI is coming fast and we need to stay ahead of the competition to be at the leading edge of the accounting profession. We need to be embracing these tools, embracing tools like ChatGPT and other large language models. Models. But we also need to figure out, well, how do we interact with them, how do we talk to them? Because it's like a new language to some extent we have to learn this skill because when we do, we're going to get much better results. We're going to get much better results from using things like ChatGPT, much better outcomes. And when you master prompt engineering, what you will find is that you will save hours and hours of time every single week when you use these tools. They're extraordinary. Our entire team uses ChatGPT. We have. Thursday is our ChatGPT day. Everybody spends the day just learning about ChatGPT, playing with it. We have a little while ago we had an hour long team meeting to share what we're doing in ChatGPT. And the reason we do that is because my team is so much more productive. They can get things done that might have taken 2, 3 hours before they can do in 30 minutes. Because of using AI. It's extraordinary what it can do. And so if you don't do this, your competitors will. If you want to build a more successful, sustainable accounting firm, more valuable accounting firm, you need to be on board with AI and understanding prompt engineering. It's one of the most important skills we should be developing because It's a communication skill, but communication skill with AI. And some of you, if you're regulars to my live streams, will know I've mentioned stories.

In for example Bloomberg that have been posted saying that some Companies in the US are willing to pay $300,000 to prompt engineers. It's that highly regarded as a skill.

If we don't learn prompt engineering then we're going to get very mediocre results. Let me just give you an example. I'm just. Let me give you an example of I used this in some training yesterday. Let's imagine I know that many people who are using who are using ChatGPT and large language models, many people are using it to answer client emails and that's a great thing that you can do. It can save you a huge amount of time answering emails. However, as many people do say who haven't learned prompt engineering, it doesn't always give great results and it doesn't and that's down to the prompts. You may have heard the phrase garbage in equals garbage out. If you don't ask great questions of the AI, you're not going to get great answers. So we have to perfect the questions, the tasks that we give it. So for example, if you went into chat GPT and said write an email to a client telling them about a filing deadline at the end of next month. If you did that, you would get within a couple of seconds a email, it might say something like this Subject Upcoming Filing Deadline Alert Action required Dear client's name, I hope this message finds you well and thriving. I'm reaching out to remind you of an important upcoming filing deadline that requires our attention. The deadline for specify the filing, e.g. annual tax returns, quarterly financial reports is set for the end of the month specific date. They're just placeholders for you to fill in the details. To ensure a smooth and stress free process, it's crucial that we start preparing now. This approach will not only enable us to meticulously review all documents, but also to identify and address any potential issues. Well advance and it goes on to say here's our proposed plan of action and it comes up with in this particular case, it comes up with three things that we tell the client they could do as part of an action plan for this filing deadline. Now this is an okay response. It's reasonably well written. It's not necessarily my writing style. It also requires quite a bit of work because there's some placeholders there. I have to fill in the details. Also, it's very generic. I actually don't like some of the wording. It's not my style, as I said. Also it gives a three step action plan which may not be appropriate. It's guessing at what an action plan might be for an email reminding of a client deadline. So whilst it's giving an okay response, it's not really a usable response because there's so many things we would need to change with that. And the reason it's not a very good response is because our prompt isn't very good. We've just simply said write an email to a client, tell them about a filing deadline at the end of the month. So we need to do better prompts, we need to figure out what do we need to tell the AI to get the results that we are looking for. So level one, level one of these six steps I'm going to go through today, level one is we need to understand what are the essential foundations of a prompt. And I think there's really five key essentials that you have to have in a prompt to get a half decent results. So if you're new to prompt engineering, the first thing that you want to do is build in these following five. So let's start with the obvious one, the task. The task is what you want it to do. And in this particular example I gave you, we've got that there. Write an email is a task and it's almost certain you will have the task in there. But there's so much more that we should build into our prompt. We have to give it context. Context is the background. We have to give it information. Without that context, it's not necessarily going to give a great result. We've given a little bit of context. We said write an email to a client. That's the task about a filing deadline at the end of the month. But that's the only information we've given, that's the only context. A filing deadline, a fine deadline for what? For a tax return, for annual financial statements, what is the month. And so all those things that we now have to go and fill in, fill in the blanks in that email. It's because we haven't given it the context, we haven't got, given it the details. So that's the second thing that you really need to be master is being great at giving context, giving some background information.

