1 00:00:00,000 --> 00:00:10,040 Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the 2 00:00:10,040 --> 00:00:15,760 latest trends, tips and tools in the world of marketing AI, helping you get the best 3 00:00:15,760 --> 00:00:18,600 results from your marketing efforts. 4 00:00:18,600 --> 00:00:23,560 Now let's join our hosts, Paul Avery and Martin Broadhurst. 5 00:00:23,560 --> 00:00:27,120 Welcome to episode 14 of Artificially Intelligent Marketing. 6 00:00:27,120 --> 00:00:28,960 I'm getting better at saying that every week. 7 00:00:28,960 --> 00:00:33,520 Martin, I've been practicing in the shower, in the mirror, in the dark, every morning. 8 00:00:33,520 --> 00:00:34,520 Absolutely. 9 00:00:34,520 --> 00:00:37,120 Just getting focused and making sure I can rock it. 10 00:00:37,120 --> 00:00:38,120 How are you, Martin? 11 00:00:38,120 --> 00:00:39,640 For episode 14, are you good? 12 00:00:39,640 --> 00:00:40,640 I'm good. 13 00:00:40,640 --> 00:00:45,360 It's Friday once again and I've had a flaffle for lunch. 14 00:00:45,360 --> 00:00:46,360 That's very exciting. 15 00:00:46,360 --> 00:00:51,600 So, you know, I'm in a good space, you know, mentally, physically, emotionally, just all 16 00:00:51,600 --> 00:00:52,600 of it. 17 00:00:52,600 --> 00:00:53,600 I'm just in a very good place. 18 00:00:53,600 --> 00:00:54,600 That's good. 19 00:00:54,600 --> 00:00:57,000 We need you centered because there's been a lot of AI news this week that we've got to 20 00:00:57,000 --> 00:00:58,600 get through. 21 00:00:58,600 --> 00:01:00,720 So lovely listeners, thanks for joining us again. 22 00:01:00,720 --> 00:01:03,400 We're going to look at our short snippets like we always do, some really good little 23 00:01:03,400 --> 00:01:05,480 bits and pieces in there to cover. 24 00:01:05,480 --> 00:01:10,400 We're going to dive into our chunky main stories, which this week is going to include Apple's 25 00:01:10,400 --> 00:01:16,640 Developer Conference in which we saw the Vision Pro for the first time, virtual reality, augmented 26 00:01:16,640 --> 00:01:21,800 reality headset, but also a ton of other things that involved AI, which they call machine 27 00:01:21,800 --> 00:01:26,340 learning because they have to be different, of course, under the hood. 28 00:01:26,340 --> 00:01:28,360 So we're going to dive into that, which is cool. 29 00:01:28,360 --> 00:01:32,160 There has been a ton of generative image and video news this week. 30 00:01:32,160 --> 00:01:35,720 We've lumped all of that into one big story for us to talk through. 31 00:01:35,720 --> 00:01:41,300 We're going to talk about DeepMind and their announcement around a generalized AI to optimize 32 00:01:41,300 --> 00:01:42,800 computing performance. 33 00:01:42,800 --> 00:01:47,280 And we're going to look at Meta's roadmap in terms of how large language models are 34 00:01:47,280 --> 00:01:52,140 going to be coming to your favorite tools that you already use, like WhatsApp and Messenger. 35 00:01:52,140 --> 00:01:57,460 And the tool of the week this week is going to be Gen2, RunwayML's Gen2, text to video 36 00:01:57,460 --> 00:02:02,740 creator, which is now easily accessible to all. 37 00:02:02,740 --> 00:02:12,360 So with that in mind, let's get straight into our short snippets with some news from AI 38 00:02:12,360 --> 00:02:20,600 startup Kohia raising $270 million to now be valued at $2.1 billion. 39 00:02:20,600 --> 00:02:22,740 We've done quite a lot of guess what? 40 00:02:22,740 --> 00:02:28,560 Another AI company got a ton of cash news, and this is another version of that. 41 00:02:28,560 --> 00:02:35,800 This puts Kohia up in the echelons of Anthropic and OpenAI and several others with very, very 42 00:02:35,800 --> 00:02:41,980 large valuations now based on the amount of cash that they're getting to fuel their large 43 00:02:41,980 --> 00:02:45,400 language model based tools. 