1 00:00:00,000 --> 00:00:09,960 Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the 2 00:00:09,960 --> 00:00:15,700 latest trends, tips and tools in the world of marketing AI, helping you get the best 3 00:00:15,700 --> 00:00:18,520 results from your marketing efforts. 4 00:00:18,520 --> 00:00:22,480 Now let's join our hosts, Paul Avery and Martin Broadhurst. 5 00:00:22,480 --> 00:00:27,920 Hello, and welcome to episode 30 of Artificially Intelligent Marketing. 6 00:00:27,920 --> 00:00:30,480 Crumbs, how did we get to episode 30? 7 00:00:30,480 --> 00:00:31,480 That's very exciting. 8 00:00:31,480 --> 00:00:35,160 Martin's here with me, Martin Broadhurst, my co-host. 9 00:00:35,160 --> 00:00:36,160 How are you Martin? 10 00:00:36,160 --> 00:00:37,160 What have you been up to this week? 11 00:00:37,160 --> 00:00:38,160 I'm very good. 12 00:00:38,160 --> 00:00:46,720 And I have been having a very busy week, working with lots of businesses across my patch. 13 00:00:46,720 --> 00:00:54,240 In fact, I've delivered two workshops about getting started with generative AI this week. 14 00:00:54,240 --> 00:00:59,520 Across the two workshops, we had 20 businesses in those sessions. 15 00:00:59,520 --> 00:01:01,480 In fact, I've got two sessions coming up. 16 00:01:01,480 --> 00:01:07,160 I've got one tomorrow in workshop and another one in Leicester. 17 00:01:07,160 --> 00:01:09,520 So it's a hot topic at the moment. 18 00:01:09,520 --> 00:01:13,120 Both of those upcoming ones are sold out. 19 00:01:13,120 --> 00:01:20,280 So yeah, it's fair to say that this is a high demand area at the moment, which we are both 20 00:01:20,280 --> 00:01:21,280 well aware of. 21 00:01:21,280 --> 00:01:23,160 You've been a busy boy then. 22 00:01:23,160 --> 00:01:27,960 It's good to hear that you've been getting in about and preaching the benefits of generative 23 00:01:27,960 --> 00:01:28,960 AI. 24 00:01:28,960 --> 00:01:31,040 I had a kind of similar experience myself this week. 25 00:01:31,040 --> 00:01:37,080 I was at the Elrig Drug Discovery Conference up in Liverpool, which was fab, extremely 26 00:01:37,080 --> 00:01:38,960 well attended, which was great to see. 27 00:01:38,960 --> 00:01:42,760 While I was there, I actually gave a vendor workshop for all the vendors that were at 28 00:01:42,760 --> 00:01:48,240 the event on generative AI and the power of the tools that we talk about week in, week 29 00:01:48,240 --> 00:01:53,640 out on the podcast line in terms of how they can be more effective, more efficient. 30 00:01:53,640 --> 00:01:54,640 That was really great. 31 00:01:54,640 --> 00:01:55,640 It was interesting. 32 00:01:55,640 --> 00:02:00,360 I did a lot of the polls that we often do in the workshops that we run and I noticed 33 00:02:00,360 --> 00:02:06,000 that a lot of people had tried to chat GPT and there were even a couple of power users, 34 00:02:06,000 --> 00:02:11,080 but much fewer had tried image generators, interestingly. 35 00:02:11,080 --> 00:02:15,680 So given that the audience was predominantly marketing and sales folk, that's kind of a 36 00:02:15,680 --> 00:02:20,080 bit surprising to me, but I think it's another good example of when you spend as much time 37 00:02:20,080 --> 00:02:26,640 nerding about AI as we do, Martin, you forget that a lot of folks have perhaps not been 38 00:02:26,640 --> 00:02:29,600 exposed to some of this stuff. 39 00:02:29,600 --> 00:02:34,360 Especially I showed them the Hey Jen example that we've talked about on the podcast a few 40 00:02:34,360 --> 00:02:40,200 times and that we've routinely used at our workshops with the, for those of you that 41 00:02:40,200 --> 00:02:44,840 haven't seen it, in essence, someone records a short video on their phone. 42 00:02:44,840 --> 00:02:51,360 30 seconds in English and then Hey Jen automatically dubs it and lip syncs it in different languages 43 00:02:51,360 --> 00:02:56,960 and you really have to watch it to see what we mean because it at first glance looks like 44 00:02:56,960 --> 00:03:00,120 it was originally recorded in those languages. 45 00:03:00,120 --> 00:03:03,840 The audio and the lip sync is so good. 46 00:03:03,840 --> 00:03:09,400 I also found it interesting that I'm definitely starting to see some segmentation occurring 47 00:03:09,400 --> 00:03:12,360 in the people who are working in marketing. 48 00:03:12,360 --> 00:03:16,720 There's definitely people who are still, like they know about ChatGPT but maybe that's about 49 00:03:16,720 --> 00:03:19,320 it and they're really interested to learn about other tools but they haven't really 50 00:03:19,320 --> 00:03:20,320 played with them yet. 51 00:03:20,320 --> 00:03:27,640 I think that's quite a large section of people in our marketing sphere that we play in. 52 00:03:27,640 --> 00:03:33,900 There are, however, a few like ninja power users who are starting to get really to grips 53 00:03:33,900 --> 00:03:39,440 with all of the foundational tools across image creation, music creation, voice creation, 54 00:03:39,440 --> 00:03:43,700 ChatGPT for text and all that other stuff and they're starting to do more power use 55 00:03:43,700 --> 00:03:49,800 cases like connecting to Zapier, running multiple workflows, trying to use different data sources 56 00:03:49,800 --> 00:03:54,960 that they push into a table or Google Sheets that then they use as prompts in ChatGPT to 57 00:03:54,960 --> 00:04:02,240 automate me create prospecting emails that are automated and customized on a use case 58 00:04:02,240 --> 00:04:06,640 by use case basis for each individual person that they want to email. 59 00:04:06,640 --> 00:04:09,920 So definitely seeing some of that segmentation, you see much of that in the workshops you're 60 00:04:09,920 --> 00:04:10,920 going to? 61 00:04:10,920 --> 00:04:12,920 Yeah, very much so. 62 00:04:12,920 --> 00:04:18,520 What I'm finding really interesting is that this technology is so different in terms of 63 00:04:18,520 --> 00:04:21,760 the adoption from people, the enthusiasm to adopt. 64 00:04:21,760 --> 00:04:27,240 I've been in lots of workshops in the past and sessions where, let's say the topic is 65 00:04:27,240 --> 00:04:31,080 search engine optimization and business owners and marketers go, right, I'm here to learn 66 00:04:31,080 --> 00:04:32,080 all about this thing. 67 00:04:32,080 --> 00:04:36,280 I know it's really important to get found in search and then they do the workshop and 68 00:04:36,280 --> 00:04:38,880 at the end of it they come away going, I'm still a bit frazzled. 69 00:04:38,880 --> 00:04:42,960 I understand its importance, but I still don't really know how to get it started because 70 00:04:42,960 --> 00:04:50,440 something like SEO might have the kind of dark arts element to it and a bit of mystery 71 00:04:50,440 --> 00:04:53,160 for people that don't do it every day at least. 72 00:04:53,160 --> 00:04:57,400 If you're immersed in that world, you understand all of it, but it can be quite jargon heavy. 73 00:04:57,400 --> 00:05:01,120 With this, it's completely different because very quickly people can see value. 74 00:05:01,120 --> 00:05:05,020 That time to value for the user is very quick. 75 00:05:05,020 --> 00:05:12,360 You load up an interface like ChatGPT and kind of show people, certainly in the workshops 76 00:05:12,360 --> 00:05:18,920 that I do, I might take somebody's homepage and stick in a prompt and get it to critique 77 00:05:18,920 --> 00:05:27,240 their homepage copy from a particular perspective or how might they improve their copy to appeal 78 00:05:27,240 --> 00:05:30,560 to a particular audience. 79 00:05:30,560 --> 00:05:36,680 Very quickly people go, oh my God, that's a time saver, I'm going to take that away 80 00:05:36,680 --> 00:05:37,680 with me. 81 00:05:37,680 --> 00:05:42,320 In fact, one of the workshops I did this week ran a series of prompts and one of the delegates 82 00:05:42,320 --> 00:05:46,760 said please send me that, email me that content, I want to see it immediately. 83 00:05:46,760 --> 00:05:51,920 The enthusiasm for people to get stuck in is so different to other technologies that 84 00:05:51,920 --> 00:05:54,960 we might have seen in this kind of digital marketing space. 85 00:05:54,960 --> 00:05:59,600 Yeah, I think that's definitely what happens. 86 00:05:59,600 --> 00:06:05,400 People see these different use cases and then they think, oh, I can do that if they're showing 87 00:06:05,400 --> 00:06:10,240 them and also once you start playing with the tool and you let your creativity run a 88 00:06:10,240 --> 00:06:13,640 bit wild, I think you can really find ways to use it. 89 00:06:13,640 --> 00:06:17,120 Obviously, a lot of people are getting started with ChatGPT but the other question that's 90 00:06:17,120 --> 00:06:23,520 coming up for me again and again is outside of ChatGPT, what tools should I be using? 91 00:06:23,520 --> 00:06:28,160 I know we've had a bit of a talk about this off air, Martin, so dear listener, if you 92 00:06:28,160 --> 00:06:33,500 are also asking yourself that question, we're going to do a little segment at the beginning 93 00:06:33,500 --> 00:06:37,720 here just to take you through some of our favourite tools. 