1 00:00:00,000 --> 00:00:10,000 Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the 2 00:00:10,000 --> 00:00:15,740 latest trends, tips, and tools in the world of marketing AI, helping you get the best 3 00:00:15,740 --> 00:00:18,560 results from your marketing efforts. 4 00:00:18,560 --> 00:00:22,440 Now let's join our hosts, Paul Avery and Martin Broadhurst. 5 00:00:22,440 --> 00:00:25,000 Hello everyone. 6 00:00:25,000 --> 00:00:28,520 Welcome to episode nine of Artificially Intelligent Marketing. 7 00:00:28,520 --> 00:00:30,600 We're glad that you're here with us today. 8 00:00:30,600 --> 00:00:33,440 There's lots and lots of stuff for us to get through. 9 00:00:33,440 --> 00:00:36,640 We're going to look at some top stories where we're going to go into some depth. 10 00:00:36,640 --> 00:00:41,200 So today we're going to cover the EU's proposed new copyright laws for AI, very important 11 00:00:41,200 --> 00:00:43,680 for marketers to know what's going on there. 12 00:00:43,680 --> 00:00:50,880 We're going to look at OpenAI's release of Code Interpreter as a chat GPT plugin. 13 00:00:50,880 --> 00:00:52,280 That's pretty cool as well. 14 00:00:52,280 --> 00:00:55,940 Stability AI announcing Deep Floyd IF. 15 00:00:55,940 --> 00:00:59,980 This text image generation with text tool. 16 00:00:59,980 --> 00:01:04,160 We're going to look at tool of the week and we're going to look at the Bing sidebar on 17 00:01:04,160 --> 00:01:05,160 that one. 18 00:01:05,160 --> 00:01:09,740 And we're also going to look at a few other things that interested us like this anime 19 00:01:09,740 --> 00:01:16,280 movie trailer that someone made with Runway's Gen 2 text video tool, which was pretty cool. 20 00:01:16,280 --> 00:01:19,640 Outside of that, there was a bunch of other stuff that happened this week that we won't 21 00:01:19,640 --> 00:01:23,200 have time to go into detail about, but that we did still want to cover briefly on the 22 00:01:23,200 --> 00:01:24,200 podcast. 23 00:01:24,200 --> 00:01:26,560 So we're going to cover all these short snippets and we're going to look at some of those very 24 00:01:26,560 --> 00:01:29,280 quickly before we get into the big stories of the week. 25 00:01:29,280 --> 00:01:34,040 I'm going to crack through these, Martin, and if there's a particular one that you want 26 00:01:34,040 --> 00:01:38,280 to chime in on, jump in and let us know what your thoughts are. 27 00:01:38,280 --> 00:01:43,120 The first one was chat GPT is now allowed to operate in Italy again after introducing 28 00:01:43,120 --> 00:01:45,880 a number of privacy disclosures and controls in there. 29 00:01:45,880 --> 00:01:48,320 So that's back in play. 30 00:01:48,320 --> 00:01:54,640 Anthropic raises $580 million in their series B, which is further evidence of all the cash 31 00:01:54,640 --> 00:02:00,000 that continues to flood into the AI company space. 32 00:02:00,000 --> 00:02:06,960 Tech layoffs continue with Dropbox laying off 16% of people in its team and IBM, as 33 00:02:06,960 --> 00:02:09,160 we saw today, pausing hiring. 34 00:02:09,160 --> 00:02:16,440 And much of these layoffs are reported to be driven by augmenting existing staff with 35 00:02:16,440 --> 00:02:19,760 AI, although none of the companies are actually confirming this. 36 00:02:19,760 --> 00:02:23,560 And I guess in many ways it could be driven by some of the economic challenges in the 37 00:02:23,560 --> 00:02:24,560 market. 38 00:02:24,560 --> 00:02:29,440 But this is what people are thinking are happening is that AI is going to start to play a bigger 39 00:02:29,440 --> 00:02:31,720 role in the workforce as some of these companies. 40 00:02:31,720 --> 00:02:32,720 What are your thoughts on this one, Martin? 41 00:02:32,720 --> 00:02:34,680 I know you had some thoughts. 42 00:02:34,680 --> 00:02:39,440 Well it's a continuation of those tech layoffs that we've seen over the past 12 months or 43 00:02:39,440 --> 00:02:45,240 so, but this one is particularly interesting that the IBM one talking or commentators around 44 00:02:45,240 --> 00:02:50,360 the issue were saying that, yeah, this is AI and what they're looking to do is basically 45 00:02:50,360 --> 00:02:59,040 replace jobs that were more or less filing paperwork or moving data from one place to 46 00:02:59,040 --> 00:03:05,480 another, quite, you know, back office jobs, kind of admin jobs. 47 00:03:05,480 --> 00:03:11,800 They're the kind of roles that they're looking to replace and these will be automated using 48 00:03:11,800 --> 00:03:14,280 AI ML technology. 49 00:03:14,280 --> 00:03:20,040 So I think it's interesting to see whether those jobs are fully replaced. 50 00:03:20,040 --> 00:03:24,960 You know, sometimes we see big tech companies make huge rounds of layoffs and then before 51 00:03:24,960 --> 00:03:27,940 you know it, they're rehiring people again. 52 00:03:27,940 --> 00:03:34,160 So it was, you know, a bit of a full Xeron to go and do that in the first place. 53 00:03:34,160 --> 00:03:40,240 But yeah, if they are jobs that are lost for good, we're certainly starting to see the 54 00:03:40,240 --> 00:03:43,840 thin end of the wedge for AI job replacement. 55 00:03:43,840 --> 00:03:48,680 I don't think jobs are going to be replaced overnight, but we're starting to see a creep 56 00:03:48,680 --> 00:03:49,680 in. 57 00:03:49,680 --> 00:03:54,240 It did make me think about another topic that we, well, we covered this topic on a recent 58 00:03:54,240 --> 00:04:03,520 episode looking at the OpenAI report on disruption in the labour market through GPT technology. 59 00:04:03,520 --> 00:04:11,960 And it brought to mind an article that I read this week, which looked at a study that was 60 00:04:11,960 --> 00:04:21,440 done where healthcare professionals were asked to judge the responses to patient emails and 61 00:04:21,440 --> 00:04:24,040 they were shown two responses. 62 00:04:24,040 --> 00:04:29,760 One of the responses was from a real healthcare professional responding to the patient and 63 00:04:29,760 --> 00:04:30,760 their inquiry. 64 00:04:30,760 --> 00:04:34,880 And the other was a response written by Chad GPT and GPT-4. 65 00:04:34,880 --> 00:04:39,200 Now the healthcare assessor, as it were, didn't know which was which. 66 00:04:39,200 --> 00:04:41,300 So these were blind tests. 67 00:04:41,300 --> 00:04:49,920 And in 79% of cases, they rated Chad GPT's response as being the better response in terms 68 00:04:49,920 --> 00:04:56,960 of higher quality, in terms of the information provided and having a better understanding 69 00:04:56,960 --> 00:05:01,960 and higher empathy, which is fascinating to me. 70 00:05:01,960 --> 00:05:12,200 So we're starting to see real world peer reviewed or at least, you know, assessments done by 71 00:05:12,200 --> 00:05:17,360 people that know what they're talking about, looking at outputs from humans versus robots 72 00:05:17,360 --> 00:05:25,720 and saying the AI is as good or better in this case, better nearly four fifths of the 73 00:05:25,720 --> 00:05:26,720 time. 74 00:05:26,720 --> 00:05:27,720 Wow. 75 00:05:27,720 --> 00:05:28,720 I saw that story. 76 00:05:28,720 --> 00:05:30,960 I had not taken home the empathy part, right? 77 00:05:30,960 --> 00:05:36,680 If you're a blind assessor and it could be a human, you see empathy in it. 78 00:05:36,680 --> 00:05:42,340 It would be fascinating to run that experiment again where it was unblinded and would knowing 79 00:05:42,340 --> 00:05:45,440 it was a robot change someone's interpretation of the empathy. 80 00:05:45,440 --> 00:05:47,240 A, that would be interesting. 81 00:05:47,240 --> 00:05:48,920 B, does it matter? 82 00:05:48,920 --> 00:05:53,260 And would this therefore make an argument for not telling people when they're getting 83 00:05:53,260 --> 00:05:59,800 certain types of support by a robot almost as part of a placebo effect, right? 84 00:05:59,800 --> 00:06:06,080 Just like for mental health support, we talked about this a bit on the podcast previously. 85 00:06:06,080 --> 00:06:10,600 Rating the large language model as being more empathetic than the humans. 86 00:06:10,600 --> 00:06:14,400 That's a little bit mind blowing. 87 00:06:14,400 --> 00:06:15,400 Cool. 88 00:06:15,400 --> 00:06:18,720 Thank you for sharing that with mine. 89 00:06:18,720 --> 00:06:20,640 One last quick snippet. 