Then we've got Persona. Now this is an interesting one. The Persona is the role that you want ChatGPT to play or the large language model. We know that with ChatGPT its default position is a helpful Assistant. It's built into the systems prompt and it does a fairly good job of a helpful assistant. But when you ask it to play a different role, a more specific role, then you get better answers. So, for example, if you're doing a rule asking to do a report based on analyzing financial data, if you ask it to play the role of a financial accountant skilled in understanding numbers, you will get a better response. It will be written from the perspective of a financial accountant and therefore more appropriate to readily use and provide to a client than if you don't give it the Persona, if it acts as a helpful assistant. So we want to have a Persona, we want to have a task. We want to give as much context as possible. We may also want to refer to the audience who's going to consume the content or the response that the LLM creates for you. Is it for your purposes? It might be, but it may be for somebody else's purposes. If you're writing a report or an email, for example, or a marketing piece, who is the person it's aimed at? Who is the audience? So the more details you give.

In the case of an email, who is the client? What's the client's name? What business are they in? If you give a little bit of information about the audience, the client, when you're asking it to create an email, it's going to be a better email.

And then we have response format, which is where we get onto level two. So level one, at the very basic level, you need to go beyond having simply a task. Give it background information, it's give it a Persona and talk about the audience that it's aimed at and you'll get a better result. But we're just on level one. That's really the basics. There's a lot more that we can be doing and should be doing. So let's move on to level two. Level two is now we want to master the response format. So when we give it a task, when we make a prompt, we want to have some control over the format in which it responds, particularly if that is a response we're going to use in some way, like an email to a client, like a report, for example. So let's talk about response format. And there's a number of things that we can think about here that will make a big difference to the result that you get. So the response format might include. These are the main things it will include. Firstly, what's the output type? What are we looking for here? Is it an email? Is it a research report? Is it going to be a tweet? Is it going to be a Facebook post? Depending on what the output type is, that will influence and affect the response you get. Now, in the case of the example I've just given, write an email to a client, the output type is an email. And so it's written it in the format of an email, which is great. So we've done that bit of it in our, in our poorly constructed prompt, we've given it the output type, an email. Because if I said instead, write a research report to a client, telling them about a filing deadline at the end of the month, you would get a different response because it would write it in the format of a research report. So that's the output type. Then we have the format now that can include things like do you want it to be formatted with bullet points, bullet point style, for example, do you want to be formatted in a table? That can be really useful so it can have some control over the layout, the format and next things that are really important is the writing tone and the writing style. These are particularly important if you are creating something you're looking to create, for example, an email to the client. You want to make sure that you tell the large language model what tone of voice and what style of writing. Because what you really want to do is get the email to be in your style of writing, in your tone of voice. And so there's a bit of experimentation here. You'll need to figure out what is your writing style, what is your writing tone. Once you understand that, you can then instruct ChatGPT or whatever large language model to then use that, and then you'll get much better results. If you want to create an email in, it'll then be in a format or in a style of writing that works for you so you don't have to edit it and change it. Now, those two things, writing tone and writing style, are very important. They can have significant impact on the response that you get. So for example, you could ask it to create an answer, an email in a professional style, a professional writing style, and that may well be appropriate to you. But if you said, if you gave exactly the same prompt but asked it to write the email in a humorous style, you'd get obviously a very different style of writing. It would be humorous, sometimes funny, sometimes not so funny, depending on how the language model is performing on that day. But sometimes it'd be very funny. Just change. And making a small change to a prompt, just changing the style can have a huge impact on the choice of words, the structure of the words. So it's important to understand what is the right style for you and the right tone. So with an email, your email might want to convey a sense of urgency. If the deadline's tomorrow, you might want a more urgent tone.