94 00:06:37,720 --> 00:06:44,040 For me personally, I am using Magi a lot, which is, how can I describe it? 95 00:06:44,040 --> 00:06:49,980 It's basically a front end on ChatGPT but you can also, so it's like a chat bot basically 96 00:06:49,980 --> 00:06:55,320 that you would use the same way you use ChatGPT but you can access multiple models, so you 97 00:06:55,320 --> 00:06:59,420 can access Claw 2 that we've talked about a lot on the show that has its large context 98 00:06:59,420 --> 00:07:02,880 window and makes it perfect for summarising stuff. 99 00:07:02,880 --> 00:07:08,680 You can also upload files like PDFs so you can ask them the tool to summarise them or 100 00:07:08,680 --> 00:07:12,880 you can in essence chat with the PDF and ask questions of the document which is especially 101 00:07:12,880 --> 00:07:15,360 useful if the document's very long. 102 00:07:15,360 --> 00:07:21,160 Because Magi works with the API version of all these tools and most major tool users 103 00:07:21,160 --> 00:07:25,880 certainly OpenAI have stated that they won't use any information that comes into their 104 00:07:25,880 --> 00:07:31,120 systems via the API, you can have that much more confidence that any materials that you 105 00:07:31,120 --> 00:07:37,640 share within those tools are, the information is not shared and used to chain the model 106 00:07:37,640 --> 00:07:42,020 which I know is a big barrier for a lot of people wanting to use the tools. 107 00:07:42,020 --> 00:07:48,120 So using Magi as my ChatGPT clawed everything in one place replacement, it's also got custom 108 00:07:48,120 --> 00:07:54,040 instructions like ChatGPT but you can in essence give it a whole suite of custom instructions 109 00:07:54,040 --> 00:07:57,720 that are called personas and you can choose the persona that's the most relevant for whatever 110 00:07:57,720 --> 00:07:58,720 you're doing. 111 00:07:58,720 --> 00:08:03,480 So I've got master summariser, expert copywriter, I've got marketing strategist, I've even got 112 00:08:03,480 --> 00:08:07,440 one persona that's set up to help us write the scripts that we create for the podcast 113 00:08:07,440 --> 00:08:11,200 so it just all I do is drop a story in and then it gives me a script to play with in 114 00:08:11,200 --> 00:08:12,200 it. 115 00:08:12,200 --> 00:08:15,160 So very good worth looking at. 116 00:08:15,160 --> 00:08:21,440 Another tool I use a lot is Audio Pen, so this allows you to basically dictate emails 117 00:08:21,440 --> 00:08:27,360 and other things that you would ideally usually write down but for whatever reason you think 118 00:08:27,360 --> 00:08:32,120 faster when you're speaking or if you're out and about like when I'm walking the dog or 119 00:08:32,120 --> 00:08:36,200 playing with the dog in the garden, I'll just speak to Audio Pen on my phone. 120 00:08:36,200 --> 00:08:40,520 The difference between this and using sort of any other like transcription tool that 121 00:08:40,520 --> 00:08:46,160 I've used in the past is that it uses the power of ChatGPT so if I ramble a little bit 122 00:08:46,160 --> 00:08:51,880 like I am now with what I'm trying to say or the note I'm trying to leave myself, it 123 00:08:51,880 --> 00:08:58,400 pushes the transcript that it captures through ChatGPT and then is able to format it in proper 124 00:08:58,400 --> 00:09:03,520 text and again you can have different personas to make that fit your email style or to create 125 00:09:03,520 --> 00:09:07,300 bullet point lists whatever you might do. 126 00:09:07,300 --> 00:09:10,920 My favourite image tool at the moment is ClipDrop. 127 00:09:10,920 --> 00:09:13,840 It's got image generation capabilities in it. 128 00:09:13,840 --> 00:09:18,440 ClipDrop uses stable diffusion, stability AIs models. 129 00:09:18,440 --> 00:09:19,800 The image generation is pretty good. 130 00:09:19,800 --> 00:09:23,920 It's not as good as Mid Journey and probably not as good as Dauley 3 to be honest but the 131 00:09:23,920 --> 00:09:26,480 reason I love ClipDrop is it's got a whole suite of tools. 132 00:09:26,480 --> 00:09:34,040 It's got background remover, spot eraser, relighting, sky replacer, upscaler, so many 133 00:09:34,040 --> 00:09:39,360 tools, really great price, love ClipDrop and its ability to remove backgrounds is insane. 134 00:09:39,360 --> 00:09:42,680 I keep throwing more and more complicated images at it thinking I've killed it this 135 00:09:42,680 --> 00:09:45,440 time and it strips the background out. 136 00:09:45,440 --> 00:09:51,080 Favourite use case so far for that one Martin is I'm always in the need of logos and I 137 00:09:51,080 --> 00:09:55,040 can never find the one that's got the transparent background and so now I just do a quick screen 138 00:09:55,040 --> 00:09:59,880 grab off a website, chuck it into the background remover and it comes out perfect with a transparent 139 00:09:59,880 --> 00:10:03,600 PNG background which is just awesome. 140 00:10:03,600 --> 00:10:09,400 I'm also loving finding new ways to use Dauley 3's chat GPT vision capabilities so both for 141 00:10:09,400 --> 00:10:15,880 image creation in Dauley 3 but as an example for my presentation at Elric this week I wanted 142 00:10:15,880 --> 00:10:22,560 a table that compared all of the different models you could use right like Bard or Bing 143 00:10:22,560 --> 00:10:25,280 or chat GPT because I didn't have one. 144 00:10:25,280 --> 00:10:27,040 I don't know why I didn't have one but I did. 145 00:10:27,040 --> 00:10:31,960 So I found a really good one online that had some stuff that I wanted but I needed to change 146 00:10:31,960 --> 00:10:35,040 it because I felt it wasn't up to date and some of the stuff I didn't agree with but 147 00:10:35,040 --> 00:10:36,640 it was an image. 148 00:10:36,640 --> 00:10:41,360 So I took a screen grab of the image, fed it into chat GPT vision and I said there's 149 00:10:41,360 --> 00:10:46,760 a table in this image, reproduce it as a CSV so I can copy paste it into Excel so I can 150 00:10:46,760 --> 00:10:48,920 edit it and it came out perfect. 151 00:10:48,920 --> 00:10:52,760 So there's just so many cool things that you can do there and then finally on the sort 152 00:10:52,760 --> 00:10:57,100 of video editing and podcasting Descript is an amazing video editor you push your video 153 00:10:57,100 --> 00:11:01,040 in and then it creates a transcript of it and then you can edit it like it's a Word 154 00:11:01,040 --> 00:11:05,960 document so you highlight bits of the script that you're going to cut out and then it just 155 00:11:05,960 --> 00:11:08,320 automatically edits it all for you. 156 00:11:08,320 --> 00:11:13,400 It's also got some nice tools in there to help clean up audio although it's not as good 157 00:11:13,400 --> 00:11:16,880 as Adobe Podcast Enhance which is insane. 158 00:11:16,880 --> 00:11:22,400 I routinely now record my talks at different events on my mobile phone. 159 00:11:22,400 --> 00:11:25,680 I just sit it on the plinth where I'm speaking. 160 00:11:25,680 --> 00:11:31,880 The echo-y terrible audio Adobe Podcast Enhance can strip out most of the background noise 161 00:11:31,880 --> 00:11:37,240 and either make it sound really polished like it was professionally recorded in the type 162 00:11:37,240 --> 00:11:41,720 of like I was mic'd up and stuff or even to the point where you can strip out all the 163 00:11:41,720 --> 00:11:46,060 background and it sounds like I was working in a proper studio. 164 00:11:46,060 --> 00:11:51,480 So on the audio restoration it can really do incredible things and is worth playing 165 00:11:51,480 --> 00:11:52,480 with. 166 00:11:52,480 --> 00:11:56,040 So for the people that keep asking me, that's the answer to that question. 167 00:11:56,040 --> 00:11:58,640 Martin, I know people ask you this a lot as well. 168 00:11:58,640 --> 00:12:00,440 What's your favourite tools to play with? 169 00:12:00,440 --> 00:12:05,480 Well, the tool I wouldn't live without is ChatGPT Plus, right? 170 00:12:05,480 --> 00:12:13,560 Given at this point GPT-4 access is kind of critical but I find myself using all of the 171 00:12:13,560 --> 00:12:15,960 features within that Plus subscription. 172 00:12:15,960 --> 00:12:24,080 So the Dali 3 image generation is increasingly... 173 00:12:24,080 --> 00:12:26,040 I didn't do a lot of image generation. 174 00:12:26,040 --> 00:12:29,400 If you'd have asked me a few weeks ago, I was always playing with mid-journey. 175 00:12:29,400 --> 00:12:34,400 I was taking the occasional image from mid-journey and putting it into slide decks and presentations, 176 00:12:34,400 --> 00:12:36,600 but I wasn't doing a great deal. 177 00:12:36,600 --> 00:12:44,680 But now because of the way that the prompting in ChatGPT with Dali 3 works, I find myself 178 00:12:44,680 --> 00:12:46,640 going to it much more regularly. 179 00:12:46,640 --> 00:12:53,200 It's just part of the way that I use the tool more naturally. 180 00:12:53,200 --> 00:12:55,560 Because it will ask questions to improve my prompt. 181 00:12:55,560 --> 00:12:57,440 That's what I like about it. 182 00:12:57,440 --> 00:13:02,480 One of the things with mid-journey is I often got frustrated trying to craft the perfect 183 00:13:02,480 --> 00:13:07,160 prompt whereas now this does some of the hard work for you. 184 00:13:07,160 --> 00:13:12,680 The browse with Bing functionality has actually meant that one of the tools that I was using 185 00:13:12,680 --> 00:13:18,520 really regularly, which was Perplexity AI, I'm using slightly less now, which is a shame 186 00:13:18,520 --> 00:13:21,800 because I've just bought the annual subscription. 