44 00:02:45,400 --> 00:02:50,600 Kohia, many of you won't have heard of as much as some of the other companies we just 45 00:02:50,600 --> 00:02:57,040 mentioned, but their technology does underpin some of the writing tools that you might use. 46 00:02:57,040 --> 00:03:00,560 Like I think is it Jasper that uses Kohia? 47 00:03:00,560 --> 00:03:03,480 I think HyperWrite might use Kohia as well. 48 00:03:03,480 --> 00:03:09,920 Yeah, I think a lot of the tools are taking different models from all of the different 49 00:03:09,920 --> 00:03:10,920 providers. 50 00:03:10,920 --> 00:03:11,920 So they're part of the mix. 51 00:03:11,920 --> 00:03:16,360 You'll inadvertently have been using them if you're using any of these tools, I'm sure. 52 00:03:16,360 --> 00:03:17,360 Absolutely. 53 00:03:17,360 --> 00:03:23,400 One of Kohia's things is to try and be somewhat agnostic and allow lots of different tools 54 00:03:23,400 --> 00:03:26,800 to basically leverage it and deploy it how they see fit. 55 00:03:26,800 --> 00:03:33,720 Yeah, they're very much positioning themselves in recent months as being for the enterprise 56 00:03:33,720 --> 00:03:34,720 as well. 57 00:03:34,720 --> 00:03:40,440 That is where they've been doing a lot of their messaging, if nothing else. 58 00:03:40,440 --> 00:03:41,440 Absolutely. 59 00:03:41,440 --> 00:03:48,040 So again, watch that space as we see all of these companies getting ever more money as 60 00:03:48,040 --> 00:03:53,760 private equity and other investment vehicles bet on AI as the next big thing. 61 00:03:53,760 --> 00:03:57,000 Let's hope it doesn't turn out to be too much of a bubble. 62 00:03:57,000 --> 00:04:01,820 In terms of the next short snippet, we're going to talk very briefly about how Mark 63 00:04:01,820 --> 00:04:08,720 Walters from Georgia has decided to sue OpenAI over claims made by ChatGPT that he embezzled 64 00:04:08,720 --> 00:04:15,240 funds, which is absolutely false and it was a hallucination by ChatGPT and it's not real. 65 00:04:15,240 --> 00:04:24,560 So further examples of what happens when ChatGPT makes stuff up and the carnage it can cause. 66 00:04:24,560 --> 00:04:29,840 So that fits in with the story that we mentioned last week, Martin, when a lawyer was using 67 00:04:29,840 --> 00:04:30,840 it for research. 68 00:04:30,840 --> 00:04:35,440 Yeah, the other side of the coin. 69 00:04:35,440 --> 00:04:40,200 I haven't dug too deep into this story, but one could imagine that if you prompted the 70 00:04:40,200 --> 00:04:46,040 tools just right, I reckon I could probably get ChatGPT to make some stuff up about me 71 00:04:46,040 --> 00:04:48,920 potentially and then maybe I can come up with a cool case. 72 00:04:48,920 --> 00:04:51,200 Well, that's the detail that's missing from the story. 73 00:04:51,200 --> 00:04:57,760 I read the report and I've read a couple of them and that's the detail that is ultimately 74 00:04:57,760 --> 00:04:58,760 missing. 75 00:04:58,760 --> 00:04:59,760 What was the conversation thread? 76 00:04:59,760 --> 00:05:03,360 That is the whole essence of it. 77 00:05:03,360 --> 00:05:11,760 I know that it was a question around, I don't have it to hand now, but it was a specific 78 00:05:11,760 --> 00:05:17,680 legal case and this radio host wasn't involved in it at all. 79 00:05:17,680 --> 00:05:21,840 And I think the person went back and asked it to clarify the details and it kind of doubled 80 00:05:21,840 --> 00:05:26,680 down on it as well, which I think made the situation slightly worse. 81 00:05:26,680 --> 00:05:28,040 Right, right. 82 00:05:28,040 --> 00:05:29,040 Cool. 83 00:05:29,040 --> 00:05:34,880 Another short snippet is another funding announcement, so this is Lightmatter that received $154 84 00:05:34,880 --> 00:05:36,880 million in new funding. 85 00:05:36,880 --> 00:05:41,160 This is an interesting one because Lightmatter is developing a form of optical computing, 86 00:05:41,160 --> 00:05:45,640 so it's a mixture of hardware and software that aims to be more efficient, faster and 87 00:05:45,640 --> 00:05:49,480 cheaper but to run AI models quite specifically. 