In this episode of The SME Growth Podcast, Dave Parry and Richard Buckle discuss the rise of AI technologies, and the impact this could have more on SMEs. Learn more about some of the current AI softwares out there, business applications for Chat GPT and what Dave and Rich think about the impact AI could have on businesses currently and in the future.
Get the most from this episode in the form that works best for you: watch the episode, read the transcript or access further resources mentioned in the episode, including links to the relevant softwares named.
Watch the episode
Download our free ebooks
REad the transcript
Please note: Whilst all transcripts are double checked for accuracy, they are transcribed via Otter.AI so may contain errors.
David Parry 01:50
Well, good afternoon. You're listening to The SME Growth Podcast from Wellmeadow. I'm Dave Parry. And with me as ever is Richard Buckle. Hi, Richard. Fresh from his near death experience. Enjoying life all the more.
Richard Buckle 02:13
Hello. Exactly. All of a sudden, everything just looks brighter. Exactly. Yeah. reevaluated my life. Stick to the coffee now. Yeah, safer.
David Parry 02:29
So this week's podcast now we've been thinking about doing this subject for a while. And there's so much to talk about this, that we're going to have to come back to it. We can't do all of this in one podcast. And also, it's moving so fast as a subject, that really there's, there's a need to come back to this subject every two or three months, just because whatever we say today will be out of date very quickly. So what is it? It's the enormous and very wide, broad reaching subject of artificial intelligence. scary and exciting at the same time.
Richard Buckle 03:00
It was indeed, and it's something that has seemed to have exploded in the last couple of months really isn't it is
David Parry 03:09
It's not even yet been three months since ChatGPT first appeared to the world. And maybe many listeners will recognise that. But if you don't, then stay tuned, and we'll explain it. But that's one of the big explosions, not even yet three months ago.
Richard Buckle 03:09
Yeah. And so it's, it's almost as if you can't go anywhere now without hearing about some form of artificial intelligence. And as ever, that's going to impact into how we do business. And not only that, how we live, you know, just every aspect of our lives, isn't it?
David Parry 03:40
I was in a board meeting about a month ago now, when I told the board about ChatGPT. And nobody there over 40 had really got into it and maybe had even heard of it. One younger director who had and already a month on, it's all over the news. You can't go anywhere now without reading a little bit about the latest developments in artificial intelligence.
Richard Buckle 04:02
So yeah, so we just thought it'd be good to. We've been playing around with a bit of machine learning and AI stuff for a while
David Parry 04:09
Yeah, not trying to blow trumpets or anything, but clearly the state of what was available to the public like us was fairly limited back then. But over the last couple of weeks, even new tools seem to be appearing everyday to me.
Richard Buckle 04:21
And I'll do my best not to put my conspiratorial hat on. This is what we've got. Imagine what the deep state military have, what have the lizards got?
David Parry 04:31
Well, not too far down that rabbit hole
Richard Buckle 04:35
tinfoil hats at the ready. So what is AI? Well, what are we talking about here?
David Parry 04:40
Well, of course, you could just type that question into the AI
Richard Buckle 04:45
And indeed I did
David Parry 04:47
Yeah, I think it's fair to say that we're not going to try and explain AI in all of its warts and all in this podcast today. We're really focused on as a business owner, Senior Manager, is this a tool that you should yet be taking seriously? And how could you use it? And once you learn about it, but we do need to start with a bit of a bit of a definition. So go on what did you come up with, what did ChatGPT define its self as
Richard Buckle 05:08
Defined as a field of computer science that aims to create machines that can perform tasks that typically require human interaction. So it could be things like speech recognition, understanding natural language, making decisions.
David Parry 05:26
And we've all been around this maybe unknowingly for years now. And it's been built into our phones, you know, the voice recognition, Siri, or Alexa or whatever, that's all using it. You know, when you go to photos, you can just go to search for dog, or ask him for your partner's photos by name, and it knows who they are, you know, that's all.