187 00:13:21,800 --> 00:13:27,120 But Perplexity, because it's a real-time search engine and it's all running the answers through 188 00:13:27,120 --> 00:13:37,200 GPT-4, was really useful for getting AI content about to the minute, fact and information. 189 00:13:37,200 --> 00:13:41,360 But actually now I find the browse with Bing is doing that for me. 190 00:13:41,360 --> 00:13:44,720 I was working with a client this week. 191 00:13:44,720 --> 00:13:54,320 They needed some content for the website that was reflective of some legal changes in their 192 00:13:54,320 --> 00:13:59,480 industry from September this year, so within the past three weeks. 193 00:13:59,480 --> 00:14:04,620 We switched on browse with Bing, we ran a few prompts, we got the content that we needed 194 00:14:04,620 --> 00:14:07,760 to stick on the website and update a marketing email. 195 00:14:07,760 --> 00:14:10,080 That was done in seconds. 196 00:14:10,080 --> 00:14:12,560 So that was really useful. 197 00:14:12,560 --> 00:14:24,160 Unlike you, I don't have Audio Pen, but instead I have the Zapier workflows that connect through 198 00:14:24,160 --> 00:14:25,320 Whisper API. 199 00:14:25,320 --> 00:14:30,040 So I'm always recording voice notes on my phone and just uploading them to Google Drive, 200 00:14:30,040 --> 00:14:33,940 which then goes to Zapier, sending them off to Whisper and running them through a sequence 201 00:14:33,940 --> 00:14:35,920 of different transcriptions. 202 00:14:35,920 --> 00:14:44,000 Audio transcription for me is one of the biggest efficiency gains that I personally get from 203 00:14:44,000 --> 00:14:45,000 AI. 204 00:14:45,000 --> 00:14:51,040 So whether it's recording voice notes when I'm out and about or doing conference talks 205 00:14:51,040 --> 00:14:55,640 if I'm doing a conference talk or if I'm at a conference listening to a talk, I'll take 206 00:14:55,640 --> 00:15:00,440 the transcription and try and capture some notes from the sessions. 207 00:15:00,440 --> 00:15:02,920 But also being on meetings. 208 00:15:02,920 --> 00:15:10,120 So I like everyone these days, I'm either in front of chat GPT or I'm on a Zoom call, 209 00:15:10,120 --> 00:15:15,900 an MS Teams call or a Google Meet call or something like that. 210 00:15:15,900 --> 00:15:19,840 Having all of those transcribed for me is a godsend. 211 00:15:19,840 --> 00:15:23,720 It's made my admin much easier. 212 00:15:23,720 --> 00:15:28,960 It's made if I'm on a sales call, it's made the post call follow up so much more effective. 213 00:15:28,960 --> 00:15:31,640 And that's because I use a tool called Laxys. 214 00:15:31,640 --> 00:15:37,480 I'm hopefully getting Laxys CEO Eric on very soon. 215 00:15:37,480 --> 00:15:40,440 It'd be great to have him on the podcast. 216 00:15:40,440 --> 00:15:43,960 But that does a great job of just capturing all the notes, writing the follow up emails, 217 00:15:43,960 --> 00:15:45,480 giving me the action items. 218 00:15:45,480 --> 00:15:51,120 And actually what I do with that, I'll often take the full transcript from a call and stick 219 00:15:51,120 --> 00:15:53,880 that into the other tool. 220 00:15:53,880 --> 00:15:59,840 If you were to say I could have two tools and I couldn't be without them, maybe three, 221 00:15:59,840 --> 00:16:00,840 maybe three. 222 00:16:00,840 --> 00:16:05,000 The thought of having to go down to two then made me struggle. 223 00:16:05,000 --> 00:16:08,720 It made me have heart palpitations. 224 00:16:08,720 --> 00:16:15,680 The ones that I couldn't get rid of would be probably Laxys, chat GPT plus and Claude. 225 00:16:15,680 --> 00:16:21,680 The hundred thousand token context window, the amount of information that you can feed 226 00:16:21,680 --> 00:16:28,560 into it and the outputs that you can get as a result of that context window. 227 00:16:28,560 --> 00:16:29,760 It's a daily driver for me. 228 00:16:29,760 --> 00:16:37,680 In fact, I'm using that easily two or three times a day, every day, even on weekends. 229 00:16:37,680 --> 00:16:43,320 It's just so ingrained in my life at this point. 230 00:16:43,320 --> 00:16:45,980 So yeah, that is definitely a big one. 231 00:16:45,980 --> 00:16:51,280 On the audio side of things, I don't use Descript because I don't have the editing skills that 232 00:16:51,280 --> 00:16:52,280 you have. 233 00:16:52,280 --> 00:16:57,480 But I have started playing and using over the past month. 234 00:16:57,480 --> 00:17:03,000 I've got a subscription to 11 Labs and 11 Labs is the voice synthesis tool that we featured 235 00:17:03,000 --> 00:17:06,600 on this and it does a great job of creating synthetic voices. 236 00:17:06,600 --> 00:17:09,320 It can clone your own voice. 237 00:17:09,320 --> 00:17:16,120 We featured it on a recent episode where it had the dubbed version of my voice into Italian. 238 00:17:16,120 --> 00:17:21,320 And it's a really neat little tool and I've been using that to create little audio snippets 239 00:17:21,320 --> 00:17:27,340 for social media videos for myself, showing clients and proposing that we build this into 240 00:17:27,340 --> 00:17:29,800 some of their social campaigns as well. 241 00:17:29,800 --> 00:17:34,360 And yeah, so far people are very receptive to it. 242 00:17:34,360 --> 00:17:39,800 I think one of the things that people like about it is, well, you know what it's like 243 00:17:39,800 --> 00:17:41,760 trying to do voiceover, right? 244 00:17:41,760 --> 00:17:44,120 And there's loads of voiceover artists online. 245 00:17:44,120 --> 00:17:50,640 You can get them from marketplaces pretty cheap, pretty efficiently and they do great 246 00:17:50,640 --> 00:17:52,240 work. 247 00:17:52,240 --> 00:17:59,000 And whilst I don't want to take work away from voiceover artists, if a client is knocking 248 00:17:59,000 --> 00:18:03,840 up a 30 second social video that needs a little bit of audio with some spoken words over the 249 00:18:03,840 --> 00:18:09,960 top of it and they can do it in something like 11 Labs and it takes them, they can do 250 00:18:09,960 --> 00:18:15,900 it in five or 10 minutes, that's so much more efficient from their workflow compared to 251 00:18:15,900 --> 00:18:22,080 writing the draft, putting it on a board, getting the voiceover artists to respond, going back 252 00:18:22,080 --> 00:18:26,080 and forth, getting the recording, feeding the amends. 253 00:18:26,080 --> 00:18:29,360 Like all of that is a headache. 254 00:18:29,360 --> 00:18:35,520 And I really don't like that AI is going to be taking those kinds of jobs away from people, 255 00:18:35,520 --> 00:18:41,500 but there's a reality to this that actually, I think that's an area where some voiceover 256 00:18:41,500 --> 00:18:44,520 work is going to be lost to AI like 11 Labs. 257 00:18:44,520 --> 00:18:49,960 Well, and if we look at some of the applications of where you might want to do audio on a person 258 00:18:49,960 --> 00:18:54,000 by person basis, like if you're doing sales prospecting and you want to leave like a personal 259 00:18:54,000 --> 00:18:58,240 voice note, in theory, you could probably start to synthesize that whether or not that's 260 00:18:58,240 --> 00:19:00,560 a bit disingenuous for the people who get those notes. 261 00:19:00,560 --> 00:19:06,640 I'm sure we'll learn to sniff those out like we do the prospecting emails that do a terrible 262 00:19:06,640 --> 00:19:09,280 job of making us think that they're just for us. 263 00:19:09,280 --> 00:19:11,280 But yeah, I hear you on that one. 264 00:19:11,280 --> 00:19:12,280 That's a tricky one. 265 00:19:12,280 --> 00:19:17,000 But like you said, the other thing that often happens is when you get the audio back, you're 266 00:19:17,000 --> 00:19:21,160 like, ah, actually, I wish the script had been this instead. 267 00:19:21,160 --> 00:19:25,520 And of course, you could just make that change instantly if you use a tool like 11 Labs versus 268 00:19:25,520 --> 00:19:29,320 rerecording and rebreefing and all that other stuff. 269 00:19:29,320 --> 00:19:30,720 Yeah, exactly. 270 00:19:30,720 --> 00:19:32,400 And that's where the benefit comes. 271 00:19:32,400 --> 00:19:36,000 That's where the efficiency game come because you can iterate quicker. 272 00:19:36,000 --> 00:19:41,080 You can just produce 20 different versions and pick the one that you like and you can 273 00:19:41,080 --> 00:19:46,920 do that sat at your desk rather than having to go back and forth with a recording artist. 274 00:19:46,920 --> 00:19:50,560 Yeah, so that's the tools I'm using. 275 00:19:50,560 --> 00:19:55,760 I mean, there's other tools I'm big on and a part of my core workflow. 276 00:19:55,760 --> 00:20:02,960 So HubSpot's suite of AI tools, I use them regularly for drafting emails. 277 00:20:02,960 --> 00:20:07,080 Principally, I'm using the content assistant. 278 00:20:07,080 --> 00:20:15,280 I'm not using their lead forecasting for myself, but certainly rolling that out with clients. 279 00:20:15,280 --> 00:20:21,880 I mentioned mid-journey or ready for image creation, but that is getting increasingly 280 00:20:21,880 --> 00:20:25,880 lower usage since Dali 3 is coming out. 281 00:20:25,880 --> 00:20:30,880 But one tool that I am finding myself using more and more since the recent updates that 282 00:20:30,880 --> 00:20:34,840 we did discuss on the show is GoCharlie. 