ChatGPT Introductory Session

ChatGPT looks set to join the family of office skills such as Word and Excel that we all need to function effectively.

Early adoption of this remarkable new tool can really help to supercharge levels of efficiency.

In this very special introductory webinar we’ll introduce you to a
fascinating new world that looks set to transform how we work forever.

Areas covered include:

• Using ChatGPT + AI whilst keeping confidential information safe and secure.
• AI essentials – understanding what is Artificial Intelligence (AI) and cutting through the jargon jungle.
• Prompt engineering – what is it and why is it so important that everyone seems to be talking about it? How to do it well.
• Understanding current limitations of ChatGPT and how soon they will be overcome.
• Top five AI tools to help you do your daily tasks quicker and better every time.

Click here to access the cheat Sheet referred to in the webinar: WebinarCheatSheet.pdf

Recording:

Transcript:

Geoff: Ladies, gentlemen, and AI enthusiasts, we have a real treat for you today. Let me introduce the incomparable chairman of Legal-Island, a man who, since 1998, has made it his mission to straddle the line between law and landmass. He's as Irish as a pint of Guinness, as English as a cup of tea, and his accent's as fresh as the day.

He landed on the Emerald Isle 25 years ago. If you are struggling to spot him, just listen to the sound of traditional Irish music or look for the most eye-wateringly vibrant shirt in the virtual room. Yes, folks, he uses his attire like he uses his words, loudly and with full impact.

Buckle up, because the fusion of law, AI, and sartorial splendour starts right now. Ladies and gentlemen, please welcome Legal-Island founder, Barry Phillips. 

Barry: Good morning to everybody, and you're very welcome to this webinar. My name is indeed Barry Phillips, and thank you to Geoff, or "gee-off" as he likes to be called, who is my virtual reality assistant.

Now, that was a bit of a cheesy introduction, but that was an introduction that was actually composed or drafted by ChatGPT. And Geoff, or "gee-off", my virtual assistant, is actually generated by an AI tool called Synthesia, and that's something that I will be showing you all a little bit later on.

The reason why I wanted to start with Geoff is that this AI tool that I used to build him in some sort of Frankensteinian way six months ago wasn't very good six months ago. So I had a go with him, and I really dropped the project because I just thought, "He's never going to be to a particular standard that we can use him in the company".

And then about a month ago, a friend of mine who does a lot of this sort of thing said to me, "Have you tried Synthesia? I actually think it's quite good". And I went back to it, I had another go, and it was significantly better.

Now, I'm sure you'll agree, and I'm sure Geoff won't be saying this about him, he's still a little bit creepy. He's not quite there yet, is he? But my guess is that if I were doing this webinar, let's say, at the end of the year, or certainly this time next year, Geoff is going to be so convincing that you'll be looking at him and you'd be thinking, "Actually, is that an AI-generated person that I'm looking at? Or is it actually a human being?" So fast is the development in this space.

What I want to say to you today is that in the brief time that we have together, I will be going through this and I will be flagging up areas in which I think it's not quite there yet. It is going to be warts and all.

I've become a bit of an AI evangelist since the beginning of this year. Some may say I've become something of an AI bore. They might be right. But today, I will be flagging up issues to you to show you where I think it's not quite right there, it's not quite there, but with the understanding that soon it probably will be.

Therefore, don't dismiss something if it appears to you to be not quite to the standard that you would expect, because the chances are, in a few months, it will be of a much better standard, and it will be ready for use in your workplace.

So we have a number of people here, mainly from the HR community, and therefore, what I'm going to demonstrate is going to be with HR in mind. But if that is not your background, then please bear with me because I'm hoping you will see lots of things that are going to be immediately useful for you either in the workplace or perhaps in your own personal world as well.

There are a few friends of Legal-Island here, which I'd like to just name-check and say welcome Marian Jennings and also Stuart Wilson, who we know from the Education Authority. 

We've got a lot to get through, so I'll just get to that in a moment. But before we do, let me start with something about Cri De Coeur for those who have joined us from the HR world. And it's simply this. I've been doing Legal-Island for the last 25 years, and in that time, I've heard quite a lot of HR people, particularly directors, say, "I enjoy my work, but I wish our department was afforded more status in my company. While I'm round the boardroom table, I don't feel that I'm afforded the same sort of gravitas that perhaps the FD would get or the operations director and so on".

And I think this is a real opportunity for HR to really embrace AI and to really step up to the plate here, because AI in coming into a company is either going to come in and HR are going to have to help with the implementation of it and it's going to be completely relevant to their work, or it will come in in different departments and it will require at least consultation with HR. So one way or the other, it is going to concern HR.

And I think those that are in HR that look at AI and really embrace it have a career opportunity here to really advance themselves in their company by being all over it, understanding it, understanding the value of it and the opportunities it's going to bring to your organisation.

So that's me off my little soapbox, so to speak. Let's have a look at the programme this morning.

I'm going to start by trying to get us all in the same space, understand a bit of jargon, a bit of context with AI and ChatGPT.

After that, I'm going to take you to look at some prompt engineering. You've probably heard the term, maybe you're not too familiar in terms of exactly what it is, but I'm going to show you a few techniques to help you make it really powerful for you.

We're going to take a break at about 10:45. Then after the break, we're going to be looking at keeping data safe as we experiment with AI, and in particular ChatGPT.