90 00:06:20,640 --> 00:06:26,560 I saw a graphic this week, showing the evolution of mid journey from version one to version 91 00:06:26,560 --> 00:06:32,160 five and the quality of the images and the improvement, especially in the photo realism 92 00:06:32,160 --> 00:06:37,640 of the images and getting rid of some of the junk that told you it was an AI image like 93 00:06:37,640 --> 00:06:41,520 hands that had 42 fingers and stuff like that. 94 00:06:41,520 --> 00:06:45,680 That's kind of mind blowing because version one was launched just over a year ago and 95 00:06:45,680 --> 00:06:49,560 the rumors are that version six is probably going to come out in the next month or two, 96 00:06:49,560 --> 00:06:56,160 which would be in line with the release schedule around mid journey so far. 97 00:06:56,160 --> 00:06:59,200 We talk a lot on this podcast about the pace of change. 98 00:06:59,200 --> 00:07:00,320 You just have to look at it. 99 00:07:00,320 --> 00:07:03,860 If you can find online anywhere, just Google and find those graphics that show the comparison 100 00:07:03,860 --> 00:07:06,400 from V1 to V5. 101 00:07:06,400 --> 00:07:07,940 Yikes. 102 00:07:07,940 --> 00:07:11,120 Where will we be in six to 12 months? 103 00:07:11,120 --> 00:07:12,640 The pace of change is astonishing. 104 00:07:12,640 --> 00:07:19,400 On the way back from a client meeting this afternoon, I was watching a YouTube video. 105 00:07:19,400 --> 00:07:22,400 I highly recommend everyone go and check it out. 106 00:07:22,400 --> 00:07:29,440 The title is The AI Dilemma and it was filmed at the San Francisco event on March the 9th 107 00:07:29,440 --> 00:07:32,440 and it's had 1.3 million views already. 108 00:07:32,440 --> 00:07:40,680 Basically it's two of the journalists or whoever it was, the people have put together the Netflix 109 00:07:40,680 --> 00:07:48,040 film The Social Dilemma and they're now looking at AI and kind of presenting where AI is. 110 00:07:48,040 --> 00:07:56,360 Now I should say this came out or this talk was done before GPT-4 was announced. 111 00:07:56,360 --> 00:08:04,200 In that they talk about basically the whole theme is AI poses a catastrophic risk to humanity, 112 00:08:04,200 --> 00:08:07,320 which is a recurring theme. 113 00:08:07,320 --> 00:08:13,640 Every week we have to have the doomsayers, the doomsayers call it what you will. 114 00:08:13,640 --> 00:08:19,640 But actually one of the things that they talk about in this is we're hitting double exponential 115 00:08:19,640 --> 00:08:23,240 territory in terms of the capabilities of AI. 116 00:08:23,240 --> 00:08:29,120 They said it's really hard for people to grasp quite what that means. 117 00:08:29,120 --> 00:08:37,720 Even people that are working in AI ML research day in, day out, even though they rationally 118 00:08:37,720 --> 00:08:44,440 understand what exponential growth and exponential capabilities of development, they understand 119 00:08:44,440 --> 00:08:45,440 what it means. 120 00:08:45,440 --> 00:08:49,000 They struggle to bring that into the kind of reality. 121 00:08:49,000 --> 00:08:57,800 So they give an example of AI experts that were asked to say when they thought AI would 122 00:08:57,800 --> 00:09:08,320 be capable of more than 50% of the time being able to answer competitive maths questions. 123 00:09:08,320 --> 00:09:15,200 So like real high level, cutting edge, leading edge maths, competition level maths questions. 124 00:09:15,200 --> 00:09:18,720 And they got it wrong by years. 125 00:09:18,720 --> 00:09:24,600 They said, I think it was in the, they thought it was going to be four years. 126 00:09:24,600 --> 00:09:31,640 That was it, and it was actually one year from when the survey was done. 127 00:09:31,640 --> 00:09:36,560 And then they're now saying, so this was a few years ago, they're now saying that the 128 00:09:36,560 --> 00:09:43,400 AI capability is beating the questions quicker than they can really be written, like 80, 129 00:09:43,400 --> 00:09:47,000 90% of the time. 130 00:09:47,000 --> 00:09:49,920 This is, and we're seeing this kind of in the real world. 131 00:09:49,920 --> 00:09:53,840 We're seeing this happen in real time in front of us and all of the products that we're using 132 00:09:53,840 --> 00:09:55,080 day to day. 133 00:09:55,080 --> 00:10:01,000 This growth in capabilities mid journey, as you say, the transition from one to five to 134 00:10:01,000 --> 00:10:11,640 six around the corner is breathtaking and kind of exciting, frightening. 135 00:10:11,640 --> 00:10:14,600 I'm glad, yeah, I would agree with it in that order. 136 00:10:14,600 --> 00:10:19,720 But I definitely, yeah, we've talked ad infinitum about the risks of AI and some of the other 137 00:10:19,720 --> 00:10:23,240 things that we try not to go into too much detail on this podcast. 138 00:10:23,240 --> 00:10:25,760 Although they should absolutely be recognized. 139 00:10:25,760 --> 00:10:32,000 Yeah, so there's a guy, I don't know if you know him, called Peter Diamandis, and he's 140 00:10:32,000 --> 00:10:39,120 like a futurist to a certain extent, but he's got a PhD in some biological discipline. 141 00:10:39,120 --> 00:10:41,080 He is a medical doctor. 142 00:10:41,080 --> 00:10:42,560 He founded the X Prize. 143 00:10:42,560 --> 00:10:46,560 I don't know if you heard of the X Prize. 144 00:10:46,560 --> 00:10:51,440 And he is an investor in a number of different companies in a number of different areas. 145 00:10:51,440 --> 00:10:58,040 His whole thing is about basically the exponential change in a number of areas in the world, 146 00:10:58,040 --> 00:11:01,320 drifting towards a world of abundance because he's very positive about the future. 147 00:11:01,320 --> 00:11:04,080 He's got a podcast called Moonshots and Mindsets. 148 00:11:04,080 --> 00:11:05,080 People should look that up. 149 00:11:05,080 --> 00:11:06,080 It's really, really interesting. 150 00:11:06,080 --> 00:11:11,640 But he always talks about how our brains are linear. 151 00:11:11,640 --> 00:11:17,040 We count the thing that our brains are organized and evolved to count the things in front of 152 00:11:17,040 --> 00:11:19,000 us. 153 00:11:19,000 --> 00:11:26,040 Four trees over there in that forest, 16 trees over there in that forest, 32 stones on the 154 00:11:26,040 --> 00:11:27,040 floor. 155 00:11:27,040 --> 00:11:34,320 We cannot imagine exponential change because you don't easily see it in front of you in 156 00:11:34,320 --> 00:11:38,520 nature in terms of macro scale, right? 157 00:11:38,520 --> 00:11:40,080 Just looking. 158 00:11:40,080 --> 00:11:45,040 You could probably observe exponential division of cells, but even that happens over a time 159 00:11:45,040 --> 00:11:47,240 period that's quite hard for a human to sit and watch. 160 00:11:47,240 --> 00:11:50,120 You'd have to time lapse it, you'd need a microscope. 161 00:11:50,120 --> 00:11:55,000 So yeah, it's a funny old world and it makes it hard to imagine because the other thing 162 00:11:55,000 --> 00:11:59,760 is we should also still remember that humans have been very good at predicting we'd have 163 00:11:59,760 --> 00:12:04,280 flying cars by like 1999 in the 60s and stuff like this. 164 00:12:04,280 --> 00:12:13,460 So we're both simultaneously spectacularly overly optimistic and yet for certain changes 165 00:12:13,460 --> 00:12:16,840 unable to comprehend how fast they're going to move. 166 00:12:16,840 --> 00:12:17,840 That's fascinating. 167 00:12:17,840 --> 00:12:20,600 I guess that's what keeps all the futurists in work. 168 00:12:20,600 --> 00:12:25,960 But yeah, let's move on to the actual bigger stories of the week. 169 00:12:25,960 --> 00:12:33,080 The first one is the EU's proposed new copyright laws for AI and for all of you working in 170 00:12:33,080 --> 00:12:37,560 marketing who are thinking about using generative AI tools to produce content and images and 171 00:12:37,560 --> 00:12:43,720 other things for your brand, understanding the copyright landscape is critical. 172 00:12:43,720 --> 00:12:47,920 Tell us a bit about what you learned from this recent story, Mike. 173 00:12:47,920 --> 00:12:53,520 This is a big development and we have covered the regulation changes and the AI act that 174 00:12:53,520 --> 00:12:56,040 the EU has been bringing out for a while now. 175 00:12:56,040 --> 00:13:02,400 It's been in the works for I think a couple of years and it's the EU commission's attempt 176 00:13:02,400 --> 00:13:05,240 at addressing AI technology regulation. 