If it's because you're writing to the client, because I haven't paid you, you might want a more aggressive and angry tone, for example. So think about the tone of voice and the writing style that you want to use. Other things that you might want to think about in terms of the response format is what about the quantity? In the case of an email, perhaps not so appropriate, but it may be that you're asking for, you're using it for some ideation. You want to come with some ideas. You want some ideas for what you could do to grow your accounting firm. Well, how many ideas do you want? 5 ideas? 10 ideas? 15 ideas? Think about the quantity of the output. Then you've got the length. Do you want it to be a long email or short email? What are you looking for? You will tend to find that with things like emails, if you don't specify, it can ramble and be really, really long. And then you have to spend some time shortening it. So you need to give it some instructions on the length. And then last but not least is also the language. Now, this may not be too important other than the fact that if, for example, you're in the UK, then you might want to specify British English. Whereas if you're in, let's say, the US and you put American English, you will get slightly different responses. So when I say language, it's not necessarily French, German or whatever. It could just simply be the dialect or the country that you're from. American English versus British English is a good example. Okay, so that's the second level. Level one is the core, the essential bits of a prompt. Level two is let's now build into our prompt. How do we, how do we control the response itself, the outcome, how do we get the outcome? So it's much more usable for our purposes, whether it's an email, a report, a ideas, etc. So let's now move on to the third thing that we need to master. The third level is, and for most simple prompts, that's going to be great for you, but very often we're doing really important things. It may be you want to do a report for a client about tax planning, you want to analyze financial data. And so the more important the task you are giving, the more Important it is to really think about and understand what are going to be the right prompts. And so let's go on to the third area. The third level is now moving on some of the advanced components of a prompt, the advanced elements that we build into our prompts. And so those advanced elements will be a goal. Examples, templates, steps, and avoid. So let's start off with the goal first. If you tell ChatGPT what the goal is, what's the out, the end result that you are looking for? When it understands better what you are looking for, the end goal.

It'S going to come up with better responses because it knows where you're going, it knows what we're trying to aim towards. That's particularly important if we're going to be doing a prompt where we're going to have some back and forth and building on something, working towards something. So your goal might be to build a marketing strategy for your accounting firm. That's the goal, the marketing strategy. But the first step might be to think about, let's start to analyze the business first and understand our services, our products, the features, the benefits. But we understand by telling it the goal, it knows where we're going. So a goal is a great thing to put into a prompt. Then another great thing would be steps, giving it the steps to go through. Sometimes this for a simple prompt that's not always required. It may be that for some of your prompts, they're simple. It's do this one thing, write this email. But let's imagine that you are using it to help you build your marketing strategy for your firm for 2024. Then there's a series of steps that you would want it to go through. You can't just say build me a marketing strategy and suddenly by magic, you've got this amazing marketing strategy. You have to build it through a conversation by going through steps. And so you might have Step number one is analyze my firm, analyze my business, make sure you understand it. Step two, tell me what the features and benefits are of my accounting service. Step three might be let's work on my ideal client profile. Step four might be, let's start to understand what are the, what are the wants, the fears, the aspirations.

What are the frustrations of my ideal client, which is really important for marketing? And then the next step might be, having now understood all of that, let's start to pull together some words to go on the website, some words that we can use in our marketing. So a series of steps. When you set out in your prompt the steps you wanted to go through. Then you'll get much better answers. Particularly for more complex tasks where you want it to think things through, things like analyzing financial statements, there's probably a series of steps you'd want it to go through. So we have a goal and we have some steps to get towards that goal. What else might we build into our prompt? Then we might build in what to avoid, what to avoid, what we don't want it to do. That could be the language. For example, one of the things I find is that, or used to find until I fixed it, is that by default, when it's writing stuff, it often uses language that I find really cheesy. I hate it. I used to find that whenever I was writing a blog post, if I was writing a Facebook article, if I was writing the description of a live stream I'm doing, it would often use words like, this is a game changer. This is going to revolutionize your accounting firm. And there's these words that I just, personally, I hate them. They're overused all the time. And you can tell sometimes when somebody's used ChatGPT to create something because it's full of These words that ChatGPT tends to use this cheesy language. And so one of the things that you can do is in your prompt is say, never use the word game changer. Avoid using the word game changer. So think about putting into your prompts what you want it not to do. Two more things in this level, three that I think are the most important, and that's examples and templates. So a template is how do you want it setting out? Let's say it's a research report for a client. Then if you wanted to do a research report, perhaps into tax planning, for example, then you may want it to do a number of sections in a certain order. You might want to have a heading that's an overview you might want at the end a heading that's a conclusion. You might want to have at some point the main content, there's a heading, and then you want it in bullet point format. You want to specify exactly what you want that report to look like. And we can do that by giving it a template. We can say, I want you to create a research report using the following structure, using the following template. And then we outline the template and. And you'll get a really good answer. That is, it's one of the most powerful things that you can do. I use this in most the prompts that I use, I specify exactly what I want it to look like. Now you can take it another step further and that's by giving examples. Giving examples. So this is important, particularly if, let's say you wanted to write in your writing style, then as part of your prompt, give it an example of something that you've written that you really like. An example of writing styles. If.