And then I'm going to finish up by showing you five AI tools, which in our marketing said will blow your mind.

So not very much, you might think, in an hour and a half, but it's going to be packed with lots of things that I hope you'll be able to practise and to take away.

Now, what I will say is we will send to everybody a cheat sheet after this webinar. So there's no need to write down things that I'm going to demonstrate and to show to you.

And I don't think it makes sense either for you to have ChatGPT up and just playing with this as I demonstrate, because I'm going to go through it at some speed today.

So, if you don't mind, I would suggest that you have a go with ChatGPT after this webinar, or at the break time if you really can't wait that long because you want to have a go with some of the things that I show you.

So I think we're ready to start. The only other thing that I want to mention is that if you have any queries as we go through, if you could put them into the Q&A session box, I will get to them. I'll try and get to them at 10:45 and then again quickly at 11:30.

So anything as I go through, if you're unclear about, or if it's relevant to some of the material that I show you, please drop it into the Q&A and I'll do my level best to get to it after that.

So let's take a quick look at the jargon and some of the expressions that you might be hearing from the AI people. You may have heard of AI, AGI, and ASI. So what is it? What are those things, and what's the difference?

Well, AI simply stands for artificial intelligence, AGI stands for artificial general intelligence, and ASI, which may not be something that you've heard yet, but you will hear about that soon I'm sure, stands for artificial super intelligence.

Now, what's the difference between these three? If you like, they're on a continuum. On the left, AI and artificial intelligence. It's what is sometimes referred to as narrow intelligence. It's something which is built for a specific purpose. And it may be something like sorting out your email in the morning, so all the spam goes into a spam folder, and everything else drops into your inbox. It may be something like Alexa.

AGI, however, is something quite different. I asked ChatGPT for its definition this morning, and it said, "AGI is engineered to perform any intellectual task that a human being can do. This includes, but is not limited to, learning, reasoning, problem-solving, and adapting to new situations". So it's basically any intellectual task that a human being can do.

And ASI stands for artificial super intelligence, which is that time when we really have to accept that what we've created here is more intelligent than we are. And that's a big thing that we might be able to touch on later.

Singularity, what is that? Well, that's really the moment that we've created something which is as at least as intelligent as a human being. So if you like, that may be considered to be the start of AGI on our continuum.

Singularity, some people would say we've kind of achieved that already with things like ChatGPT, because ChatGPT is like having 20,000 geniuses on your desk. You can ask it about history, about science, about mathematics, all sorts of things. So there's an argument in terms of whether we've reached singularity yet or not.

Alignment is basically the idea that whatever we create is aligned with the interests of human beings. And more specifically, for a company, if we create our own version of a ChatGPT, we want to make sure that it's aligned with the values of our organisation.

You may have heard the term or the acronym LLM. That simply stands for large learning module. And I don't think we need to kind of know anything more about that other than a typical LLM is ChatGPT, which we're going to be looking at today. But there are other forms of that, things like Llama 2 you may have heard of. It's now Llama 3.

There are different things that these LLMs can do. They can be text-to-text like ChatGPT is. In other words, you put in text and it generates more text, usually lots more.

Or it could be text-to-audio. So you could go to an LLM and say, "Give me this, an audio file with a tweeting bird on it". Or you can go to something like Midjourney, and I'll show you Midjourney a little bit later, and you basically describe an image that you would like it to create for you. You can also do text-to-video as well.

And finally, you may have heard about regulation of all of this. A lot of people have been saying that it needs to be regulated. And there's a big argument going on at the moment in terms of whether you can actually regulate AI and whether it's actually too late.

But many are sort of saying that we must, we simply must, and we've got to find a way, and we've got to regulate it like perhaps we've tried to regulate nuclear weapons thus far.

So that's a bit of jargon, but what is artificial intelligence? It's simply three things. It's data, usually a whole load of data. Websites, book databases, things like Wikipedia, and so on are all gobbled up by these machines as they're training themselves on an awful lot of data.

Sometimes it's difficult to know what data is going into these things, because ChatGPT and their competitors are not very transparent in terms of what they are actually training these machines on.

It's also an algorithm which is interpreting the data. And then the third thing is compute, which is basically GPUs, graphic processing units, that are, if you like, the engine of all of this that drives the algorithm, which interprets the data.

What about the evolution of AI? Well, it started really kicking off in the 1950s as a result of a lot of work that was going on. And Turing was very busy not only cracking the Enigma code in the '40s, but also working in this space. So his name often comes up as well, but there were many others that may be attributed to the development of AI.

But by 1961, we had the first industrial robot formed in the USA. And that was actually invented more out of need than anything else, because in the production process, there were a lot of injuries having a human being in that particular part of the process. They felt they've got to invent machine to do it, and hence they invented a robot.

By 1986, the Germans had invented the first self-driving vehicle.

By 1997, Deep Blue had defeated Kasparov, which was considered to be a big seminal moment when one of the cleverest human beings on the planet, certainly the best of the time at chess, was beaten by a machine.

I met Kasparov a few years ago in London. I queued up to buy his book called something like "How Chess Imitates Life". Or is it the other way around? And in the two minutes I spent in his company, I've never felt so chilled in all my life. His eyes. Or never had such a chill, I should say. His eyes just seemed to penetrate and burn right through me as he looked at me, and I thought, "Goodness me, if I had to face this man over a chess board over many hours, possibly days, I think it would really, really freak me out".