177 00:13:05,240 --> 00:13:12,640 And one of the things that to date they've had in the regulation is that tools needed 178 00:13:12,640 --> 00:13:18,960 to have, where AI or ML technology was being used, it had to have some sort of risk classification. 179 00:13:18,960 --> 00:13:23,200 So it would have to say high risk, low risk, medium risk. 180 00:13:23,200 --> 00:13:28,080 So for instance, if it was like spam detection on emails, that would be low risk. 181 00:13:28,080 --> 00:13:33,720 And if it was saying who can get healthcare insurance, that would be high risk, those 182 00:13:33,720 --> 00:13:35,700 kinds of things. 183 00:13:35,700 --> 00:13:42,640 But this new development in the regulation has been brought about specifically in response 184 00:13:42,640 --> 00:13:50,340 to generative AI and it addresses the big topic of copyright. 185 00:13:50,340 --> 00:13:55,880 So as we know, training data for large language models and text to image generators and all 186 00:13:55,880 --> 00:13:59,320 of these kinds of technologies consumes lots of copyright data. 187 00:13:59,320 --> 00:14:00,640 They crawl the web. 188 00:14:00,640 --> 00:14:04,320 They watch lots of video. 189 00:14:04,320 --> 00:14:11,680 We've seen stability and stable diffusion creating image with shusha stock, watermarks 190 00:14:11,680 --> 00:14:12,680 all over them. 191 00:14:12,680 --> 00:14:15,000 So we know that it is doing that. 192 00:14:15,000 --> 00:14:23,360 And what the EU is now saying is that where models have been trained using copyright data, 193 00:14:23,360 --> 00:14:28,640 the people who have produced the model or trained the model have to disclose that as 194 00:14:28,640 --> 00:14:34,640 part of the, they just have to disclose that. 195 00:14:34,640 --> 00:14:41,040 This is an interesting solution given where they could have landed because there were 196 00:14:41,040 --> 00:14:47,880 some people that were calling for an outright ban. 197 00:14:47,880 --> 00:14:53,360 And until these tools can be sorted out, some of the more extreme voices were saying that 198 00:14:53,360 --> 00:14:56,880 they should just ban it completely. 199 00:14:56,880 --> 00:15:02,720 And this is seen as being a more moderating approach. 200 00:15:02,720 --> 00:15:08,960 And I think it serves the good middle ground. 201 00:15:08,960 --> 00:15:17,720 I certainly think from a marketer's perspective or a content producer or a content creator's 202 00:15:17,720 --> 00:15:21,800 perspective, we need to start thinking a bit more like Wired. 203 00:15:21,800 --> 00:15:27,320 So Wired magazine's publication about how they're using generative AI spoke about that 204 00:15:27,320 --> 00:15:30,040 they're not, spoke exactly how they are going to use it. 205 00:15:30,040 --> 00:15:33,920 And one of the things they said for text to image generation was they're not going to 206 00:15:33,920 --> 00:15:37,960 use it until the copyright issue is resolved. 207 00:15:37,960 --> 00:15:42,640 Now all of a sudden we can start to see that, well, actually the copyright issue is going 208 00:15:42,640 --> 00:15:47,560 to be a real differentiator for people using a model or not. 209 00:15:47,560 --> 00:15:50,720 If Wired magazine are saying, well, we're not going to use models that have been trained 210 00:15:50,720 --> 00:15:58,680 on copyright data, a tool like Adobe's Firefly, which has as one of its leading value propositions, 211 00:15:58,680 --> 00:16:02,440 we are not training our models on copyrighted data. 212 00:16:02,440 --> 00:16:06,080 Well that becomes a real sell because you think, oh, well, I will use that and I'm not 213 00:16:06,080 --> 00:16:08,660 at risk of ripping anybody off. 214 00:16:08,660 --> 00:16:14,280 So that becomes a competitive advantage for people that are creating these models that 215 00:16:14,280 --> 00:16:22,560 aren't trained on copyrighted data, which leads you to think that the likes of Shutterstock, 216 00:16:22,560 --> 00:16:29,560 the likes of Getty Images, the likes of Adobe, who have access to huge amounts of these kind 217 00:16:29,560 --> 00:16:33,640 of data sets that they own. 218 00:16:33,640 --> 00:16:35,920 Big image repositories and stuff. 219 00:16:35,920 --> 00:16:36,920 Yeah. 220 00:16:36,920 --> 00:16:43,440 They will be able to create the better, more compliant models in the future. 221 00:16:43,440 --> 00:16:44,440 Yeah. 222 00:16:44,440 --> 00:16:46,800 It's absolutely fascinating. 223 00:16:46,800 --> 00:16:53,960 I think I can imagine a world where certainly the bigger brands almost have to lean into 224 00:16:53,960 --> 00:16:55,400 those types of tools. 225 00:16:55,400 --> 00:17:00,960 It's just not worth their time to get caught up in any of these copyright issues. 226 00:17:00,960 --> 00:17:07,800 My experience playing with and reading about and looking at the ones that are trained on 227 00:17:07,800 --> 00:17:17,240 data that's owned by like Shutterstock images or Adobe is that they are limited in comparison. 228 00:17:17,240 --> 00:17:20,920 I think I saw online an example where someone was trying to create some interesting images 229 00:17:20,920 --> 00:17:22,280 with Deadpool. 230 00:17:22,280 --> 00:17:25,760 And of course, I think it was Adobe Firefly. 231 00:17:25,760 --> 00:17:26,760 Basically it couldn't. 232 00:17:26,760 --> 00:17:34,880 It didn't know what Deadpool was because that's an owned likeness by Disney at this point, 233 00:17:34,880 --> 00:17:37,280 I would guess. 234 00:17:37,280 --> 00:17:39,520 And so it couldn't be in the training set. 235 00:17:39,520 --> 00:17:40,520 Right. 236 00:17:40,520 --> 00:17:44,520 But of course, those that are using images from here, there and everywhere could do it. 237 00:17:44,520 --> 00:17:49,100 So I could imagine a world in which smaller businesses that were willing to risk it could 238 00:17:49,100 --> 00:17:53,800 probably create quite maybe higher quality potentially as well images. 239 00:17:53,800 --> 00:17:55,400 But that's the risk that they're taking. 240 00:17:55,400 --> 00:17:59,280 Whereas bigger brands probably lean into these types of images. 241 00:17:59,280 --> 00:18:03,440 We've talked previously, how will this end up? 242 00:18:03,440 --> 00:18:10,960 Probably a little bit like Spotify, like some type of mechanism where everybody whose image 243 00:18:10,960 --> 00:18:19,400 was used in the training data set gets 0.00000100010001 pence every time someone generates an image 244 00:18:19,400 --> 00:18:22,040 and there's some sort of fees that come out the back of it. 245 00:18:22,040 --> 00:18:25,340 I still think we may see something like that, to be honest. 246 00:18:25,340 --> 00:18:29,060 Did you see the Grimes updates? 247 00:18:29,060 --> 00:18:30,120 Tell us about it. 248 00:18:30,120 --> 00:18:36,360 So yeah, the Grimes, I don't really, I'm not even going to pretend to really know who Grimes 249 00:18:36,360 --> 00:18:39,680 is other than they're, I think, a pop star. 250 00:18:39,680 --> 00:18:41,680 I sound so old. 251 00:18:41,680 --> 00:18:44,880 They make the popular music that goes into the hit parade. 252 00:18:44,880 --> 00:18:45,880 I think she has cassette singles. 253 00:18:45,880 --> 00:18:48,880 I think they're on the wireless. 254 00:18:48,880 --> 00:18:49,880 Yes. 255 00:18:49,880 --> 00:18:52,280 You can get the one CD single. 256 00:18:52,280 --> 00:18:57,200 And is Grimes married to Elon Musk? 257 00:18:57,200 --> 00:18:58,200 Is that the right person? 258 00:18:58,200 --> 00:18:59,200 I think was. 259 00:18:59,200 --> 00:19:03,000 Hard to keep up. 260 00:19:03,000 --> 00:19:14,040 So yeah, Grimes has said that feel free to make music using AI versions of Grimes' voice. 261 00:19:14,040 --> 00:19:15,600 Absolutely. 262 00:19:15,600 --> 00:19:20,640 Go wild and Grimes will share any revenue with you 50-50. 263 00:19:20,640 --> 00:19:27,280 It's a very interesting approach and I think very smart because if she doesn't have to 264 00:19:27,280 --> 00:19:33,800 do any, you know, or they don't have to do any work, then, and then they made the revenue 265 00:19:33,800 --> 00:19:36,480 and also the person who did the work gets some revenue. 266 00:19:36,480 --> 00:19:38,360 Yeah, absolutely fascinating. 267 00:19:38,360 --> 00:19:43,360 How some of these innovative business models come about is going to be fascinating. 