Join hosts Dave Parry and Richard Buckle from Wellmeadow Limited as they explore the world of generative AI, also known as gen AI, and its implications for small and medium-sized enterprises.
In this quarterly AI update, discover how AI can drive growth and innovation for SMEs, enhance operational efficiency, and improve decision-making processes. Dave and Rich share their personal experiences with AI integration, discuss the latest AI news and emphasise the importance of creating AI policies for safe and ethical use. Gain valuable insights into leveraging AI's potential for SME growth in this podcast episode.
Get the most from this episode in the form that works best for you: watch the episode, read the transcript or access further resources below.
Find out more about what we do at www.wellmeadow.co.uk
You can listen to the full episode here
Want to listen later? Head to Spotify, the episode if also available to listen on Apple Podcast, and click here for all other available platforms.
Please note: Whilst all transcripts are double checked for accuracy, they are transcribed via Otter.AI so may contain errors.
David Parry 01:31
we know who you are. If you start listening. Don't listen anymore. Alright, let's do today's on them. Cool. Hello, and Welcome again to The SME Growth podcast. And it's Dave Parry here from Wellmeadow Limited with me is Rich Buckle.
Richard Buckle 01:47
Hello.
David Parry 01:48
Good morning, Rich. Good afternoon,
Richard Buckle 01:50
afternoon, lunchtime. Later. Yeah.
David Parry 01:52
If you hear that Tommy's rumbling our way through, this
Richard Buckle 01:55
I missed breakfast this morning.
David Parry 01:56
So that's not good for you.
Richard Buckle 01:58
It's not good.
David Parry 01:58
Although, anyway, we're gonna get into the whole
Richard Buckle 02:01
intermittent fasting
David Parry 02:02
dietary advice, intermittent fasting. It's not what today's podcast is about. Now, we said we'd come back to this subject. I didn't look up the exact date of it. But I think it was about February, we last covered this topic. And it's so fast moving that really, we need to come back to it every two, three months or so. It's new stuff all the time. And it's the subject of AI, artificial intelligence, more specifically, generative AI or Gen AI, which is what everybody's talking about. Ever since chatGPT was launched on us unsuspecting public last November. And we're taking a particular angle on this, we could spend all day as people do talking about it. And we do want to cover a bit of the news, don't we have what's happening of late because it's in the news, again, it's always in the news. But we want to try and take a particular focus on how SMEs could and should be taking advantage of it. What should they be doing already? If they're not? What are the risks?
Richard Buckle 02:53
Yep, absolutely. So yeah, I mean, it's something we're on a bit of a journey with AI
David Parry 02:59
well. I think we're probably top 10% Maybe of small companies using it maybe even top 1%? I don't know. Yeah.
Richard Buckle 03:07
It's hard to tell, isn't it? Really, but it's, but we like to try and we you know, we're I think we're increasingly incorporating AI into our working practices into our day to day life. I think I'm still you know, it's a rough guess. But maybe saving an hour a day on, you know, not every day, but over the course of a week, probably an hour a day in terms of productivity saving
David Parry 03:30
It's impressive, I'm certainly saving time. It's hard to put a finger on it, I suppose. Because it's starting to become a natural thing, almost like going to the internet and Googling for something. Now you I've always got a window open for for both of the main chat versions, ChatGPT and Bard. And yeah, it's definitely saving time,
Richard Buckle 03:49
saving time. I mean, I think because even becoming a bit more, so I could tell a story about my daughter using it. That's kind of a, you know, it's becoming, I wouldn't say ubiquitous, but it's much more common. And saying the other day, my daughter's applying for become head girl, or whatever it is in her school. And they had to write a personal statement, which she then informed me that she wrote, and it was, you know, good, but she put it through ChatGPT. And before I was able to kind of proof it for her, she informed me she'd already sent it. But you know, you've got Americanised spellings in there. And there were things that you probably expect a 12 year old to not quite say but articulate it, but she prompted it to say it wasn't just rewrite this it was rewrite it with a focus on leadership style. So you'd actually engineered the prompt a little bit, which is why it came out as something that probably was a bit beyond.
David Parry 04:42
But it's certainly got a style isn't it even when you ask it to adopt different styles, you can start to spot now when there's an AI.
Richard Buckle 04:51
Yeah, I think the American spelling's are a very good giveaway, some of the language it uses. So, in the end, what we decided to do was to actually send in the original piece of work to her teacher, because you'd already obviously sent the AI version to say you told us about AI in an assembly. So I've used it to enhance my work. But the original idea are my own
David Parry 05:15
Good save. Just gonna show you got a credit ChatGPT when using, I think it's, it's going to catch people out if they don't if they think this is just a shortcut. Well, yeah.
Richard Buckle 05:24
And that might my point to her was that it's all well and good to use this to augment and enhance what you're doing. But what you want to show is that the original ideas were your own. Yeah, not that you've had AI because I said to Alexa, that I can get AI to give me the same ideas that you've you've come up with, just by asking the right questions. And how does your teacher know whether or not it's you or the AI? So there is that kind of, you know, what do they call it? Is it province or something? It wasn't? There was a big word of the NFT Community,
David Parry 05:56
provenance?