2020, OpenAI releases GPT 3, and we've since seen two more iterations of that. And that was seminal because it appeared to be human. It was like we're interacting with a human being rather than a computer, and it really brought AI into the mainstream.

So what about typical examples of AI? We've got production line robots, self-driving cars. There are self-driving taxis going around now in San Francisco, and you simply call it on your iPhone, it rocks up, you put in a code so the door unlocks, in you hop, and off you go. This sounds really scary, but it's getting good reports, and the safety of these things is actually very good.

Somebody said to me as an example, or I heard it on podcast actually, they said, "Look, when a human being crashes into another human being in a car, the two human beings learn from it. The difference within AI is that when an AI self-driving robot car crashes into another self-driving robot car, all robot cars learn". That's the huge difference, and why all of this is accelerating quickly. It's just the learning that comes from it and can be distributed so powerfully through AI.

You've also got facial recognition on your phone. That's AI. Spam filter, as I mentioned. Email prediction in your email bar. As you go in, and if you were to type and you start with M, it will come up with all those that have a first name beginning with M. That's AI. Autocorrecting a word on a mobile device. So it's not only huge, fantastic robots that do very big jobs, but it's simple things as well that perhaps we take for granted every day.

And it's interesting because I read a survey report last week, and I think it was by McKinsey or one of those big companies, which was saying that . . . Their survey suggested 30% of companies have still to adopt AI, which, of course, is not right.

We're all doing AI. Perhaps what we really meant was to significantly accept it as a strategy for their further development or to roll out something like ChatGPT as a device to really help drive efficiencies.

So why is ChatGPT so important? It's so important, as I touched on before, because . . . I think last year everybody was talking about HR or reading about it, AI in the HR world, but most people are sort of still thinking, "Well, that's still to come. It's still out there. It's not on my desk. It's not really in this company in any big way. It's something that I can get to at another point".

This has really driven it into the workplace, so that now HR are really having to look at this and think, "First of all, how can I use it in our department to drive efficiencies, but also what about the headache of how we regulate this in the workplace and to make sure that it's used properly, fairly, and in a way that protects the data that we have in the company?" And I'll be touching on a few things relevant to that shortly.

Now, how does it work? I don't think we really need to know the nuts and bolts in terms of how it works, but it's probably helpful to know that it's simply nothing more than a letter-prediction machine.

And what it's doing after an input is given to it from the user is churning an awful lot of data and trying to predict what the next letter or word should be that is going to be of use to the user.

Some people seem to think it is an intelligent being, but it's not really. Speak to the IT people saying, "No, there's no intelligence there. It's just a prediction machine". But it's easy to step into that feeling that you are dealing with something that's intelligent and also human-like.

I was working with it last week, and I put in quite a fluffy question prompt. It wasn't very accurate, and it got back to me and basically told me that, and said, "Look, I need more information. It's not a clear question". I felt myself getting a little bit embarrassed as I got back, and I said, "Oh, I'm sorry. Here's another go". It got back and it said, "No, you don't need to be sorry".

I was beginning to get seduced by this thing into actually feeling that it's a real human being on the other side of the screen, which it's not. It's a machine. There are advantages to working with a machine over a human being, which I'll touch on in a moment.

So we're going to go to a quick poll here, and we're going to ask this question. "Is AI a friend, or is it a foe?" So I'm going to pass over to my real assistant in the real world, not Geoff, but Maria, and just ask Maria to drop in this poll. So over to you, Maria.

And on my screen, because I'm in present mode, I don't actually see what's going on here, but you should be seeing three questions, or rather one question with three possible answers. "Which of the following statements is true about world poverty today?" Is it getting better? Is there less poverty? Has it remained more or less the same over 50 years? Or is it actually getting steadily worse? Those are the three solutions that you should see.

And once you've answered there, then hopefully Maria will appear and just announce the results to us. So if you could do that now, that would be useful. And if I could then ask Maria just to pop up and share the answers with me, that would be great. Maria, are you there? Can you just pop up and share?

Maria: Hello, Barry. So 67% said it's actually getting better, less poverty, 33% say it's remained more or less the same over the past 50 years, and 0% thought that it's actually getting steadily worse.

Barry: Thank you. These stats are really interesting because, normally, it's the other way around. So well done to everybody in attendance who got the right answer, which is that it's actually getting far better. In other words, less poverty. Normally, when I ask that question, it's the other way around, and everybody answers, "Oh, it's got to be getting worse".

And the reason I ask the question is because . . . still hear me. If you can't, Maria will hopefully pop in and tell me otherwise.

But basically, there is so much bad news, and if you read Steven Pinker, he's a professor in the United States, he'll say, "Really, for s development to make the news, it has to be bad". And so it's really easy to look at everything through very dark glasses and to be persuaded that we're all doomed and that the world is going to come to an end quite quickly because of this new development of AI and it falling into the wrong hands, or a piece of AI machinery going rogue and working against us.

So I would encourage everybody to say, "Yes, there are risks here, but let's just focus on the wonderful opportunities that come with AI".

I was listening to a podcast a few weeks ago, and it was the head of Accenture who was talking. He said about AI, and I thought this was really quite good, "Look, AI is actually going to give us all a promotion in the workplace". And by that, he meant basically it's going to help us cut back on all the admin tasks that we have to do and will free us up to basically do the sorts of things that we should be doing in the workplace that we are employed to do.