88 00:05:49,480 --> 00:05:52,320 I'm sure there'll be other applications, but the one they're focusing on at the moment 89 00:05:52,320 --> 00:05:53,880 is that. 90 00:05:53,880 --> 00:05:57,840 This is hot on the heels of some of the discussions we've been having, Martin, about how much 91 00:05:57,840 --> 00:06:03,780 energy it takes to run the compute power needed to run these models, how much water is required 92 00:06:03,780 --> 00:06:07,280 to cool the server farms and all that stuff. 93 00:06:07,280 --> 00:06:12,920 So this is a company that's trying to help solve for some of that and obviously convincing 94 00:06:12,920 --> 00:06:15,640 a number of other people with money that it's a goer. 95 00:06:15,640 --> 00:06:19,320 So they're hot on the heels of some of those challenges. 96 00:06:19,320 --> 00:06:24,720 There was a report from Salesforce this week that found that 51% of marketers are using 97 00:06:24,720 --> 00:06:30,960 generative AI at work with the most popular applications being, as you might have expected, 98 00:06:30,960 --> 00:06:33,280 content and copy creation. 99 00:06:33,280 --> 00:06:38,440 But I think this is part of an emerging trend that I'm starting to see when I'm talking 100 00:06:38,440 --> 00:06:43,200 to more and more people in the market that they not only are aware of these tools, but 101 00:06:43,200 --> 00:06:46,560 they've tried them and they're playing with them and they're seeing how they might fit 102 00:06:46,560 --> 00:06:48,040 into their workflows. 103 00:06:48,040 --> 00:06:56,120 Yeah, I saw something else this week that said people are on average using ChatGPT six 104 00:06:56,120 --> 00:06:59,760 times and then going, I'm done with it now. 105 00:06:59,760 --> 00:07:00,760 Yeah. 106 00:07:00,760 --> 00:07:04,360 Which certainly isn't my, I'm using it six times a minute at the moment. 107 00:07:04,360 --> 00:07:10,160 But that was an interesting thought that actually the early adopters are all in on it. 108 00:07:10,160 --> 00:07:14,840 But now as we're starting to see more and more people get familiar with it, they're 109 00:07:14,840 --> 00:07:18,760 using it, but not really diving in with both feet. 110 00:07:18,760 --> 00:07:26,560 That's a good question for me to ask folks because I definitely think the tools are quite 111 00:07:26,560 --> 00:07:27,560 iterative. 112 00:07:27,560 --> 00:07:31,200 They're good at certain things and it's quite hard to get good outputs out of them for other 113 00:07:31,200 --> 00:07:32,200 things. 114 00:07:32,200 --> 00:07:35,440 And I do think you have to have a bit of sticking power in terms of figuring out how you're 115 00:07:35,440 --> 00:07:38,120 going to make them work for you. 116 00:07:38,120 --> 00:07:40,840 So I guess that fits because if you don't have that sticking power, you might be like, 117 00:07:40,840 --> 00:07:43,500 oh, it's pretty useless. 118 00:07:43,500 --> 00:07:48,120 As an example, I've been experimenting, I might have mentioned this before. 119 00:07:48,120 --> 00:07:51,640 I'm using a tool at the moment called Audio Pen, which is pretty cool. 120 00:07:51,640 --> 00:07:56,920 You record a short audio snippet, like I think it's up to three minutes. 121 00:07:56,920 --> 00:08:00,080 So I'm basically using it for idea collection when I'm walking the dog on the stuff at the 122 00:08:00,080 --> 00:08:01,320 moment. 123 00:08:01,320 --> 00:08:06,720 And it doesn't just transcribe it, but it uses probably GPT-4, I don't know, a large 124 00:08:06,720 --> 00:08:10,400 language model to turn it into coherent text. 125 00:08:10,400 --> 00:08:13,760 So if you wanted to share it with someone else or read it yourself later and have it 126 00:08:13,760 --> 00:08:18,280 make sense rather than the rambling that you get when you do a voice note, it's actually 127 00:08:18,280 --> 00:08:20,680 really quite good for that. 128 00:08:20,680 --> 00:08:24,840 But what I found is I really wanted that same thing, but to be able to draft emails for 129 00:08:24,840 --> 00:08:30,760 me, but the email, so I'll record an audio snippet and then I'll push it into chat GPT, 130 00:08:30,760 --> 00:08:33,080 but the emails are never in my voice. 