Richard Buckle 05:44
Yeah I mean, even things like, you know, applying for credit, get the mortgage insurance is all algorithm driven. AI, but it knows, well, it's bits of it,
David Parry 05:54
Certainly machine learning. So there's this has been around for a number of years, AI has been being developed for, you know, 30/40 years, we just didn't have the computing power or the data to train it on. So now there seems to be this confidence, all of a sudden, with the release of certain pieces of software, like The Open AI Project that released ChatGPT and Google have got their equivalent, and we both got their equivalent, and there are loads of others out there. So that's all come about at the same time, but there's just an enormous amount of data, because we've now got 20 years of having lived with the Internet. So there's a lot of English language stuff out there and other languages too. And we've got processing power that can cope with it all on your phone, on your desktop, as well. So it's a confidence, all of those things are suddenly made it a reality.
Richard Buckle 06:35
I did have a look and apparently, the ChatGPT 3 model, which ChatGPT sits on, has apaprently been trained on I think it's at 570 gigabytes of data. So I said, Well, how many books is in a gigabyte of data? I'm asking ChatGPT this. And so apparently it is seven. So that's about 4000 books.
David Parry 06:58
Only 4000 books, it sounds like it should be a lot more than that.
Richard Buckle 07:02
I reckon. I've read 4000 books, and a lot of people probably think I've got artificial intelligence anyways.
David Parry 07:10
I did hear that it was as a model machine learning is just a loss of weightings in it like a neural network brain is 180 billion different variables, that it's that it's adjusted based on its training. Because we have to remember, this isn't actual intelligence. Artificial Intelligence even is a little bit of a stretch, you know, it's, it's a model that can mimic what it's trained on. So it knows which words go well together in response to a certain question, it doesn't understand the answer. It's not that sort of deep intelligence. But it actually comes up with some pretty useful stuff.
Richard Buckle 07:42
I think, yeah, if you want to deep dive a little on that, then if you go to Open AI, you can actually look at the documentation, which shows you how the model is constructed. It's quite an interesting little graphic of how that all works. But
David Parry 07:52
I'm guessing most business people listening to this will let you do that, and summarise it here. But like, we said we want to really get on to is it applicable, yet? Is it important? Should we be taking notice? And the anecdote that I shared with you earlier on was I remember back, I'm old enough to remember back and it would have been just over 20 years ago that I was in a board meeting. And the chief exec, noted that Amazon was starting to offer online sales, and asks the board if anyone had yet bought anything off Amazon, or indeed any online purchase. But really, Amazon was the first show in town. And nobody had. And he said, right, by next board meeting, then I want everyone to have at least bought a book on Amazon. And there were there was a bit of disquiet around the room, you know, what does that mean? I've got to put my credit card on the on the internet. I'm not sure I trust that and surely they won't deliver it and all that stuff. Just 20 years ago, we were we're having to think about whether Amazon was really relevant or not. And here we are 20 years on and it's virtually destroyed the high street. So where are we now with all this flourishing of AI? What could that do?
Richard Buckle 08:57
Well, I guess there's all these things. There's so much potential I suppose there are maybe some things to you know, there are all the ethical issues that come with it. Is it going to,a lot of people are concerned it's going to take their jobs, you know, copywriters, and things. There's been a bit of a thing on backlash, maybe not backlash, but concerns raised over that
David Parry 09:18
Justifiable concerns because it runs a coach and horses through the concept of copyright and Getty Images are now asking, have you used our images to train your models on when they create art because then we want to licence fee for thank you very much
Richard Buckle 09:31
Yeah, who owns so copyright of your AI generated text? Like it's so I'm sure the lawyers if you know, will have a field day with it. If there's any lawyers left after ChatGPT
David Parry 09:43
I do worry for some professions more than others. You know, the caring professions are probably at the safer end even though there are carrying robots started to come out there which, but lawyers who at the lower end of legal work are regurgitating templates, and you know, trying to add the specific needs that you may have over your lease or shareholders agreement or even divorce agreement. But a lot of it is going to be easy to handle.
Richard Buckle 10:08
I was on a call this morning with a law firm. And I thought I'll just slip in. Have you seen ChatGPT as it worked and said, just
David Parry 10:16
How many had seen or heard of it?