One of the things I teach is marketing, so I've written, I have a standard system now for coming out with headlines for emails, headlines when I'm promoting events. And what I do is I give it a series of examples of great headlines that work really, really well. And then it will just follow that and come up with similar, similar headlines, but using the new context, whatever it is the session's about. I can't understate how important examples and templates are in getting just the right response that you want. Okay, that's level three. Level three is to use those more advanced components of a prompt. Let's now take our prompt engineering to another level, to level four. And at this stage, I'm talking specifically about ChatGPT and, and I'm talking about using, becoming a ChatGPT+ user. And I know there's many people here who are a ChatGPT plus user. If you are, just let me know in the chat box, let me know if you are using ChatGPT plus and also put in the comments what you think, why you use the plus account. So it's not the free account, you do have to pay for it. But one of the big reasons why you would want to pay for a Plus account, ChatGPT plus, apart from the fact it's crazy cheap, is it allows some amazing things, including some great tools. So level four is when you start to use other tools effectively in as part of your prompts. So what are those tools?

So one of them is dall e dall e3. So within ChatGPT you can create images. I'm going to brush over that because that's probably the least important tool for us in the accounting profession. But nevertheless, it's really easy to, to create images. Perhaps the way you might do it and the way I use it is if I've got a prompt to analyze an article and then write in my writing style a summary of that to be used as a Facebook post, to post into Facebook. Then one of the things we know about Facebook is if you add an image as well, then people tend to click on and they'll see your post much more likely if there's an image. And so I would just simply, if I've asked it to go to Facebook post. I would then have an extra step that says based upon this Facebook post, create a suitable image for Facebook. And in a matter of 10 seconds or so there's an image I can download and I can add to Facebook to go along with the Facebook post. So Dall E3 is really cool. Probably limited use in the accounting profession, but there's still some things you could use it for. Second tool you have access to is what they called GPT Vision when it came out last fall, last autumn. GPT Vision is the ability to upload images into ChatGPT, which may not sound like much. It's extraordinarily powerful. A picture paints a thousand words. So I talked about context. Giving context is really powerful and one of the ways to give context is using an image. So for example, I did a session recently in the Value Pricing Academy, teaching marketing back in January and I have a slide for it. The slide is a graphic, an image which explains the four key things to build, the four steps to build into a Facebook ad. And I simply just took that image, I uploaded it into ChatGPT and then based on the context I'd built up before that, which was an ideal client profile, I said, right, I would say I said writing this style, a Facebook ad for this ideal client. And it did an extraordinary good job. It read the image, could see what I meant, what it meant in there by the way, to structure the words and so on from that graphic. Vision is so, so powerful. It's amazing how it can understand things. Now from a more practical point of view as an accountant, that might mean, for example, you could upload graphs and charts and get it to interpret it. If you've got some sort of dashboard, an Excel based dashboard with graphs and charts for your clients with their numbers, you can copy and paste and put that in as the, as the graphic and then it can understand that you can ask questions about it. Next tool that's really powerful that you get with the plus account is browsing. Browsing. With Bing you can browse the Internet and that's really, really powerful. Particularly if you're doing things like you're using ChatGPT to research, do some research to help you with some tax planning. Without the browsing capability, the problem you have is the knowledge Cutoff that with ChatGPT4, the information, it's only been trained up until April 2023 and much earlier for the free account. But browsing allows you within a chat to then browse the Internet. Now you do need to invoke it, you do need to specifically say, use your browsing capability to research this. And if you tag in, use your browsing capability, it will then go and look to the Internet for the answer. Very powerful if you're doing things like tax research. Because if you don't say that, it may just simply use its knowledge up until April 2023, which might be okay. But if it's the latest tax legislation you want, it might be out of date. Okay, so the browsing functionality is really powerful. Even more powerful for us in the accounting profession is code interpreter. They did rename it last year to Advanced Data Analysis and now they seem to have gone back to Code interpreter inside their GPT. So it's a bit confusing. I'm going to stick with Code interpreter as the name, but some people might know it as Advanced Data Analysis. Now this is really powerful. Some people say that ChatGPT can't do math. It, it can do extremely good math when it uses code interpreter. And so this is powerful. If you upload, for example, an Excel spreadsheet financial data, what you can do is you can upload financial data and then you can ask ChatGPT, analyze it and write a report based on that. And as long as you specify to use code interpreter, what it does is it will then write Python code. In seconds, it'll write Python code to understand your Excel spreadsheet, to do all the calculations it needs to do, and then it will give you the answer and then you can quiz it and you can have a conversation around that. It's so powerful for financial data analysis. And then the last thing that you can use with ChatGPT plus is you've also got access to third party GPTs. I'm not too excited about those, but I'm very excited by GPTs as you'll soon see. So let's move to level five. So level four is using the tools effectively to go beyond just simple word based prompt. That's level four, but level five. Level five is where we now start to create templates. This is where you're going to save hours and hours of time. You see, we've talked about all the components of a great prompt and you might be thinking that's a lot of stuff. In fact, I was teaching back last month, I was teaching prompt engineering in the Value Pricing Academy and I showed people a prompt template.