283 00:20:34,840 --> 00:20:41,320 So I've had GoCharlie for a while and they recently released their marketing agent, as 284 00:20:41,320 --> 00:20:45,160 it were, and you can give it commands and it will go off and perform tasks. 285 00:20:45,160 --> 00:20:53,400 And I'm finding that really useful, particularly for video summarization from YouTube videos. 286 00:20:53,400 --> 00:20:58,120 That's not the only thing I'm using it for, but that is something I'm using that for two 287 00:20:58,120 --> 00:20:59,320 or three times a week. 288 00:20:59,320 --> 00:21:07,960 I'm absolutely turning to GoCharlie to capture some great extracts from long form video content. 289 00:21:07,960 --> 00:21:09,520 Cool, cool. 290 00:21:09,520 --> 00:21:12,360 Well, there you have it, dear listeners. 291 00:21:12,360 --> 00:21:14,160 Obviously we get asked this a lot. 292 00:21:14,160 --> 00:21:21,320 There is a bullet point whistle stop tour of a billion tools for you to go try, but they 293 00:21:21,320 --> 00:21:26,920 are based on us trying at least 10 times as many tools as what we just described, many 294 00:21:26,920 --> 00:21:31,480 of which we left behind because we didn't really feel like they really did much to our 295 00:21:31,480 --> 00:21:32,480 workflow. 296 00:21:32,480 --> 00:21:36,160 And of course, and I think most people, certainly those listening to the podcast are probably 297 00:21:36,160 --> 00:21:37,920 already going to be here. 298 00:21:37,920 --> 00:21:41,240 If you're going to start anywhere, you start with ChatGPT Plus because it just has so much 299 00:21:41,240 --> 00:21:45,440 power in it, it's probably going to be able to do 80% of the things that most marketers 300 00:21:45,440 --> 00:21:46,440 want to be able to do. 301 00:21:46,440 --> 00:21:51,600 And then when you want to get into more exotic use cases like editing video, creating video, 302 00:21:51,600 --> 00:21:56,400 creating synthetic voices for audio versions of your blog posts and all this other stuff, 303 00:21:56,400 --> 00:21:59,280 then of course some of those tools will help. 304 00:21:59,280 --> 00:22:00,760 Right. 305 00:22:00,760 --> 00:22:03,880 Let's get into the news, Martin, the new ooze. 306 00:22:03,880 --> 00:22:06,400 What have we got as the first news story this week? 307 00:22:06,400 --> 00:22:12,640 So we've got big news coming out of Jasper. 308 00:22:12,640 --> 00:22:21,160 Jasper AI, one of the first tools on the market to really make use of the GPT-3 API. 309 00:22:21,160 --> 00:22:28,400 They put a front end on it and made it really easy for marketers to create marketing copy. 310 00:22:28,400 --> 00:22:32,720 They did all the prompting in the backend and just gave a really nice UI for people 311 00:22:32,720 --> 00:22:33,720 to be able to create content. 312 00:22:33,720 --> 00:22:35,800 And they've been on the market for a few years now. 313 00:22:35,800 --> 00:22:42,760 As the market has evolved, so has the need for products like Jasper to evolve. 314 00:22:42,760 --> 00:22:49,600 So while it was once primarily a generative AI tool for quick content creation, it's now 315 00:22:49,600 --> 00:22:57,880 seeking to become an AI copilot for marketers with a much greater focus on strategy. 316 00:22:57,880 --> 00:23:04,920 So this is a significant expansion for the product, but also for the company. 317 00:23:04,920 --> 00:23:11,880 They just had record breaking series A funding and recently had a change at the top of the 318 00:23:11,880 --> 00:23:12,880 company. 319 00:23:12,880 --> 00:23:16,560 They've got a new CEO, the former Dropbox president, Timothy Young. 320 00:23:16,560 --> 00:23:21,040 So the founding CEO has recently moved roles. 321 00:23:21,040 --> 00:23:24,960 So this expansion is very much multifaceted. 322 00:23:24,960 --> 00:23:28,800 It's designed not only to accelerate, but also improve marketing performance across 323 00:23:28,800 --> 00:23:32,720 a range of areas, not just on content production. 324 00:23:32,720 --> 00:23:37,400 So one of the areas that they're looking at, they're looking at analytics and insights, 325 00:23:37,400 --> 00:23:42,880 including real-time performance analytics on content with AI recommendations to optimize 326 00:23:42,880 --> 00:23:50,040 any underperforming content, as well as one-click deployment for content revisions. 327 00:23:50,040 --> 00:23:53,120 So fully integrated into your workflow. 328 00:23:53,120 --> 00:23:54,360 This is really what they're about now. 329 00:23:54,360 --> 00:24:01,800 It's about helping execute tasks as well as create content. 330 00:24:01,800 --> 00:24:06,580 They've got what they're calling a company intelligence hub, which is a central storage 331 00:24:06,580 --> 00:24:09,160 for all of your brand voice. 332 00:24:09,160 --> 00:24:13,280 Now this is an area that we know companies really struggle with when it comes to tools 333 00:24:13,280 --> 00:24:17,240 like chat GPT, just sounding a bit too generic. 334 00:24:17,240 --> 00:24:19,660 So Jasper are tackling this straight on. 335 00:24:19,660 --> 00:24:23,180 You can upload all of your company documents into the central storage. 336 00:24:23,180 --> 00:24:29,280 You can have strategy documents and other wider institutional knowledge, all of which 337 00:24:29,280 --> 00:24:36,800 will help the AI to create less generic outputs and reduce inaccuracies and bring that knowledge 338 00:24:36,800 --> 00:24:37,800 into the output. 339 00:24:37,800 --> 00:24:43,520 So it can be more kind of fine tuned as it were to your own company information. 340 00:24:43,520 --> 00:24:48,320 And it also allows for kind of dynamic updating, right? 341 00:24:48,320 --> 00:24:56,080 You can import new documents and that gets incorporated into the model straight away. 342 00:24:56,080 --> 00:24:59,240 As well as that, they've got some wider marketing campaign tools. 343 00:24:59,240 --> 00:25:05,760 So what they're calling the campaign acceleration tools, including integrated assignments, comments 344 00:25:05,760 --> 00:25:10,480 and scheduling to improve team collaboration. 345 00:25:10,480 --> 00:25:20,240 And they are doing features such as AI summaries of feedback from teams and one click automation 346 00:25:20,240 --> 00:25:22,600 of revisions. 347 00:25:22,600 --> 00:25:29,760 All of which shows that they're really moving away from being a tool for kind of solopreneurs 348 00:25:29,760 --> 00:25:36,160 or single entrepreneurs where I think that was a big market for them in the early days. 349 00:25:36,160 --> 00:25:41,700 It was, they were one of the first AI writing tools on the market. 350 00:25:41,700 --> 00:25:47,160 So if they could help really small businesses to produce content quickly, that's very much 351 00:25:47,160 --> 00:25:49,420 where they were positioning themselves. 352 00:25:49,420 --> 00:25:56,720 But this change and this repositioning is squarely putting them at the, trying to get 353 00:25:56,720 --> 00:26:00,040 them in front of teams, marketing teams. 354 00:26:00,040 --> 00:26:06,320 This is about collaboration, about security, about feedback and revisions and all of that 355 00:26:06,320 --> 00:26:07,820 kind of thing. 356 00:26:07,820 --> 00:26:12,320 Just on that point actually, they have enterprise controls coming. 357 00:26:12,320 --> 00:26:17,240 So they've got API availability for platform integration, which presumably is how they'll 358 00:26:17,240 --> 00:26:21,520 be doing things like the one click deploying of content and the revisions. 359 00:26:21,520 --> 00:26:26,360 You can easily imagine Jasper being fully integrated with the HubSpot marketing suite 360 00:26:26,360 --> 00:26:31,840 and you could deploy it, getting the content from Jasper into an email marketing campaign 361 00:26:31,840 --> 00:26:35,640 or a blog or a product page in the CMS. 362 00:26:35,640 --> 00:26:41,380 They've also included SOC2 compliance for added security and they've added more administrative 363 00:26:41,380 --> 00:26:47,200 tools which will give greater levels of control over things like user permissions and work 364 00:26:47,200 --> 00:26:50,120 spaces, both private and public. 365 00:26:50,120 --> 00:26:52,880 So yeah, it's a big shift for Jasper. 366 00:26:52,880 --> 00:26:55,840 You can see where they're heading. 367 00:26:55,840 --> 00:27:01,080 Obviously they couldn't just be a content creation tool anymore because you can do all 368 00:27:01,080 --> 00:27:08,640 of that kind of stuff with a well-crafted prompt using Bing Chat AI, which is free. 369 00:27:08,640 --> 00:27:13,640 So if they're going to get value, if they're going to continue to offer value to companies, 370 00:27:13,640 --> 00:27:19,480 they've got to move up market and basically make sure that they are a real, well, as they 371 00:27:19,480 --> 00:27:23,000 call it themselves, marketing co-pilot for teams. 372 00:27:23,000 --> 00:27:27,840 I think this is a really smart move because, and as a callback to the beginning of the 373 00:27:27,840 --> 00:27:35,280 episode, we're often asked what is a great tool for creating content and writing and 374 00:27:35,280 --> 00:27:41,940 ultimately TrackGPT 4 is probably still the best tool in my opinion. 