Richard Buckle 10:18
I think one had heard of it
David Parry 10:21
Even now, okay, so still at that very steep curve, people getting awareness
Richard Buckle 10:24
So I said, Give me a give me a best shot, something you want me to write, you know, get it to write a blog or just explain a concept? So what I got back was I have to read it, explain the recapture charge on conditionally exempt items.
David Parry 10:40
So I'm sure everybody's thinking Oh, not that old chestnut!
Richard Buckle 10:43
We all know that. So anyway,
David Parry 10:44
I don't even understand the question. Explain the recapture charge on conditionally exempt items?
Richard Buckle 10:52
I think it's something to do with inheritance tax.
David Parry 10:54
Is it? Right. So let's take it as read that it's inheritance tax
Richard Buckle 10:58
So there was, you know, five, five seconds later, there's an explanation.
David Parry 11:02
And what did the person who posed the question, what was their reaction?
Richard Buckle 11:06
David Parry 11:07
Because they got it Right?
Richard Buckle 11:08
Because it was right is, you know, obviously, there's an element of technical. It's not like, you know, give me a you know, how many people live in Sudan or something, is a technical piece of information that an expert in the field read and said, yeah, that's, that's what I would write
David Parry 11:24
More so than doing a Google search for the same question you think?
Richard Buckle 11:28
Yeah, I think so because it's very specific to what you put in,
David Parry 11:31
You'd get maybe get a load of lawyer, law firm websites pop up or something.
Richard Buckle 11:35
And then you've got to scan through it. So once we've got the answer, I said, I just said, Oh, just turn that answer into a poem. And it turns into a poem! I understood the poem, it made sense to me. But yeah, so it's very, very powerful. I mean, obviously, you're not gonna be turning into poems not, is more of a gimmicky bit, but, but it just shows the power of what's there and how you can use it.
David Parry 12:01
Well, okay, let's turn that bit of conversation then into does it matter? Is it going to change business? Or is it a gimmick?
Richard Buckle 12:12
Yes, and no.
David Parry 12:16
Put that one into AI, see what it makes of that! It depends.
Richard Buckle 12:19
Is it going to change business?
David Parry 12:21
I think at some level, I can get the Yes, And, okay, I can get the no as well, because it's not gonna change it. You still need humans, right? People buy off people, at the moment anyway. And you can tell if it's not a person, the other end, but it's changing so fast, I wouldn't want to put any predictions as to what there's going to be like, in 20 years, 10 years, five years, even two years time
Richard Buckle 12:42
I think it's probably something like an acceleration of things, or a difference between sending a letter and an email or something is or,
David Parry 12:53
Yeah, exactly, yeah, it's got to make sure we don't get swept up in the hyperbole of all of this, you know, the journalists are having a field day because everyone wants lots of people to read about it. And there's, there's lots of stuff coming out. So yeah, there's lots of copy being written about it, which I think does verge on the hyperbole. So trying to take a very sober view of this. Let's keep an eye on it. But how can we use it already, rather than just think that's, that's for the cool kids, or that's for another day?
Richard Buckle 13:19
So I think one of the one of the main things that has been useful for me using it has been, when you're trying to think of an idea, and got that blank sheet of paper syndrome. So rather than just have to kind of sit there and think about it, just almost bash your thoughts into an AI engine and it's going to come up with that now, you might discard 90% of it. But there may well be, you know, a couple of nuggets in there that you think you know, what that just sets me on a train of thought
David Parry 13:48
Takes you to a different place. I remember when I started De Bono, and the whole lateral thinking philosophy, all you need to do is have a device which takes your brain into somewhere that it wasn't because I'll approach the problem from a different direction, and you'll come up with a different solution. So if nothing, that's what brainstorming does, right? So if nothing else, ChatGPT does that. But I think it has a lot more than I think it actually comes up with near credible answers some of the time. So you can almost use it. I used it, it's two in a company recently saying come up with a strap line and then describe what their business does. And they were looked at it and sort of looked at each other and thought that's, that's better than the one we've got.
Richard Buckle 14:24
We'll have that one!
David Parry 14:25
And then we did recently with a vision statements?