So key players in this. There are a number, but Amazon, Apple, Google, IBM, Meta, Microsoft, and OpenAI are probably the key players. There are a few in China that are beginning to really invest in this space that we can expect to enter the market or to make an impact on the market at some point.

I might just pause here for a moment, and I'll explain that the presentation system that I'm using is an app that I've been testing for today called Tome. And Tome is basically an app to help you generate presentations. It doesn't help you do PowerPoint slides better. It is an alternative to PowerPoint.

I'll show it to you later, but what it does is it produces a slide plus an accompanying image. And I asked it to produce an image to illustrate AI key players.

Now, one thing it did was to produce four guys. There are equality issues there, and I'll touch on that a little bit later.

But if you have a look closely at these figures, they've got quite a good ethnic balance going on there, but obviously not a good gender balance. But look at the hands of the characters sitting down, and particularly the gentleman on the left, his left hand. You'll see that it's not a pretty sight. The guy on the right, as you look, sitting down I think has something like seven fingers.

So you can see that AI is struggling in areas when it comes to visual representation of things and it's still got a bit to go. The poor fellow in the middle standing in the back, I don't think, if he was from the real world, that he would be very happy with how he looks.

So ChatGPT, basic or premium subscription? Which should we go for and what's the difference? Well, the difference is that ChatGPT basic is for free, and premium is a monthly payment of $24. It was $20, but it's gone up.

And the other difference is that ChatGPT basic does not guarantee you access to the latest version, which is ChatGPT 4. Occasionally, it gives you it, but a lot of the time, it's only ChatGPT 3.5, which is not so good.

And it doesn't come with a facility to add in plugins, and plugins are an increasingly important feature of ChatGPT. So if you are serious about using ChatGPT, I would say to you that the premium subscription is worth paying for.

Now, sometimes ChatGPT basic will give you access to ChatGPT 4, but it doesn't guarantee it. Whereas with premium, it guarantees it to a point, as I found out this morning, because when it's really busy, it only gives you a certain number of prompts in a particular time period.

I didn't see that this morning when I was playing with ChatGPT, and it basically threw me out and said, "You'll have to come back in three hours’ time, because you've used your limit for this window, which is 50 prompts".

Now, this is a little bit cheeky because it's a bit like paying for a day pass at Tayto Park . . . I forget what it's called now. Is it Emerald Park or something like that? And then by lunchtime, being told, "Well, actually, you can't go on all the rides. You can only go on three or four".

But most of the time that I've been using it, it's usually without any limitations. It tends to be peak periods during the day.

So that's the difference. What I want to talk about quickly before we look at the interface of ChatGPT, and I'll show you some examples of prompting, is just to talk about four prompting styles.

And these prompting styles are simply "act as if", which basically means you tell it it's an HR expert, it's a marketing expert, and you expect it to behave accordingly.

Chain prompting is when you keep drilling it, and you say, "Okay, yes, but I want you to develop the title, for example", or the first paragraph or the summary.

Critique prompting is when you drop in some information and say, "Critique it. Check it for syntax, for spelling. Be brutal and give me three improvements to it".

What I was saying as I introduced the interface to you is that you can go down to the bottom left of the screen and you can go into the back of ChatGPT and you can fiddle around with a few buttons and adjust a few settings.

One of the settings that we have here is ability to put in plugins, which is that on there, and I'll explain those later. And also to enable or disable chat history and training. If you disable that, and not many users know this, but it basically means that ChatGPT will not use anything that you put into the prompt bar for its own training purposes.

That's important because if you don't know that and you put in some information, it could gobble it up and spit it out somewhere else quite often and quite possibly in an answer to another user.

So if that's personal data or commercially sensitive data, then big alarm bells should be sounding there.

So that's a quick look at the interface, and you'll see that you've got ChatGPT 3 there and ChatGPT 4. And that's effectively a little tour of that.

What I want to do now is just to go into ChatGPT, the main screen, and practise a few prompts before we go into a break.

So what I'm going to do, I'm just going to ask it a basic question, and I'm going to use the microphone that I've dropped in. In the ChatGPT cheat sheet, you will get instructions on how to drop in your own microphone. It's very simple, and you can then talk to it rather than to type. I prefer to talk.

So this is what I'm going to do. I'm going to say, "Draft me a 100-word blog post on HR as a career choice today. Optimise it for SEO". And there it is, immediately produced a blog post that can be used for your website or something like that.

I can ask for a title. So, "Give me a title for the blog post".

Then I can say, "Okay, give me three alternatives". So this is what we talked about with chain drilling. "Please provide three alternatives to the title you have given".

Now, I could look at that, and I can think, "Okay, I don't like any of them, but I like the phrase 'unlocking your future', so let's use that". "Please provide three more alternative titles, but using the phrase 'unlocking your future'".

That's basically the way that you keep going and you keep trying to improve things.

A lot of people who've tried ChatGPT will say to me, "Well, I tried it, but the answer wasn't very good". And I usually say to them, "That's probably because of the prompt that they used or because they basically accepted that as the first and only answer".

The golden rule, I think, with using ChatGPT is not to accept the first answer. Try to improve on it by using some of the techniques that I've shown to you.

So what I would say is, "Now, please return to the article and make it more funny and wittier". We don't have time, I'm afraid, to read through the alternatives that it's giving, but you can see it's made an effort there, we think, to try and be funnier and wittier.