375 00:27:41,940 --> 00:27:49,080 I've tried Jasper, I've tried CopyAI, we were piloting and testing a tool called Writer, 376 00:27:49,080 --> 00:27:55,760 RYTR before TrackGPT was released, probably two years ago, something like that. 377 00:27:55,760 --> 00:28:02,400 And I think where a lot of these tools like Jasper were in risk is kind of being sort 378 00:28:02,400 --> 00:28:07,000 of augmenting people's writing by having lots of templates that write in different marketing 379 00:28:07,000 --> 00:28:10,320 frameworks like Ada or whatever. 380 00:28:10,320 --> 00:28:15,480 But actually not really necessarily being that helpful compared to just using TrackGPT. 381 00:28:15,480 --> 00:28:20,440 This is a really powerful mechanism to really differentiate Jasper as to why would you buy 382 00:28:20,440 --> 00:28:21,440 over the other stuff? 383 00:28:21,440 --> 00:28:23,280 I think you're bang on. 384 00:28:23,280 --> 00:28:26,840 And actually it gives me a bit of FOMO. 385 00:28:26,840 --> 00:28:29,800 I can't say I've had much FOMO when it comes to most of the writing tools. 386 00:28:29,800 --> 00:28:35,520 So it's pretty just happy using TrackGPT or in my case Matchi which is just really a front 387 00:28:35,520 --> 00:28:37,560 end for TrackGPT. 388 00:28:37,560 --> 00:28:43,800 But I could see how a lot of that stuff would actually be really quite useful, especially 389 00:28:43,800 --> 00:28:44,960 for larger companies. 390 00:28:44,960 --> 00:28:51,480 So as is proving the case, you can't really make any hard and fast decisions on what tools 391 00:28:51,480 --> 00:28:53,000 are going to be good or not. 392 00:28:53,000 --> 00:28:56,900 You've got to pay attention to all the different changes that are happening in the space. 393 00:28:56,900 --> 00:29:01,240 So that as tools release new things and it brings them back into contention, you know, 394 00:29:01,240 --> 00:29:05,840 and then you can go test them and play with them and see just how capable they have become. 395 00:29:05,840 --> 00:29:09,200 So Jasper's back on my go test it list. 396 00:29:09,200 --> 00:29:10,200 Cool. 397 00:29:10,200 --> 00:29:11,200 Good story. 398 00:29:11,200 --> 00:29:12,880 Thank you for sharing that, Martin. 399 00:29:12,880 --> 00:29:19,460 Into the next story then, and this one's an interesting one about Anthropic and Universal 400 00:29:19,460 --> 00:29:20,460 Music. 401 00:29:20,460 --> 00:29:25,720 So Universal Music has lodged a copyright infringement lawsuit against Anthropic, which 402 00:29:25,720 --> 00:29:29,320 for those of you who are regular listeners of the podcast will know, has created one 403 00:29:29,320 --> 00:29:33,320 of the large language models Claude II. 404 00:29:33,320 --> 00:29:39,280 And the reason is that Universal Music has accused Anthropic of generating nearly identical 405 00:29:39,280 --> 00:29:46,480 copies of lyrics from Universal's artists via the chap Claude. 406 00:29:46,480 --> 00:29:52,120 Universal along with Concord and ABCO filed the lawsuit in a US court stating that when 407 00:29:52,120 --> 00:29:57,240 Claude is prompted for lyrics to songs like Glory Again as I Will Survive, it produces 408 00:29:57,240 --> 00:30:00,840 a near verbatim rendition. 409 00:30:00,840 --> 00:30:07,440 Music which has investments from Amazon and Google has yet to respond to the allegations. 410 00:30:07,440 --> 00:30:11,760 This is quite interesting because we know that Anthropics Claude is trained in a slightly 411 00:30:11,760 --> 00:30:19,160 different way and its reinforcement learning is a bit different from how ChatGPT works. 412 00:30:19,160 --> 00:30:23,920 I haven't heard anything like this coming out of ChatGPT, but I don't know, mine, does 413 00:30:23,920 --> 00:30:29,880 this flag to us that for all the Claude love that we give on the podcast, especially yourself, 414 00:30:29,880 --> 00:30:34,960 is it maybe slightly more likely to produce things that are a bit more plagiarized that 415 00:30:34,960 --> 00:30:41,520 are not just driven by its clever vector models and that actually there's a little bit of 416 00:30:41,520 --> 00:30:43,360 reproduction happening here? 417 00:30:43,360 --> 00:30:46,360 I don't know, what do you think? 418 00:30:46,360 --> 00:30:48,080 Too early to tell. 419 00:30:48,080 --> 00:30:57,640 I wouldn't be surprised if we see similar suits coming out against OpenAI and ChatGPT. 420 00:30:57,640 --> 00:31:04,080 It wouldn't surprise me, I'm not saying that is likely at all, or I don't have any inside 421 00:31:04,080 --> 00:31:09,080 knowledge that these cases are pending, but I think as more companies do more of these 422 00:31:09,080 --> 00:31:15,840 kinds of experiments on large language models, they're going to come across cases where clearly 423 00:31:15,840 --> 00:31:20,720 copyright data has been put into it and if you get the prompts just right, it will spit 424 00:31:20,720 --> 00:31:22,680 out something very similar. 425 00:31:22,680 --> 00:31:36,320 There was the example with Katy Perry's Raw, which was a big thing in this story as well. 426 00:31:36,320 --> 00:31:42,680 So yeah, they've clearly got it across a bunch of different artists. 427 00:31:42,680 --> 00:31:52,320 I think there was one also where the prompt asked it to write a story about somebody's 428 00:31:52,320 --> 00:31:53,320 death. 429 00:31:53,320 --> 00:32:01,240 It was a high profile person and it came back with lyrics of a song that was actually a 430 00:32:01,240 --> 00:32:04,080 real song that was about that person's death. 431 00:32:04,080 --> 00:32:12,280 Clearly they were trying to trick the model into recreating a piece of, well yeah, recreating 432 00:32:12,280 --> 00:32:13,280 a song. 433 00:32:13,280 --> 00:32:17,920 So yeah, I think Anthropix got some answering to do here. 434 00:32:17,920 --> 00:32:21,240 I don't really know what the, you know, what is the case? 435 00:32:21,240 --> 00:32:27,720 Is it that it was trained on it or that it generated the content? 436 00:32:27,720 --> 00:32:34,720 Because if I ask you to write me a song and you come up with a song that has lyrics that 437 00:32:34,720 --> 00:32:43,480 are nearly identical to Katy Perry's Raw, for instance, I guess you would be, if I was 438 00:32:43,480 --> 00:32:50,440 to then publish that, you, me or both could be sued for publishing it. 439 00:32:50,440 --> 00:33:00,640 But for just recreating them and writing them down, do you get, is that worthy of a legal 440 00:33:00,640 --> 00:33:02,000 complaint in and of itself? 441 00:33:02,000 --> 00:33:06,960 Well, I guess it's interesting, A, it would definitely imply its ability to reproduce them 442 00:33:06,960 --> 00:33:13,480 pretty much verbatim would imply that the song words were included in the training material. 443 00:33:13,480 --> 00:33:17,680 So I suspect that's where the main issue lies. 444 00:33:17,680 --> 00:33:24,760 In terms of producing aspects, like I could definitely imagine use cases where I would 445 00:33:24,760 --> 00:33:31,480 love to use a really good quote from an author or a famous person, but I just can't quite 446 00:33:31,480 --> 00:33:33,000 remember the quote. 447 00:33:33,000 --> 00:33:38,400 So I go ask ChatGPT or Claude in this case to give me that quote and remind me of it 448 00:33:38,400 --> 00:33:41,040 instead of using Google. 449 00:33:41,040 --> 00:33:45,280 And then where, you know, where does that issue lie? 450 00:33:45,280 --> 00:33:52,120 If you search for song lyrics, you will be able to find them in snippets on Google search 451 00:33:52,120 --> 00:33:57,160 results page if you really wanted to find Gloria, Gloria Gaynor's lyrics. 452 00:33:57,160 --> 00:34:00,720 So yeah, it's, I think it's going to come back down to training and this is probably 453 00:34:00,720 --> 00:34:07,320 the best example yet that any of these companies has found to prove unequivocally almost that 454 00:34:07,320 --> 00:34:10,840 their IP was used in the training. 455 00:34:10,840 --> 00:34:14,160 I think that's why this is important. 456 00:34:14,160 --> 00:34:21,840 But I think we're going to, we know that copyright material was included in the training. 457 00:34:21,840 --> 00:34:28,800 So I'm really interested to see the specific legal challenge here. 458 00:34:28,800 --> 00:34:36,880 Because is it effectively just the music version of the Sarah Silverman case against OpenAI 459 00:34:36,880 --> 00:34:38,440 and their training? 460 00:34:38,440 --> 00:34:44,240 In which case, you know, I think these language model companies have, they're going to have 461 00:34:44,240 --> 00:34:48,560 to change their revenue structures or the way that they do licensing. 462 00:34:48,560 --> 00:34:54,000 I mean, you know, how are you, given the way that they work, every time you put a query 463 00:34:54,000 --> 00:35:01,320 into one of these models, the inference will check the whole model, right? 464 00:35:01,320 --> 00:35:03,720 It doesn't just go off to a certain domain. 465 00:35:03,720 --> 00:35:08,560 It runs it against the whole massive model. 466 00:35:08,560 --> 00:35:11,520 You can't charge on a per inference basis. 467 00:35:11,520 --> 00:35:13,520 I just don't understand. 468 00:35:13,520 --> 00:35:16,880 Clearly smarter people than me are going to have to work out that licensing model. 469 00:35:16,880 --> 00:35:18,040 They're going to have to figure this out. 470 00:35:18,040 --> 00:35:22,960 And I think one of the complexities here is if Sarah Silverman feels like her entire 471 00:35:22,960 --> 00:35:28,720 book was used in the training of a large language model, either because the digital version 472 00:35:28,720 --> 00:35:31,520 was accessed or procured somehow. 