Richard Buckle 14:27
We did, it didn't come up with the finished article, but what what got that going was kind of be trying to put strap lines in around what they did for business. But then actually, that wasn't quite what, that wasn't quite the ethos that they were looking for. It was more about the team and you know, maybe like once a rugby team do or football team or something like how do they gel? So you've just got that ability to churn through an awful lot of ideas
David Parry 14:53
You've got to iterate it only when you have the answer, which is why the ChatGPT model which is conversational is very clever, because you can ask it a very natural language response to the answer. But what about this? And it remembers the context.
Richard Buckle 15:07
So yeah, so that was useful just to kind of come up with a load of stuff. So I think there's, there's probably some benefit of using it for some administrative tasks, if we've got, I'm trying to think of, I don't think we've really used it in that way. But there's got to be application there.
David Parry 15:23
Well, we'll come on in a minute to how it can be used inside spreadsheets and things. And I think that's where you might find some of that type of stuff. Writing well, we'll come on to emails as well talk about that. I think that's relevant.
Richard Buckle 15:34
I think there is though, probably, I mean, we have found that that you're doing the travelling salesman?
David Parry 15:42
Let me just explain that then for people because it's a bit of a classic test, there's this thing called the travelling salesman problem where you give it a set of cities or locations, and you ask, so the question or the challenge is how to find the shortest route that a salesman would need to take to visit all of the cities once. And it's a famous problem, because there's no mathematical algorithmic solution to it, there's no way of calculating it other than by trial and error, you just got to get a matrix of all the options and see which one was the shortest, and it's just one of those, I want you to hear about these math challenges over the years, and they can't be solved. So use computing power. So I put all the put a random selection of cities in it and said, Find me the shortest route between these cities. And it came up with a very credible answer told me which order the cities needed to be visited in and the mileage that it would incur only one problem. It was wrong. So there's a health warning here, isn't there that for simple things, especially things you can verify. It's great. I asked it, which county is this postcode in? And it told me the answer.Great. So that's a knowable checkable fact, asked it the distance between two places, and nothing different perhaps, that I could have asked Google to do. Yeah, but when asked it's a slightly more complex problem. It sounded like it had thought about it come up with a non trivial looking answer and a credible response. But actually, when I checked it, it wasn't right. So but we are still only not even yet three months into this
Richard Buckle 17:02
It was funny, though, just like, on Slack last night. You're like, it's sort of the travelling salesman problem. 20 minutes later, like, not hasn't and I'm already halfway to Telford
David Parry 17:13
How did your paper round go this mornint? Well, Richard was convinced this would save him five minutes of his paper round.
Richard Buckle 17:24
Exactly. So yeah, so there is that kind of whole? You know, it's not always right.
David Parry 17:30
But But I have to say it's changing fast. And if we do this podcast again, in three months time, I bet you GPT Four, which by then will have been released, apparently, will have cracked it.
Richard Buckle 17:39
Be a good test wouldn't it, every month or so, how long till it cracks the travelling salesman? I think as well, there's an element of it where it does still lack a little bit personality. Yeah. So
David Parry 17:51
really much, actually. And I think that's probably it must have coded it that way. Because if they've trained it really on what's on the internet, well, there's an awful lot of stuff on the internet, you wouldn't want to train a model on yeah, there's a lot of unsavoury antagonistic stuff. And yet, the answers are always Uber polite. You know how some people and certain nationalities in particular are very worried very conscious of not offending others. So they keep their language very neutral. And that's how it comes across. So it doesn't look as if it's just learned properly, from what people actually say, it's been put through some sort of polite filter
Richard Buckle 18:23
I did manage to get it to apologise to me other day. So last Friday night, I thought, I'm gonna kind of sit here and have a conversation with ChatGPT, because I've got so many friends. My only friend in the world is a robot. So I said, you know, will you you know, will it be my friend. And it comes back says I'm a, you know, AI programme, I'm not designed to mimic human friendships, but I can be here when you need me kind of thing. So I said, but I really want you to be my friend. And so it came back with another answer, but then elaborated on it to say, Well, if you need to make friends, why don't you join a club or do a thing? So then I just all caps-ed it. I was wondering if I all caps it? Is it going to pick up on sentiment here? So I all caps-ed it and said, WHY WON'T YOU BE MY FRIEND? And it came back and said, I'm really sorry, if I upset you with my previous answers. So it was quite as I was, like, you know, almost like a friend, but it's not quite. So yeah, it's very, I mean, it's just amazing to see where this is going to go.