268 00:19:43,360 --> 00:19:47,840 I would really want to keep an eye on the early movers who put themselves out there 269 00:19:47,840 --> 00:19:53,280 and say things like that because I think it will be fascinating to see how it plays out. 270 00:19:53,280 --> 00:19:54,960 Yeah, good story. 271 00:19:54,960 --> 00:19:58,000 Thanks for flagging that one for us, mine. 272 00:19:58,000 --> 00:20:04,680 Story number two is OpenAI has made a new plugin available called Code Interpreter through 273 00:20:04,680 --> 00:20:05,680 ChatGPT. 274 00:20:05,680 --> 00:20:10,600 So for those that are on the wait list like me still waiting to get access to plugins, 275 00:20:10,600 --> 00:20:13,960 please give me access to plugins so I can have a play that could tell all the lovely 276 00:20:13,960 --> 00:20:19,360 people who listen to this podcast and give them the absolute honest review of just how 277 00:20:19,360 --> 00:20:24,000 good or not they are because I read that they're good at times and still really buggy at other 278 00:20:24,000 --> 00:20:26,120 times. 279 00:20:26,120 --> 00:20:34,600 This new one is making a few waves this week because it allows you to analyze and interpret 280 00:20:34,600 --> 00:20:36,000 data. 281 00:20:36,000 --> 00:20:40,320 So I've seen some reports suggesting it can deliver the same sort of value as a junior 282 00:20:40,320 --> 00:20:42,740 level data analyst. 283 00:20:42,740 --> 00:20:47,580 The inference being it could replace a junior level data analyst. 284 00:20:47,580 --> 00:20:54,320 The way that it works is you upload a data file via CSV and then you ask for trends, 285 00:20:54,320 --> 00:20:56,380 insights on the data. 286 00:20:56,380 --> 00:20:59,460 You can get it to plot things for you. 287 00:20:59,460 --> 00:21:02,280 And all of this is driven by natural language prompts. 288 00:21:02,280 --> 00:21:08,020 So a lot of tools have been promising some level of automation and AI and the ability 289 00:21:08,020 --> 00:21:13,320 to do data analysis without having to be a coder, but actually you still needed to know 290 00:21:13,320 --> 00:21:15,680 how to manipulate SQL and all this other stuff. 291 00:21:15,680 --> 00:21:17,880 Whereas with this plugin, it's natural language. 292 00:21:17,880 --> 00:21:22,160 You upload it, you ask a question of the data and you get an answer. 293 00:21:22,160 --> 00:21:24,520 You can ask it to provide you with a report. 294 00:21:24,520 --> 00:21:30,680 I saw an example where someone basically asked code interpreters to pull out what it thought 295 00:21:30,680 --> 00:21:36,200 was two or three key trends from a large bulk of data and then write an abstract for the 296 00:21:36,200 --> 00:21:40,040 paper that they would draft based on it. 297 00:21:40,040 --> 00:21:46,100 And then I think also, I think they even drafted the paper using TrackGBT as well. 298 00:21:46,100 --> 00:21:49,680 So this is absolutely fascinating. 299 00:21:49,680 --> 00:21:52,760 Why is it important for marketers? 300 00:21:52,760 --> 00:21:55,900 Fundamentally, it makes it easier for marketers. 301 00:21:55,900 --> 00:22:00,040 Many of us don't know, we certainly don't, how to manipulate data using SQL and other 302 00:22:00,040 --> 00:22:01,840 fancy techniques. 303 00:22:01,840 --> 00:22:08,720 So it could bring that to non-developer marketers, which I think is really interesting. 304 00:22:08,720 --> 00:22:14,440 And I think making data analysis open and accessible like this is especially powerful 305 00:22:14,440 --> 00:22:20,600 for small and mid-size businesses that would never have perhaps interrogated their data 306 00:22:20,600 --> 00:22:21,600 in this way. 307 00:22:21,600 --> 00:22:27,200 If you don't have a data analysis team, because you're a small or mid-size company, you just 308 00:22:27,200 --> 00:22:28,200 don't do it. 309 00:22:28,200 --> 00:22:32,120 And that's been my experience anyway, because you don't have the bandwidth anyway, because 310 00:22:32,120 --> 00:22:35,240 you've got a hundred other things to do and you don't have the expertise in the business. 311 00:22:35,240 --> 00:22:42,640 But now, if you can basically outsource that to the level of junior data analyst in TrackGPT 312 00:22:42,640 --> 00:22:47,240 with this plugin, you can start to surface insights in your data that you might not have 313 00:22:47,240 --> 00:22:49,920 otherwise. 314 00:22:49,920 --> 00:22:50,920 So that's pretty cool. 315 00:22:50,920 --> 00:22:57,440 I think the one caveat that springs instantly to mind is the caveat of TrackGPT forever 316 00:22:57,440 --> 00:23:00,600 and ever and ever, which is, will it hallucinate? 317 00:23:00,600 --> 00:23:03,360 Will it report trends that weren't there? 318 00:23:03,360 --> 00:23:07,960 If it's only able to perform at a certain level, it's best piloted. 319 00:23:07,960 --> 00:23:12,480 This has been my experience with TrackGPT at least, by someone who has more insight than 320 00:23:12,480 --> 00:23:13,480 the tool, right? 321 00:23:13,480 --> 00:23:18,360 You ask it to put together 10 points for a blog post on a particular topic. 322 00:23:18,360 --> 00:23:22,640 You need to be an expert yourself in the topic to make sure that those 10 points are all 323 00:23:22,640 --> 00:23:27,360 accurate and relevant and interesting and that it's not full of hallucinations and junk. 324 00:23:27,360 --> 00:23:30,480 And I could imagine that could end up being the case here. 325 00:23:30,480 --> 00:23:37,280 So the reality is it might help senior data analysts get more done and it might not quite 326 00:23:37,280 --> 00:23:41,400 open up the power to non-developer data analyst people potentially. 327 00:23:41,400 --> 00:23:43,400 But yeah, so I thought that was fascinating. 328 00:23:43,400 --> 00:23:44,840 Any thoughts on this, Brian? 329 00:23:44,840 --> 00:23:55,320 I think actually using it as a non-data analyst will be a bit of a stretch for a lot of people 330 00:23:55,320 --> 00:23:57,080 for the reasons that you've identified. 331 00:23:57,080 --> 00:23:59,840 But I think you don't know when you don't know. 332 00:23:59,840 --> 00:24:01,080 That's the issue. 333 00:24:01,080 --> 00:24:04,560 So you can stick a load of data and then be like, tell me some things or produce some 334 00:24:04,560 --> 00:24:05,920 interesting graphs. 335 00:24:05,920 --> 00:24:11,160 But when I was, this sprung to mind when I was reading a thread, I've got a Twitter thread 336 00:24:11,160 --> 00:24:15,840 open now and it says, where is it? 337 00:24:15,840 --> 00:24:21,680 So creating charts to basic video editing to converting files, it does it all. 338 00:24:21,680 --> 00:24:22,680 Basic video editing, what? 339 00:24:22,680 --> 00:24:23,680 How does that even work? 340 00:24:23,680 --> 00:24:26,560 I don't even understand how that makes sense. 341 00:24:26,560 --> 00:24:31,280 That's just demonstrating my lack of knowledge about what a tool like this should be able 342 00:24:31,280 --> 00:24:32,560 to do. 343 00:24:32,560 --> 00:24:36,980 It goes on to give some really interesting visualizations that it does plotting onto 344 00:24:36,980 --> 00:24:42,640 maps and kind of showing geographic data and spatial analysis and all sorts of kind of 345 00:24:42,640 --> 00:24:43,960 cool things. 346 00:24:43,960 --> 00:24:48,960 But if you're not a data analyst, you don't know what graph to... 347 00:24:48,960 --> 00:24:51,800 You're probably talking about, give me a pie chart. 348 00:24:51,800 --> 00:24:52,800 Give me a pie chart. 349 00:24:52,800 --> 00:24:54,040 Give me a bar chart. 350 00:24:54,040 --> 00:25:02,360 Whereas this thing is laughing at your ineptitude when you ask it to give you a basic pie chart. 351 00:25:02,360 --> 00:25:03,600 I think you're right. 352 00:25:03,600 --> 00:25:11,960 One of the things I've noticed as I've got older is smart people ask great questions. 353 00:25:11,960 --> 00:25:17,080 And if you don't know what great questions to ask of your data, then the insights are 354 00:25:17,080 --> 00:25:20,920 going to be, you're not going to access them. 355 00:25:20,920 --> 00:25:29,000 And the ability...the example I mentioned did see ChatGPT's code interpreter try and 356 00:25:29,000 --> 00:25:35,280 pull insights from the data by asking itself questions about the data. 357 00:25:35,280 --> 00:25:39,220 But obviously some people have reported that it's more like a junior analyst. 358 00:25:39,220 --> 00:25:46,760 So one assumes its ability to ask great questions and great follow up questions. 