So you can change the style of the answers, and you can also say to it something like, "Please provide me with three tweets I can use to promote this blog post".

So you can immediately use Twitter, or it's called X now, to promote the blog that you've just written.

You've already asked it to be optimised for SEO, in other words, search engine optimisation, and therefore, hopefully, people will be able to find it because it's dropped in keywords early in the article that it knows are going to be popular with a search engine.

Let's try out a ChatGPT "as if" prompt. So as follows, "You are a university HR lecturer. Please provide me with a list of the different types of generations now at work. Please provide a way in which I can remember them".

So there you are. You have a list of the generations. I can never remember the generations and when they start and so on and so forth, so I see that they've given me an acronym, which is VIPBOM, to remember it by.

But I might say to it something like, "I prefer my information in a table. Please represent the information in a table".

Finally, you might do this. "I prefer my information visually. Please represent the information bearing this in mind".

When I've asked it to do that before, it's actually dropped in a few icons of things like animals or toys, things like that. This time, it hasn't, and I think that's because I'm using ChatGPT 3 on this, having been thrown out this morning using 4.

But that gives you an idea of how you can generate the information and keep drilling so you get better versions of it.

Finally, before we go into a quick break, I'm just going to ask it to draft an AI use policy. So here we go. I'm going to start by telling it it's an HR expert. "You're an HR expert based in Ireland. Please draft an AI use policy for a company working in Galway in the IT sector with 100 employees". And off it goes.

Now, as I alluded to this morning, the great thing about working with a machine is you can tell it to keep improving the results that it's giving to you.

Imagine doing that with a human being. If you had a colleague and said, "Get me a draft AI use policy", and you sent it back three times, by the third time they would be pretty angry with you, and you probably wouldn't bother for causing offence. With a machine, you can keep going back and keep asking for more.

So, with this, I might say something like, "Please critique this brutally and suggest three immediate improvements, highlighting what they are." What you'll find there, certainly it can suggest three immediate improvements, and it's mentioned those at the bottom.

Now, in terms of my prompting, notice that I mentioned the jurisdiction because, obviously, when it comes to HR policies, that's important.

Notice that I mentioned the size of the company, because if this is a multinational company with thousands of employees, you would expect a different length of an AI use policy than a smaller company that's just got 100 staff.

Notice I mentioned the industry that it was working in, because you might expect that to be taken into account in drafting an AI use policy as well.

Now, what I would say is this. It's really important to know that whatever it comes back with, and however many times you've asked it to improve and to critique its answers, you must check it yourself. You must never think it's reliable, because as we'll see later on, it is prone to be inaccurate at times.

So, as I show you how to do this, I do it with a little caveat that you must never take it for granted that it is immediately useful and you can sign it off without checking it through, particularly something as important as an AI use policy.

So, finally, what we're going to do is I'm going to show you how I can sort things. I'm going to tell it as follows, "Draft a table based on the information I will provide you with. Please drop in the dates on the left-hand side of the table and the events on the right". So I've told it that, and it's asked me for the information.

What I'm doing is I've just dropped into it a chronology of important developments at Legal-Island. We've been going for 25 years, as I said, and we're celebrating our anniversary this year, and these are the milestones. So I'm dropping those in, and let's see how it does.

So it's immediately put it into a table. And you'll see that that took just a few seconds to put it into a table.

What you'll see is that I deliberately left out on the second line a date. So I said, "In 1998, Legal-Island was incorporated. The following year, it established its employment law email service". So I didn't tell it that it was in 1999, but you can see it's worked it out there for itself.

You can get it into a more presentable format. Let's see if we can do that now. "Please redo this table in a more presentable format and one that I can cut and paste and drop into Word".

So it hasn't done that very well. And again, I think that's because it's on ChatGPT 3.5. If I'd done it in 4, it would've gone to something which would be immediately useful to come and paste and drop it into the Word document.

But whatever, I think you can see there that that is a task that if you had to do that in HR, it would probably take you 20, 30 minutes to sort out, type into a table, and to swap out all the information. And this has generated it in just a few seconds. So I think it's quite easy to see that there are a lot of time-saving opportunities by using ChatGPT, and also using tables and different things like this.

Finally, before we break for very quick comfort stop, I'm just going to ask it to summarise something, because summarising, I think, is useful for a lot of people.

So I'm going to say, "There's a very useful book which is called 'The Power of When'. So please provide a summary of the book 'The Power of When' in eight bullet points". Off it goes.

Now, normally, that would take you . . . or it would certainly take me probably about 8, 10 hours to read that book. I'm quite a slow reader. It summarised it in just a few seconds for me.

If I wanted to play and mess around, I could do this. "Please redo the summary in the style of William Shakespeare". Of course, you won't be doing something like that, but I'm doing it simply to use an extreme example to help illustrate in the time available.

And finally, "Please redo the summary, understanding that I am a 12-year-old boy". So you can see it's completely changed how it's pitched the answer, having been told that I'm a 12-year-old boy.

Now, you may kind of think, "Well, why would I want to do that in my world?" If you're in HR and you come across the transfer of undertaking regulations, and you've forgotten all about them, and you needed a very quick summary of them, then you might want to think, "Let's dumb it down a bit because I just need a very quick summary of it again before I go into this meeting. I'm time-poor. It has to be quick and cheap and cheerful". And there you go.