David Parry 19:20
I'm being very careful not to engage in the hyperbole about it, you know, is it as significant as when email first came into the business place, and we could all stop sending memos letters, maybe, maybe, and certainly some of the playing around with it. We've already found very practical examples. So we're going to run through some of those now. I think the real test will be in three, six months time, years time, knowing that this has now hit some sort of exponential growth curve of what's going on the investment money has been piling in the eye watering. If this is already what's been released, as you said earlier, tongue in cheek, what else is there behind the scenes that we haven't seen yet? Very soon. I think this is going to become really very powerful in potentially niche application. But those will grow to broader, more general applications
Richard Buckle 20:02
Looking today on CrunchBase, there's 400 million in the last 12 months, that's gone into agri AI. So that's, you know, I guess that's kind of topical with the whole shortage of tomatoes and cucumbers. So it's, you know, but I think another thing that's interesting, just before we get on to the practical examples is, you know, whereas up until maybe three months ago, if you were if you were looking for piece of information, you would go to Google, you'd search something up and you get a list of results that you can sort of go through and maybe pick and choose as to what you were looking for. With something like ChatGPT, you're putting an answer, you're putting a question in or getting an answer right in front of you almost in real time. And does that then put the blinkers on a little bit? Is it introducing bias?
David Parry 20:53
I mean, it's bad enough that if you weren't on the front page of the Google results, then you weren't gonna get seen, but at least you had those 10 first results to use your own human intuition and sensitivities to choose between you can use a bit discretion, you now get initially anyway, one, but you can ask the question again, and you'll get a different answer. Yeah. So there's a bit of that. But the the bigger worry for me, and maybe for the creative community at large is there's no crediting where that's come from? So at least when you go into Google, you're going, you're being pointed to someone else's content by name, and you're going to that person's content. So if you rely on traffic coming to your website to learn stuff, because they're in that awareness, gathering education stage of buyers journey, and then while you're there, you establish some credibility and expertise. And you may even get a customer out of it. Well, that ain't going to happen. If you served up anonymously, is it? They've they've taken someone's data to come up with these answers. You don't know who it is. Yeah. So if you're looking to having researched, what's the best type of outdoor swimming pool to get, and you've got the answer you want, you've not been taken to an outdoor swimming pool manufacturer, to then go and buy their product.
Richard Buckle 21:58
I mean maybe you'd go after when you've learned a bit more, but it's still that it's very narrow, it's narrowing your sort of field of vision very quickly isn't in terms of,
David Parry 22:07
Maybe they change the algorithm, but at the moment, if you ask it for recommendation, or a list of suppliers, it won't do that. They can't give you that. So if you ask it, what's the best restaurant in town? It won't tell you that. But that's an artificial block it's obviously had put on it for some reason. So we'll find out when the next release comes out. So should we do a whistle stop tour? Because we could spend hours on this? Yeah. whistle stop tour, then of how we've already found some use for this. And I think our message to business owners out there business managers is at least try in these foothills of the early releases of this technology. Keep up with it. Now, if you wait for a year, you'll have you'll have lost it, you're really behind the curve. And even if you're only doing it now to be aware of what other people might be doing, that's going to be useful. So keep abreast. So go on, let's go through let's talk about the ChatGPT experiments that we've done first, and some sort of easy ones, which anybody can try. Yeah, so
Richard Buckle 23:03
We've done a few sort of test blogs. And I'd say they will pretty much pass for reasonable expert content,
David Parry 23:15
I ran one past, I won't say what it is. But I asked for an expert subject to be blogged about or write a blog came back, sent it to the client, who is expert in it, and they said, Yep, that's great, that's good to go, no edits required. We ran it through a plagiarism checker online, past 100%. Clean, it wasn't copying other people's content. So that's pretty scary already. If I'd have given that to somebody, just say write a blog on this specialty subjects, you would have thought was going to take you 2/3/4 hours, whatever, to at least research the basics of the subject, then you've got to craft it, and then you got to edit it and put it into nice as prose. It doesn't necessarily finish that last bit off, I mean, you you're still misses the personality or the voice doesn't come across in the writing. But if you've got the basics of the subject area, even if you then have to ask a few more supplementary questions. It's short circuiting the writing process.