131 00:08:33,080 --> 00:08:38,560 It sounds, they're either way too corporate or if I'm like, hey, could you chatty this 132 00:08:38,560 --> 00:08:43,080 up a bit, it's like, hey dude, some ideas. 133 00:08:43,080 --> 00:08:45,240 So it's like, it goes all the way the other way. 134 00:08:45,240 --> 00:08:51,360 Like I'm either like running, I'm either CEO of like a $25 billion company or I'm running 135 00:08:51,360 --> 00:08:54,080 a surf shack and it doesn't seem to know. 136 00:08:54,080 --> 00:08:57,720 There is no in between, but I think that that's how I see you. 137 00:08:57,720 --> 00:09:01,160 You're a $25 billion CEO that is running a surf shack. 138 00:09:01,160 --> 00:09:05,760 And I'm always like, he just seems poorly primed for either of those positions to be 139 00:09:05,760 --> 00:09:06,760 honest. 140 00:09:06,760 --> 00:09:07,760 I couldn't agree more. 141 00:09:07,760 --> 00:09:09,480 I think it depends on what day I run it. 142 00:09:09,480 --> 00:09:15,040 It knows I'm big time Charlie on a Monday, but I'm down the beach cats and some waves 143 00:09:15,040 --> 00:09:16,040 on a Tuesday. 144 00:09:16,040 --> 00:09:18,480 So that's probably how it's, it's figuring it out. 145 00:09:18,480 --> 00:09:22,860 It's a great example of the voice note, because we spoke about this with Whisper and the Whisper 146 00:09:22,860 --> 00:09:25,880 integration and Zapier. 147 00:09:25,880 --> 00:09:30,480 And I think that's maybe where you can play around with that a little bit. 148 00:09:30,480 --> 00:09:35,200 Try it with Zapier, try the voice note, save it to a Google drive, stick it into, because 149 00:09:35,200 --> 00:09:42,720 that, if you, I guess, if you were to in your prompt, put some examples of your existing 150 00:09:42,720 --> 00:09:45,640 email tone, it would then start to reflect that. 151 00:09:45,640 --> 00:09:47,240 So that would be an interesting one to play with. 152 00:09:47,240 --> 00:09:50,840 We should, we should, we should take this offline and give it a go. 153 00:09:50,840 --> 00:09:51,840 Yeah. 154 00:09:51,840 --> 00:09:57,960 I think definitely there must be some sort of tool or Zapier that you can build that 155 00:09:57,960 --> 00:10:03,640 in essence rhymes at every time on your voice by actually automatically giving it quite 156 00:10:03,640 --> 00:10:05,880 a comprehensive prompt. 157 00:10:05,880 --> 00:10:12,400 And then of course, at some point when Copilot goes live and Bard is integrated into Gmail, 158 00:10:12,400 --> 00:10:14,920 one assumes this won't be a problem at all anyway. 159 00:10:14,920 --> 00:10:17,760 So certainly not a tool I would build to try and commercialize because I think in three 160 00:10:17,760 --> 00:10:19,320 months it would be dead in the water. 161 00:10:19,320 --> 00:10:23,920 But in terms of a quick Zap to create that might do some cool stuff that currently I'm 162 00:10:23,920 --> 00:10:27,560 not able to do, completely agree. 163 00:10:27,560 --> 00:10:32,440 While we're talking about Bard and Copilot and all these things, the other short snippet 164 00:10:32,440 --> 00:10:38,240 this week is that Bard can now execute code as well as write it. 165 00:10:38,240 --> 00:10:39,480 Execute code as well as write it. 166 00:10:39,480 --> 00:10:40,640 Martin, you pick this one up. 167 00:10:40,640 --> 00:10:43,760 Can you elaborate a bit for us on that? 168 00:10:43,760 --> 00:10:50,760 I mean, it's as the story says there, to be honest, it's able to, it's given it more reasoning 169 00:10:50,760 --> 00:10:56,600 capabilities which is just extending its powers really. 170 00:10:56,600 --> 00:11:00,480 I didn't get a great deal more from it than that. 171 00:11:00,480 --> 00:11:01,480 Where is the article? 