And this prompt template is, think of it as a checklist for all the elements to have in a great prompt and how to structure that.

And you might be thinking, yeah, that's a lot of stuff to think about that's going to take me so long to write the prompt, I might as well do the answer in the first place. And you would be absolutely right. If, if you're just doing something fairly simple, if you just wanted to come up with five ideas for growing your accounting firm, you could do that in a simple prompt. But sometimes we want to do things that are really, really important, and sometimes we want to do things that we are also going to do them over and over again. We're going to do them frequently. And so this is where creating prompt templates is really powerful. Let me give you a non, a non business use case for a minute. So one of the things that you, you probably do from time to time is travel. Perhaps you go on vacation, you travel, and perhaps you want to do some research. You want to know where should you visit, where should you go to. And the old fashioned way of doing that would be to go to Google and say, I'm going to Croatia.

What are great things to go and visit in Croatia? And it'll give you the response in Google. It'll give you all these different posts. The first bunch you have to ignore because they're sponsored posts, they're adverts. We have to wade through the adverts to find things like people's blog posts. But then we're overwhelmed with all these different blog posts on people have been to Croatia. And we start reading them and we realize they all recommend different things because everybody who writes a blog post about things to see in Croatia is writing it from their point of view. Somebody who loves museums will tell you about all the great museums to go to. Somebody else who loves walking will tell you about all the great national parks. And so we can spend hours going through Google searches and going through blog posts to find the things that we want to go and see. But with ChatGPT, we can be very specific about what we want to do. And the more specific we are, the more, the better the prompt we build, then the better results we're going to get. And if you're going to be doing traveling frequently, then it's worth creating a great prompt that you can use over and over again. So one of the prompts that I did recently was a prompt for coming up with a travel plan where I would simply give it a few details, like how many days going to Canada. Then I have my role, my contacts, responsibility, instructions of what it wants to do step by step. I then have an example output, because examples then control the layout and you might think this and some rules and so on. This is a crazy long prompt. But what we're doing here is recognizing these are things I might do regularly. Let's create a prompt and make it so it's a template that we can use over and over again. So all I have to do is cut and paste that answer a few questions at the top here and it will create an answer something like this. So I've got. I'm going to Canada. Sarah and I go to Canada in September and we want to go and visit some areas between Toronto, Montreal and it's come up with a travel plan of places to go and visit in exactly the format that I'm looking for. It's a great response. There's actually more things I asked for as well. It gives me a draft ask, give me a draft one. And if I'm happy with that draft, I'll say, yep, that sounds like it's pretty good for me. Then it'll do a more detailed one and so on. It's a great prompt for getting a really good quality, detailed travel plan that suits my needs because in the prompt at the top it's got things like my preferences and what I dislike. I don't like museums. I'm an outdoors person. I love architecture. I love botanical gardens. I love beautiful scenery. I don't want to be cooped up inside. So by specifying that, it will come up with an itinerary that suits your specific wants and needs from a travel plan. So this is level five. Level five is about thinking about all the things that you might want to do on a frequent basis and building a brilliant prompt that you can cut and paste in whenever you want to do that prompt. So in the accounting profession, practice practical examples might be. Let's say you are doing financial data analysis. That might be something you do regularly. So let's think about what is a great prompt for analyzing a profit and loss account. Let's come up with a great prompt with giving it some steps, giving it some context, giving it the way that we want it laying out. Let's think about that turn into a template that then in our firm, anyone in our firm can then do that. They can follow the template and they'll get just as good results as you. Because it's all. It's because it's all been thought through carefully now on level five, let's just take this to another. Let's take level five to another level because my favorite feature in ChatGPT, if you have a plus account is back in November they introduced GPTs. GPTs are extraordinary. Now there are third party GPTs I can't get excited by, but GPTs, the real power of them is think of them as mini systems. Mini systems you might think of as a chatbot. So if I just go back into. So if you go into ChatGPT, you'll see that if you've got a Plus account you will find that on the left hand side as well as all of your history of chats you will find at the top you'll have icons for GPTs and you can create, you can create GPTs. It's really easy to create them. A GPT is one way of thinking about it is it's a prompt template that does a specific thing. And so anything that you do frequently you just call up the GPT and have a specific chat with that and it's already been pre programmed with a prompt that will give you the output you want. So examples might be and something I'm working on right now for members of the Value Pricing Academy. Let's imagine for argument's sake that you're a bookkeeper or even an accountant and you ready to clean up work. And you'll know that one of the things I teach when I teach pricing cleanup work is that you should always have a price. You should charge a fee for doing the initial diagnostic to enable you to then do a proper quote for the full job. It's the diagnostic review. Now somebody asked me a little while back in the Value Pricing academy, how do we make to justify the fee, how do we make that diagnostic review more valuable? And I said, well, you created a really valuable diagnostic report and one of the things that you can do is you can have a GPT to do that, where you just tell it what you found wrong in the records and it will create an amazing report in minutes with real value and the way it's structured. So if you do lots of diagnostic reviews, then you could create a GPT for that. The equivalent, if you're an auditor, might be the letter of weakness. Well, you could build a GPT to create a really powerful and valuable letter of weakness in a consistent format over and over again. It might be you create a GPT to analyze financial statements in a particular way. It may be that you create a GPT to answer client emails and you've trained it in your writing style and how you do. You say, dear John, or is it hi John, hello John. That's all just built into the way you've trained the GPT within your prompt and it will then do much better answering client emails. It may be that you want a GPT to build an ideal client profile. If you're into marketing or complete marketing strategy even. I've got GPTs for members of the Value Pricing Academy on how to create their packages. So if you're not sure how to create a bronze, silver, gold level of let's say your cleanup service based on I've given it details of my training so that it will then go through a process and build your packages for you using a GPT.