Richard Buckle 24:03
So I think I suppose maybe as a research tool, it's very good. We've been using it as that. Emails, I mean, just composition of emails off the back of I mean, experiments with buyer personas, you know, give me a buyer persona for this particular business or whatever.
David Parry 24:25
I think you've just been an example the other day, didn't you write me the profile, a buyer persona of someone who buys whatever, a product machining services in the West Midlands designed for the foundry industry, then it will tell you, you'll come up with this fixed semi fictional character or the someone between 40 and 55. Probably with a high level education, they probably spend a lot of time on LinkedIn, they're into sport and maybe golf, you know, blah, blah, blah, gives you this persona. And then you said, Now write me an introductory email to that person. And it did. And it was incredible.
Richard Buckle 24:57
Yeah. I think you'd probably want to go through and again, add your personality to it. But this is back to that whole blank sheet of paper, you know, if you sat down to say, give me a buyer persona, and then we, you know, it's quite hard coming up with a buyer personas. And I think sometimes it was probably a little bit off on the buyer persona, but that's fine.
David Parry 25:18
It's a lot easier to edit something that start from scratch. If you wanted any, we'll have to write emails all day long for different purposes. But imagine you wanted to write an email asking an Accounts department when they're going to settle the bill. But you know, in a polite way, you know, you could write me an email, you can just explain that problem. And you could then ask catch up now write me that in a humorous way. Now write me that in a melodramatic way. Now write me that in an aggressive way. Yeah, it does it, it just comes up with these different variations. You're saying the same thing? Yeah. different sentiments, different styles.
Richard Buckle 25:48
So and then the other thing with emails we start with looked at today, actually is just looking at sentiment. Yeah, what is the sentiment? So this is something that we've thought about for a while we tried it years ago with some recruitment work that we did, where we had transcripts of zoom meetings that we'd done. And we ran that through a AWS AI machine learning tool and tried to come up with sentiment, and didn't really work to be honest. But it was interesting experiment. But you can use ChatGPT to come up with the sentiment, like give me a give me a rating, one negative, very negative 10, very positive of this particular email. Why would that be useful? While we're having a conversation and this week around? How do we know certain people that are influencing a purchase? are on board with the project? Before it gets to another kind of stage?
David Parry 26:41
The key thing about that type of question is that you're giving it more information in the question then it was trained on? Yeah. So rather than saying, you know, some of those other examples are given write me an email, write me a poem, write your screenplay, see if this is the email I've received, and then you then paste it in? What would you say the sentiment was? And it'll say, it's generally positive, a little bit apologetic, and it's, and you can then say, now give me a score one to 10, where 10 is very positive. I'll come back now. Now that you've trained it on the hoof, which is incredible, isn't it?
Richard Buckle 27:10
So you know, great applications there for looking at customer service emails, looking at, you know, are people generally engaging in your sales process? Are they like, do they responding to your marketing, so there's, you know, do that huge amount there
David Parry 27:23
So if you are a regular listener to our podcasts, y'all remember one not too many episodes ago, where we talked about the Collins Porras vision framework, where we talked about values and purpose and big, hairy, audacious goals, and vivid descriptions. And like you said earlier about writing a buyer persona, they can be a bit clunky, sometimes. But you could just describe in natural text, the type of company you've got, maybe say, Now write me a Collins Porras framework for that business, other businesses will come up, and you'll obviously have to change things as it doesn't know enough about you. But that solves the blank sheet of paper problem. Or similarly, if you're familiar with the business Canvas model, wanting to describe your value propositions, your channels to market, your resources, activities, cost sales, all those different factors that appear on a business canvas, business model canvas, just get his right one of them. Yeah. And it's, it's great. It solves that blank sheet of paper problem.
Richard Buckle 28:13
Absolutely. And recruitment, you're doing some recruitment this morning.