172 00:11:01,480 --> 00:11:06,000 Google has unveiled a major new update to its AI chatbot, Bard, that significantly improves 173 00:11:06,000 --> 00:11:07,880 its powers of logic and reasoning. 174 00:11:07,880 --> 00:11:13,320 It has a new technique called implicit code execution that enables Bard to detect computational 175 00:11:13,320 --> 00:11:15,800 prompts and run code in the background. 176 00:11:15,800 --> 00:11:20,400 So that's what it's doing and the result is that Bard should theoretically be able to 177 00:11:20,400 --> 00:11:26,280 respond more accurately to mathematical tasks and coding questions as it will have already 178 00:11:26,280 --> 00:11:28,200 tested the outcomes that it proposes. 179 00:11:28,200 --> 00:11:35,120 So in terms of really ramping up its capabilities, I think that's quite a step up. 180 00:11:35,120 --> 00:11:37,280 That does sound interesting, but it's not. 181 00:11:37,280 --> 00:11:39,160 It deploys code in the wild. 182 00:11:39,160 --> 00:11:43,320 It's like, here's some code and by the way, make it so that it will actually run on a 183 00:11:43,320 --> 00:11:45,600 server or as a web app or whatever. 184 00:11:45,600 --> 00:11:46,600 Yeah, correct. 185 00:11:46,600 --> 00:11:49,960 It's just kind of running it in the background and throwing the result back to you. 186 00:11:49,960 --> 00:11:50,960 Cool. 187 00:11:50,960 --> 00:11:54,800 While we're on this topic, because we love a tangent here at artificially intelligent 188 00:11:54,800 --> 00:11:59,520 marketing, as you will know, dear listener, and we appreciate your patience because of 189 00:11:59,520 --> 00:12:05,600 that, I was playing with chatbot, so HubSpot's chat tool this week and I hadn't really gone 190 00:12:05,600 --> 00:12:08,920 back to it for like three or four weeks because basically everything I asked it to do, it 191 00:12:08,920 --> 00:12:09,920 couldn't do. 192 00:12:09,920 --> 00:12:13,360 It kept sort of saying, I know what you want, but I can't do that yet. 193 00:12:13,360 --> 00:12:16,840 Or it went, I don't know what you want and neither of which are responses that were particularly 194 00:12:16,840 --> 00:12:18,480 useful to me. 195 00:12:18,480 --> 00:12:20,400 But they've changed how it works. 196 00:12:20,400 --> 00:12:22,600 So it's not on the old URL anymore. 197 00:12:22,600 --> 00:12:25,120 It's been redeployed in some other way. 198 00:12:25,120 --> 00:12:28,400 I think they're moving into that part where maybe they might be ready to move from alpha 199 00:12:28,400 --> 00:12:32,520 into beta because it just got a ton better. 200 00:12:32,520 --> 00:12:33,520 Interesting. 201 00:12:33,520 --> 00:12:34,680 Like miles. 202 00:12:34,680 --> 00:12:39,600 So I had a new business call this week and I was like, I just wrote in natural language, 203 00:12:39,600 --> 00:12:47,760 tell me about company X and my goodness, it told me how many employees, revenue. 204 00:12:47,760 --> 00:12:50,760 It told me a short summary of a company. 205 00:12:50,760 --> 00:12:55,200 It told me what technologies they were using to underpin their marketing and sales tech 206 00:12:55,200 --> 00:12:56,200 stacks. 207 00:12:56,200 --> 00:12:59,800 It told me that they were on WordPress and all this stuff, all in one output. 208 00:12:59,800 --> 00:13:09,040 That's interesting because that's the content that you get from the AI insights that HubSpot 209 00:13:09,040 --> 00:13:11,840 has been providing anyway for a long time. 210 00:13:11,840 --> 00:13:15,800 So the data enrichments that HubSpot has been providing through HubSpot Insights, they're 211 00:13:15,800 --> 00:13:21,120 obviously integrating that knowledge base and database into Chatspot now. 212 00:13:21,120 --> 00:13:23,400 I've not had to play with it like you. 213 00:13:23,400 --> 00:13:26,160 I kind of left it for a few weeks. 214 00:13:26,160 --> 00:13:27,160 So that's interesting. 