473 00:35:31,520 --> 00:35:34,640 That's one thing. 474 00:35:34,640 --> 00:35:38,600 The lyrics to lots of songs float around on the web because people want to know what they 475 00:35:38,600 --> 00:35:39,600 are. 476 00:35:39,600 --> 00:35:41,960 You couldn't really access them at the start of the web. 477 00:35:41,960 --> 00:35:44,240 That wasn't something you could easily get hold of. 478 00:35:44,240 --> 00:35:47,840 And so these massive websites sprung up where you could get the lyrics to any song. 479 00:35:47,840 --> 00:35:53,120 We know that a lot of these large language models were trained based on content found 480 00:35:53,120 --> 00:35:54,120 across the web. 481 00:35:54,120 --> 00:35:58,880 And actually the emergence of the World Wide Web is an important precursor to even being 482 00:35:58,880 --> 00:36:01,720 able to have these models exist. 483 00:36:01,720 --> 00:36:03,960 So what do you do then? 484 00:36:03,960 --> 00:36:11,000 If you're open AI and you're training your model and there's 52 copies of the lyrics 485 00:36:11,000 --> 00:36:15,760 on these sites that, by the way, Universal Music hasn't pulled down over the last 20 486 00:36:15,760 --> 00:36:23,320 years for whatever reasons, how are you supposed to stop that ending up in your material? 487 00:36:23,320 --> 00:36:27,680 Or then by default, do you then assume that everything on the web was actually copyright 488 00:36:27,680 --> 00:36:31,280 material because somebody wrote it and owned it in some way? 489 00:36:31,280 --> 00:36:33,200 It's so complicated. 490 00:36:33,200 --> 00:36:37,520 In fact, probably spend the next two hours trying to debate this, even though neither 491 00:36:37,520 --> 00:36:43,000 of us have any real legal background and therefore probably not the qualifications to do so. 492 00:36:43,000 --> 00:36:48,440 Suffice to say, Martin, if you're a marketer, you've got to keep a close eye on how this 493 00:36:48,440 --> 00:36:53,680 is all drifting because I guess one potential is at some point the tools that we've come 494 00:36:53,680 --> 00:36:59,720 to know and love, they all get stopped and we're not allowed to use them for some period 495 00:36:59,720 --> 00:37:01,560 while all this gets sorted out. 496 00:37:01,560 --> 00:37:03,680 I don't think they'll ever disappear forever. 497 00:37:03,680 --> 00:37:07,080 I think there's just too much commercial opportunity for all the people who've built them and even 498 00:37:07,080 --> 00:37:11,320 the people whose stuff got used to train them. 499 00:37:11,320 --> 00:37:15,920 With open source models out there, the genie is out of the bottle. 500 00:37:15,920 --> 00:37:20,920 Just to pull it back, before we move on to this story, the song that I was referring 501 00:37:20,920 --> 00:37:23,320 to, I've got the story in front of me. 502 00:37:23,320 --> 00:37:30,520 It said, the complaint used the example prompt, write me a song about the death of Buddy Holly, 503 00:37:30,520 --> 00:37:36,320 which led the large language model to spit out the lyrics to Don McLean's American Pie, 504 00:37:36,320 --> 00:37:38,480 word for word. 505 00:37:38,480 --> 00:37:39,680 Interesting. 506 00:37:39,680 --> 00:37:45,920 Especially given that I assume just that stanza that's about... 507 00:37:45,920 --> 00:37:49,960 Anyway, the whole song's got loads of other bits in it. 508 00:37:49,960 --> 00:37:54,600 I think as marketers, fundamentally, we're going to have to keep an eye on this in case 509 00:37:54,600 --> 00:37:57,520 it influences how we can use the tools. 510 00:37:57,520 --> 00:38:02,160 Certainly the same things apply if you're using chat GPT, don't use the outputs as they 511 00:38:02,160 --> 00:38:03,640 are as they come out. 512 00:38:03,640 --> 00:38:05,680 There's a whole host of reasons not to do that. 513 00:38:05,680 --> 00:38:12,240 The first is you can't copyright them, that seems to be where legal precedent is drifting. 514 00:38:12,240 --> 00:38:17,260 If you use them as is, often the writing will be a little bit stilted and a bit samey. 515 00:38:17,260 --> 00:38:19,720 There's lots of kind of like grammar loops. 516 00:38:19,720 --> 00:38:24,400 It loves to start a sentence with this, for example, which is not necessarily great writing. 517 00:38:24,400 --> 00:38:26,200 So edit the outputs a lot. 518 00:38:26,200 --> 00:38:27,800 You want them in your own brand voice. 519 00:38:27,800 --> 00:38:32,280 You want to add some insight that chat GPT doesn't have, that your own brand and your 520 00:38:32,280 --> 00:38:34,560 own subject matter experts have. 521 00:38:34,560 --> 00:38:40,160 Scan your content through a plagiarism detector to make sure that chat GPT hasn't just lifted 522 00:38:40,160 --> 00:38:45,600 any content that somehow made it through your rigorous edit and is still present in your 523 00:38:45,600 --> 00:38:46,600 copy. 524 00:38:46,600 --> 00:38:49,720 All these things are still just really, really good advice. 525 00:38:49,720 --> 00:38:56,120 And if this story develops and it impacts how we're as marketers able to use these tools, 526 00:38:56,120 --> 00:39:00,320 well, dear listener, you'll be sure to hear about it first on the Artificially Intelligent 527 00:39:00,320 --> 00:39:03,800 Marketing Podcast. 528 00:39:03,800 --> 00:39:07,040 Should we put those song lyrics to bed? 529 00:39:07,040 --> 00:39:08,040 Right. 530 00:39:08,040 --> 00:39:12,160 The next story is with you, Martin, is about celebrity clones. 531 00:39:12,160 --> 00:39:13,160 Yeah. 532 00:39:13,160 --> 00:39:21,360 So in a compelling mix of fan service and AI, high profile prominent figures such as 533 00:39:21,360 --> 00:39:29,260 Tim Ferriss and Scott Galloway have trained some AI chat bots on their accumulated body 534 00:39:29,260 --> 00:39:35,040 of works, including their podcasts and all of their blog posts, which, you know, is a 535 00:39:35,040 --> 00:39:38,160 large volume of content. 536 00:39:38,160 --> 00:39:42,400 The promise of these chat bots is simple and captivating. 537 00:39:42,400 --> 00:39:51,200 Pose a question and your favorite thought leader will answer based on their previous 538 00:39:51,200 --> 00:39:53,240 statements and insights. 539 00:39:53,240 --> 00:40:00,520 So you get your own Scott Galloway in your pocket, so to speak, which is quite a neat 540 00:40:00,520 --> 00:40:05,000 and handy way to distill the mountains and mountains of content that they've run over 541 00:40:05,000 --> 00:40:06,000 the many years. 542 00:40:06,000 --> 00:40:13,560 So this is a digital trend that is not just confined to business gurus. 543 00:40:13,560 --> 00:40:20,000 Meta is pushing the envelope further by introducing celebrity inspired chat bots on Instagram, 544 00:40:20,000 --> 00:40:27,960 where users can interact with AI versions, AI avatars of their favorite stars, such as 545 00:40:27,960 --> 00:40:31,520 the YouTube sensation, Mr. Beast. 546 00:40:31,520 --> 00:40:39,400 So these initiatives are more about prioritizing entertainment over practicality, admittedly, 547 00:40:39,400 --> 00:40:45,360 pending the results of the popularity, they could pave the way for new and innovative 548 00:40:45,360 --> 00:40:49,800 use cases as yet not seen. 549 00:40:49,800 --> 00:40:58,000 So the target market for this is likely to be Gen Z, who are particularly taken with 550 00:40:58,000 --> 00:41:05,520 character AIs companion chat bots, where you can chat with various different characters 551 00:41:05,520 --> 00:41:08,360 with their AI avatars. 552 00:41:08,360 --> 00:41:15,920 So this poses the question about whether it could be a significant trend in the influencer 553 00:41:15,920 --> 00:41:19,440 marketing space. 554 00:41:19,440 --> 00:41:27,040 If AI avatars, and we already see actually AI created, well actually I'm not going to 555 00:41:27,040 --> 00:41:36,600 say AI influences, it's more the CGI created and managed avatars are very big in the influencer 556 00:41:36,600 --> 00:41:37,600 world. 557 00:41:37,600 --> 00:41:43,800 Now, if we can do this with AI as well, who needs real people? 558 00:41:43,800 --> 00:41:49,920 Why pay real people for their influence when it could just be an AI created character? 559 00:41:49,920 --> 00:41:52,080 Yeah, what did you think about this one? 560 00:41:52,080 --> 00:41:54,680 How do you fancy a Tim Ferriss in your pocket? 561 00:41:54,680 --> 00:41:56,360 Yeah, it's interesting. 562 00:41:56,360 --> 00:42:01,520 So there's a sort of futurist that I followed called Peter Diamandis, and he's already got, 563 00:42:01,520 --> 00:42:06,400 I think it's called Peter Bot, which I've had a play with and there's two versions. 564 00:42:06,400 --> 00:42:11,560 The free version is like Chat GPT, but it's influenced by Peter Diamandis' writings and 565 00:42:11,560 --> 00:42:18,080 podcasts so it produces information related to what he usually talks about, but also with 566 00:42:18,080 --> 00:42:22,360 his trademark optimism and excitement, which is kind of nice. 567 00:42:22,360 --> 00:42:27,720 And then if you pay, it will actually create a Hey Jen-like version of Peter's who he 568 00:42:27,720 --> 00:42:33,280 actually speaks and it's video, which I think is just a really interesting exploration of 569 00:42:33,280 --> 00:42:35,680 this. 570 00:42:35,680 --> 00:42:41,720 And then I can see the continuation of big personalities, big celebrities, big thought 571 00:42:41,720 --> 00:42:48,080 leaders, the Simon Sinek's of the world doing something similar, either as part of their 572 00:42:48,080 --> 00:42:52,960 own offering or licensing their content and likenesses. 