So,that is a quick tour of, first of all, AI jargon, of prompting with some examples there of how to prompt, which I'm hoping will be immediately useful to you.

I'm sorry about the AI glitch, but I hope we've sort of re-tracked there and we're back on course. Rather than a 15-minute break, can we go for a 5-minute break? I will see you back here at 11:00 when we can finish off with two items, both of which are very important. One is about how to keep your data safe, and the other is about how to use five other tools to really improve your efficiency.

So let's see you back here at 11:00. Thanks. See you. Bye-bye.

Hello to everybody, and welcome back. I hope you had a good break and your caffeine levels have been suitably replenished.

So, in this section, what we're going to do is take a look at keeping data safe whilst doing all our work on ChatGPT. So let's take a look at that now before we look at some AI tools.

One of the concerns about using ChatGPT . . . well, the big one is accuracy. I've touched on that earlier this session. And it's the propensity of ChatGPT to make things up.

Now, it's a curious habit, but if you compare it with Alexa, if you ask Alexa something and she doesn't know the answer, she'll say so. She'll say, "Sorry, I don't know the answer to that". Not so with ChatGPT. It clearly has an awful lot of pride and would rather try to make something up rather than saying, "Hands up, I just don't know the answer".

So until they cure this issue, and trust me, there are an awful lot of engineers working on the matter at the moment, then we have to be very suspicious of the answers, be cautious, and totally check things for accuracy.

Next is the data security side of things, and it's as simple as this. I think that you must not use ChatGPT for anything which includes inputting personal data or commercially sensitive data. You have to assume that it will use that data, churn it up, and it could be used somewhere else.

And even if you have the training function switched off, which I explained earlier today, there is still the risk that you are putting information into another computer that is outside the IT walls of your organisation.

So irrespective of what it promises to do or not to do with the data, you shouldn't use it if you have personal data in your input bar or commercially sensitive data. End of.

Now, what some organisations are doing is creating their own versions of ChatGPT and bringing that version in within their own IT walls. And they're doing that because they know you can treat it like your computer on your desk, and you can use it for a lot more than if it's something which is outside of your IT walls.

And for anyone that's involved with HR and regulation of AI, and in particular ChatGPT, you really need to know what your organisation plans to do in terms of rolling out a device like this. Whether they're going to roll it out to ChatGPT, where it's hosted at the moment in San Francisco, or whether they're going to have their own model and bring it inside the walls of the particular organisation.

Copyright issues. There are a number of copyright concerns because this information that these machines use, particularly those in the visual world that use text-to-visual work, are scraping data from the internet. And there is the concern that that data and information may be represented in a way that others might recognise.

So there are a number of legal actions that are taking place at the moment, like iStock photography are suing ChatGPT because they're saying that it was trained on some of its photographs and they are beginning to recognise some of the photographs and images that it's generating, which is a breach of copyright.

And finally, there are equality concerns. There are so because something is only as good as the information that it's trained on. And if it's trained on information that comes with lots of biases, then you would expect that these are going to be represented in the answers.

And when it comes to ChatGPT, if we're accepting that it's trained largely on information gobbled up from the internet, then we have to immediately sort of say, "Well, hang on, this information is largely northern hemisphere, Western Europe-centric". And a lot of the information is generated by males.

For example, I think 90% of contributions to Wikipedia are from males. So you would expect that that bias that's being written in simply because of where it's being sourced and the type of person it's come from is going to be amplified in some of the results.

Let me show you just a few examples of that.

Midjourney, which I'm going to show you in a moment, is a text-to-image outlet. ChatGPT is text-to-text. Midjourney is text-to-image.

Recently, I was speaking for the Corporate Governance Institutes based in Dublin, and I asked them to do an image which illustrates corporate governance. This is what it came up with. It gives you four answers each time.

And you can see top right, there are an awful lot of men in suits, but not many women, possibly two in that. Top left, you've got a few women possibly represented. Bottom left, no women. Bottom right, no women. So, immediately, you can see the biases that are creeping in.

And when I went to ask the same question of DALL-E 2, which is another text-to-image generator, this is what it produced. Six very happy-looking people, but all males. I'm not sure if corporate governance would make me so happy, but maybe. I'd need to do the course again. But there we are, something in terms of the quality concerns.

So AI other tools, let's have a look around. I mentioned DALL-E, so let's go to DALL-E and see what it's all about.

DALL-E is the sister tool to ChatGPT 4 that we've been using. It's produced by the same company, OpenAI.

And this is what you do. You would simply go to the bar and say, "Please generate an image of four people having a picnic". This is not my favourite. I don't like it. Maybe it's getting better. It's a while since I I've used it, but the images that it tends to generate are not too good.

But these are not bad, actually. The one on the right, if you look at the lady's face, I'm not sure if that's quite right, the second lady on the right. But you've got cartoon images there, which might be useful to you. So that's not a bad go.

Let's have a look at a competitor to DALL-E.

DALL-E is the same, it's free, but there's the premium version that you can pay for as well.

So if we go along to Midjourney, which is the main competitor of DALL-E, I think the results are better. This is its interface, and let me see if I can show you just a few commands that have been put in and what it has generated.

So somebody's gone to it and said, "Produce me a map of America and Canada in the 1700s, and probably the style of the cartographer". That's what it's produced there.

You can see it is a preference for perhaps those in design and the art world because it is very arty in its results.

Here, "composition of a vintage Volkswagen camper driving in an English coastal town, vintage style, watercolour illustration, illustrated on white waterscape, hand drawn".