It's endless, the possibilities, these little GPTs that will perform a very specific task exactly how you want it in minutes that you can then get anybody in your team. I've just done, for example, I've done my daughter, one of my daughters has done a 30 page book review of my last book. It's this one, value pricing for Accounting Professionals.

Which if you've not read it, watch this space. We'll be sharing a 30 page book review. That 30 page book review is extraordinary and it was created from a prompt. We spent a long time crafting the prompt because our plan is to then do other book reviews because these are valuable that we can share with our audience. So that's level five. Let's move on to level six and then I'll answer questions. I am looking and I can't see many questions going in at the moment. Remember, put two capital Q's in front if you have a question. Level six. Okay, level six. Then the next level is be at the leading edge. Be at the leading edge. And what we need to understand. If you want to really master prompt engineering, then you need to understand a few things. You need to understand the limitations of large language models. The limitations and how to get around that. There are so many limitations, but we can get around them. For example, hallucination is a fun one. Hallucination is. It has a tendency to make things up because what large language models do is they are based on probabilities. They predict what is the most likely next word. They don't actually understand what they've written. It just predicts the most likely word. So let me just give you a fun example. I wrote a blog post a few months back on clearing debt and I decided to do a prompt where I said to ChatGPT, Read and analyze this. And I then posted the link to my blog post on clearing debt. Now I was using 3.5. ChatGPT 3.5 does not have access to the Internet so it could not go and look up that blog post. And since the blog post was written after the knowledge cutoff, there's actually no way in the world that ChatGPT could know what that blog post was about. But nevertheless, the answer was the article titled Clearing Six Steps to Financial Freedom provides a comprehensive analysis, comprehensive guide to managing and eliminating debt effectively. Here's a breakdown and analysis of its key points and it goes through and it analyzes the six key points in the blog post. However, that was not the title of the blog post. The blog post had five points, not six, and they were different five points. It made it up, but in a convincing way. It comes across if it really understands what that blog post is about. Now, what it did because of the way it works is it saw my prompt read and analyze this and then a URL. The URL had blog clearing debt. So it knows it's a blog post, it knows about clearing debt. So it used its probability to work out what might be the most expected title of that clearing six Steps to Financial Freedom. And what might be about. It made it up completely. In fact, I asked it, I said, how do you know with absolute certainty that is what this blog post is about? And it said with absolute certainty, I confer the content of the blog post based on its title, which it's not the title. And it went on to say, while I don't have direct access to the content of the blog post, I can accurately deduce its main topic and key insights by analyzing the title and drawing from my knowledge an understanding of common financial advice. It was making it up. Okay, it's never read my blog post, has no idea what it's about. That's hallucination.