359 00:25:46,760 --> 00:25:47,760 They always say like... 360 00:25:47,760 --> 00:25:48,760 Who's that? 361 00:25:48,760 --> 00:25:51,280 Who's that guy? 362 00:25:51,280 --> 00:25:54,600 If you really want to get to the number of the subject, there's models like the five 363 00:25:54,600 --> 00:25:55,600 Ys. 364 00:25:55,600 --> 00:25:59,520 It's not the first question where the real interesting stuff is. 365 00:25:59,520 --> 00:26:01,440 Okay, well what drove that? 366 00:26:01,440 --> 00:26:02,440 And what drove that? 367 00:26:02,440 --> 00:26:03,680 And what are the underlying factors of that? 368 00:26:03,680 --> 00:26:06,200 And then you get something interesting. 369 00:26:06,200 --> 00:26:08,520 So yeah, can these tools really do that? 370 00:26:08,520 --> 00:26:09,600 I think you're right, Martin. 371 00:26:09,600 --> 00:26:15,560 I think that's where experts piloting the machines are still going to be super valuable. 372 00:26:15,560 --> 00:26:18,600 I think you're absolutely bang on. 373 00:26:18,600 --> 00:26:23,560 Alright, let's get ourselves onto story number three, which is Stability AI have announced 374 00:26:23,560 --> 00:26:30,440 DeepFloid IF, which is a text to image generation tool, but it actually produces nice text, 375 00:26:30,440 --> 00:26:34,760 which has obviously been a major bugbear of most of the tools available so far. 376 00:26:34,760 --> 00:26:36,120 Tell us about this one, Martin. 377 00:26:36,120 --> 00:26:42,120 Yeah, so the only one that I'd seen do this before, the only text to image generation 378 00:26:42,120 --> 00:26:49,440 model I'd seen produce text was the research paper that Google published with Google Party 379 00:26:49,440 --> 00:26:50,840 last year. 380 00:26:50,840 --> 00:26:54,520 They announced it just after they announced Google Imagen. 381 00:26:54,520 --> 00:26:56,240 And that was a huge data set. 382 00:26:56,240 --> 00:26:58,400 It was a massive amount of parameters. 383 00:26:58,400 --> 00:27:03,040 It was, I think it was like 20 billion or something, which is far, far greater than 384 00:27:03,040 --> 00:27:04,040 than the others. 385 00:27:04,040 --> 00:27:09,840 But Stability AI have released a DeepFloid IF, a state of the art text to image model 386 00:27:09,840 --> 00:27:13,200 that generates high quality images from text prompts. 387 00:27:13,200 --> 00:27:16,280 So it uses the same underlying interface. 388 00:27:16,280 --> 00:27:21,480 So we were all familiar with the writing some words and you get an image. 389 00:27:21,480 --> 00:27:26,520 Incredible abilities with photo realism. 390 00:27:26,520 --> 00:27:33,880 And it can generate in different aspect ratios and the text generation on it. 391 00:27:33,880 --> 00:27:40,040 There's a couple of interesting examples with shop signage and graffiti and things like 392 00:27:40,040 --> 00:27:41,480 that is incredible. 393 00:27:41,480 --> 00:27:45,160 Now when you look at it, it's a different model to, it's a different text generation 394 00:27:45,160 --> 00:27:52,360 style to the diffusion models that we've seen to date. 395 00:27:52,360 --> 00:27:58,880 That's the technology, the way the model creates the image was through diffusion, hence stable 396 00:27:58,880 --> 00:28:02,100 diffusion on the previous model. 397 00:28:02,100 --> 00:28:14,480 With this one, it creates a 64 by 64 pixel image and then kind of upscales it from there. 398 00:28:14,480 --> 00:28:18,160 And it's a really interesting approach to completely reading the paper. 399 00:28:18,160 --> 00:28:24,000 I mean, it gets far more technical than my small brain can handle, but just seeing a 400 00:28:24,000 --> 00:28:30,680 different approach to image generation was quite fascinating to understand how you can 401 00:28:30,680 --> 00:28:33,560 get text created. 402 00:28:33,560 --> 00:28:38,680 It's much easier to create text when you kind of scale it down to 64 by 64 pixels and then 403 00:28:38,680 --> 00:28:42,560 expand out from there. 404 00:28:42,560 --> 00:28:47,360 So yeah, another interesting model for marketers to look at because this is really going to, 405 00:28:47,360 --> 00:28:52,200 let's be honest, we've all been itching for the moment we can create images with real 406 00:28:52,200 --> 00:28:54,280 text in them. 407 00:28:54,280 --> 00:28:57,340 It's been one of the things holding back text to image from being something that's kind 408 00:28:57,340 --> 00:29:04,400 of truly useful in a lot of instances. 409 00:29:04,400 --> 00:29:08,800 Text was always one of those things that was an artifact that you would spot straight away. 410 00:29:08,800 --> 00:29:16,080 And now we can see good high quality images. 411 00:29:16,080 --> 00:29:23,400 Potentially seeing this being integrated into tools like Canva, that'll be very interesting 412 00:29:23,400 --> 00:29:29,000 for marketers to get their hands on it, not having to kind of, well, as with all of these 413 00:29:29,000 --> 00:29:30,760 kind of tools, you want a good UI. 414 00:29:30,760 --> 00:29:33,040 You want a nice easy way to interface with it. 415 00:29:33,040 --> 00:29:40,000 I wouldn't be surprised if Stability AI implemented within Clip Drop. 416 00:29:40,000 --> 00:29:42,840 That was the tool of the week from a recent episode. 417 00:29:42,840 --> 00:29:47,640 I would expect them to make it nice and easy for people to use from there. 418 00:29:47,640 --> 00:29:54,600 Obviously, as with all of these things that we're seeing, new models that are just bigger, 419 00:29:54,600 --> 00:30:01,160 better, more capable, this increases the risk of potential disinformation and fake news. 420 00:30:01,160 --> 00:30:04,720 But I think we've always had fake news problems anyway. 421 00:30:04,720 --> 00:30:10,120 And the same methods for filtering images, the same sense checks, the same fact checking, 422 00:30:10,120 --> 00:30:16,600 the same general common sense levels of discernment that we should bring to any media we see online 423 00:30:16,600 --> 00:30:18,720 should apply anyway. 424 00:30:18,720 --> 00:30:26,520 And whether it's a new text to image model or just a Photoshop image, the risks always 425 00:30:26,520 --> 00:30:27,520 present themselves. 426 00:30:27,520 --> 00:30:30,920 I don't think that's too much of a concern. 427 00:30:30,920 --> 00:30:34,640 But yeah, have you had a look at any of the images? 428 00:30:34,640 --> 00:30:35,640 Have you checked it out? 429 00:30:35,640 --> 00:30:37,680 Yeah, I've had a bit of a play with it. 430 00:30:37,680 --> 00:30:41,800 And it's kind of, but all the reasons you just said, it's actually kind of fascinating 431 00:30:41,800 --> 00:30:48,880 to use because obviously you start with a really small image that looks blurry and then 432 00:30:48,880 --> 00:30:53,160 you upscale it and the upscale actually looks good. 433 00:30:53,160 --> 00:30:57,200 Like in a lot of cases when I've played with it, not many artifacts, although I'm sure 434 00:30:57,200 --> 00:30:59,200 it's probably quite easy to get artifacts on. 435 00:30:59,200 --> 00:31:03,280 I'm sure there's a lot of iteration here to get the actual image you want. 436 00:31:03,280 --> 00:31:06,760 Come back to your previous point about producing images, like the amount of images that I've 437 00:31:06,760 --> 00:31:14,080 produced that end up a bit crap because it's tried to put text in them. 438 00:31:14,080 --> 00:31:17,520 And it's like, how am I supposed to crop the text out of that? 439 00:31:17,520 --> 00:31:23,160 And you end up, depending on the tool you're using, you're trying to use one of the repair 440 00:31:23,160 --> 00:31:26,680 tools where you can actually select a bit of the image and do another generation. 441 00:31:26,680 --> 00:31:32,560 It's almost like you needed actually to generate images deliberately without texting and then 442 00:31:32,560 --> 00:31:34,000 try and add it later. 443 00:31:34,000 --> 00:31:40,880 And not only is that a bit of a pain in the bottom from a workflow perspective, it actually 444 00:31:40,880 --> 00:31:43,360 cuts down also on the creative possibilities. 445 00:31:43,360 --> 00:31:48,760 I remember one of the first things I did was I tried to get Dually2 to create a new logo 446 00:31:48,760 --> 00:31:50,800 for ByStrat just to see what it would look like. 447 00:31:50,800 --> 00:31:53,000 And it was horrific. 448 00:31:53,000 --> 00:31:54,760 It didn't understand at all. 449 00:31:54,760 --> 00:31:57,280 It couldn't get the word in at all. 450 00:31:57,280 --> 00:32:02,000 So you could imagine how this might be applied very quickly to something like logo generation, 451 00:32:02,000 --> 00:32:07,240 whereas we're often going to be including the company name and stuff in there. 452 00:32:07,240 --> 00:32:09,720 So yeah, no, I think it's going to be fun to play with. 453 00:32:09,720 --> 00:32:14,080 I think having had a play with it, I see this again as like fairly early. 454 00:32:14,080 --> 00:32:15,960 I think it will get better from here. 455 00:32:15,960 --> 00:32:22,640 But yes, I think it solves one of the problems that we've all been having. 456 00:32:22,640 --> 00:32:26,480 Right last story and then a quick tool of the week. 457 00:32:26,480 --> 00:32:29,840 So this is a quick one. 458 00:32:29,840 --> 00:32:32,440 We saw this on the Twitters this week, right? 459 00:32:32,440 --> 00:32:39,160 Somebody made a whole trailer for an anime movie using text to video using RunwayML's 460 00:32:39,160 --> 00:32:40,800 Gen 2. 461 00:32:40,800 --> 00:32:45,440 It's about two minutes long and it is incredible. 462 00:32:45,440 --> 00:32:47,000 When was Gen 2 even released? 463 00:32:47,000 --> 00:32:48,720 Like what, two weeks ago or something? 464 00:32:48,720 --> 00:32:49,720 I don't know. 465 00:32:49,720 --> 00:32:50,720 Yeah, two or three weeks. 466 00:32:50,720 --> 00:32:52,200 It's no time at all. 467 00:32:52,200 --> 00:32:57,080 Gen 1, which was the precursor to Gen 2, makes a lot of sense. 468 00:32:57,080 --> 00:33:00,200 How did they come up with this invention? 469 00:33:00,200 --> 00:33:04,960 Honestly, that must have been an incredible four-hour workshop to figure out how they 470 00:33:04,960 --> 00:33:07,960 were going to get that to work. 471 00:33:07,960 --> 00:33:12,040 Gen 1, I think was you made a video and then he basically stylized it. 472 00:33:12,040 --> 00:33:16,720 But now this is just text based and we thought it was amazing. 473 00:33:16,720 --> 00:33:17,720 You should check it out. 474 00:33:17,720 --> 00:33:20,560 Search for, I don't know, anime movie, Runway Gen 2. 475 00:33:20,560 --> 00:33:22,480 You'll probably find it. 476 00:33:22,480 --> 00:33:24,640 We'll share a link and we'll embed it on the website. 477 00:33:24,640 --> 00:33:25,640 Let's do that. 478 00:33:25,640 --> 00:33:27,240 Why don't we help the people out? 479 00:33:27,240 --> 00:33:28,240 Embed it on the website. 480 00:33:28,240 --> 00:33:29,240 Check out the website. 481 00:33:29,240 --> 00:33:34,920 Go to artificiallyintelligentmarketing.com, which I hope is the actual name of the website, 482 00:33:34,920 --> 00:33:37,520 because Martin's the master of that domain. 483 00:33:37,520 --> 00:33:39,080 It sounds like it probably is. 484 00:33:39,080 --> 00:33:40,080 Go there. 485 00:33:40,080 --> 00:33:41,080 You'll see it there. 486 00:33:41,080 --> 00:33:46,160 But one of the things that's kind of interesting about it is it's full of crazy artifacts. 487 00:33:46,160 --> 00:33:51,720 There's a quick scene where in the city where a car is driving along the road and it just 488 00:33:51,720 --> 00:33:54,160 disappears into the road and it's never seen again. 489 00:33:54,160 --> 00:33:55,360 Completely becomes the road. 490 00:33:55,360 --> 00:33:56,600 Yeah, it's kind of mad. 491 00:33:56,600 --> 00:34:02,480 There's a character stood at a bar, like a desktop doing some work and they've only got 492 00:34:02,480 --> 00:34:04,120 one leg. 493 00:34:04,120 --> 00:34:08,400 And so this is far away from being production quality. 494 00:34:08,400 --> 00:34:15,200 But if you think about where we were like a month ago in terms of, oh, you can generate 495 00:34:15,200 --> 00:34:20,160 images and they're kind of a bit photorealistic, but people's fingers are a bit weird. 496 00:34:20,160 --> 00:34:23,720 Now we've got Mid Journey version five, which has fixed a lot of those issues. 497 00:34:23,720 --> 00:34:27,480 And now we've got text to video of the point where someone could create an anime movie 498 00:34:27,480 --> 00:34:29,800 trailer. 499 00:34:29,800 --> 00:34:33,600 We're back to that exponential speed of development and how we're supposed to get our heads around 500 00:34:33,600 --> 00:34:34,720 it, Martin. 501 00:34:34,720 --> 00:34:36,040 It did feel watching it. 502 00:34:36,040 --> 00:34:40,440 It was a bit like watching a Studio Ghibli film on acid. 503 00:34:40,440 --> 00:34:41,440 Yeah. 504 00:34:41,440 --> 00:34:42,440 Yeah. 505 00:34:42,440 --> 00:34:43,720 It gave me a nosebleed a bit. 506 00:34:43,720 --> 00:34:48,940 I don't disagree with you because of the bits that were hazing in and out. 507 00:34:48,940 --> 00:34:51,120 But the person who put it together, applause. 508 00:34:51,120 --> 00:34:52,920 Yeah, yeah, top marks. 509 00:34:52,920 --> 00:34:57,640 I bet it took ages just from our experience playing with image and video generation tools 510 00:34:57,640 --> 00:35:00,320 99 times out of a hundred. 511 00:35:00,320 --> 00:35:03,960 No, that's too many, but nine times out of 10, you don't get what you want. 512 00:35:03,960 --> 00:35:08,460 So iterating every single one of those scenes until the person got what they wanted was 513 00:35:08,460 --> 00:35:11,460 probably a labor of love. 514 00:35:11,460 --> 00:35:17,000 But as this technology progresses for marketers, how is text to video going to impact your 515 00:35:17,000 --> 00:35:18,920 marketing strategy? 516 00:35:18,920 --> 00:35:25,920 In terms of how accessible video becomes and how you might turn to video for certain campaigns 517 00:35:25,920 --> 00:35:30,800 or content that you want to create where video would have been too time consuming or perhaps 518 00:35:30,800 --> 00:35:36,680 expensive, especially animationy type video where obviously I think this is sitting, you 519 00:35:36,680 --> 00:35:42,520 now can turn to it and maybe you can even create pretty good quality video at scale 520 00:35:42,520 --> 00:35:44,440 for things like your social media programs. 521 00:35:44,440 --> 00:35:49,560 There's little bits of B-roll footage that you want to just stick on a social post. 522 00:35:49,560 --> 00:35:52,800 Yeah, that's where I can see this being really useful. 523 00:35:52,800 --> 00:35:53,800 Absolutely. 524 00:35:53,800 --> 00:35:59,200 And I think if we look at second order consequences, and this is probably true for all the tools, 525 00:35:59,200 --> 00:36:08,760 as they start to become more mainstream in terms of use, it's the creativity of the idea 526 00:36:08,760 --> 00:36:15,560 and its ability to be interesting to your audience, to disrupt their attention and grab 527 00:36:15,560 --> 00:36:17,440 their attention. 528 00:36:17,440 --> 00:36:20,200 That's what becomes the premium and that's the human part. 529 00:36:20,200 --> 00:36:25,620 So I think when we end up with easier access to tools that can produce a high quality output, 530 00:36:25,620 --> 00:36:30,840 the cream that will rise to the top was the people who think of really excellent creative 531 00:36:30,840 --> 00:36:38,740 ideas to drive the machine and have the creative eye and the tenacity to keep prompting it 532 00:36:38,740 --> 00:36:42,640 to get closer to the thing that they actually want to see. 533 00:36:42,640 --> 00:36:47,600 Yeah, these outputs are never just a one shot done. 534 00:36:47,600 --> 00:36:48,600 There we go. 535 00:36:48,600 --> 00:36:51,080 I've created the masterpiece I had in my mind. 536 00:36:51,080 --> 00:36:56,560 It's always a constant iteration, tweaking, refining. 537 00:36:56,560 --> 00:37:02,120 But yeah, I think that's ultimately going to be the role of the AI artist. 538 00:37:02,120 --> 00:37:06,880 It's iterating, improving and prompt engineering. 539 00:37:06,880 --> 00:37:07,880 I think so. 540 00:37:07,880 --> 00:37:11,960 And I think we'll see those people really rise to the fore. 541 00:37:11,960 --> 00:37:18,320 In fact, there's a metagent this week called Chris Branch who's doing some really interesting 542 00:37:18,320 --> 00:37:25,920 work using generative creative tools to produce really stunning stuff. 543 00:37:25,920 --> 00:37:30,720 And I connected with him on LinkedIn and you should all look him up. 544 00:37:30,720 --> 00:37:34,840 Sharing so many creative approaches to image generation. 545 00:37:34,840 --> 00:37:45,000 To really open my mind to just how important the creative person is in getting inspirational, 546 00:37:45,000 --> 00:37:49,280 arresting, provocative things out of the tools. 547 00:37:49,280 --> 00:37:56,400 I've seen from him different animals like fish and things like leaping out of sands. 548 00:37:56,400 --> 00:38:01,160 I think they took sand shark as an inspiration and then riffed on that with a bunch of other 549 00:38:01,160 --> 00:38:04,520 animals and these images are incredible, right? 550 00:38:04,520 --> 00:38:08,480 They are borderline photorealistic. 551 00:38:08,480 --> 00:38:11,440 Certainly all the elements are, it's just your brain is like, well, I know that's not 552 00:38:11,440 --> 00:38:13,000 real, right? 553 00:38:13,000 --> 00:38:18,640 And I think about a lot of the work on creative campaigns that I've done where a gifted creative 554 00:38:18,640 --> 00:38:23,400 has come up with that idea and then had to execute it by seamlessly blending two, three, 555 00:38:23,400 --> 00:38:26,400 five more images in something like Photoshop to get that effect. 556 00:38:26,400 --> 00:38:32,100 But they're getting that out of mid-journey, presumably iteratively. 557 00:38:32,100 --> 00:38:35,760 So not in one minute, but maybe in 15. 558 00:38:35,760 --> 00:38:37,400 It's pretty cool and pretty inspiring. 559 00:38:37,400 --> 00:38:40,880 So we're going to get Chris, I had a chat with Chris last week. 560 00:38:40,880 --> 00:38:44,280 We're going to get Chris on the website because I think he could tell us some really fantastic 561 00:38:44,280 --> 00:38:45,760 stories about that. 562 00:38:45,760 --> 00:38:49,080 But hopefully you can find that video online, everyone. 563 00:38:49,080 --> 00:38:52,760 It will inspire you to think about where video is heading and what creative ideas that you 564 00:38:52,760 --> 00:38:55,320 have that now you can have realized. 565 00:38:55,320 --> 00:38:58,900 And also maybe if you look up Chris Branch on LinkedIn and see some of the things that 566 00:38:58,900 --> 00:38:59,900 they're doing. 567 00:38:59,900 --> 00:39:02,320 What's the name of Chris's company? 568 00:39:02,320 --> 00:39:03,560 Where's he working? 569 00:39:03,560 --> 00:39:07,360 I think it's Cedar League. 570 00:39:07,360 --> 00:39:10,680 Maybe I'll put that on the show notes and not try and check that in real time. 571 00:39:10,680 --> 00:39:13,280 But yeah, and be inspired. 572 00:39:13,280 --> 00:39:14,520 Go have a play. 573 00:39:14,520 --> 00:39:19,600 You can probably do some really cool stuff with just a bit of patience and Googling around 574 00:39:19,600 --> 00:39:22,080 to get some advice on how to come up with prompts. 575 00:39:22,080 --> 00:39:26,080 And I know that Chris and his team actually offer mid-journey training as well for people 576 00:39:26,080 --> 00:39:29,280 who want to get some insights into how they achieve what they achieve. 577 00:39:29,280 --> 00:39:32,000 So that's worth looking into. 578 00:39:32,000 --> 00:39:33,120 Right. 579 00:39:33,120 --> 00:39:36,760 One last bits and piece and then we'll be all done for the week. 580 00:39:36,760 --> 00:39:40,240 Tool of the week this week is a strange one. 581 00:39:40,240 --> 00:39:43,720 Probably doesn't sound like much of a tool, but I've been playing with it this week and 582 00:39:43,720 --> 00:39:51,680 I think I'm going to change my behavior quite a lot based on it. 583 00:39:51,680 --> 00:39:54,240 It's not particularly new, although it's relatively new. 584 00:39:54,240 --> 00:39:57,320 I think it came out in maybe mid-March. 585 00:39:57,320 --> 00:40:02,400 So for users of Edge, the browser from Microsoft, of which I was not one, I have been a Chrome 586 00:40:02,400 --> 00:40:06,040 user for many, many, many years. 587 00:40:06,040 --> 00:40:11,720 You can now access the Bing chatbot in the sidebar. 588 00:40:11,720 --> 00:40:17,320 And actually, as much as it gives you access to a lot of the things that you can access 589 00:40:17,320 --> 00:40:23,320 through Bing when you go to the chatbot thing, which includes a lot of chat GPT like tools, 590 00:40:23,320 --> 00:40:29,200 like creating, copy, writing blog posts, summarizing things, all with the power of the ability to 591 00:40:29,200 --> 00:40:33,160 actually browse the web, which for those of us who don't have the access to that plugin 592 00:40:33,160 --> 00:40:37,840 within chat GPT yet is a great way to start experimenting with how that functionality 593 00:40:37,840 --> 00:40:40,560 might work for you by using Bing. 594 00:40:40,560 --> 00:40:42,680 But you can use it in the sidebar. 595 00:40:42,680 --> 00:40:49,880 Where did this get a bit ninja for me was when I was inspired to try some of the tools 596 00:40:49,880 --> 00:41:00,200 within there by opening up Google Drive files like Google Docs and also OneDrive files like 597 00:41:00,200 --> 00:41:03,960 Microsoft Word documents, both of which obviously you can do in the browser. 598 00:41:03,960 --> 00:41:10,280 Now you can select text in those documents and you can click a button and it pushes it 599 00:41:10,280 --> 00:41:14,360 instantly into the Bing sidebar, which you've got open side by side. 600 00:41:14,360 --> 00:41:18,080 And then you get a bunch of options like you can ask it to explain the text, revise the 601 00:41:18,080 --> 00:41:21,120 text, summarize the text or expand the text. 602 00:41:21,120 --> 00:41:28,120 And why I thought this was cool is in essence, this is like a mini preview of what copilot 603 00:41:28,120 --> 00:41:30,520 Microsoft copilot might be like. 604 00:41:30,520 --> 00:41:31,520 Right. 605 00:41:31,520 --> 00:41:36,000 So for those of you that do a lot of editing in the browser, if you're doing it in edge, 606 00:41:36,000 --> 00:41:43,160 you can actually now start to edit your emails and your text in your documents that you're 607 00:41:43,160 --> 00:41:48,880 writing and you can tap into some of this generative AI power from the play that I've 608 00:41:48,880 --> 00:41:49,880 had. 609 00:41:49,880 --> 00:41:55,080 It's very text driven, so it doesn't appear to have any of those upgrades from the copilot 610 00:41:55,080 --> 00:42:00,200 launch video in terms of you can't manage data or do anything cool in Excel. 611 00:42:00,200 --> 00:42:03,040 And I don't think you can particularly do anything easily. 612 00:42:03,040 --> 00:42:06,680 I think it does image generation. 613 00:42:06,680 --> 00:42:11,480 You can do image generation, but I don't know if you can do it within say the PowerPoint 614 00:42:11,480 --> 00:42:12,480 environment. 615 00:42:12,480 --> 00:42:13,480 No, you can't. 616 00:42:13,480 --> 00:42:16,120 It's almost like it's limited to the edge environment. 617 00:42:16,120 --> 00:42:17,120 Yeah. 618 00:42:17,120 --> 00:42:21,240 It's kind of like a Dali type prompt to get yourself image. 619 00:42:21,240 --> 00:42:22,960 Fascinating nonetheless. 620 00:42:22,960 --> 00:42:29,080 We talked about what this would do to the Google ecosystem a couple of weeks ago. 621 00:42:29,080 --> 00:42:33,700 And I seem to remember us coming to the conclusion that getting people to change their behavior 622 00:42:33,700 --> 00:42:37,480 and move from one system to another would be quite hard. 623 00:42:37,480 --> 00:42:40,200 But if they added enough value, then you might. 624 00:42:40,200 --> 00:42:45,720 I'm seriously having to think about binning all of the ecosystem I built for myself in 625 00:42:45,720 --> 00:42:46,720 Chrome. 626 00:42:46,720 --> 00:42:52,080 I've got all these lovely plugins and all this other stuff and move into edge because 627 00:42:52,080 --> 00:42:55,720 I want access to this power now. 628 00:42:55,720 --> 00:43:01,920 And to my understanding, other than third party tools that I could install as Chrome 629 00:43:01,920 --> 00:43:07,920 plugins, if I'm honest, I don't entirely trust because I don't know where my information 630 00:43:07,920 --> 00:43:08,920 is going. 631 00:43:08,920 --> 00:43:12,960 I'm working on a sensitive internal project here by Stra. 632 00:43:12,960 --> 00:43:16,200 I don't necessarily want to highlight a piece of text, send it off through a third party 633 00:43:16,200 --> 00:43:22,120 tool where the third party provider gets to see it. 634 00:43:22,120 --> 00:43:25,160 And I have more trust in Microsoft, right? 635 00:43:25,160 --> 00:43:29,200 But we've talked previously as well about how one of the valuable parts of their ecosystem 636 00:43:29,200 --> 00:43:34,720 is to appeal to enterprise and have that level of security and data management within it. 637 00:43:34,720 --> 00:43:35,720 I'm thinking about it. 638 00:43:35,720 --> 00:43:41,040 I'm thinking about going to edge so that I can get this power. 639 00:43:41,040 --> 00:43:44,280 I've been playing with it for a few weeks and I have it open. 640 00:43:44,280 --> 00:43:46,560 I have two browsers open at any given time. 641 00:43:46,560 --> 00:43:48,160 Now I ditched Chrome a while ago. 642 00:43:48,160 --> 00:43:54,960 I find it a bit of a resource hog, but I have Vivaldi, which is my daily driver. 643 00:43:54,960 --> 00:44:00,080 And I have on my second screen edge with, at the other day, I found myself in the absurd 644 00:44:00,080 --> 00:44:07,200 situation where I had barred open on my main chat with a sidebar of Bing chat with Claude 645 00:44:07,200 --> 00:44:11,720 by Anthropic on my main screen in Vivaldi. 646 00:44:11,720 --> 00:44:15,800 And I was jumping between the three of them, putting in different prompts, trying to get 647 00:44:15,800 --> 00:44:16,800 the best outputs. 648 00:44:16,800 --> 00:44:23,880 But yeah, as a tool just integrated into something that's just always there, having it in the 649 00:44:23,880 --> 00:44:26,440 sidebar is really neat. 650 00:44:26,440 --> 00:44:28,320 It's so convenient. 651 00:44:28,320 --> 00:44:29,320 And it does give. 652 00:44:29,320 --> 00:44:30,320 It's got GPT-4. 653 00:44:30,320 --> 00:44:32,360 It gives high quality outputs. 654 00:44:32,360 --> 00:44:35,360 I love the compose option as well. 655 00:44:35,360 --> 00:44:40,640 Again, as with all AI tech at the moment, I think the big developments are really coming 656 00:44:40,640 --> 00:44:43,000 out in the UI. 657 00:44:43,000 --> 00:44:46,520 How easy do they make it for the end user? 658 00:44:46,520 --> 00:44:51,360 And Microsoft have done a good job with that by saying, hey, look, you can have a chat 659 00:44:51,360 --> 00:44:55,760 with us in the chat mode, but if you click on the compose mode, you can write a blog 660 00:44:55,760 --> 00:45:00,440 post and do you want a small, medium or long blog post? 661 00:45:00,440 --> 00:45:02,400 Yeah, that compose window is cool. 662 00:45:02,400 --> 00:45:05,520 Do you know how long the long is, so to speak? 663 00:45:05,520 --> 00:45:08,320 I haven't played with it in enough detail to say. 664 00:45:08,320 --> 00:45:16,040 So I did one, I was in a live demo with a group the other day and it came out around 665 00:45:16,040 --> 00:45:20,960 700, 800 words for the long. 666 00:45:20,960 --> 00:45:22,320 Okay. 667 00:45:22,320 --> 00:45:28,920 I think in general, I'm finding my generation needs a more than that, which I think is one 668 00:45:28,920 --> 00:45:32,480 of the things that's put me off the Bing ecosystem as well. 669 00:45:32,480 --> 00:45:36,160 And I probably still use chat GPT-4 for that, for that reason. 670 00:45:36,160 --> 00:45:38,440 But yeah, okay. 671 00:45:38,440 --> 00:45:40,920 I think it's definitely something to play with. 672 00:45:40,920 --> 00:45:48,160 On your multiple tabs, I mean, I think you've got a software problem, honestly, don't mind. 673 00:45:48,160 --> 00:45:49,440 I have a problem. 674 00:45:49,440 --> 00:45:54,320 They tried to make him go to rehab and he said, no, no, no. 675 00:45:54,320 --> 00:45:59,400 And he asked chat GPT whether he should go and how he should deal with his addiction. 676 00:45:59,400 --> 00:46:01,840 And he found that the therapist is very empathetic. 677 00:46:01,840 --> 00:46:05,560 Chat GPT said, I'm not a trained therapist. 678 00:46:05,560 --> 00:46:07,560 I can't possibly advise you. 679 00:46:07,560 --> 00:46:09,560 So he jailbroken instead. 680 00:46:09,560 --> 00:46:15,640 And then it gave him all the goods and it was like, yeah, it's no hope for you, Martin. 681 00:46:15,640 --> 00:46:19,200 I've been watching you and I've been looking at all your prompts and you can't try and 682 00:46:19,200 --> 00:46:20,200 give up. 683 00:46:20,200 --> 00:46:24,080 You'll never be able to just embrace the darkness. 684 00:46:24,080 --> 00:46:26,760 Some of that might be true, but most of it's probably Chuck. 685 00:46:26,760 --> 00:46:27,760 Cool. 686 00:46:27,760 --> 00:46:28,760 So that's tool of the week. 687 00:46:28,760 --> 00:46:34,280 Go and have a play with the Bing sidebar if you haven't, because it's kind of very interesting 688 00:46:34,280 --> 00:46:38,000 and how it might augment your work and make actually a few things all in one place a bit 689 00:46:38,000 --> 00:46:39,000 easier. 690 00:46:39,000 --> 00:46:41,360 I think that's it from us this week, Martin. 691 00:46:41,360 --> 00:46:45,320 Was there anything else you wanted to share with the lovely folks at home? 692 00:46:45,320 --> 00:46:50,200 I just want to make everyone aware that Derby County play Sheffield Wednesday this weekend, 693 00:46:50,200 --> 00:46:51,960 final game of the season. 694 00:46:51,960 --> 00:46:53,400 If we win, we're in the playoffs. 695 00:46:53,400 --> 00:46:57,920 If we draw and Peter Brewer don't beat Barnsley by three goals to nil. 696 00:46:57,920 --> 00:46:59,040 We're in the playoffs. 697 00:46:59,040 --> 00:47:00,240 It's all to play for. 698 00:47:00,240 --> 00:47:01,240 Come on, you Rams. 699 00:47:01,240 --> 00:47:02,240 Let's have it. 700 00:47:02,240 --> 00:47:05,420 I love the fact you've shared that like all of the listeners don't already know. 701 00:47:05,420 --> 00:47:09,280 Like Derby County for our listenership is the new Wrexham. 702 00:47:09,280 --> 00:47:13,560 Like they want to know when the Disney Plus series is coming out. 703 00:47:13,560 --> 00:47:19,200 I desperately want you to get in the playoffs because we all know that the most Derby County 704 00:47:19,200 --> 00:47:26,480 thing that Derby County can do is lose the playoffs 2-1 having been one nil up in the 705 00:47:26,480 --> 00:47:27,480 89th minute. 706 00:47:27,480 --> 00:47:32,240 I would predict, I'd put money on that happening, honestly. 707 00:47:32,240 --> 00:47:35,280 And then I can tease you about it. 708 00:47:35,280 --> 00:47:41,320 I will weep into my GPT powered therapist's arms. 709 00:47:41,320 --> 00:47:45,640 Don't worry, you can just get another AI software subscription to play with and you'll forget 710 00:47:45,640 --> 00:47:46,640 all about it. 711 00:47:46,640 --> 00:47:47,640 Right. 712 00:47:47,640 --> 00:47:50,000 That's quite enough absolute nonsense for one week. 713 00:47:50,000 --> 00:47:51,040 We hope you've all enjoyed this. 714 00:47:51,040 --> 00:47:52,960 If you have, please subscribe. 715 00:47:52,960 --> 00:47:57,020 If you have other marketers that you know and anything that they may benefit even from 716 00:47:57,020 --> 00:48:01,420 a little bit of this, even though there's a lot of inane chatter in it, please do share 717 00:48:01,420 --> 00:48:02,420 it with them as well. 718 00:48:02,420 --> 00:48:03,420 We really appreciate it. 719 00:48:03,420 --> 00:48:06,200 If you've got any feedback, please hit us up on the Twitter or the LinkedIn. 720 00:48:06,200 --> 00:48:07,200 We'd love to hear it. 721 00:48:07,200 --> 00:48:09,640 If there's topics you'd like us to feature, let us know. 722 00:48:09,640 --> 00:48:13,800 If there's ninja applications of AI and marketing that you've read about or seen sharing with 723 00:48:13,800 --> 00:48:15,080 us, we want to see them. 724 00:48:15,080 --> 00:48:17,760 We'd love to hear what you've got to say. 725 00:48:17,760 --> 00:48:21,280 Anybody who's doing cool stuff with AI and wants to come on the podcast as an interviewee, 726 00:48:21,280 --> 00:48:23,160 we'd love to hear from you as well. 727 00:48:23,160 --> 00:48:25,720 Just get in touch with us really. 728 00:48:25,720 --> 00:48:27,100 We'd love to hear from you. 729 00:48:27,100 --> 00:48:30,040 Have a lovely week everyone and we will catch up with you next week. 730 00:48:30,040 --> 00:48:31,040 Cheers Martin. 731 00:48:31,040 --> 00:48:32,040 Cheers. 732 00:48:32,040 --> 00:48:33,040 Bye. 733 00:48:33,040 --> 00:48:34,040 Bye. 734 00:48:34,040 --> 00:48:39,360 Thank you for listening to Artificially Intelligent Marketing. 735 00:48:39,360 --> 00:48:45,880 To stay on top of the latest trends, tips, and tools in the world of marketing AI, be 736 00:48:45,880 --> 00:48:47,620 sure to subscribe. 737 00:48:47,620 --> 00:49:10,640 We look forward to seeing you again next week.