215 00:13:27,160 --> 00:13:28,600 I will definitely get into it. 216 00:13:28,600 --> 00:13:31,720 And that was the type of question I was asking it previously and it was saying, I know what 217 00:13:31,720 --> 00:13:35,080 you want, but I can't do it. 218 00:13:35,080 --> 00:13:37,920 I think the other thing is it's the little things that it's adding in. 219 00:13:37,920 --> 00:13:42,160 So it also tells you how many followers they've got online. 220 00:13:42,160 --> 00:13:44,280 It tells you how many locations they've got. 221 00:13:44,280 --> 00:13:49,240 Then there's some handy buttons so you can click SEM summary and it will tell you what 222 00:13:49,240 --> 00:13:54,560 SEO, so organic keywords they rank for, but it will also tell you what paid ads they bid 223 00:13:54,560 --> 00:13:57,040 on. 224 00:13:57,040 --> 00:13:59,760 And that's just by clicking a button. 225 00:13:59,760 --> 00:14:06,680 So you can quite quickly really gather some interesting data on a company just from its 226 00:14:06,680 --> 00:14:10,120 name and quite iteratively. 227 00:14:10,120 --> 00:14:15,960 And so if you had a play with it when the alpha first launched and you were like, crumbs, 228 00:14:15,960 --> 00:14:18,480 it promises loads of cool stuff, but it can't do it. 229 00:14:18,480 --> 00:14:24,360 Go have a play because if my play this week is anything to go with, it has vastly improved, 230 00:14:24,360 --> 00:14:25,360 vastly improved. 231 00:14:25,360 --> 00:14:27,880 So that's pretty cool. 232 00:14:27,880 --> 00:14:29,040 Right. 233 00:14:29,040 --> 00:14:31,200 Let's get into the actual main stories. 234 00:14:31,200 --> 00:14:36,040 Although hopefully those snippets were a few deep dives that were still useful. 235 00:14:36,040 --> 00:14:37,280 And we're getting to story number one. 236 00:14:37,280 --> 00:14:40,720 And this is Apple's Developer Conference, which was real cool. 237 00:14:40,720 --> 00:14:43,840 I'm a bit of a tech nerd. 238 00:14:43,840 --> 00:14:45,440 I know you enjoyed this, Martin. 239 00:14:45,440 --> 00:14:50,680 Tell us, what did you take away from Apple's news and launches this week? 240 00:14:50,680 --> 00:14:57,640 Well, one hot take that I read on Twitter was that Apple is in trouble because they 241 00:14:57,640 --> 00:15:00,080 are not focusing on AI right now. 242 00:15:00,080 --> 00:15:01,080 Really? 243 00:15:01,080 --> 00:15:03,440 It's got an issue. 244 00:15:03,440 --> 00:15:04,840 It was the hottest of hot takes. 245 00:15:04,840 --> 00:15:05,840 I thought, wow. 246 00:15:05,840 --> 00:15:09,680 They've got this machine learning thing. 247 00:15:09,680 --> 00:15:12,720 I don't know what that is, but they've got no AI. 248 00:15:12,720 --> 00:15:14,680 They're not an AI first company. 249 00:15:14,680 --> 00:15:15,680 Yeah. 250 00:15:15,680 --> 00:15:20,960 And it was interesting that they didn't mention AI throughout the keynote once, but I don't 251 00:15:20,960 --> 00:15:26,120 know what the count was for the number of times they said machine learning. 252 00:15:26,120 --> 00:15:29,680 But machine learning is coming to pretty much everything. 253 00:15:29,680 --> 00:15:33,240 I mean, not that it wasn't baked into the whole system anyway, but there's some really 254 00:15:33,240 --> 00:15:35,560 interesting new examples. 255 00:15:35,560 --> 00:15:40,800 One of which that everyone will be, everyone within iOS device will be happy about, particularly 256 00:15:40,800 --> 00:15:46,360 if they're a little bit fruity in their language, will be the new auto correct coming to iOS 257 00:15:46,360 --> 00:15:49,800 17, which is powered by on-device machine learning. 258 00:15:49,800 --> 00:15:56,360 And it actually has a transformer based language model on device capable of not just auto suggest 259 00:15:56,360 --> 00:16:00,080 for a language, but auto suggest for language generation. 