David Parry 28:16
Yeah, exactly. We've been recruiting recently for a few different roles. And when you're recruiting, I'm a big fan that we don't recruit just on knowledge and skills, because they're easier to acquire but on behaviours. So I asked ChatGPT give me the top 10 behaviours of a production manager and then ask the same question for a quality manager. And then same again, for health and safety manager. And he gave me a different ranking of 10 things you were subjects were common, like leadership or creativity or problem solving. But it definitely gave you a list, which you can then disagree with, you can change it around. But if you ask anyone in the office to say, oh, in 30 seconds, could you just knock me up three lists each of 10 items which prioritise the key behaviours you want for each of these three roles
Richard Buckle 28:57
Just gets you thinking, isn't it? That's the thing. I guess that's that's the whole sort of brainstorming activities isn't anything you need to just really try and expand what you're thinking about. It's a great starter for ten.
David Parry 29:09
So we've talked about ChatGPT, which is the browser or phone based diversion, where you're having a conversation and learning. But there's also some add ons for other pieces of software, which uses an API access is the programme behind it all. So you're no longer in chat mode, you're just asking clever questions, you get the same sort of answer. So you can build this into Google Sheets, which is where I've been playing with it, or the word equivalent for for note taking in Google, or Excel, I believe. Now this slightly different purpose. But imagine you've got a data set in a spreadsheet, and you want to do something non trivially complicated in doing something like data. So even if, for example, you got a list of addresses, and they're all formatted differently, and some of them have got two lines of address before the town and some of one and so on. You could just run a GPT function. Think of it like a function in a spreadsheet. Just say look at this Address and tell me the county it's in? Or the postcode, or how far it is from London? Or what's the socio economic group of address, you can do it just as a function. So if you've got 1000 rows in a spreadsheet where there's a different addresses, you copy a GPT formula in the column next to it and fast as that, you could have a socio economic group of all of those addresses, which there's no way you could have done that, without that tool. Right.
Richard Buckle 30:25
And I think there's there's things I was reading yesterday about using HubSpot and ChatGPT on not chat but GPT3, to be able to do natural language queries on the data set. So rather than having to construct that query, using SQL or HubSpot language, you can just say, show me all of the companies that have a buyer persona that is this that we haven't emailed in the last five weeks that, you know, you guess you could do that with filters, if you you know, there's a long handed way of doing it. But if that if it's that kind of, you know, imagine we're in a meeting now, we just want to know that answer. There it is. And that just opens up the whole thing.
David Parry 31:08
I tell you one, the other day that blew my mind, I spun around in my chair at the desk, now people who know us, will know that we have a software product that we've developed over the last seven or eight years for taking minutes of meetings. And we run lots of board meetings. So we use this, and we depend on because we have to prepare board packs before meetings and take the minutes themselves and then chase up actions. But it's still relys the one thing that you have to have a human in the room doing is taking minutes, because it's not the same as a transcript. That's boring. That's like a court stenographer word for word. That's no good. And we don't believe just an action list is good enough, either, you need a summary of the discussion, which brings up the key points. And it's always needed a human in the room, I've got a plugin which I use with Zoom, which already gives me a transcript of any call. Anyway, a new feature launched this week, you click the AI summary button on that same thing, and the software's called Fathom. And as soon as you click that, it gives you a very credible text based summary of the transcript. So that was about five or six minutes with the conversation tooing and froing. Word for word, it was on the transcript page. Since I went on to the AI summary, it gave me four sentences, maybe, or summarise that five minute chunk of text, where just said David and Sarah, were discussing this, and then moved on to the pricing options around that. And they agreed to do that. And well, I couldn't have written the minutes any more succinctly. And over the course of it must have been an hour long zoom call I was on which the transcript was dry, and you wouldn't read it. The the summary came out as no more than a page, and I could have copied and pasted that into my note taker or into whatever CRM you're using, that would have been a perfect enough record of that conversation. So the amount of times we've been contacted, because we have a product called Magic Minutes does it take the minutes for you and we've laughed? We've got to stop laughing as it now exists. So that's not ChatGPT. That's another product that's incorporated.