Now, what it's produced there, I think you'd have to admit it's pretty good, isn't it? It's followed those directions pretty closely, and the results, I think, are really quite impressive.

Now, what I don't like about this is that it's not particularly user-friendly. The help service on it is really poor. And very often, you can just put in the prompt bar your prompt, but because so many others are using it, the results go quickly flying off the screen off the top, and you end up having to chase it to find out where the results are.

But you can see it's pretty realistic in what it comes up with. Here's a prompt, "realistic photo, very sharp, multiracial university students sitting together at a table with books and laptop". So what it's come up with there is pretty good.

And again, I returned to the point of not being satisfied with the first result. You can see with my first play at DALL-E that it was generating stuff that perhaps wasn't too balanced from a diversity point of view. If that is the case, then you just specify and say, "I want more diversity", or, "I want better representation of ethnicity", and you keep going.

So I suppose in the visual world, it's easier to see the biases, and it's perhaps in the text world that you have to be a lot more vigilant and you have to be a lot more careful.

So that is Midjourney. Let's go along to Synthesia and take a quick look at Synthesia.

So Synthesia is the software that generated Geoff that introduced me this morning. And how you would operate this is you would go log to the website, and then you would click on a template, use this template, and it should pop up for you.

And what you can do is you can go and change the avatar. So let's say you didn't wage this one, let's say you want to go for Gary there. If you've got somebody who's going to talk about buildings and cranes behind it, yeah, maybe he's the sort that you'd expect to do that.

And you've got a chance here, on the left, to change language, or if not the language, then certainly the accent. So let's have a look to see if we can go from English U.S. to something else.

It seems to have frozen. Oh, no, it doesn't want to play at the moment. So let's see if we can go back from that. We can.

So down the bottom, we have the content that you want to put in. If we put in there, "Hello to my guests on the webinar today". Can we go there and see if we can change . . . Oh, yes, here we go. So we're going to go from American to English Confident, choose that, and click here.

Gary:   Hello to my guests on the webinar today.

Barry: He sounded a little bit like Geoff to me. Let's see if we can go English-Irish. Irish Youthful, let's try that. And it's taking a little while to . . .

Gary:   Hello to my guests on the webinar today.

Barry: I'm not sure how convincing that Irish accent was there. It sounded like an American practising an Irish accent, but there we are. You can select and choose the one that's most useful.

Now, you'll see from that his lips are not moving. That's because you basically get it right until such time as you are ready to publish it. Then you click Generate, and once you click that, it will send you back in a few minutes' time the final version, which will have an active figure with the lips syncing to the content that you put below. And that is how that works.

How we would envisage using this at Legal-Island is when you have a need to have some sort of spoken text visually presented on screen, and perhaps you don't have time to get a human being in to do a piece with a camera or those who are available don't feel comfortable in front of a camera. It's quite simple to go for this option.

Are we going to see this increasingly in the future? I think we are, but who knows? It's difficult to predict.

So that's the Synthesia. And is there anything else I need to explain to anybody there? I don't think there is. Let me go now to Tome.

Tome is the AI that I've been using for this presentation. And I've been practising with this to see how useful it is as a means of generating something really quite quickly.

So you'll see that as you go into it, you're given lots of options, and you've got lots of different templates that are around.

So let's choose this one just for the purposes of demonstration. I've clicked on this template, and you'll see at the bottom there's a template there where it's going to ask me for things. So I'm going to say to it here, "Create presentation about how to draft an AI use policy". And I'm going to tell it here, "I want it in . . ." Yeah, let's stay with six pages just for the purposes of this.

I click Return, and you'll see that it's going to generate an outline. It's going to give me six slides. Then I click Continue, and it's going to then produce those, and it's now done that on the left-hand side. For some reason, it tends to drop them in in quite a strange order, but let's see if I can just go to it.

So "Welcome to the future of work". You can see there are six slides up and with various headings. Now, you could have given it your own headings. You could have said, "These are the six headings. Please create content for each of these".

Now, why I like it is that the images that it produces generally are really quite colourful and quite spectacular, and they're easy to use. If you don't like a particular image, you could just go in, and you could say, "Create image", and describe the image. Let's say, "Human resources". So generate something to illustrate that, and you'd wait a few seconds, and hopefully it would come up with some samples for you.

This one seems a little bit slow at the moment. So just as we are waiting for them . . . Okay, here we go. You've got this one. Human resources? Yeah, not bad, I suppose. That one, I don't know what really links into human resources in particular.

Now, one thing it does struggle with is with letters. So with this, it's not quite good at capturing letters into the images. As we can see here, it's got "Human Sure". So that is a limitation of it.

Now, to sum up in terms of whether I would use this, I don't know that I would for many presentations. You see it also generates the content as well. So if you were to use this . . . Let's say you were required to do a presentation at short notice internally at 2:00, and you're given the brief at 1:30. If you've got to do something which is fast, cheap, and cheerful, I think it might work for you. But other than that, I'm not so sure at this stage. So that's my view on this.

Can we now go to the Summarizer? This YouTube Summarizer will do exactly what it says on the tin. It will summarise a YouTube clip.

So here's a clip from Jordan Peterson, who's appearing in Belfast next month. And he was talking about the importance of gratitude. I'm just going to ask it to summarise this video, which I think was about 10 minutes long from memory.