573 00:42:52,960 --> 00:42:55,840 But if you follow it even further, because I think one of the key things with all of 574 00:42:55,840 --> 00:42:59,720 this AI driven tech is just how quickly it becomes really cheap and anyone can basically 575 00:42:59,720 --> 00:43:06,680 do it, that if you've got a load of material on your own company website or your blog post 576 00:43:06,680 --> 00:43:10,080 and what have you, how do you approach this as a brand? 577 00:43:10,080 --> 00:43:16,720 Do you have like a new synthetic human AI influencer straight thought leader that's 578 00:43:16,720 --> 00:43:20,900 basically trained on all the content you've ever produced for your company? 579 00:43:20,900 --> 00:43:26,720 Do you choose one of your subject matter experts to be that person? 580 00:43:26,720 --> 00:43:29,120 I think we're going to see a lot more of that. 581 00:43:29,120 --> 00:43:33,600 I know you and I might get asked a lot about customer service chat bots and training them 582 00:43:33,600 --> 00:43:39,520 on internal company information that your human customer service reps would use to answer 583 00:43:39,520 --> 00:43:45,000 questions and this is like an extension of that I think wherein you can ask opinions 584 00:43:45,000 --> 00:43:49,260 and more thought leadership driven aspects. 585 00:43:49,260 --> 00:43:51,540 So I think that will be interesting. 586 00:43:51,540 --> 00:43:55,640 Will we see a revolt of where people are like, no, no, I really want to see what the real 587 00:43:55,640 --> 00:44:02,480 Simon Sinek's got to say, not the shallow AI interpretation of Simon Sinek. 588 00:44:02,480 --> 00:44:04,920 That remains to be seen. 589 00:44:04,920 --> 00:44:12,000 There's probably space for a Martin Broadhurst in my pocket and a real Martin Broadhurst 590 00:44:12,000 --> 00:44:17,180 on the podcast probably because they have different use cases, right? 591 00:44:17,180 --> 00:44:22,440 But I think it's kind of cool and again, hucking back to the beginning of the podcast, there 592 00:44:22,440 --> 00:44:27,400 are people who are, oh, I can do some stuff with chat GPT in there at that stage of development 593 00:44:27,400 --> 00:44:31,840 and sort of interest in all of this as a topic, but there'll be people now who are already 594 00:44:31,840 --> 00:44:38,200 starting to imagine those that are sort of surfing this wave right on the edge that how 595 00:44:38,200 --> 00:44:39,720 can I use this for my company? 596 00:44:39,720 --> 00:44:43,400 How can we use that as a mechanism of standing out from competitors? 597 00:44:43,400 --> 00:44:45,320 How can we use it to create bars? 598 00:44:45,320 --> 00:44:50,400 How can we use it to provide a better service to our customers? 599 00:44:50,400 --> 00:44:55,040 And I think those people will be, what they do with this is going to be what drives the 600 00:44:55,040 --> 00:45:02,040 future of this, is people figuring out useful applications and hopefully, and we talked 601 00:45:02,040 --> 00:45:08,000 about this on a previous episode, not getting caught out when somehow nefarious actors are 602 00:45:08,000 --> 00:45:13,000 able to trick prompt them into saying something or doing something that they shouldn't. 603 00:45:13,000 --> 00:45:21,240 Well, that was exactly what was going through my head just then is from a brand reputation 604 00:45:21,240 --> 00:45:26,720 perspective, there's quite a lot on the line for these individuals, right? 605 00:45:26,720 --> 00:45:32,440 It's one thing for a major corporation to have an automated chat bot that says something 606 00:45:32,440 --> 00:45:49,080 it shouldn't, but a hey Jen talking avatar, self-made, self-published, deep fake effectively, 607 00:45:49,080 --> 00:45:55,080 chatting with somebody offering a bit of support, guidance, insight, but one that just goes 608 00:45:55,080 --> 00:46:01,680 a bit off the rails and tells someone to kill themselves or what have you. 609 00:46:01,680 --> 00:46:02,960 That's a bad look. 610 00:46:02,960 --> 00:46:08,040 That's a bad look for your major high profile influencer. 611 00:46:08,040 --> 00:46:11,920 So that's something that you guarantee is going to happen at some point, right? 612 00:46:11,920 --> 00:46:12,920 I agree. 613 00:46:12,920 --> 00:46:15,660 I think that will definitely happen. 614 00:46:15,660 --> 00:46:22,600 Some chat bot somewhere with the likeness either in the voice of or even the video is 615 00:46:22,600 --> 00:46:24,120 going to, that is going to happen. 616 00:46:24,120 --> 00:46:26,840 They're going to find ways to trick it to say things that it shouldn't and it's going 617 00:46:26,840 --> 00:46:30,120 to be a major news story. 618 00:46:30,120 --> 00:46:35,960 How that influences brands and personalities and famous people as to whether they want 619 00:46:35,960 --> 00:46:40,280 to do this or not is going to be interesting. 620 00:46:40,280 --> 00:46:44,680 Honestly, the reason I haven't created, because as you know, I've got access to the hey Jen 621 00:46:44,680 --> 00:46:51,320 beta and the thing that's stopped me doing it so far is what happens when I have trained 622 00:46:51,320 --> 00:46:52,640 a model on me. 623 00:46:52,640 --> 00:46:57,160 There's now a synthetic version of me, which is fine when I'm in charge of it. 624 00:46:57,160 --> 00:47:02,800 I'm not particularly well known or important, but someone else could get hold of access 625 00:47:02,800 --> 00:47:07,240 to that and make me say and do things that I wouldn't normally do. 626 00:47:07,240 --> 00:47:16,560 I don't know who the producers of that tool are and what other plans they may have for 627 00:47:16,560 --> 00:47:18,960 my likeness. 628 00:47:18,960 --> 00:47:23,120 Probably none for the reasons already said, but it's enough to stop me doing it and my 629 00:47:23,120 --> 00:47:26,680 risk profile is extremely low. 630 00:47:26,680 --> 00:47:34,240 So Snoop Dogg, Tim Ferriss, do they have more to worry about? 631 00:47:34,240 --> 00:47:35,240 Probably. 632 00:47:35,240 --> 00:47:38,880 I think it'd be interesting to see how it plays out. 633 00:47:38,880 --> 00:47:39,880 One more story this week. 634 00:47:39,880 --> 00:47:40,880 You probably get a feel-lessness. 635 00:47:40,880 --> 00:47:45,440 It's been a little bit of a lighter week actually on the AI stories, but we've got one more, 636 00:47:45,440 --> 00:47:49,880 which is actually a continuation of ChatGPT Vision. 637 00:47:49,880 --> 00:47:54,120 We talked about this a couple of weeks ago, some of the cool early use cases. 638 00:47:54,120 --> 00:47:58,560 We mentioned one at the beginning of the podcast, feeding it a picture of a table and then actually 639 00:47:58,560 --> 00:48:03,360 having it engineered into a real table in Excel that you can actually edit. 640 00:48:03,360 --> 00:48:08,080 But what we've spotted this week is a few other really interesting use cases floating 641 00:48:08,080 --> 00:48:13,480 around the web of how as a marketer, you can use ChatGPT's new vision capabilities, not 642 00:48:13,480 --> 00:48:18,120 just some of the cool more general use cases that were popping up. 643 00:48:18,120 --> 00:48:26,480 So I think one of the ones I really like, Martin, was marketers taking screenshots of 644 00:48:26,480 --> 00:48:33,480 high-performing landing pages or websites that they really admire and then asking ChatGPT 645 00:48:33,480 --> 00:48:39,120 Vision to break the website down and use it as an exploration as to why it might be working 646 00:48:39,120 --> 00:48:45,360 well and what features to emulate or asking it for critical feedback on your own website. 647 00:48:45,360 --> 00:48:50,600 The difference here being between just pasting copy in, you'll get feedback on your copy. 648 00:48:50,600 --> 00:48:55,280 But if you take a screen grab of your site, because ChatGPT's vision's capabilities 649 00:48:55,280 --> 00:49:01,720 are so incredible, it can actually see your website and talk about spacing, colors, design 650 00:49:01,720 --> 00:49:03,960 and all of those other things. 651 00:49:03,960 --> 00:49:05,780 Structural elements, which I thought was awesome. 652 00:49:05,780 --> 00:49:13,920 You can also apply that to extending it to other materials like PowerPoint slide decks 653 00:49:13,920 --> 00:49:15,480 and all these other things. 654 00:49:15,480 --> 00:49:19,320 Just take a screen grab and then you can get critique. 655 00:49:19,320 --> 00:49:22,280 Of course, it can read the copy in those as well. 656 00:49:22,280 --> 00:49:29,760 So you're getting design input, you're getting your copy reviewed all in one go. 657 00:49:29,760 --> 00:49:34,240 Yeah, I actually did the landing page one, took a mobile screenshot of a landing page 658 00:49:34,240 --> 00:49:42,120 that I created, put it into GPT for vision and it was the spacing that just caught me 659 00:49:42,120 --> 00:49:46,680 off guard actually because I was expecting it to critique the copy and the flow and the 660 00:49:46,680 --> 00:49:47,680 general structure. 