And whilst that's a fun example, hallucination is worrying if you are using it for doing things like research, like tax research. The good news is there's a ways around that. By getting it to cite, for example, citations, by giving you the facts, it's assumed there are ways that you can work around limitations. So you do have to be aware of the limitations so you can create the workarounds. And you also have to be aware of the fact that this is such a new space, it's moving so fast, it's changing. And so at the level six, you need to be aware of the research of what's happening. So for example, just to give you a few examples, back in December of Last year.

Google DeepMind did a research paper 7th of December 2023 and looking at large language models and one of the things that they were talking about was this idea of chain of thought. So if you want to do something where you want to think things through in more detail, asking it to think step by step gets a better result when you insert that into the process. It's called chain of thought. I don't go into the detail right now, but it's an advanced level. But some other cool things that we found is in research, there was some great research done 26th of December 2023. And in this research they studied what are some of the things that can get better results from prompts. And one of the things that was really fascinating from this research study done by Sharat Mirzakhan and Shen Back in December 2023, it was called principled. And principled instructions are all you need for questioning LLAMA and GPT. And one of the things that was really interesting they found in that was, and there's been other people have confirmed this as well. For example, a Twitter post is that if you say in your prompt, I'm going to tip $300,000 for a better solution, they found by testing it, you get a better solution, a more detailed answer. Now that sounds absolutely crazy, but then again, it perhaps isn't because large language models are trained on.

On humans. They're trained by humans and on humans and so they recognize some little quirks. And, and so we are now discovering through this research that sometimes something is crazy sounding like, like giving, offering to give it a tip. It will give you a better answer. Now this is such a new space right now. New stuff's coming along all the time. So level six is about being at the leading edge. It's about being aware of all the research, what's going on and what's working today in large language models. And AI isn't necessarily going to work in six months from now. There are things being discovered all the time and also large language models are developing all the time.

If you found anything useful in this last 50 minutes, please do me a favor and hit that like button. I know I'm going to answer questions next in a second because we have got some questions coming now, but let me know if you've liked it. Also just put in the comments box. What's been the number one insight for you of all the things that we've talked about? We've covered a lot of stuff in 50 minutes, but what's been the number one insight? Put that in the comments box because I'll read those later. I'd love to know what you have has been the eye opening thing. What's been the Idea, the tip, the thing that you want to try out and start using when you are doing your prompts. Let me just before I answer questions, remind you of a couple of things. Sarah might put the links in the chat box again. We are running the AI and ChatGPT virtual summit next week, Wednesday, March 27. If you want a ticket, then there's a link going into the chat, into the comments that you can grab your ticket for next week's event.

And I think I have one more link for you. Yes, if you're fascinated by this prompt engineering and want to develop your prompt engineering skills. A few months ago I did a course. It's a two hour course so you can get through it fast with a 70 page workbook called ChatGPT for Accountants. And in that I walk you through ChatGPT in a bit more detail and I go through specific prompts and a workbook full of prompts. It's a great next step. If you are new to this and want to take the next level of understanding prompt engineering and how to use ChatGPT. Okay, let me deal with questions. I normally have more time for questions, but a I had lots of content to share and B I noticed at the corner of my eye there weren't too many questions coming in. So let me see.

There'S a lot of comments I'm just looking through.

Douglas Boaz says, can you give us an example of a prompt model that we can use as a format? Can you give us some examples that we can use to write our own?

If you come along to the AI Summit next week, both Emily and I will be doing lots of practical examples of ChatGPT and we will be giving out some free resources during the course of the three hours. Some resources, some example prompts. And also I mentioned that course as well, that two hour course. I do lots of prompts in there. In terms of building out a prompt template, it's something that I teach in level three of the Value Pricing Academy. I did a session yesterday where I built out the template with a lot of advanced stuff in there. But you want to. What I recommend, Douglas, is start somewhere simple and you can build on that.

What I might do is I might do a Facebook post with a prompt template with all of the core things that you need to have.

I need someone to remind me of that because I'll probably forget. Perhaps Douglas, if you're in the Facebook group, either my Facebook groups. If you just put a post saying mark, could you give me an example of a prompt template? If you can do that one of my Facebook groups. I will in the next few days give you something. Okay. I have nothing to hand right now, but if you ask that in Facebook, I can do that. Feel free to tag me in my Facebook group. I might not do it till Monday because we take Fridays off, but I'll see what I can do.

Okay, what else have we got? Questions? Okay, here's an interesting question. Annette says, how do you share your PowerPoint and video at the same time? Okay. I use a tool called ecamm for that, Annette. It's a tool. It's one of my favorite pieces of software. Ecamm, ecamm Live. And it's for Mac users only. You have to be a Mac user to use it. There are other tools as well. There are browser based tools like streamyard will allow you to do it. I forget now all the other ones. Streamyard, Restream, There's a number of these tools now that will enable you. They're really aimed at streaming where you can then set up scenes in advance so that you can toggle between.