Richard Buckle 05:57
provenance, provenance to show that your idea? Yeah, and your,
David Parry 06:01
I mentioned, augment there as well. And we're gonna mention in a minute, an interview with Chris Young, who's head of strategy for Microsoft to very much banging on about that being the intended use of augment not to replace. I thought it would be useful. but that's good example. And maybe come on to bits of that, it comes under the what it's useful for, but where the risks are. But I thought it might be useful to do a bit of a roundup of stuff we've both been reading and listening to in podcasts and just being aware of because there's a lot of stuff going on at the moment. I mean, at the time of reading, it's it was headline, I think, yesterday on the BBC News, because the Prime Minister's adviser was suggesting that AI would have the power to kill millions of people within two years. So that's certainly a good headline grabbing story, if ever there was one, or enough that come from, and I think that was in the context of Sunak going across to America to speak to Biden, and they're trying to come up with some sort of initiative, which will start a regulatory process of some sort.
Richard Buckle 06:57
So what how, how was he imagining that it would kill millions of people?
David Parry 07:00
Well, that's first thing I searched for. I was gonna do someone else as well. And his particular point was that it could help to create either cyber weapons or biological weapons. Okay, still need a human.
Richard Buckle 07:12
So indirectly, not kind of a Skynet? Well,
David Parry 07:14
I don't think at the moment, we've connected AI up to anything, yet. I don't think anyway, that has the power to actually pull the metaphorical trigger. I guess once it starts driving cars for you, it could drive you off a cliff or something. But it's I don't think yet we've given it the control. We haven't given it the red button for the nuclear arsenal have we?
Richard Buckle 07:33
Like Irobots or something, isn't it? When they take over.
David Parry 07:36
But there was an honour to come back on this because that really echoed a go to statement put out last week. So this is also newish was put out by the Centre for AI safety, which is an organisation but it was co signed by 1000s of luminaries in the AI world, including the CEO of both open AI and Google's subsidiary, which is doing the same sort of AI work. And so it's got some weight, there's only 23 words to it. And I thought it was almost worth reading out for those that didn't pick it up. But the statement that was supported by Open AI, Google, Google DeepMind and others, was as follows that mitigating the risk of extinction from AI should be a global priority, alongside other societal scale risks, such as pandemics and nuclear war. That's their study. That's the statement 23 words. Now I guess it's probably a bit maybe they spent a long time crafting those 23 words such as such a thing that everybody can put their name to, doesn't go into great detail, but it's certainly a cry for giving it attention and giving it to the very people that have created this technology that are saying so
Richard Buckle 08:43
Pandora's box and we opened it, lifting the lid peaking well
David Parry 08:47
Well and truly lifted surely
Richard Buckle 08:49
I think one of the problems with trying to regulate this I don't know if they'll be able to because how are you going to limit you know, kind of Western you know, democracies or whatever which most of this call for is coming is the new call and regulations coming from versus Russia China "SME podcast gets political"
David Parry 09:11
we've just been cut off in China
Richard Buckle 09:14
But they're not going to stop are they,the Chinese aren't going to stop developing AI.
David Parry 09:18
I've been recommended a book which I just started reading which is the history of the development of the atom bomb, okay, from pre war you know, the background of the physics and then obviously the creation through through the war and so on. And it's a I'm drawing strong parallels so that the once you've invented it, no matter what you try and do through all the start treaties and everything else to try and get the control over it, you can't and it's been done humans have done it they've now created at least maybe two tools that can kill all of humanity at the press of a button. Nuclear bombs and now maybe AI I don't know. We're not given it yet as I say the power to physically do it. But in the in that website, the Centre for AI safety they mentioned that Four specific risks, which they think are very real in terms of what it could do, maybe not to entirely destroy humanity within two years, which is what Matt Clifford, Sunak's advisors said, but they list the four as being that it could be weaponised, such as drug discovery tools could be used to build chemical weapons. That's one risk. Another one is around misinformation. Now, given that, we saw a lot of bots applying through both of the recent elections and referendum, and so on, and they've been accused of skewing different democratic outcomes, then imagine if you could do that at scale in a very convincing way in so I guess that's one of the concerns.
Another one is that it may not be democratised in the sense that not everybody would have equal access to it. So if it became more in the hands of the few, rather than the many, then that's going to potentially create conflict, because some people then get to use it for their, their benefits, whether it be rich countries, or powerful people, or whatever it may be. And the other one was more esoteric really talks about the enfeeblement of humans. And I think this is a concern I've heard expressed elsewhere that if we tend to rely on AI too much, we will just stop thinking for ourselves. And there was even a thought that, given the AI, generative AI is trained, on example, data that humans have created over the whole history of time. But as recorded on the internet, if we stop creating data, and output of our own, because we're using ChatGPT, or run out of new stuff to teach itself on. So then there's a question of can it create enough innovative content of his own that it can train itself on its own output? It needs some quality measure, you need to know whether what it produces is acceptable or not. And it's still learning, isn't it? But when does he get to the point where it can judge for itself, whether it was produced is good enough?