260 00:16:00,080 --> 00:16:04,480 So I guess it's going to be similar to auto complete on Gmail and things like that. 261 00:16:04,480 --> 00:16:09,860 So finally, that auto correct that everybody gets when they type in the F word and it switches 262 00:16:09,860 --> 00:16:14,920 it to ducking, well, that will no longer be the case. 263 00:16:14,920 --> 00:16:21,160 I was fascinated here, Martin, by the on-device machine learning because there was, so maybe 264 00:16:21,160 --> 00:16:25,040 we'll get into the headset in a moment and the Vision Pro, but there are a number of 265 00:16:25,040 --> 00:16:30,520 things there where they really touched on protecting user security and information while 266 00:16:30,520 --> 00:16:32,340 they were talking about how it works. 267 00:16:32,340 --> 00:16:39,000 And we've talked previously about how on-device machine learning models might be needed to 268 00:16:39,000 --> 00:16:43,280 help people feel comfortable that their personal data and other things are not being shared 269 00:16:43,280 --> 00:16:47,120 in the cloud and used to train models or accessible by others. 270 00:16:47,120 --> 00:16:53,200 And this is the first time I can remember really a key tech player or software player 271 00:16:53,200 --> 00:16:57,120 talking about how their models are being run locally rather than in the cloud. 272 00:16:57,120 --> 00:16:58,120 Yeah. 273 00:16:58,120 --> 00:17:04,160 And we saw it coming, didn't really, we kind of forecast this with the investment and the 274 00:17:04,160 --> 00:17:08,360 capabilities that were never quite fully tapped in the M chips. 275 00:17:08,360 --> 00:17:09,360 Yeah, yeah, absolutely. 276 00:17:09,360 --> 00:17:12,640 And that was indicative of it. 277 00:17:12,640 --> 00:17:18,760 So we've also got Dictation in iOS 17, getting a new transformer based speech recognition 278 00:17:18,760 --> 00:17:19,760 model. 279 00:17:19,760 --> 00:17:24,800 iPad OS PDF features use machine learning models to identify form fields. 280 00:17:24,800 --> 00:17:25,800 That's pretty neat. 281 00:17:25,800 --> 00:17:26,800 That's pretty cool. 282 00:17:26,800 --> 00:17:30,440 Yeah, I mean, there's loads of these things. 283 00:17:30,440 --> 00:17:31,440 So, airpads. 284 00:17:31,440 --> 00:17:34,440 Where can I get those? 285 00:17:34,440 --> 00:17:35,440 Airpads. 286 00:17:35,440 --> 00:17:36,440 Yeah. 287 00:17:36,440 --> 00:17:37,440 Not sure what that was. 288 00:17:37,440 --> 00:17:42,360 That sounds like some sort of basketball performance enhancing drug, doesn't it? 289 00:17:42,360 --> 00:17:44,560 Or sort of very special trainers. 290 00:17:44,560 --> 00:17:48,480 Can I get me some Nike Airpads, please? 291 00:17:48,480 --> 00:17:53,920 So AirPods adaptive audio features use machine learning to understand your listening preferences 292 00:17:53,920 --> 00:17:59,400 over time, the smartwatch, the Apple watch is getting a smart stack widget to show the 293 00:17:59,400 --> 00:18:05,120 relevant information, journal app on iOS 17 users on device machine learning to provide 294 00:18:05,120 --> 00:18:10,520 personalized suggestions based on photos, locations, music and workouts. 295 00:18:10,520 --> 00:18:17,600 You've got the animated eye display on the aforementioned Apple Vision Pro headset, which 296 00:18:17,600 --> 00:18:25,720 is coming in at a bargain price of $3499 USD. 297 00:18:25,720 --> 00:18:29,760 And that's being created using the most advanced machine learning techniques on an advanced 298 00:18:29,760 --> 00:18:33,420 encoder decoder neural network. 299 00:18:33,420 --> 00:18:39,520 And then the M2 Ultra chip in the Mac Studio and the Mac Pro has been announced and it's 300 00:18:39,520 --> 00:18:44,800 powerful enough to train, this was interesting, to train large transformer models and other 301 00:18:44,800 --> 00:18:48,120 massive machine learning workloads. 302 00:18:48,120 --> 00:18:50,600 As we did discuss previously, right? 303 00:18:50,600 --> 00:18:52,680 They're absolutely leaning into that. 304 00:18:52,680 --> 00:18:54,840 I thought that was a fascinating bit of noise. 305 00:18:54,840 --> 00:18:57,680 Hopefully not noise, but a bit of news. 306 00:18:57,680 --> 00:19:00,840 A bit of noise that the Apple team started making. 307 00:19:00,840 --> 00:19:01,840 Lovely noise. 