You can toggle between. For example, this is a screen share. This one here is a PowerPoint slide. I've got one different place on my screen and I've just got buttons. I can just press and switch between them seamlessly. So the tool I use for that is ecamm Live. If you're a Mac user and if you do live streams or webinars, then I think it's an extraordinary piece of software. Okay, we have five minutes left. I'm struggling to see another question. Actually, I normally have lots of questions. So, Carolyn, you said be specific, but I don't know if that was to me or not or just a comment. So I'm not sure I can address that one.

Joey says I use ChatGPT4 because it accesses the Internet and images.

Personally, I can't see why you wouldn't use ChatGPT4. It's a better model as well. It's $20 a month, which I think is crazy low for all the amazing things it does.

And for the hours that you save. So I use it almost every day. I must save.

In a week. I probably save five or six hours of time every single week. I probably save two to three days a month. And so you have to ask yourself the question, is a day of my time worth more than 20 bucks? Well, I really hope it is. And so it's going to be. To me, it's a massive return investment. I think it's an extraordinary tool. Whether or not I'll still be using ChatGPT for a year from now, I don't know because this is changing so, so fast. I think I did my last livestream a few weeks back was looking at the future of the accounting profession in 2024. If you haven't caught that, it's on my YouTube channel. I do recommend you watch it because I talked about some of the big things that are going to happen in our profession with this AI and where things are going. It's extraordinary. The speed at which it's changing is just mind blowing. I think it's great. I think it's a huge opportunity if we just embrace this technology and use it.

But whether it'll be ChatGPT4, I don't know. There was some news announced this week, I think it was yesterday. I don't know if you heard this, that Apple are in conversations with Google about their Gemini. I said in one of my predictions for 2024 is that by the end of this year, Apple will be probably one of the main players in this space. They are never the first to market. They make sure that what they're focusing on is embedding a large language model into their iPhones, for example, and they'll make it much more user friendly. ChatGPT is great, but it's not very user friendly. Its user interface isn't the best because they weren't setting out to be a for profit company originally. So it could be much, much better. Like being able to file your past chats or tag them in some way and put them in folders would be really useful, but they don't do that. But Apple are going to be very big in this year and it's interesting that Apple are currently debating about do they build their own large language model or do they team up with. They're talking to Google with Gemini. I believe they're also been talking with ChatGPT with OpenAI and so that's gonna be really interesting. Watch this space. I'm a big Apple fan. I love the usability, I love the functionality, the ease of use. And if they can embed something as powerful as ChatGPT inside an Apple product, that's what I'll be using. That's what I'll be using. Okay.

All right. Okay, thank you, Carolyn. I get that Be specific was your takeaway. Be specific in your request. Okay. I think I've dealt with all of the questions Bill Robinson says this was an answer to. Annette. VMIX is great for Windows. So Annette, what Bill was referring to is VMix is the Windows version of ECAMM Live. So if you're not a Mac user, check out vmix vmix, right? I think that's all the questions. There weren't many this time. My next live stream is going to be in about four weeks. I forget the exact date, but it will be some more stuff on what's happening in the world of AI and ChatGPT. I can't tell you what it will be because the world will be different in four weeks time. This is moving so crazy fast. There'll be some new stuff happening in the next few weeks. I'm expecting very soon something brand new inside Chat GPT that I hope to be able to tell you about because that's going to be really interesting as well. Okay, thanks for coming along. Remember, if you enjoyed it, hit the like button. I'm going to play with some music as I normally do so I can check out and see what other comments there are. Remember to put in the comments box what was the most valuable thing you learned today? What? What stood out for you? See you soon. Bye for now.

I wasted all my time I wasted all my nights I wasted it on someone who's indifferent he didn't love me right he told me I was blind but I never really wanted to listen.

You have been here since day one see my ups and downs.

Never left my side once you st around.

Who knew it was you you have been here all along you have always been the one yeah who knew.

I never saw the signs it didn't treat me right but I just couldn't to let him go you waited patiently you waited for me you show me what it was to love.

You have been here to stay one.

Never left my side once you st around.

You knew it was you you have been here all along you have always been the one who knew it was you you have always been the one.

You have been here.

You have been here since day one I I.

Never left my sister. I know it was you you have been here all along you.

You have always been the one that's the TR.