Richard Buckle 11:47
Well, it's like the kind that was on Google music, where you can get it to produce its own music now. So if you say, so rather than like going to Spotify or Apple Music and saying, Okay, I want to listen, some jazz, you can go to this AI generated music, and it will say, original composition of some, you know, Danish jazz in the style of the 1950s or whatever. And it will give you the sax. Yeah. So it's like, yeah, it's when almost
David Parry 12:18
Yeah, well, artists have certainly come up with lots of concerns about it, because of its ability to create art. And we were talking earlier about the ability now of some of these tools like Midjourney to produce fantastic images. You don't need a photographer or an artist, you can just create it music is the same as per the written word which is recited out. So the enfeeblement of humans, it's a, it's a real worry, isn't it? And I remember back when I was younger, I used to be able to remember people's phone numbers, because you had to dial them. Yeah, you don't anymore. It's maybe a trivial example. But it just goes to show how, what we regard as being important, we now no longer bother remembering. And I used to have a road atkas than the car because I had to navigate in places. But now I've got some have, you know, my my memory for road numbers and where junctions are, you know, I'm getting lazier. I just follow it. And digital natives have never had to do that. They just
Richard Buckle 13:08
every time I bring this up, and my kids, they always quote back to some like Albert Einstein quote, around like, you know, if it's written down somewhere, why do I need to bother remembering and they're like, we'll have Einstein said that. You still need to learn your times tables,
David Parry 13:24
I think the bigger risk for us to worry about, given the scope of what we can do about it, is to be aware, make sure that you and your teams are aware of the more practical risks of it, rather than its ability to destroy humanity within two years, there's not a lot we can do about that. Yeah. And they're probably more around the release of confidential data as people start playing with it. So if you wanted to give it a bunch of data and get it to analyse that for you, that's all very well and good unless that data happened to be very confidential personal data, or commercially sensitive data you've just released. So very good to try these things. But maybe just have a pep talk with your team to say that if you're going to use it, and you're welcome to anonymise the data first, which
Richard Buckle 14:04
is interesting poison, it says, like, I guess there's all these, you know, Centre for AI and other bodies now starting to talk about policy around AI. But I wonder how many SMEs have an AI policy? And is that worth thinking about? Like, what data can go into it? What can it used for
David Parry 14:19
it, lots of larger companies are developing this to the point where some of the very largest companies in the world have now about AI departments. And it's not just a spin off of the IT department. It's there as a service to all the business units. And they're starting to look at how they can redesign their workflows, or their business processes.
Richard Buckle 14:36
So I saw somewhere that they that there was a prompt engineer job going six figure salary, prompt engineer,
David Parry 14:43
I've heard that talked about a lot of prompt engineering, and maybe that in time doesn't become as necessary because there's so much intelligence around the prompting that the AI can work out what you really meant to say, but for now, you need to ask quite specifically. That's true. So I think there's certainly the problem around confidentiality and they will Because of that, I think another big risk is the hallucination problem. And that's what they call it when AI bots come up with a very plausible, convincing answer. It's just completely made up. Yeah. So I'll make up references, for example for where it's got information from, or it'll make up information itself. So that can be very dangerous, if you haven't yet internalised the fact that these things are just huge statistical models trained on massive amounts of language. And all they're doing is trying to work out what is the most likely next word? And the answer doesn't have to be right. It just has to be the most likely
Richard Buckle 15:34
answer. Yeah, I think there was somebody who recently tried to defend themselves in court using GPT. Maybe you've been a law firm that used to, and then they they had this kind of like case references of Brown versus Yeah, whoever. And it made one up, and that's the quota. So the judges then okay, well, what's the you know, what, what happened? And then they kind of went and looked at it and ChatGPT just made up made up the case. And it sounded plausible to wincing. But you know, when you dig into it, it's like this thing is, there is no record in any actual kind of legal precedents that this thing
David Parry 16:11
It's been released without those checks and balances in place, that's going to have to happen. I was talking to a senior partner in a law firm last week. And he agreed that the type of law which is around creating leases, or conveyancing for house purchase, or doing wills, template stuff, you know, shareholders agreements, that's something that is definitely going to be much more easily replaced by AI and other types of technologies than high-end advocacy and opinion.
Yeah, sure. So lots of wherever. So I think the confidentiality problem, the hallucination problem, for sure, I think there's also a very real risk, not quite of industrial unrest is maybe too strong a word, but anxiety, industrial anxiety, where I think people in various jobs are going to start to get worried about whether their job is at risk because of this, and be looking over their shoulder all the time. And I'm gonna refer to a couple of interviews in a moment because I'd refer people to them if they want to find out more about it.
But there's, there's one thought that there's two types of mistakes that companies are making that are mutually exclusive, you're making either one or the other. And either you shut down all use of ChatGPT, and Bard and everything else, and you've got a company policy not to allow it to be used because it's just potentially too dangerous, and you don't want to get there. And we know organizations literally that have done exactly that we know people who work in them. And okay, I understand that that's a very risk-averse, safe position to take. Now, apart from the fact that you're saying you don't trust your people very much. And you're not giving them access to the latest tools. You're asking them to work with the hands tied behind the back compared to other people who are allowed to use it. So there's definitely a risk of avoiding, you know, it's out there. It's like saying to people 20 years ago, you're not allowed to use the internet.
And the other risk, which is almost the opposite end of the spectrum, is where we've completely unfettered access, no thoughts at all, as to the risks, that you just let people use it, however, and you don't train people or inform people. And I think one of the aspects of that is to make sure everyone realizes that if we do make a success, when we make a success of AI in our businesses, you have a choice. If productivity increases, productivity means doing more with less. So are you going to use it to do more? Or are you going to use it to have less people doing, so you're going to cut jobs when you do more stuff.
And the advice certainly from those propagating the use of AI is you've got to continue to think about creating more value all the time. So let's say you could multiply your productivity by 10, you know, and you've already saved an hour a day, even with fairly primitive AI tools, compared to what they'll be in a few years' time. You're not having an hour off a day or cutting your week short by a day, you're doing a lot more stuff. And not just doing more of the same, but thinking of other ways of adding value in more innovative methods, approaches. And that's what we've got to make sure everybody realises is the likelihood office, okay, when certain industries were retired, let's say the mining industry in the UK or something, there was a short-term adjustment period where people had to retrain and they do and jobs. But now we have a labor shortage, million vacancies, and people just retrained into other things. So I'm relatively optimistic about this, I think there will be ever-increasing amounts of work to be done that adds value. And if we get another tool that comes along, that accelerates us so much, then great, but let's take control of it and be aware of it and be on the front foot and make sure that we use it to help us rather than be caught out by it because we've been very active
Richard Buckle 19:46
Yeah, it's interesting isn't it gets philosophical quite quickly, isn't it in terms of what what do you do with these tools which which sectors are going to be hit? Hardest by this? You know, I saw something the other day around. Somebody using AI to generate architectural plans. And then using it to create 3d model of the house that rendered and then, you know, you could then I guess get it to do a bill of materials or something for you if you wanted to and all of those sorts of things, I think, well, that's an architectural example. You've talked about the legal sector. Accounting could be, you know, very much lend itself to AI,
David Parry 20:20
The more its trained on it. It's still in his training, and it can be trained on examples. Yeah.
Richard Buckle 20:23
So it's, yeah, I think it's the professional services sector could be
David Parry 20:30
used for sure. Especially if it's template driven. Yeah, like maybe it can see, but certainly some elements of the legal profession, just in the way the internet fundamentally changed travel agency, as a business model, and maybe even estate agents, you know, that access to information being democratised is a good thing, you start expanding that now into an AI way of interpreting that. And then you do a whole different, but I think
Richard Buckle 20:52
on that note is an interesting thing to think alike. I guess I see a lot of stuff in my Instagram feeds and things like that around AI. And if you're reading around the subject, again, your news feeds are going to come up with an AI so you almost get this kind of this bias kind of kind of built in, isn't it? Like kind of confirmation bias? The AI is everywhere. Yeah. And it's ubiquitous, and it's going to take over and it's gonna happen, you know, next week type of thing. Whereas sometimes I speak to clients or the people who aren't plugged into AI at all, there Absolutely. Is a fad. AI is never going to take off. Nothing. You know, it's just all something that everyone's getting hyped up about. Anything. Okay. Well, there's maybe something in that, that you can, you know, overplay the definitely speed at which this is gonna hit the mainstream and is this still, but I think that in and of itself then represents a great opportunity for SMEs to say, right, if you can get on the front foot of this. Yeah, there's probably five, maybe 10 years?
David Parry 21:50
No, I think it's a lot faster.
Richard Buckle 21:51
You think?
David Parry 21:52
Okay. So I said, I was coming back to a couple of interviews. I wanted to refer to them. They're both from the Harvard Business Review podcast, which you don't have to be a subscriber to Harvard Business Review to listen to their publicly available. So go and have a listen. They've done a series during May have four different podcasts around AI. And they interviewed the two most interview interesting interviews for me, were Sundar Pichai, who's the CO CEO of Google alphabet, whatever. He was really interesting on it. And the other one was Chris Young, who's head of strategy at Microsoft. And they were talking about the various different evolutions of their products that are coming up soon. Yeah. And Sundar Pichai said, It think of this not just as important as when the internet came along, and how that changed. Now, maybe he's ever playing this. He said, think of this as like my fire came along. You know, it's going to be that seismic a change to the way we work. And it's gonna happen really, really fast. So it's hard to believe that reasonable access to the internet only started happening 20 to 25 years ago. Yeah. And there wasn't a lot on it. Anyway, that's a very short period of time. And look where we've come to now, you know, with the billions of page. So I think what we're seeing that development will now happen in two or three years, that amount of improvement, but in the in the AI space,
Richard Buckle 23:09
think they will become mainstream, though.
David Parry 23:10
In the same way the internet has. Yeah. And if you're interested in a book on this, Ray Kurtz point of singularity addresses this when he published it that God knows how long ago I remember it was recommended to us. And he set out to try and work out when in the future is the year that computers are able to match humans in every aspect of ability, and long short that he came up with the year 2047. And the way he did that is to try and normalise the amount of progress that humans have made in each century. And market how much of the following century did they need to make an equivalent amount of advancement, right? So we had to come up with had you equivalents, things like Steam versus flying. And the broad rule of thumb we came up with is that every century's worth of improvement happened in the 20 years at the beginning of the following century. And it was exponential, because that 20 years didn't happen in the, whatever four years remaining, and so on. So all of what happened in the 20th century has already happened in the 21st century in terms of progress. And bear in mind of the 20th century went from not even being able to fly or make a telephone call. Yeah, right through to you know, just about having a few mobile phones kicking about?
Richard Buckle 24:14
Well, I guess there's looking something today on quantum computing is making particularly large strides at the moment.
David Parry 24:21
Yeah. And Google are really investing a lot in that they really sense that's going to be one of the big next thing and there
Richard Buckle 24:26
was there was talk of it being almost like they're not quite there yet, I don't think but that stage where the interface like because clearly most people aren't going to be able to own a quantum computer. They are, you know, fairly specialised in massive pieces of equipment.
David Parry 24:41
Well, that's what IBM said in the 60s too.
Richard Buckle 24:44
But the idea is that you would be able to kind of access that as kind of a SAAS. Here's my issue.
David Parry 24:50
Yeah, who wants a computer anymore anyway
Richard Buckle 24:52
Quantum computer to be able to solve the things which will do it. I mean, I don't know quite what the number is, but it's 1000s if not millions of times. is more processing power than, than a normal computer today?
David Parry 25:04
What's crazy about that is no matter how fast it gets, we're still relying on cables that are on the seabed. So the Russians can go along and snip with their submarines. There are gonna be these points of frailty. Imagine if we get so dependent on using AI systems just the same way we're all dependent on, you know, our mobile phones now let's say, or the internet. As soon as that turns off we'll turn into zombies staring into space.
Richard Buckle 25:26
Whether it's someone saying or something yesterday with an EMP in the UK, if an EMP strike happened 90% of the population would be dead within three months because it would just like cripple everything, takes all the systems out all the electricity because of electromagnetic. And then someone else about this thing, how the nuclear, because if you soon as you kill all the electricity, once the nuclear power station reactors, backup power fails, all of those cores just overheat.
David Parry 25:56
That's what we're worried about in Ukraine at the moment
Richard Buckle 26:00
So yeah, so yeah, It's all getting a little bit dystopian
David Parry 26:02
Let's just bring it back a little bit, I want to do a quick roundup on the tech and where we're at. So Google have just announced the combination of Deep Mind and Brain which there are two organisations working on this, they've come together, and they've unveiled their next big product called Gemini, which is designed to take on open AI as chatGPT head on rather than bard at the moment. And when Sundar Pichai was asked, What's going to be different about that, then two things, one is going to build a lot more safeguards on it and tried to be more responsible about quoting sources. But the main two technological improvements are one that is that'd be multi media, rather than just text based, it won't just be able to input video and images, and maybe audio better output them as well. So if you ask it to write an illustrated essay, or write a PowerPoint presentation, it will also come up with images that go in AI generated as part of a maybe even a soundtrack of somebody talking about. So that's one. But the other one, which I was much more excited by in a way, it's you as well, is that it will be programmed to make use of other tools just like humans can, right. So we've all got loads of other SAS products available to us, whether it be a CRM product, or project management tool, or whatever, are tools out there that can convert JPEGs to PNGs, and all that stuff, then it will also be able to access those. So that suddenly becomes a very powerful prospect. Now he's hoping that that's gonna come out in about October this year. And then we'll see where that goes. Bard is already there, Bard has recently changed to allow you to access the internet Live, which ChatGPT for didn't do until recently. If you're a paid member, you can switch that on. So whereas you used to get responses all the time saying I am an AI language model, I can't tell you that because I was trained up till September 21. Well, now it can access new stuff. So I typed in earlier, our web address wellmeadow.co.uk, what can you tell me about that company? And it told me an estimate of the number of employees what we did. And so some of that you can say that's almost Google searchable. But then it said, what are our strengths and weaknesses? Well, we don't list our strengths and weaknesses on our website. It's inferred that from stuff, maybe comparing it with other things. So already, we can get a bit of that. And interestingly, both it and chatGPT are starting to put the source information against their responses to link it through to the website, where it got most of its content from
Richard Buckle 28:23
That's got some pretty powerful applications for either competitor monitoring or prospecting. You know, just what information could you get AI to serve up, give you insights
David Parry 28:35
Given this company X, Y, Z limited. Tell me who are their competitors or if this is one of our top customers, tell me who else is like that customer in that region? You can see why GDPR is going to become evermore focus here, if that's being used for personal data, using if a company's slightly different. So that's where they're at ChatGPT, you know, t they've launched four obviously that we talked about. Sometime 4.5 is in the pipeline but not due until the autumn. So that'll probably come out at about the same time as this Genesis thing from Deep Mind. And they're even talking about a ChatGPT five, but wouldn't be till next year sometime. But bear in mind that when they rolled out four, and they did the video showing what it could be capable of a very much bandied around the use of image input. And that still isn't broadly available only to the beta tester people. So we can't now use the chat interface for ChatGPT Four, and give it an image. But what I did try earlier and because started getting there, if you give it the URL of an image and say, Tell me what this image is. It starts off by saying IMO text base large language model. And then it says, and so I did it earlier on and the URL I sent it was a picture of a broken egg. And then it says after saying I'm a large language model, I can't tell you anything about this you need to go and look, he then says but if this is a picture of a broken egg, I can tell you about the cholesterol content of eggs or what the shells made of all that. So I don't know why they choked off. Clearly, it's made an inference that this might have broken egg. But it's not telling me this is a broken egg on a black surface with lighting. So maybe that will come it sounds like it's it's sort of almost there where it's just about how you prompt for it. So that's where we're at with all that. I said, we come back to the Chris Young interview, who's the head of strategy at Microsoft, because he's very much pushing this concept that we all need to regard AI as being a co worker augmenting you. And he said that in their own fact, Sundar Pichai said this as well back up the point, they found that a pair of programmers work better than two programmers in isolation, because they spar off each other and said, Imagine, that's how you have chat, GBT sat by your side. So it's just someone else prompting you with ideas, brainstorming, correcting your English. And they talked about the fact that it brings, if you've got a range of abilities, it brings the lower end of knowledge workers up to at least the average, if not better. So the improvement in your team could well not be that it's taking the most able and taking them further. It's bringing the lesser able ones up to a comparable level. So that's quite an interesting sociological point.
Richard Buckle 31:13
you could go so many places with this, just in terms of just it's almost like a quasi employee, isn't it? Yeah, having artificial intelligence,
David Parry 31:22
like a personal assistant with all sorts of power. So you give everybody a personal assistant,
Richard Buckle 31:26
but you then you would then say, right, okay, read our office manual. This is our office culture. This is our, you know, SOP, this is all of the stuff about our business. Put that into the AI, learn, train yourself off our information, and now be a personal coach to whoever's in the business. And their role is this. So what would be the best development plan, you know, remote applications? What would be the best development plan for someone in this role? There we go, you know, our business, you know, what we're talking about, you've done a website you've seen, you know, social media, imagine
David Parry 32:00
if you had access, you could give it access to your diary and your emails and whatever other databases you use, and just ask it, what's the most important thing that I do next?
Richard Buckle 32:09
Scheduling would be Yeah, that that type of thing?
David Parry 32:11
What's distracted me the most in the last week? Or what should I have done that I didn't do? Why, where am I failing as a leader, you know, it can only tell based on what you're training with. But there's all sorts of augmentations. It's almost like the Ironman suit for the worker, it still relies on your brain
Richard Buckle 32:32
that I just showed you that clip of Humane, which is a spin off or not spin off. But there's a load of Apple employees who create this company called Humane, we're now creating a wearable device that is powered by AI, link to AI. And I think they're, they've done a TED talk recently, or something on that and showed some of the capabilities. But just on that point, you could imagine something that you're wearing this device, you can see what you see, right, so it can see what's on your screen. It could analyse what you're looking at on the screen. compare that against, you know, what your diary said you were gonna do? And say, well, it looks like you were working on this piece of work, which isn't the same as what this was. What about that distraction thing, or?
David Parry 33:14
Well, even, there must be countless occasions when you're trying to improve your management style all the time. And after a conversation, you think to yourself, I could have probably approached that habit differently. Imagine if that was almost fed to you live, you know, wouldn't it's not far of a stretch to have an earpiece connected to this. And have your your AI assistant saying to you what, don't forget to mention this or, you know, don't go down that road. Or don't forget, this person has got sensitivity around that subject because of this other event. Have you asked them about their holiday last weekend. So given the time we've got left, why don't we do just a quick reminder of maybe where companies could be implementing this. So this will be done so far is just a general update on and a waffle on what's going on in the world of AI because it's so such big news. But there are still very practical things that SMEs can be doing.
Richard Buckle 34:03
I suppose the caveat here is that we're still on the journey with this or starting the journey with this, aren't we we're we're trying to implement some of these things and try out new technology. And yeah, so we're just yeah, for you go through what we're sort of working on at the moment.
David Parry 34:18
Yeah, we've done some interesting things. And I would encourage other SMEs, listening to this to come up with their own maybe based on what we've done or others, and just get some people to try it. But doing a reasonably structured way, tell him what it was going on. Make the point that we're learning, you know, this isn't gonna put people out of a job. It just allows us to do our jobs better and therefore do more of it. But try some of it.
Richard Buckle 34:38
I think picking up on a point that we made earlier about the policies around AI after this podcast and might get a chatGPT write one for us.
David Parry 34:46
Well it's not a bad idea. What is our policy on it? Are we going, we haven't said anything to the team yet about whether they can use chatGPT or Bard, which of course we would let them do. We will encourage them to do it. But with what safeguards what bumper guards we put around that You know, we haven't told everybody that if you're going to upload data to get it analysed just to see if you can you've got to anonymise the data. And that's so basic stuff don't reveal personal information. Yeah, you're right, we should write up an AI policy. That's not a bad thing, maybe for all companies. In fact, if we do and we'll refer to it in a future podcast, and maybe put that on our resources page. As an example, I don't know if other companies have got one yet interesting to ever look around. We should ask some recent clients have you got an AI policy, they've all got bring your own device policies and internet use policies. But this is a different order of things. So there's that. I think we're using it quite a lot for writing stuff, text based output. I think I mentioned in our last podcast that I took a an email I'd received, put it into ChatGPT and asked for it to write a response and it did. So that's kind of an easy one. You can get it to rework stuff you've done but you feel you've done the wording a bit clumsily and you want it polished up a little bit. Be careful of those American spelling's
Richard Buckle 36:01
Yeah, experimenting with style. I think I did that. This is trivially jokey example, but somebody was quoting me in an article for a piece of work that we we had done and they wanted me to approve the quote. So I was like, okay, that's fine. But you could also try it, as in the version that like Clint Eastwood would do what they'd said, put it into chatGPT and said, Give me an example of this, as per Clint Eastwood, and did another one is Elon Musk, and it did another one as an eighth century monk from Shrewsbury Abbey. And they were all like, you know, just to prove the point that the seconder can take that one bit of text and change the style.
David Parry 36:38
So I think it's good to know what it can do. That reminds me of the three phases of AI so far since November last year, where to start with we were playing with it, then we tried to break it. And now we're starting to use it.And I think we we're starting to get a bit more serious about it. It's not just a passing fad, you know, we said a hundred million people had signed up for it in the first three months. Well it's over 500 million now. Wow. It's crazy. And there's only 8 billion on the planet, how many of them have even got a computer so it's quite scary. So yeah, we're starting to take a bit more seriously. So writing text based stuff, I heard about someone the other day creating a little automation loop using Zapier, which some listeners may know about that links other systems together. And through their automated candidate management system, for dealing with job applicants, they were able to just click a button to say send a rejection letter to that candidate. And it composed the letter based on all the records that have been received. So far, the cv, the scores the interviewers are given or the assessment they're given anything that had been determined that far in the process, and wrote a reasonably comprehensive rejecting letter, which clearly wasn't a standardised, dear John template. And they said, the amount of feedback, they got a very positive nature, you know, thank you for taking my application seriously. And I found your response letter very helpful for me and all that sort of thing.
Richard Buckle 37:53
So the GDPR implication of that it's taking someone's personal data and sticking it into an AI
David Parry 37:57
That's back to your policy again, because if you're naive about it, you've just released that as it's now open source. And, you know, almost future language,
Richard Buckle 38:05
learning on it saying, as you know, by applying for this job, you almost need some sort of policy statement in there. And by submitting this form, or applying for this role,
David Parry 38:14
I wouldn't have gone as far as automating the link through to chat GBT on that I would do, like we said earlier, anonymise the data, you know, for candidate A, for job B, Company C, this was the facts. And then you have to take that
Richard Buckle 38:26
and put it in there some interesting implications there was.
David Parry 38:29
And I was with someone last week who was planning a conference in the retail sector, and they wanted that they tasked their team with coming up with some good ideas for plenary topics, so that after the main seminar, they could break out into breakout rooms and talk about these, these other subjects as well. And so I said, Well, why don't you ChatGPT for that, and it was still new ish aware of it, of course, but not having used it. So fired it up on my phone, because I was out and about, gave it that problem statement, and then immediately produced a full running order for the conference, simple stuff, which we were talking about three or four months ago, there's still people aren't, aren't using it for well enough
Richard Buckle 39:07
That that's back to that kind of that's, that's a good example of how you can save an hour a day type of thing on average, because
David Parry 39:14
it's gonna save several team members, maybe an hour a day, you know, so
Richard Buckle 39:17
you've got, how long would it take to come up with all of those ideas and get everybody coordinated on it.
David Parry 39:24
And within a few seconds, you ask it, then give me three key questions to ask in each of these plenary sessions, and how do I prepare and so on. So you can save an awful lot of time
Richard Buckle 39:33
an awful lot of that kind of ideation time out of the
David Parry 39:36
And make it better? Brainstorming, you know, give me 10 ideas for this, give me the top 10 issues around that and build that into what you're doing. Risk Assessments, what are the most likely risks and mitigation factors when I'm running a business entering this new market? So it doesn't have to be right as we said before, but it will give you prompts and you can turn half right answers into
Richard Buckle 39:57
it's gonna get you going.
David Parry 39:58
I think writing code tell us about that. because we did an interesting project,
Richard Buckle 40:01
we've done interesting thing around this specific application, as many of our listeners will know we're HubSpot partners. And so we have an interest in sort of finding other companies that are using HubSpot to offer our services. And so how do you do that kind of needle in the haystack type of thing. So one of the approaches we looked at was to come up with a way of writing some code to identify whether a website is using HubSpot or not based on the tracking code code that you get in HubSpot, sort of similar to Google Analytics type code. And so yeah, we use chatGPT to enable us to write that code, and then iterated that. Get it to do certain bits and pieces that we wanted it to do. We were to go and find the web addresses what to do with them once it had found you know, what values to return? If it found that a website had a HubSpot link? And, yeah, like something that probably six months ago, 12 months ago, you'd have had to get a developer to come and help you do. Yeah, it was sort of citizen coding, half a day or so of time.
David Parry 41:06
And that's what amazes me about it, really, we don't you know, for everybody just think he doesn't know, we don't have coders, developers, software people on the team. And so we were able to do without that. ChatGPT was our augmented people. We had techy people, maybe who can work their own way around websites and things but only like a lot of people can. And then they managed to take the instructions given to us by GPT, almost like assembling the IKEA wardrobe and produced the finished product.
Richard Buckle 41:32
I think sort of rough numbers, I'd say that there's probably 15% of the so we took our kind of contact database to look at. And probably 15% or so were HubSpot users. And of that probably 7/8%. We didn't know that they were obviously some of them are clients and whatever but so as a lead gen tactic using ChatGPT coding,
David Parry 41:59
now that exact example won't relate to anybody else. I think the point you're making is that we constructed a problem statement, had a, something that if only we had a programer that could do this for us and then thought, okay well let's see if ChatGPT can do it for us
Richard Buckle 42:11
And that yeah, that's the approach, I encourage people to take with it to say, oh, it's almost that kind of, if you had a magic wand, what would you do? And it's like, okay, if you go to ChatGPT and say, I don't know how to solve this problem, give me 10 ideas of how five ideas of how I could, you know, find HubSpot companies, and then start just sort of riffing off that, as it were. And one of those might be, well look for the tracking code. Okay, well, how would I find the tracking code? Well, you could write some Python to do that. Okay, what do I do write the Python assembly? Where it'll tell you? Yeah, how do I then implement this.
David Parry 42:45
And so maybe on a more accessible level, for some people listening, most people are familiar with Microsoft Excel, we mentioned that a couple of episodes ago, didn't, we talked about productivity tools. And most people know how to enter a formula, even if it's a fairly simple formula, like equals sum, A to C or whatever. we'll just as chatGPT, to write a formula for you. And it can do some pretty clever stuff, you know, say I just want to know how many of those data items on the left, we've got this word in, when that kind of, you know, which is constructed in natural language, your problem, and it will come up with a formula for you. You can't quite link spreadsheets as a data source yet. But you can input the data into the prompt. Yeah, so ChatGPT If you're on that takes 25,000 words, that's quite a lot. So you can cut and paste a fairly big section of a spreadsheet and paste it and make sure you've anonymised it all first, so that you don't mind. Paste it in. And I did this earlier with some of our deal data all suitably anonymised and just asked it, What are the major influences? And it said, Well, it looks like the more times that you have a sales activity with a customer that higher chance of it closing a deal successfully? Well, fancy that may be a trivial answer, but it worked it out. When you can get it to analyse correlations and so on gives you the the chance of something being better or worse based on something else you're doing. So we're still at the foothills really of exploring how to use it responsibly for data analysis. But imagine you just put in company accounts every month, anonymised again and say, what can you tell me? What are the key? Top three things we should be talking about this month based on that set of management information? And it should probably do no more I would imagine, then if we put it through a spreadsheet and look for the major variances ourselves, but they'll do it as quick as you may, if you're really clever about it, and you gave it historical information, it may say, Oh, well, this happened last April as well. And that we were pretty clever. So we can do more with that.
Richard Buckle 44:42
Absolutely. And you could you could, I suppose the other thing with that is that if you put that company data when you could say like give me the results as though I'm a you know, senior qualified accountant, what would I be looking for give me the results as a, you know, just sales director or give me a what what am I gonna That's the kind of power it can have in terms of what that you'll get a very different response.
David Parry 45:05
Yeah, sure. And if you had access to other accounts, don't forget, as AI evolves, you'll be able to privately train it. So apparently, ChatGPT is already coming up with a potential option of a privacy mode, where you're going to be able to create prompts that it doesn't allow to go into its repository for future training data. So it helps to solve some of these confidentiality concerns. But you'll also be able to have the function to train it on your own data, which won't go out of your organisation. And this is already happening, in some cases for some advanced users, like on legal databases. So if you trained it on all of your own, let's say financial information over the last 10 years, nobody else would get access to it. And then it could say, the last time this happened six months later, that happened. Yeah, or this happened. And we found it was this last time, you know, you can, it will just have perfect recall, and the ability to draw correlations, you know, at a speed and you wouldn't have time to do it yourself. So lots, that's all in the future. But for now, I think the top tip, as I talked about where I was saying about these interviews earlier, earlier on, and Chris Young was saying this, from Microsoft, the head of strategy there, he said, The best tactic for SMEs is just start to use it. Just try something. Just get familiar with the fact of it can do stuff. And then as further advances come out, and it's built into other products, you'll be less prone to it. More to follow on this, I think this could well be at least a quarterly update, maybe on our own journey, and what we're doing with it.
Richard Buckle 46:31
Well, if we're here next quarter, by the sounds of it,
David Parry 46:34
We've got two years. Matt Clifford said we've got two years.
Richard Buckle 46:41
Need to ask ChatGPT and say if I was a prepper, what should I be? Give me a checklist. Beans. Don't get me started. i'm a wannabe prepper
David Parry 46:54
tonnes of beans in the kitchen. Okay, so we'll do that in about three months time, then we'll give an update on any further developments we've made using various AI tools. And another update on what's happened in the world of AI. Interesting, though. Thanks for your input on that. Good. So once again, you've been listening to The SME Growth podcast and thanks for listening all the way through. You've done well to get this far through us waffling on about AI. As ever, we asked you to share the fact that we're doing this podcast on a range of topics and please share that with other friends and contacts you have in business. I know we aim it for the SME market but I'm sure businesses from startup to build these can get something of value out of what we're talking about. If you could subscribe to us on the various podcast outlets that you see click the notifications button and I know several of you are listening. Every week as soon as the podcast drops. Thank you very much for your support. Good luck with the businesses.
Interested in learning more about using AI tools in your SME? Have a listen to our other podcast episode focusing on personal productivity where Dave and Rich discuss AI and other tools to work more efficiently.