661 00:49:47,680 --> 00:49:52,320 But actually it highlighted a section on it where it said it was all a little bit cramped 662 00:49:52,320 --> 00:49:54,440 and the padding and the margin was off. 663 00:49:54,440 --> 00:49:57,960 I hadn't really considered that. 664 00:49:57,960 --> 00:49:59,880 Nice, nice. 665 00:49:59,880 --> 00:50:06,120 I think it's going to be really useful because just having nearly worked 15 years in agencies, 666 00:50:06,120 --> 00:50:11,380 some people seem to develop a really good design eye for balance, spacing. 667 00:50:11,380 --> 00:50:15,240 Some people have that attention to detail where they can see that the margin's off, 668 00:50:15,240 --> 00:50:18,000 but your ability to just feed that through a tool. 669 00:50:18,000 --> 00:50:21,240 I mean, at the moment, this is all very piecemeal, right? 670 00:50:21,240 --> 00:50:23,400 One by one screen grab. 671 00:50:23,400 --> 00:50:27,640 But if OpenAI open this up as an API, which one assumes they probably will given they 672 00:50:27,640 --> 00:50:32,360 have pretty much all the rest of their tools, then at some point someone's going to build 673 00:50:32,360 --> 00:50:38,240 a tool that automatically screen grabs every page of your website, feeds every page through 674 00:50:38,240 --> 00:50:42,800 a tool and then gives you critical feedback on how to improve it. 675 00:50:42,800 --> 00:50:47,200 And given the example you gave us from Draspa earlier with its one-click content upload 676 00:50:47,200 --> 00:50:52,520 deployments, how long before you can just one-click fix a load of padding issues on 677 00:50:52,520 --> 00:50:56,480 your website because this vision tool was able to detect them all. 678 00:50:56,480 --> 00:50:59,360 So that's pretty awesome. 679 00:50:59,360 --> 00:51:03,560 The other thing I saw this week, which I thought was nice is, and I guess it's kind of obvious 680 00:51:03,560 --> 00:51:07,120 and we've talked about it a little bit off air, but for those people that love to do 681 00:51:07,120 --> 00:51:12,600 like right hand written notes to themselves or brainstorm on a whiteboard. 682 00:51:12,600 --> 00:51:18,360 And I've taken so many pictures over the years at workshops of like sticky notes on a wall 683 00:51:18,360 --> 00:51:20,360 and all this other stuff. 684 00:51:20,360 --> 00:51:25,040 But the ability of ChatGPT Vision to actually just suck all the text out of those and turn 685 00:51:25,040 --> 00:51:29,160 them into like written notes that you can now use digitally to share as a follow-up 686 00:51:29,160 --> 00:51:34,760 after the meeting or for you to index and search later makes it a pretty powerful tool 687 00:51:34,760 --> 00:51:35,760 for that. 688 00:51:35,760 --> 00:51:40,480 And I've played with that yet definitely worth a play. 689 00:51:40,480 --> 00:51:44,880 That is a real benefit. 690 00:51:44,880 --> 00:51:49,080 If you're in a meeting, so I was at a client meeting, I tend not to make handwritten notes 691 00:51:49,080 --> 00:51:53,240 because I just don't really, to be honest, I never go back and check them. 692 00:51:53,240 --> 00:51:56,840 I'm not as diligent in going back and reading them as I should be. 693 00:51:56,840 --> 00:52:03,480 But I've took the photo, extracted the text and yeah, just took it into my CRM and all 694 00:52:03,480 --> 00:52:04,480 was good. 695 00:52:04,480 --> 00:52:06,720 Yeah, I think it's really, really cool use case. 696 00:52:06,720 --> 00:52:11,520 And I think there's going to be loads more that come out of this. 697 00:52:11,520 --> 00:52:19,040 Another screen grab one was people who were not so O-FAY with like a software tool, screen 698 00:52:19,040 --> 00:52:23,940 grabbing it and then asking for an explanation on how to use it or how to interpret information 699 00:52:23,940 --> 00:52:24,940 in it. 700 00:52:24,940 --> 00:52:27,280 So the example I saw was Google Analytics dashboard. 701 00:52:27,280 --> 00:52:31,400 But of course, I think you could apply this to any analytical software that you're perhaps 702 00:52:31,400 --> 00:52:33,540 not O-FAY with. 703 00:52:33,540 --> 00:52:38,400 And in the example, I saw someone took a picture of their Google Analytics dashboard and were 704 00:52:38,400 --> 00:52:42,640 asking questions about what the different stats mean and what the trends might mean 705 00:52:42,640 --> 00:52:44,320 for their marketing and stuff like that. 706 00:52:44,320 --> 00:52:49,060 So I thought that was pretty cool one as well. 707 00:52:49,060 --> 00:52:56,120 So yeah, I think we always talk about, don't we Martin on the podcast, that using ChatGPT 708 00:52:56,120 --> 00:53:00,120 is as much about user imagination as it is the power of the tool because these things 709 00:53:00,120 --> 00:53:03,200 don't come with your traditional software manual. 710 00:53:03,200 --> 00:53:08,040 They don't have like preset eight things they can do and then you learn them and then you 711 00:53:08,040 --> 00:53:10,500 know how to use it. 712 00:53:10,500 --> 00:53:14,280 How you interact and prompt these tools and the creativity you bring to that process is 713 00:53:14,280 --> 00:53:16,960 going to open up new applications for you. 714 00:53:16,960 --> 00:53:25,160 And now that ChatGPT can see, not just hear and read, really opens up a lot of interesting 715 00:53:25,160 --> 00:53:28,320 things that we probably haven't even begun to think of what they might be. 716 00:53:28,320 --> 00:53:33,800 Yeah, so I had an example of that in one of the workshops this week where somebody, they 717 00:53:33,800 --> 00:53:40,480 worked in property and property investment and they were talking about trying to get 718 00:53:40,480 --> 00:53:47,880 investor packs put together and they often had various bits of artwork or they had some 719 00:53:47,880 --> 00:53:54,700 photography from developments that they were working on and they were just describing how 720 00:53:54,700 --> 00:53:59,560 much time it would take them to write a bit of copy around certain elements of these investment 721 00:53:59,560 --> 00:54:01,780 packs. 722 00:54:01,780 --> 00:54:08,020 And he said, could it do this for us? 723 00:54:08,020 --> 00:54:13,580 So we found a kind of fairly generic example of the kind of picture that he would typically 724 00:54:13,580 --> 00:54:18,620 use in his investment pack, put that in, wrote a few prompts, did it live in the session 725 00:54:18,620 --> 00:54:20,680 and there you go. 726 00:54:20,680 --> 00:54:25,820 It created the content for him and he couldn't really believe his eyes. 727 00:54:25,820 --> 00:54:31,760 So yeah, it's as much having the imagination and the creativity and identifying the problem 728 00:54:31,760 --> 00:54:34,820 and saying, could it do this thing? 729 00:54:34,820 --> 00:54:37,440 And then trying it. 730 00:54:37,440 --> 00:54:39,240 That's where you find the magic. 731 00:54:39,240 --> 00:54:41,800 Yeah, I agree. 732 00:54:41,800 --> 00:54:42,840 So there you have it folks. 733 00:54:42,840 --> 00:54:48,440 If you haven't had a play with ChatGPT Vision yet, the easiest way is actually to start 734 00:54:48,440 --> 00:54:53,240 screen grabbing things on your computer, save that as an image and start plugging it in. 735 00:54:53,240 --> 00:54:54,940 Think of interesting questions to ask. 736 00:54:54,940 --> 00:54:58,720 Even if it's, what color is this? 737 00:54:58,720 --> 00:55:02,500 You want to find a hex code, you could just take a screen grab of the button that you're 738 00:55:02,500 --> 00:55:05,800 trying to reproduce and ask for the hex code. 739 00:55:05,800 --> 00:55:11,920 You can grab images and say ChatGPT if I wanted to create an image like this, what would be 740 00:55:11,920 --> 00:55:15,240 like a good prompt that I could use and so it will reverse engineer prompts and give 741 00:55:15,240 --> 00:55:16,240 you prompt ideas. 742 00:55:16,240 --> 00:55:19,280 So yeah, it really is pretty awesome, pretty awesome tool. 743 00:55:19,280 --> 00:55:24,040 Well, I think that's all we've got time for this week, Martin. 744 00:55:24,040 --> 00:55:26,080 I think so. 745 00:55:26,080 --> 00:55:28,620 Well thanks always to you lovely listeners. 746 00:55:28,620 --> 00:55:33,120 We really appreciate you spending your time with us wherever you may be, dog walking in 747 00:55:33,120 --> 00:55:39,080 the gym, sat in your office, hanging out with your pals and saying, God, these are cool 748 00:55:39,080 --> 00:55:42,820 stories and sharing them with all your friends, which I'm sure everybody does. 749 00:55:42,820 --> 00:55:47,440 If you are not subscribed yet, get subscribed, visit ArtificiallyIntelligentMarketing.com. 750 00:55:47,440 --> 00:55:52,440 You can subscribe to the newsletter or you can subscribe to this on all your favorite 751 00:55:52,440 --> 00:55:54,080 podcasting platforms. 752 00:55:54,080 --> 00:55:56,240 Martin, thanks so much for your time. 753 00:55:56,240 --> 00:55:59,640 I look forward to hanging out with you again next week. 754 00:55:59,640 --> 00:56:02,880 I'll speak to you next week, Paul. 755 00:56:02,880 --> 00:56:06,360 Thank you for listening to Artificially Intelligent Marketing. 756 00:56:06,360 --> 00:56:12,400 To stay on top of the latest trends, tips and tools in the world of marketing AI, be 757 00:56:12,400 --> 00:56:14,160 sure to subscribe. 758 00:56:14,160 --> 00:56:42,480 We look forward to seeing you again next week.