It's now summarising it, and it's telling you the highlights. It's giving you a detailed summary, and it's giving you a bullet-point summary of that as well. So it's pretty powerful, I think, and it's quite a useful tool.

So I've shown you that, I've shown you Discord, I've shown you DALL-E. Let's go back to ChatGPT, and let me show you briefly plugins.

To do that, I've just got to bring over this screen and make sure that the plugins are ready to go. And they are.

It currently has a message thing. Let me make sure that plugins are enabled. They are, so this should work out now. For some reason, I don't think it is. Not quite sure why that is.

But in that case, let me go and show you one other thing, which is Poe, which is to be found here at poe.com. And Poe is interesting because this is where it collects all of the LLMs, like ChatGPT, its competitors, like Claude, Llama, etc. So you have in one place here an opportunity to effectively compare them all.

And you could put even one prompt, and across it, you could actually test it out and just basically see which is your preferred machine.

Claude, I'm understanding from people that use that regularly, they are saying it's actually very good, and in some circumstances, significantly better. And I've heard that from a legal point of view, when it comes to using it for legal information, it's actually more reliable than ChatGPT.

I think that when it comes to legal stuff, I have big doubts and reservations with this because it is really good in certain parts but unreliable in others. And therefore, I wouldn't recommend it to be used for legal purposes.

If you've got a legal query and you are not familiar with employment law in your jurisdiction at the moment, I would say, by all means, play and experiment for a bit of fun, but don't use it at the moment in a serious way.

So we're coming up to 25 past. I was hoping to show you ChatGPT plugins, but unfortunately, on the system that I have, it doesn't seem to want to play ball. And this is the thing with plugins. Sometimes they are quite unreliable, and they don't actually deliver and operate. That's why they're in beta for it, and they still have a bit to go in terms of delivering reliably.

I've got a question in, just as we come close to 11:30. "Could you please give a brief outline of costs of these softwares?"

So these softwares generally per month are about $20 to $25. That's the premium version. The basic versions are usually free, but they obviously give you enough in the premium version or not enough in the basic to tempt you to upgrade, so they are clever in that. It gets a bit addictive, and if you're not careful, you can end up shelling out $20 all over the place.

There are corporate or enterprise versions coming out of ChatGPT, but no details are available on that at the moment. So that might be that they will offer 50 or 100 users at a certain rate for a particular company. But until such time as they publish those details, it's difficult to know.

I know some organisations are sharing in access. I'm not really sure if that should be done, and certainly, from a data protection point of view, I'm not sure that that is advisable.

If anybody else has got any questions, please do pop them into the Q&A box, and I can just get to them in the last few minutes.

I'm just looking to see if there's anything else perhaps that I should finish on. Resources, these will be in your cheat sheet. You can see there is a podcast which is called "AI Breakdown for Daily Developments in Around 5 Minutes". You'll have details on that. If you take your information by Twitter, you'll have information about Andrew Ng. There are blogs, LinkedIn, websites, email that you could subscribe to future tools, which you'll find very useful. And of course, there's Legal-Island's Hub email service, which does have a lot of information at the moment.

Can I just wrap up, then, by saying that there is a full day on the application of AI to the workplace, which is happening next Thursday, the 21st. Loads of organisations have booked to attend that, so if you haven't, you may want to consider attending that next week.

We also have a "Streamlining HR Tasks with ChatGPT" workshop, which is happening on 3 October. That's a full half-day. So this has been a very brief overview. We'll be looking at HR techniques and tools in much more depth on that day. So if you'd like information on that.

And finally, of course, we have the Annual Reviews in the north on 7 November, and in the south will be 29 November. And we have a very special full session on how to regulate the use of ChatGPT in the workplace.

So Marian Jennings, just something from Marian there. "I hope you’re keeping well.  I’m delighted you're taking the lead on AI". That's very nice of you. Thank you. "Your energy is fantastic". Thank you, Marian. That's a lovely comment to finish on.

I should apologise again for the lacking of audio. I'm sorry. I'm not quite sure what happened there, but I hope that you found this free introduction to it useful. And anything that you might have missed will come through in the cheat sheet, which will be filed through to you sometime today.

And so we've got one final question. "Hi, Barry. Are you using the paid version of the ChatGPT, and are you using it more on the . . ." I'm not quite sure what you mean by the last bit there, but yes, I'm using the ChatGPT paid version because it guarantees access to ChatGPT 4, which is significantly better than 3.5.

The reason I struggled with it today was that I've been practising before going live and I wasn't aware of the 50-prompt restriction on it. So it shut me off and put it back onto 3.5.

But I'm using this as a co-pilot, and I'm finding it very useful. It really is cutting down a lot of my admin and reducing a lot of the admin time. So ChatGPT and I are working quite closely on a lot of projects.

The next thing I'm looking to do is to get my company to actually import a ChatGPT version within our IT walls so I don't have to worry about personal data and commercially sensitive restriction and that we, all staff, can use it, full out, and get the advantages and hopefully the competitive advantages that come with ChatGPT when it's used intelligently and well.

That is 11:30 just gone, so it leaves me to say thank you to everybody for attending this morning. I do hope you found it useful. I do hope it's whetted your appetite to explore ChatGPT and other AI instruments and tools further.

I look forward to seeing you at some future stage, hopefully at some future Legal-Island event before the end of the year. Thank you. Bye.

Sponsored by: