Chris Norton: 0:52

Welcome to Embracing Marketing Mistakes. A little bit of a revert back to where we were a year ago, because in the studio today I am joined with the lovely, lovely, lovely, lovely Andrew Bruce Smith. Welcome back to the show.

Andrew Bruce Smith: 1:08

Andrew, my goodness. Well, thanks for having me back. First of all, didn't put you off first time round.

Chris Norton: 1:15

No, and my God, how much has happened in a year. First things first, from our side, we've changed the name of the podcast. The podcast changed about well, it's changed a little while ago. We've literally only just rebranded it to Embracing Marketing Mistakes, because we're aiming to want people to understand what the show is about and and socially unacceptable was a cool name, cool brand name, but actually you were like okay, what does that mean? Does Chris swear too much? Which, let's face, it probably does. But obviously a year ago that we've changed the name of the show, loads of downloads, but you have just gone nuclear, like what's been going on in your world in the last 12 months Gosh.

Andrew Bruce Smith: 1:49

Well, where do I start? I could be quite cheeky and say, well, 12 months is like 12 lifetimes in AI. It is kind of hard to try and think back to 12 months ago and to remind yourself what the state of the world and AI was then and try and reflect on everything that's happened since. Kind of need ai to help you out, to try to try and sort of make some sort of sense of the, you know, just insane um technological development, particularly in the world of kind of marketing and uh, communications, the adoption of it, or you know, arguably, the slow rate of adoption, possibly, uh, you can't avoid talking about AI but trying to find actual, honest to God, real practical applications and use cases. It's not as easy as you think. So there's all of that, so a lot a lot to take in.

Chris Norton: 2:47

Yeah, so much has changed in the last 12 months in terms of AI. We had Stuart Bruce. I was just telling you before we started recording that we had Stuart back on the show because he was our first guest and, obviously, my ex-boss. So we had him on the show and I was like, let's go through, what have you been doing the last 12 months? And he was saying like so there's yourself, andrew. For those of you who don't know, andrew's company is called Esherman, esherman, esherman.

Andrew Bruce Smith: 3:11

Have I pronounced that right? Esherman, that's correct. And what does?

Chris Norton: 3:13

Esherman, I don't know if we covered that in the first podcast last year. Oh dear.

Andrew Bruce Smith: 3:17

Well, I'll try and give you the short version of the name, what it stands for. Well, 40-odd years ago I read a book called Hofstadter, 1979, won a Pulitzer Prize. It's still regarded as the Bible of artificial intelligence. And I read that all those years ago. And 15 years ago, when I was trying to think of a name for the business, I'd sat around for a couple of days and my wife, who's very practical, said pull the finger out. Just, you got to come up with a name. So think of the things that you like. So I don't know, I like that book, gird. So think of the things that you like. So I don't know, I like that book Good. Oh, escher, I quite like that Escher fan. Oh, Escherman, that's a name. I went on and checked out the domain Dot com was available. I thought, oh, what the hell, let's go with that, right, ok? So actually there are.

Chris Norton: 4:00

AI roots of us doing social media right at the beginning.

Andrew Bruce Smith: 4:01

That's because we're old.

Chris Norton: 4:16

And people. Sorry, no, I am, that's how people see us. You know we were like some of the first. There was a certain group, in fact, there was a Twitter group on PR, and I use this sometimes back in the day. I remember me and Stuart Bruce, and we joined Twitter in a similar time because I worked for him, and I swear down, you're one of the people. There was people like Stephen Davies and like some people that we've forgotten about, like Tim Huang and Jed Hallam who's also been on the show.

Andrew Bruce Smith: 4:41

Yes, of course, yeah.

Chris Norton: 4:42

And there was only like 60 people on Twitter.

Andrew Bruce Smith: 4:49

That's right. Yes, Right back in the day when it first launched.

Chris Norton: 4:51

Those were the days. Those were the days I remember we'd all keep jumping on it and we kept getting the fail whale and it kept breaking and everybody was like this is never going to work, never going to catch on. And then Elon got hold of it.

Andrew Bruce Smith: 5:00

Eventually yes, there we go.

Chris Norton: 5:03

So, in terms of my point about Stuart being on the show a couple of weeks ago is basically Stuart was telling me. He said a year ago I said what, what's changed for you in terms of ai? Because he's covered a few things, but ai is obviously something I'm interested in as well as because it's the big thing, isn't it? It's the new social media. And he said a year ago, when I saw you, um, we were doing like we were saying to big brands, you need to be doing ai workshops and like looking at your processes and breaking it down to see how you can. And he said a year ago, people were saying, yeah, let's see a proposal, and then it'd go the ghost in a minute, go quiet. And then now he said he's literally been on stage speaking at something, come off. And he's got three, three clients.

Chris Norton: 5:47

He did something the other week. He didn't even send a proposal someone. They just they just said how much is it? And he said and he said, oh, my boss has signed it off straight away. That that's the, the difference. People have just gone. It's got to be a dot, have you? Have you found this, something similar to that?

Andrew Bruce Smith: 5:59

uh, yes, um, I mean, quite frankly, the last, yeah, the last 12 months I have epped, slept and drank, you know, ai related work, similar to stewart. Um, I mean, I've been to Budapest and Belfast, barcelona, birmingham I mean that's all the bees. Uh, speaking at conferences, delivering training, consulting with PR firms of all types, shapes and sizes, in-house teams, uh, yeah, there is an absolutely insatiable, you know, demand to try and figure out. You know how the hell do we practically make sense of AI and apply it in a kind of practical way? So great, loving it, but I think, interestingly, in spite of all the unbelievable hype, I still think that the actual kind of day-to-day adoption of it is not as widespread as you might might imagine, because the technology is one thing, but getting human beings to kind of, you know, change their habits, um, you have to look at this not just from a tech standpoint but from a, uh, a culture standpoint, you know, uh, implementation standpoint.

Andrew Bruce Smith: 7:05

You know change is hard. Change is hard for, for organizations, and this is this does require, you know, often, uh, radical change. You've got to take into account people's concerns rightful concerns and fears. Um, how often do I go into a pr firm, say, and you're talking to the whole, whole audience and there's a row of account execs at the front and I can tell within 10 minutes they're looking at me thinking is andrew putting us all out of the job? You?

Andrew Bruce Smith: 7:28

know they've said that today, and there you go you see, that's, that's the effect I have, um, and I have to kind of reassure them. No, no, I, I don't believe. You know, senior management are sitting there thinking this is brilliant, we can now just get rid of half the workforce. Um, although some may be tempted to think that's an appropriate thing, um, but again, consider it it. It's inevitably going to change the nature of the work that people do. So it's a bit of a cliche now, but 12 months ago you know Sam Altman's line about AI will replace tasks, not jobs, and everyone's used that one now, but it's still true. I think in that respect he is correct. It will replace certain aspects of people's work, but you can either choose to view that as well great, I don't need as many people or, now that time that's freed up can be redeployed. You know people can use that time that they've got back to do more useful, valuable, creative, strategic things, which I suspect everyone would love to do. That you know, if the AI can do the stuff that, quite frankly, if you didn't have to do it, you wouldn't do it. Well, why wouldn't you want that? But yeah, I think it's interesting. Maybe we shouldn't be surprised that this has all happened so quickly, we shouldn't be surprised that organizations everywhere, I think, are still finding it quite difficult to really roll it out and implement it in a systematic way and understand well what is it good for. And that's part of the challenge.

Andrew Bruce Smith: 9:03

These are omnipurpose technologies. That's a phrase from Mustafa Suleiman, one of the co-founders of DeepMind, great book that he wrote, published last year, the Coming Wave. Recommend it to you. It'll scare the pants off you, but it's required reading. And he used that phrase. You know omnipurpose, which is a blessing and a curse. It means wow, there's literally no limit to things you could use it for.

Andrew Bruce Smith: 9:24

But where do we start? You know we can do anything with it, but we don't know what we should be using it for. Know what we should be using it for. And that gap, that gap between the desire to be able to use it and the absolute crying out demand for people to train people how to use it. Well, I mean, all the management consultants, for example, have done big surveys of their user bases and that's usually right at the top of the list of their big concerns. We've got no one that we can find to actually train our people to use this sensibly, even though the management consultancies, of course, are making money hand over fist by going in and running gazillions of pilot projects for clients. But again, the number of actual proper, valid case studies saying yep, this is a real practical rollout and application of AI Pretty still a bit thin on the ground, if you ask me.

Chris Norton: 10:53

I mean it's interesting you say about management consultants, because obviously the main people listening to the show. There'll be people listening to the show of all levels of marketing, all the way up to CMO, and they'll be like AI. I've heard enough about it. I've had a play around with GPT, I might have had a play around with Gemini, but I don't really see. I don't think, the quality of the content. We've heard it. Today in the workshop we were discussing it when we were earlier people saying the content that's coming out is a bit vanilla and to which we said what did you say when people say the content's a bit generic and vanilla?

Andrew Bruce Smith: 11:28

That's what often you get an argument from people, isn't it? Yeah, I mean, it's the age-old computer acronym of garbage in, garbage out. And, in a similar vein, it's the quality of the prompt or the quality of the brief that you give to the AI that absolutely plays a massive role in what you get out. If you give it a vanilla prompt and a vanilla brief, clever as it is, the AI can't magically read your mind or try and sort of second guess or fill in the gaps for what you want. If you don't know what you want, how can you expect the AI tool to give you something.

Andrew Bruce Smith: 12:02

So it really does kind of flush out. I think a lot of the time it's, it's, uh, it's, it's the actual kind of human being in the process who themselves don't have clarity of what, what they, they think they want, and they somehow expect that the technology to miraculously do it all for them, but it it doesn't.

Chris Norton: 12:19

So there's still work involved so how are you implementing it in? So? Are you doing half day day sessions, two day sessions, do sessions? Do you work with a big? So if you're an in-house marketer, you've got a team of 20, would they get you in to come in and work a workshop to identify processes or something?

Andrew Bruce Smith: 12:36

Yes, I mean, as I say, it's been a relentless 12 months of working with organizations in every sector, pretty much all types, shapes and sizes. I've run for you know huge, multi-billion dollar corporations training hundreds of people on the use of AI, marketing and comms. You know, typically that's running sort of a series of you know seven, eight, nine, one to two hour workshops giving people the kind of the basics all the way down to you know tiny little agencies. Or I did a session kind of at the kind of the basics all the way down to you know tiny little agencies. Or I did a session kind of at the beginning of the year with a bunch of PR freelancers said, oh look, you know, individually we don't think we can afford it. But if we all club together, you know 18 freelancers put us all in a room. Could you say, sure, let's do that. So it's extends from from kind of you know solo operators all the way up to huge organizations. And so it's yes to all the above, half day, full day, do some training.

Andrew Bruce Smith: 13:34

People then say can you come back and help us with the implementation and the rollout of the technology? It almost is a quasi-management, consultancy-style role because there is a technology element, because there are loads of considerations about which tools should we use. And, again, depending on your overall IT policy, if you're a Microsoft shop, maybe you might go co-pilot route. If you're Google, you're clearly going to want to look at Gemini and so forth. There's all the obvious issues surrounding data governance. That's one of the number one questions that comes up, particularly from people in PR and comms and marketing. It's you've shown us all these amazing things, but we can't do most of it because we're not allowed to. Really, yeah, that's still very prevalent where the IT departments are saying well, we still don't feel comfortable about you using this stuff, because what if you know? How do we know that you put in some highly confidential document and it sends it off to whoever OpenAI and you know they might do something with it. We don't like that.

Chris Norton: 14:41

So with that theory, then does that mean that you'd say never put anything that's highly confidential into any of the aforementioned?

Andrew Bruce Smith: 14:50

AI tools. Well see, I mean it's the consultant's answer I mean it depends. I mean this is a fairly standard and obvious rule of thumb that if you're using a free tool you know, using FreeChat, dbt or Claude or whatever, or Claude or whatever, odds are firmly on the fact that you may not be paying any money. But the AI company, on the other end, if you read the fine print of the Ts and Cs, will say you know everything you give us, you know we reserve the right that we might do something with it. You know.

Speaker 1: 15:19

It's like Google, we might yeah absolutely Google and.

Andrew Bruce Smith: 15:22

Facebook You're paying with. You know, ask for forgiveness, not permission. That hasn't gone away. So, understandably, it's like hang on, don't like that idea that you're going to train your model on it or whatever you're going to do with it. So you don't want that. But again, if you want assurance on that, it can be had, but you have to pay for it. So OpenAI offers ChatGPT, enterprise gives you all the assurances on, and it's like people go, oh great, how much does that cost? It's like, well, a lot, you know, that's as I understand it. You're going to be paying $60 per user per month. You have to commit to a minimum on your contract and you have to commit to a minimum of 150 users.

Chris Norton: 16:07

What is amazing me right, this is this is amazing me, like for everybody out there listening to this. They've got like just to have a play around with it. People, every I'm I'm gonna guess that 80 of people listening to this will be saying I've bloody used it, but they might not use the paid one. And people that have used the paid one maybe we're down to 40 of people that have used the paid one but the actual implementation of, like, a microsoft co-pilot for those of you that don't know, because I own blooming prohibition and I've looked at it and I thought, yeah, let's get co-pilot, is it okay? Is it 30 or 60? It's 30 a month, I think yeah, it's a minute, it's.

Andrew Bruce Smith: 16:41

I think it's 360 a year. You have to pay up front, up front for you don't know how good it is. 360, yeah, 360 dollars a year up front per person, per user, on top of the license fee you're already paying for, yeah, your word and your excel, etc. So it, yeah, I don't think any, any organizations anywhere that have made a wholesale commitment. What I mean? Why would you? You want to say, well, hang on, well, we might try out with one or two people. Um, when?

Chris Norton: 17:06

you've got to pay up front. It always makes me suspicious that they're not confident in the product, because if you give a free trial and it's good, then you're going to pay a small amount $30 each per person. $360 a year is a lot.

Andrew Bruce Smith: 17:21

It's my own personal opinion, but anecdotally you talk to others who are using it in anger and I just get this general sense that people are kind of you know, they're disappointed. It is a lot of money and the concept is amazing. Wow, you know, ai embedded into Word and Excel and PowerPoint and laptops now and laptops now and all that malarkey. So the demonstration video looks fantastic, but the reality of does it actually deliver that? Well, no, I mean just an example the AI copilot in Excel. You see, I just assumed naively it would immediately replicate the data analysis capabilities in ChatGPT.

Chris Norton: 18:03

Yeah.

Andrew Bruce Smith: 18:04

It does not. That was very-. First question is why? Well, quite so you're immediately thinking well, hang on, why would I fork out all that? And it's not. I'm paying $20 a month for ChatGPT and it does a whole bunch of stuff and it gives me that quite well, not quite very, I think useful data analysis capability, and I just assumed you'd get that, wouldn't you, yeah, definitely.

Andrew Bruce Smith: 18:29

Anyway. So, yeah, I mean, who knows? I'm sure things will change and I'll get, but right now it's not, I think, surprising to see that I think the number of organizations that are kind of committing wholesale to it I'd be surprised there are that many, because that's a pretty big, big gamble to make, I think, right now. But I mean it's again, it's, it's a cliche, but true of the the, the pace with which the technology is developing. I mean it is literally every day there's something new that comes along and you just go, oh great, I mean it's a bit of a joke in my training workshops that I always have a disclaimer at the start of the session that says, by the end of the session, almost certainly something I said at the start will be out of date by the time we finish.

Andrew Bruce Smith: 19:14

Well, another example I did a session back in February um, actually it was Manchester University did a whole session with the marketing comms team there at Manchester University, 60 odd people, and that was the day. We finished the session at midday and I think at quarter past 12. Um, sora, from open ai was revealed and there was, I hours before saying, well, you know the ai video generation tools like runway ml. Yeah, they're quite impressive, but I didn't feel that they were kind of ready for prime time, as in you know would. Would you ever use that to actually generate marketing? You know video that you would use in anger, and then there you go, 50 minutes later they're sore. It's like hold my beer, oh you know. Yeah, ignore what I said at the start of the day. That completely changes the whole picture.

Chris Norton: 20:02

One thing that when we were going through our workshop today, which was quite fascinating, was nothing to do with work, I just thought was brilliant and was the end bit, was the um, the music. So for those of you who don't know, uh, andrew has a a secret talent where he's quite good on the uh guitar and he's done a few riffs and he's combined his world of work and entertainment together and you've what the tool was. What was it that you were?

Andrew Bruce Smith: 20:28

using this particular tool is called Udiocom.

Chris Norton: 20:32

And it was a beta, wasn't it?

Andrew Bruce Smith: 20:34

Yeah, well, like most of this stuff, it's probably in perpetual beta. But I mean, there's another tool called Suno as well. There's Suno and Udio, but of the two, I'm a big big fan of Udiocom, but I I've um. Of the two, I'm I'm I'm a big big fan of, of, of Udiocom, um, but yes, but the power.

Chris Norton: 20:51

Well, sorry, not sorry to interrupt you, but my point was like, um, what I liked about this? So, if I'm um one of the things about the stuff that we've done here, we're doing a video for a client, um, you might do a really emotive video and then the client we've done TV ads and stuff like that for a client, for instance and then you need a really great. If it's just a, if it's just a web video, you need a really great piece of music. Now, for those of you you know people out there that have bought copyrights of music, they'll know it's an absolute fortune. But Udio, the example you've shown us. You could just create tracks based on your brief that sound like real emotive amazing tracks.

Andrew Bruce Smith: 21:28

Yeah, I mean, I think I mean well, I mean first of all, uh, anyone should really try out udio, because I, I think it it really is quite, quite mind-bending what, what, what it can do. I mean you, literally, as you say, you put in a prompt uh, you know, I'd like a folk song about you know, whatever, um, etc. With these lyrics, and 10 seconds later it comes back with the sort of a 30 second clip and, if you like it, you say, well, make it, turn it into a full-blown song. And within minutes you've got the full, the full kind of thing, um, the, the little nuance that that they added a few weeks ago, which is from a, from a kind of admittedly very amateur musician standpoint. Um, I'm sure there are, there are literally thousands of kind of bedroom guitarists out there.

Andrew Bruce Smith: 22:09

You know you bought the amp and you've been noodling around the guitar for years and you've been using kind of Apple's GarageBand to record your riffs and you think, oh, that's quite a nice little riff, but you know, I can't actually seem it's a lot of effort to actually turn that into a full-blown song.

Andrew Bruce Smith: 22:23

And if I a full-blown song and if I've turned it into a song, you know I'm not, I can't sing really very well then blah, blah. So with udio you can upload your own riff and then it builds a song from the riff that you have supplied. So you put in your guitar riff and then say you know, I'd like a song in this kind of style, and it takes your, your music and then builds a song out of it, and that I'm just obsessed now because I'm like I've got well, like many amateur garage band users, you've probably got years worth of these riffs just sitting there doing nothing, and now you can actually turn them into songs and maybe you could go off and play that in your band and whatever. So I personally see that as a really good creative use of technology to allow you to do things that you couldn't have done previously, and it's not relying entirely on the AI to do everything for you.

Chris Norton: 23:15

Let's listen to said riff. We can stick it on the pod, so stick it in next and the guys can hear what we've been talking about. But why does it feel like there's chains on me? No, it's not delusion. Now, the weight of this world, it keeps us down and we might just struggle till we die While shadows creep, hiding in plain sight. Also, you know that was based on um, your riff.

Chris Norton: 23:59

Then then in the session you said give us a topic. And we, we came up with a topic and obviously, unfortunately, I should have mentioned at the beginning but my co-host isn't here today, william is, isn't? William ockenden isn't feeling very well. So because he wasn't feeling very well, we, you know, did what any supportive company and friends would do and wrote a song about him using and because will is famously known for triple booking his meetings and having a diary that is more shocking than anything you've ever seen. He used to have a paper-based diary and blame everything on that. So so we said let's create a song about William Ockenden and diary chaos. And then you said what style of music? And it was gangster rap.

Andrew Bruce Smith: 24:37

Gangster rap was the consensus, I think, from the audience, so fair enough.

Chris Norton: 24:41

We went with that Gangster rap.

Speaker 1: 25:01

And this is what it came up with Locked in, no set courses, semi-automatic K-dops, raining the schedule. Ain't no magic Got appointments flashing like titans, getting graphic, reminding those buddies still forgets. Yeah, it's kinda classic. Trying to juggle, it's weak. It's more than drastic Dropping dates like they hot, it's so erratic. Listen, listen, yeah, it's all erratic. Hey, will's dilemma's striking hard. His diary's a battlefield, it's all erratic. Hey, will's dilemma's striking hard. His diary's a battlefield. Misses meetings like a sport In clutter. He's concealed Clocks ticking double book Engagements all unreal. He's living in the moment With the FOMO that he feels. Do's mystic with his time Like he's ghosting on the real, scattered pages of his life the struggles, raw as steel, ducking deadlines on the daily, escaping like a seal juggling chaos on the ridge obviously your background, um, has been analytics like you've been involved in google analytics for years.

Chris Norton: 25:55

You, when we we had um, we had you over in 2021. We were just talking about that during lunch and we were doing google analytics to you know, to get that to to embedded in the agency which we did, which was quite interesting Google Data Studio and it's now called Google Looker Studio and everything that we were talking about today. But Google Analytics is shite now, isn't it? Let's not beat around the bush. It's terrible, isn't it?

Andrew Bruce Smith: 26:24

Yeah, well, maybe I wouldn't go quite as far as to say that, but I think there's a number of different sort of factors all kind of coming into play here. I mean without boring the audience about it, because I'm sure they're all painfully familiar with it, but over the last several years we've had this whole thing about the transition from Google Analytics 3 to 4. As we all know, google Analytics 4 is a completely different beast to what we're all used to. So many people have now spent so long trying to get familiar with the new version. It all seems to most people horribly complicated and more complex and so on and so forth, and I know this is a reflection of the wider view. But I used to have a lot of demand for Google Analytics training. It's kind of dropped off. It's almost like AI has sucked all the air out of the room.

Andrew Bruce Smith: 27:15

But equally, I think people are thinking well, hang on, why would I need to go to all the bother of figuring out all this stuff for GA when? Why can't AI do that for me? And why can't Google put that AI stuff into GA and spare me all the pain? Why can't Gemini just plug itself into analytics? Well, I'm sure it could, but in fact you could today do that. You could take the latest version of Google's Gemini model. If you use Google AI Studio, you could stick in the data and say hey, gemini, use your incredible data analysis capabilities to go and look at this data for me. You tell me what I should be looking at, you explain to me what all this means and it will come back with a pretty decent answer. Because, much like the data analysis capability of ChatGPT, there's some serious kind of analytical chops in the background there and that kind of historical barrier and this affects so many people in marketing where everyone's been told for decades probably at least a decade to be more data-driven. That's very easy to say, but it's a lot harder to actually do it. Your background is marketing. There's very few people who have actually had any formal analytics training or statistical analytical training, and you're sitting there looking at all these numbers going yeah, well, I don't know what this is telling me. I've got to go off and find somebody who does. And even then, you know I'm going to have to kind of help them to understand what I'm looking for. So, yeah, it's been a bit of a challenge.

Andrew Bruce Smith: 28:49

Now look at that, give it to the AI and ask it the questions that you aren't answering, or ask it to do those kind of deep analyses. Ask it to explain to you you know what it is doing. I love doing that. It comes back and says oh, it's used some incredible statistical technique. I'm going, well, I've got no idea what that is. Explain it to me. Explain it to me in words that a 10-year-old would understand. And it does, and it's like ah yeah, this is great. So it's reducing the time that you would ordinarily have to spend or the resources and money you would have to spend trying to get that insight. I think we're in a much better place to actually deliver data-driven marketing without having to rely upon resources and approaches that, historically, would have been out of your reach because you didn't have them, or you couldn't afford to get them, or they just weren't available to you.

Chris Norton: 29:42

So what are the big companies doing then? Are they spending massive budgets on AI Because it doesn't seem to be that expensive really? Well, I think they are and they aren't I mean other than Copilot, which, if you've got hundreds of staff, £30 each a year up front it doesn't seem that price prohibitive to get involved. So to try and get it embedded in the adaptation.

Andrew Bruce Smith: 30:05

Well, there's a whole bunch of factors involved here. I mean, yes, yes, on one level you could argue, seriously, it's, it's 20 bucks a month for a paid user license for chat, gpt, and you know, even if it's for 10 people, that doesn't seem, you know, you know overly, you know onerous. Um, you look at the management consultancy firms, all of them saying that that they're they're getting huge demand for kind of AI projects, and I don't deny that they aren't. But the money that's being spent isn't necessarily just on the tech itself. There'll be the training, there'll be the, you know, trying to scope out, well, what should we be kind of using it for and building out those kind of pilot projects. So I think we're still everywhere in that sort of very, very early phase of people just trying to figure out what should we be trying to kind of do with it. And of course, there are still only 24 hours in a day and you know the work doesn't go anywhere still got to be done. I think that's a general challenge that everyone says where do you find, find the time, andrew, today? Well, I'm a sad person who clearly spends their entire lives doing nothing else but this stuff. But you kind of have to somehow figure out how you can carve out some time and space to actually start trying out yourself. Um, trying to figure out well, what are the workflows and processes we we currently do spend a lot, a lot of time on and, quite frankly, if we could reduce the time we spend on this stuff, that would be really helpful Because, quite frankly, it's got to be done. But if a human being didn't have to do it, that would be great Because it would free up their time to devote them to far more useful and interesting and creative and strategic stuff. But it's that kind of paradox. These incredibly powerful technologies have just literally been just unleashed on the world. No guidance, no training, nothing.

Andrew Bruce Smith: 31:54

The fact that there was a sales I think Salesforce did this huge survey of people end of last year. They interviewed like 20,000 people around the world, asked them loads of questions have you tried using an AI tool? Yes, of course, everyone has. Everyone's tried using ChatGPT, generally the free version. But when they were asked, have you received any guidance or training on this? And nearly three quarters of people said no, nothing, zero, zip. We've left literally our own devices. Haha, excuse the pun um to uh, to just play around with it and who knows, we might, by by luck, end up stumbling across something that's useful. But equally, we're futzing around, spending a lot of time literally playing around with it, not getting very far, and thinking well, what's all the fuss about? Um, which is a shame, because when you are pointed in the right direction, patently there are benefits to be had immediately um well yeah it's literally like getting in a car without a sat nav and trying to find your way around again.

Chris Norton: 32:55

Yeah well, we used to do that back in the day, yeah it's the old, I mean it's a worn out joke now.

Andrew Bruce Smith: 33:01

But yes, it is like giving people the keys to the ferrari and say, there you go. No driving test, no, nothing, no, just see, you might, you might work out how to get into the car, you might actually kind of get it to move. You might end up in a ditch. But yeah, without some kind of basic guidance in the first place the, the, the ads the odds, sadly, are that you'll, you'll, you'll spend an awful lot of time, you know, literally spinning the wheels, not getting very far. I think the disappointing thing there is that it then puts people off and they kind of think well, you know what is all the fuss about. It is overhyped. You know, this is pretty meh.

Chris Norton: 33:36

There is a lot of hype, though, isn't there, of course, of?

Andrew Bruce Smith: 33:38

course there is, I mean gee it's. I mean it's incredible when you think it's still so sustained, the, the interest and the hype and the talk.

Chris Norton: 33:49

It's just like social media. This is what social media was like, but it's but it's on steroids, it's way way.

Andrew Bruce Smith: 33:54

It's gone way beyond the hype around around social. Uh, I think what's disappointing and maybe you feel the same is that you do. You do get to a point where it's like, yeah, you're hearing the same things time and again.

Chris Norton: 34:07

But the technology is getting better. So, for all those people out there that have used it, like I said, and you've thought, oh, I'm getting vanilla responses. Now, like you've said, if you've got some direction on how to use it, you're going to get a much, much better response.

Andrew Bruce Smith: 34:20

A concrete example the vast majority of people up until a few weeks ago would have used the free version of chat, gpt uh, which used an older model. It was using version 3.5, um, which, yeah, pretty good for its time, but it's nowhere near as good as gpt4 or claude or gemini, of course. Now uh open, ai has made gpt4o. Uh stands for omni. They're hopeless names for their products, by the way.

Chris Norton: 34:48

Yeah, it's not a great name, it's terrible, it's no co-pilot.

Andrew Bruce Smith: 34:51

Is it Awful Anyway, but GPT-4O, there you are, for free. You can now use the best model available. I just wonder. People who've tried the free version previously went meh, thinking well, isn't that just going to be more of the same? No, it isn't. I didn't know if it was.

Chris Norton: 35:07

I didn't know if they'd released this because in in combat of everything that's going on, in the fact that I'd heard, I think I've seen statistics on the fact that um usage of it, it was going down, so people were accessing, like you've just said there, first time, having a play around with it, like everybody who's listening probably has, and then just thinking, oh, it's not quite good enough for me and I don't really need this. And you know, I've seen it's not going to, it's not going to nuke us, it's not going to take my job, and that's kind of how people, yeah, whereas the rest of us, some of us, are using it all the time.

Andrew Bruce Smith: 35:39

So well, I think it's, it's the, it's the old, the old joke. Isn't it about the? You know the frog being slowly boiled? It doesn't realize that the water keeps getting warmer and thinks, ah, it's all right, until it suddenly realizes, oh hell, you know I'm dead because I've been boiled. Boy lie, but it just sort of, yeah, it creeps up without you even realizing it.

Andrew Bruce Smith: 35:57

You're already seeing the impact in various areas. Well, we were talking about earlier, weren't we in the workshop, about the work of translators. I mean, anecdotally, people doing translation work these days. Their job is now simply to check the output of AI-generated translation concepts. So they're still doing work, but it's of a much kind of lower level and they're paid an awful lot less than they were previously, when they were actually doing the translation work. So that's happened. There's an example, example. I forget the name of the company, but the, the company that basically, um, you know, got rid of hundreds of customer support people replaced them with ai and then rather rather triumphantly said, hey, look at the. That that's added 40 million dollars to the bottom line. It's best thing we ever did. Getting rid of all those human beings hasn't gone down to I think quite rightly getting a lot of stick for that, because that's not really, I think, a winning message to put out.

Chris Norton: 36:53

But yeah, ai is already having an effect on various Customer service definitely, yeah, blimey, definitely, because the responses it does can be better than the, and if it's, if they're carefully constructed and structured, the responses can be better.

Andrew Bruce Smith: 37:10

You've got those kinds of you know use case where, yeah, it's already, I think, sort of well well proven that sure, yeah, it can absolutely replace human beings to do a lot of this stuff. But, yes, but, but. But in the world of marketing I think you do have this kind of interesting sort of distribution curve of some people clearly have just taken it like Dr Water and gone this is the best thing since sliced bread. But again, they tend to be kind of in sort of small pockets and silos and I think the challenge everywhere is how do you kind of take that interest and enthusiasm and sort of spread it across the whole organization or organization of the whole?

Andrew Bruce Smith: 37:49

team difficult, that's a difficult bit, that's um you know maybe we should ask chump gbt to tell us. I'm sure it would come up with something. Uh, just ask it to be more, uh, more, more circumspect, um it's more punchy, more punchy, yeah that's a good.

Chris Norton: 38:05

That's a good system, prompt to give it um, okay, um, and well, one of the things that I I thought was interesting is the is apple and the fact that obviously apple has, I mean, all these companies. Google got caught with its pants down at the beginning, and then gemini's been, has been, repopulated, and what? What edition are we on on gemini now?

Andrew Bruce Smith: 38:27

it's just yeah, 1.5 pro is. That is the is the current uh version um which has got a much bigger context window, right, oh yes, uh, well, if you're, if you're lucky enough to to get on the beta you, you would now have access to it to a 2 million token context window. Um, I just to clarify what we're going on about there. It's kind of the short term memory of the model you're using. So 2 million tokens, that's roughly 1.75 million words of content.

Chris Norton: 38:58

I wonder how long the Bible is.

Andrew Bruce Smith: 39:00

Oh, shorter than that, you probably could put the.

Chris Norton: 39:02

Bible in. You could write a new chapter.

Andrew Bruce Smith: 39:04

Yeah, you could put the entire well put in the bible, you know, put in well, put in every every harry potter novel. Ask it to write a new one and it would do something. I'm not going to say it's going to generate anything brilliant, but but purely from the perspective of being able to to give it all of that content and it's actually kind of able to pay attention to all of that. And you're working with that. If Disney are, listening.

Chris Norton: 39:29

What they should do is get all their Star Wars scripts and punch them in and see what it comes out with, because some of the generic crap that they keep producing compared to the good old stuff.

Andrew Bruce Smith: 39:39

Well, I strongly suspect, or probably, or what I think. When you think of those things, you can guarantee somebody's already done it somewhere in the world.

Speaker 1: 39:45

Oh, yeah, they will.

Andrew Bruce Smith: 39:48

But anyway, but yes, to create scripts and books.

Chris Norton: 39:50

Yeah, Amazing that it can do. It could do that in that style. Do you know what I mean? And so the bigger the context window, the more accurate and relevant to what you're trying to get.

Andrew Bruce Smith: 40:01

It's again, it's one of the nuance and again, without sort of getting too technical, there are two types of memory really with with, uh, with a large language model, the so-called latent space, which is effectively all the possibilities of what it might be able to extract, you know, from its model, versus the context window, which is effectively its short-term memory, as in your, your, you give it a prompt and in that particular session, how much can it kind of, you know, pay attention to in that uh, in that particular session, how much can it kind of pay attention to in that individual session? So when ChatGPT first launched out of context, we were thinking about 3,000 tokens. It's too tiny.

Chris Norton: 40:36

Now you look back at it and think 3,000, that's like 2,000 words, 2,000 words, yeah, hardly anything, and now it can do an entire book.

Andrew Bruce Smith: 40:41

Oh, yeah, absolutely Crazy. Yeah, and that of course includes both the input you give it and the responses you get back from it. So which?

Chris Norton: 40:49

makes you think what's it going to be like when you can put the entire world, all words ever created, yeah through it and then it's, then it's going to run out of content, yeah I mean, there's been, you know, sort of research papers talking about 10 million, 20 million, 30 million token context windows.

Andrew Bruce Smith: 41:03

It all starts to get a bit, a bit blurry really. And then some people would argue well, you know, even though you can put it all in there, is it really pay whatever? From a practical standpoint, particularly in the world of marketing, you can already see some fairly obvious you know use cases for those kind of larger context windows. I mean, think about the basics of, you know, creating a marketing plan. I mean, how would you normally go about doing that? Well, typically you're going to have a brief where you're going to define some objectives. You're almost certainly going to do some kind of research and finding information about the market and all those kinds of trends, whatever.

Andrew Bruce Smith: 41:39

Blah, blah, blah. And then what do you do with all that stuff? Do you give it to human beings who kind of go off and they try and read it all and assimilate it and then come back. You know, we, we have our own neural networks, we have our own context windows and, bless us, our context windows are nowhere near as big as a gemini um yourself. Well, there you go. I do apologize, of course I've read more than one book.

Andrew Bruce Smith: 42:01

There you go, haha. Can you recite it word for word and remember every aspect?

Chris Norton: 42:05

of it I can't remember.

Andrew Bruce Smith: 42:06

remember what I did yesterday. There you go. Anyway, it's a classic example of these AI tools rapidly gaining those kinds of capabilities that we don't have, but at the same time, there's loads of stuff that we're going to continue to be much better at than they are, certainly for the foreseeable future. But they're again, again, giving all that information to Gemini and saying right, you take all that, you take on board the role of of an expert at figuring all this stuff out and come back with a first draft approach to how we might might tackle it. I'm not pretending it will come back with the perfect answer on the first go, but you're going to get something fit for purpose a lot more quickly and it's that combination of you and the AI that I think really is the sweet spot to aim for. And again, the more people understand which tools are good for certain things and certain approaches. What are the limitations? What do you need to be aware of?

Andrew Bruce Smith: 43:11

Again, it favors people who actually do know what they're doing, and there's a lot of fear from people thinking well, you know, that means anybody can just tip up and push a button and do the work of an expert and it's like well, not quite. I mean, yes, they can just walk up to it and push a button and they'll get something out of it. Is it actually going to be kind of any good or fit for purpose? Would the non-expert ever recognize? Oh, hang on, it's actually made some mistakes there. No, they won't.

Chris Norton: 43:38

Well, you say that I've got some interesting anecdotal stuff that I want to run past you that we haven't actually discussed even off air, and that was basically. I know a particular student who has just come out of university and they were telling me that numerous people on the final degree, on their final year of degree, were getting caught out for using AI in dissertations, which I was quite amazed at because I didn't think the universities would be fully up to speed with how. Because it's and I get that it can be generic sometimes, but if you're going to use it on a dissertation like that's A, that's naughty and B, you better make sure it's damn good and accurate.

Andrew Bruce Smith: 44:18

Well, I think. Well, I mean, I too have got an anecdote along those lines. I was giving a presentation to a university down on the south coast and there was an academic there who said, yes, they had. It was a PhD student who had submitted their kind of thesis. They suspected that they'd been using AI but hadn't quite been able to prove it. But they got the final draft back and there were all these references. They started trying to check out the references, going, oh, hang on, they do not exist, they've all been made up.

Andrew Bruce Smith: 44:51

Really, that was just a very lazy use of of ai. It's a bit like the lawyer in america who last year famously went to chat. Well, they went to google first to try and find some sort of uh cases to to back up his argument. Didn't find anything on google, he thought, oh, I'll try chat gpt, and she came back. So there you go, here's his 10. He's like, oh, brilliant, and he put those in his submission. Of course it made it all up, it hallucinated it the whole damn thing. So he was fine, was it 50? Not check the links? No, of course not.

Chris Norton: 45:21

How does the paid lawyer who's on several hundred pounds an hour? I?

Andrew Bruce Smith: 45:24

think that's kind of what I'm driving at. Is that that if you're, if, basically, if you're, if you're basically lazy, uh, if you use it in a lazy way like that, well, yeah, it's unsurprising that that you will get. You will get kind of, quite rightly, you'll get found out, I think, the more. The more subtle thing is that you know patently, certainly in in education, students everywhere, we all know they're using ai in various shapes and forms, and you know they're not, they're not stupid, their their ability to use it, so that actually it becomes very hard for, um, you know, a human being to kind of figure out whether they did or they didn't, uh, use ai.

Andrew Bruce Smith: 45:59

And the counterpoint, of course, is that you're getting also cases of false positives, people being accused of using ai when in fact they didn't yeah, I mean there's an argument that actually this kind of penalizes non-native English speakers because most of the AI tools are largely trained on English language content and non-native English speakers. It's not because they've been using AI, but it's just that they don't necessarily have the same kind of grasp of the language and it's very unfair that they're't necessarily have the same kind of grasp of the language and it's very unfair that they're accused of using AI when in fact they're perfectly, perfectly innocent.

Chris Norton: 46:36

You would think that language would not be relevant anymore in terms of you know, the thing doesn't have to be in English, it could be. You know, I gave the example earlier about Joe Rogan talking about the fact that his podcast is going to be available in 65 different languages, even though he can't speak any of them other than English. Sorry, if you can speak another language, Joe, I'm not sure, but anyway there's 65 different languages. It's going to be produced in that. So language pretty much is going to become irrelevant because the AIs can transfer. You know, they'll be able to turn your dissertation on how to work with PR, yeah, but I think in a similar vein.

Andrew Bruce Smith: 47:12

It's the same for, well, not just marketing and communications, but for most industries. It will force everyone to figure out well, what are the skills and the tasks that will remain the province of the human being, the skills and the tasks that that will remain the province of the human being, what are the things that we are going to basically just let the ai get on with? Or, I think, more likely, that that combination of human and ai working together to, to, to generate the, the, the outputs, uh, required. Um, and again, you, you could argue that well, the, the number of tasks that the ai, the ai can do, seems to be rapidly growing by the day. Um, you know what? Is it just going to do it all?

Chris Norton: 47:57

I hope so, well save us a bit of time, we can all go play racket sports the optimistic view clearly is is that you, you let it get on and do the stuff.

Andrew Bruce Smith: 48:05

That, quite frankly, if you didn't have to do it, you, you, you wouldn't. Um, I mean, the classic of the last 12 months has been the rise of the AI meeting assistant. I'm sure everyone not everyone, well, probably everyone's tried them out, and maybe people are. I certainly can't work without them now. No, I can't. You do have an online meeting. You've got your AI assistant. They take a faithful transcription and summarization of what you talked about. All those of what you talked about, all those minutes and hours you would have previously spent writing up those meeting notes. That's time spent doing other stuff.

Chris Norton: 48:33

It's far more interesting and useful, yeah and who used to read the meeting notes as well. Yeah, I remember getting trained how to do contact reports.

Andrew Bruce Smith: 48:42

Yeah, it was literally the most boring thing I've done in public relations and marketing. Well, I've never met anyone who got into marketing or PR. That said, I wish I could write more contact reports. That's my number one aim is to write more.

Chris Norton: 48:54

We're going to give that one to the AI. They can have that forever. Exactly, let them have that one.

Andrew Bruce Smith: 48:59

But that time that's being saved is now freed up to do other stuff. Book your holiday, whatever. That's your choice.

Chris Norton: 49:12

It whatever that's, that's your choice. It's your choice. So. So 12 months has passed, and if where we're going to be in 12 months time from now, then I mean where we're going to be in 12 days time let's be honest, if, if I, if I really did know that I wouldn't be here, chris, would I be making a fortune?

Andrew Bruce Smith: 49:21

if I had that if I had that crystal ball? Um, yeah, I mean it's it without sort of again descending into the same cliched responses that everybody else seems to trot out on this stuff. Nobody seems to know, nobody really truly knows what is going to happen. Having said that, it's fairly safe to say well, is the technology going to suddenly have a bit of a pause? No, it's not. It's going to continue, probably, to develop at the same breakneck pace as it has the last 12 to 18 months. What's going to stop the tech companies from doing that? Well, nobody.

Chris Norton: 49:58

And what are the ethics behind it? That's the really worrying thing.

Andrew Bruce Smith: 50:01

Yeah, I mean, we've definitely. Yeah, Well, some might argue that the whole ask for forgiveness, not permission, ethos of Silicon Valley never really went away. Yeah, and we're certainly seeing that in respect of all the issues we're seeing around copyright. Yes, you've got the New York Times is suing OpenAI saying hey, you took 40 million of our articles and you used to train your model and you didn't ask for permission. You've never paid us a bean. We're very unhappy about that. On the other hand, lots of other media companies are saying fine, we'll accept the price you're going to give us. The FT has said sure, we'll take the money. Here you go, here's all of our content. We're happy for you to use that to train your model, et cetera, and so forth.

Chris Norton: 50:54

What's bizarre there is like we were talking about with the music. You know, the will song that we played earlier is the copyright side of things. It's for inspiration, really, because it is original stuff. So, yeah, despite the fact that it's read 14 000 pages of the you know washington post or the new york, whatever it is, yeah, you could, andrew bruce smith or chris norton could have read all of those websites. So are we infringing their copyright by reading them?

Andrew Bruce Smith: 51:14

Well, it's the argument that's certainly been used by the AI companies in America. I should quickly stress I am not a lawyer. However, my understanding is that certainly you take the US as the example that the concept of fair use is the argument usually adopted by the AI firm, saying, yeah, the only difference is the scale. What's the difference between a human being who has listened to thousands of tracks, is inspired by listening to all of that and then writes their own stuff, versus it's a machine that's listened to all that stuff and you're asking it to generate a new song in a particular style or genre and it gives you that. And this is the thing that people have a problem with. Copyright law has been around since whatever it is 1725. And you could argue that really it hasn't really evolved to reflect modern day technological reality.

Chris Norton: 52:07

Modern day. It's only just been made legal, hasn't it, that you can record the TV legally it was illegal for like 25 years. You had VHS tape and Betamax. You couldn't record, you weren't allowed legally to record the TV.

Andrew Bruce Smith: 52:21

I think people, I think, understandably struggle with this idea. There's a clue in the name. It's copyright, but there is no copy of the original works used to train the model in the model, the model itself. Um. Andre kapathi, um, you're well known ai researcher. I like his description of large language models being a kind of gigantic zip file. Um, of all the content it's it's been used on, but it's unlike a normal zip file which is lossless, as in. You unzip it and you get the original back completely. With an AI model, there's lots missing and it's extracted the essence of what it was trained on. But I think some people think, oh, it's copied. There's a copy of it all sitting there in the AI model and it's not. It's not there. It's not there. It's an essence and it is then utilized to then generate something new.

Andrew Bruce Smith: 53:15

The whole issue of, well, how do these AI companies appropriate the content that they use to train them? Well, that's an entirely different issue altogether. Mostly well, in fact, I think all the AI tools, uh, use web content. They generally use web content. That was, uh, uh, found by a company called Common Crawl, which nobody's ever heard of, tiny little outfit in San Francisco, set up years ago as an archiving project, and for most of its existence who cared? Nobody gave a monkeys about what it was up to, but it was quietly indexing all this content, web content, and if you felt like it, anyone could download it and do what you want with it, and the AI companies go brilliant. All this web content sitting there for nothing. We'll just have that. Thank you very much. Didn't bother to ask for See. New York Times, strictly speaking, should be suing Common Crawl. They're the ones that crawled the New York Times content originally. True, true, but why sue a tiny little non-for-profit that?

Speaker 1: 54:12

employs like two people in South, yeah, exactly.

Andrew Bruce Smith: 54:16

Well, and to be cynical about it, of course the New York Times was negotiating with. I understand it was negotiating for payment, they just didn't like the price.

Chris Norton: 54:25

Oh right, okay, so they're like it's not high enough, oh right okay.

Andrew Bruce Smith: 54:26

So they're like it's not high enough, so we'll sue you instead. Okay, and maybe that will bring you back to the table and maybe you'll offer us more money. So we'll see.

Chris Norton: 54:34

Well, thanks for coming back on the show, Andrew.

Andrew Bruce Smith: 54:36

My pleasure. Lovely to have you back here again, especially in lovely Leeds. Yes, it is, it's delightful.

Chris Norton: 54:41

You keep saying that.

Speaker 1: 54:42

It is.

Chris Norton: 54:44

The weather's.

Andrew Bruce Smith: 54:45

Genuinely, the sun is shining bright, there is blue sky, it's.

Chris Norton: 54:49

Leeds. How can people get hold of you if they want to pull you into an AI webinar or get you along at a conference?

Andrew Bruce Smith: 54:54

Yeah, I mean, I'm on LinkedIn. If you search for Andrew Bruce Smith, there is only one Andrew Bruce Smith on LinkedIn. Original. Yeah, all the usual suspect channels eshmancom websites.

Chris Norton: 55:14

I'm still on OnX as well as andy smith andi smit. Yeah, I never understood why it was andy's andy's of smith, but then you're eshman and then you're andrew bruce smith I can.

Andrew Bruce Smith: 55:19

I can give you the very short version. Um, the very, very first uh online service I signed up for was the, the compulink information exchange, which I joined in 1990, which I used to be the second person in Britain to send a press release by email using CompuLink in 1991. You had to give yourself a username. I originally thought I was typing in and it was eight characters. It had to be an eight character limit on your username. I thought I typed in A-N-D-Y, andy Smith, fat finger syndrome, I'd hit the, I hit return and you couldn't change it. Oh no, so that was my original online handle, was Andy Smith with an I, and I thought, well, I'll just stick with it. And I've stuck with it ever since.

Chris Norton: 56:02

I can't believe you waited right to the end of the show to leave your fucking purchase right there. Well, that's the usual place for it, isn't it right? At the end, great punctuation.

Andrew Bruce Smith: 56:10

Yeah, there you go. That was a classic yes sort of user error.

Chris Norton: 56:13

It was literal fat finger hit I instead of Y and that's stuck with me for god 30 odd years, we've all done that where you get fixed and stuck inside a URL that you can't change. Yeah, alright. Well, yeah, thanks for coming along. Absolute pleasure.