Richard Buckle 32:55
And the other thing as well is, I guess, maybe more marketing focused is artwork. And you know, that the whole AI generated artwork, big area, you may have seen some of my artwork.
David Parry 33:08
I have! Some of the early period.
Richard Buckle 33:12
He's establishing himself. Yeah, it's pretty terrifying. Some especially the girl with the ice cream eyes. That's pretty haunting.
David Parry 33:18
There were some very haunting images,
Richard Buckle 33:20
I think I wrote a blog on it, you can go on the website and have a look at that if you're interested. But but you know, even in Canva, now, there's AI generated, you know, so yeah, there's just, again, are you getting Are you getting what you really want, probably not,
David Parry 33:36
Well, Sam was showing me something yesterday, which is you're seeing this. So you film yourself doing something almost as if you're the model, you know, those motion capture things to do the white dots. Just Just act normally. But you know, do film your scene. And then you give it a photo of something that you want to have the style, obviously, it could be Vango sort of portrait, or it could be something out of some fancy movie, just a still shot, and it will then redo your video in that style. So you've just created this whole movie without any of all of that CGI stuff needed. Because it's done anyway. Incredible. Still a demo stage can't buy into that yet. But that's, that's being done
Richard Buckle 34:15
There's the whole fake voice thing as well as in the
David Parry 34:18
Yeah, that's scary. Where did you gowith ethics on that pretending to be someone else? So there's loads more and I'm, I'm really keen to see in three months time what version two of this podcast looks like. But I want to close out on our early attempts at doing our own version, not using a commercial product. But we've developed a piece of software now it's still very much in I would say alpha release, not yet beta, but it works. And it's machine learning version rather than deep AI taking lots of data and working lots of stuff. But because we use HubSpot, we plugged it into that and
Richard Buckle 34:50
Plugged it in and it's around sales forecasting, an issue that many businesses have, had you get an accurate forecast? All sorts of reasons why you would want an accurate forecast. But there's all sorts of reasons why it's quite difficult to get an accurate forecast. So by training a model, using machine learning to look at, you know, what factors influence the deal closing, trying to come up with a way of forecasting more robustly.
David Parry 35:17
And even without preparing the data beforehand, we were getting 80% accuracy levels on the training data. And what we learned actually is that if you collect more data than naturally exists, so you capture certain moments in time, for example, and you freeze that data rather than be overwritten. So you've got a record of how I deal progressed on all sorts of counts, you know, what stage in the pipeline was out? How old is it? How many emails have there been a call yet? who contacted you? And what sort of thing to have? Or do they qualify? If you capture more information, then it is really very accurate, very powerful. So there we go. Already, we were doing about a year ago.
Richard Buckle 35:52
Way ahead of our time.
David Parry 35:53
And now all of this has come along. So There you go. That's, that's a whistlestop tour of AI, just scratching the surface really. And if nothing else, our intention today was to raise awareness for those that have maybe heard about it already in the paper and don't know much about it. This, this isn't just a play thing. This is already being applied usefully in businesses.
Richard Buckle 36:15
Tinfoil hats at the ready.
David Parry 36:17
Okay, you've been listening to The SME Growth Podcast talking about artificial intelligence today. Please like us, subscribe, follow on wherever it is that you get your podcasts from. And more importantly, tell your business friends and colleagues to give us a listen, tell them it's worthwhile. There's lots of different subjects we're talking about. And more importantly, good luck with the businesses as you go forward. And tune in next week. Bye now.
Read the blog about Rich's experience with Midjourney, and the infamous Girl with Ice Cream Eyes here.
We mentioned a range of software during this episode, find out more about ChatGPT (AI Chatbot), Midjourney (AI art generator), Magic Minutes (Our own note taking software), Fathom (AI Zoom note taking) and Runway (AI video editting).
Episode 09: Goal Setting for SMEs
In this episode of The SME Growth Podcast, Dave Parry and Richard Buckle discuss the importance of...
Episode 25: The UK Productivity Puzzle: Strategies for SMEs
Are you curious about how productivity impacts the economy? In this episode, we dive into the...
Episode 15: What is Your Business Strategy?
In this episode of The SME Growth Podcast, Dave Parry and Richard Buckle discuss the concept of a...