308 00:19:01,840 --> 00:19:03,960 Yeah, it's funny. 309 00:19:03,960 --> 00:19:09,120 You talk about $3499 for the Vision Pro and there's two camps forming. 310 00:19:09,120 --> 00:19:11,240 One's a massive camp and one's a small camp. 311 00:19:11,240 --> 00:19:15,800 Some people think, wow, that's outrageously expensive and other people are like, that 312 00:19:15,800 --> 00:19:17,240 is not expensive at all. 313 00:19:17,240 --> 00:19:18,680 That is a bargain. 314 00:19:18,680 --> 00:19:22,960 And I started in the expensive camp, but now that I really think about the technology that's 315 00:19:22,960 --> 00:19:30,400 in that headset and what it potentially could replace, it's not a price. 316 00:19:30,400 --> 00:19:35,160 It's not a bargain, but it's potentially really good value when you actually think about it. 317 00:19:35,160 --> 00:19:36,160 Yeah. 318 00:19:36,160 --> 00:19:44,040 I mean, cards on the table, it is expensive, but it's not ludicrous and it's not for everybody. 319 00:19:44,040 --> 00:19:51,520 There was an interesting tweet from Robert Scoble who tweets all sorts of stuff about 320 00:19:51,520 --> 00:19:52,520 AI all the time. 321 00:19:52,520 --> 00:20:00,800 And he put an interesting comment saying that when he was really jealous of Wozniak at Apple 322 00:20:00,800 --> 00:20:06,400 when he got his first color printer way back when and it was wildly unaffordable. 323 00:20:06,400 --> 00:20:11,560 And now you can pick up the same printer for like, or a printer 10 times better, sorry, 324 00:20:11,560 --> 00:20:15,160 or a hundred times better even for like $70. 325 00:20:15,160 --> 00:20:16,560 And that's just the nature of technology. 326 00:20:16,560 --> 00:20:20,280 When things come to the market, when real innovative technology comes to market, it's 327 00:20:20,280 --> 00:20:23,220 expensive and not everybody can have it. 328 00:20:23,220 --> 00:20:25,280 And then the price comes down over time. 329 00:20:25,280 --> 00:20:26,280 That's the nature of it. 330 00:20:26,280 --> 00:20:29,200 So we can all sit here and go, oh, that's so expensive. 331 00:20:29,200 --> 00:20:31,920 But fast forward three or four years and do you know what? 332 00:20:31,920 --> 00:20:34,600 It's probably going to be about 1500 quid. 333 00:20:34,600 --> 00:20:36,720 And that's just where we are. 334 00:20:36,720 --> 00:20:40,920 I did think actually, because you've got the battery pack as well, haven't you? 335 00:20:40,920 --> 00:20:46,040 And the battery pack in additional 1500 or something. 336 00:20:46,040 --> 00:20:51,000 I saw something saying that all in it was somewhere in the region of just over 5000. 337 00:20:51,000 --> 00:20:53,400 We have to buy the battery pack separately. 338 00:20:53,400 --> 00:20:54,960 I did not know that. 339 00:20:54,960 --> 00:20:57,960 That seems unlikely, but I guess it's possible. 340 00:20:57,960 --> 00:21:01,520 Yeah, maybe it was a spare battery. 341 00:21:01,520 --> 00:21:02,520 I don't know. 342 00:21:02,520 --> 00:21:05,800 I saw someone on Twitter saying it and I didn't do a deep dive into it. 343 00:21:05,800 --> 00:21:10,520 It didn't make me sit up and think, okay, if you're at the 5k territory, that is. 344 00:21:10,520 --> 00:21:12,520 That's getting quite insane. 345 00:21:12,520 --> 00:21:13,520 Yeah. 346 00:21:13,520 --> 00:21:18,640 I guess it comes down to the difference between cost and value. 347 00:21:18,640 --> 00:21:21,240 I completely agree. 348 00:21:21,240 --> 00:21:26,440 It's a high cost for most people. 349 00:21:26,440 --> 00:21:32,640 I think my point was more when you look at the technologies in it, it's potentially you 350 00:21:32,640 --> 00:21:36,600 get quite a lot of value for the money, but there's no getting away from it. 351 00:21:36,600 --> 00:21:41,600 It's a high cost, especially when you compare it to other VR headsets on the market and 352 00:21:41,600 --> 00:21:48,760 also similar technologies like high definition 8k TVs or whatever. 353 00:21:48,760 --> 00:21:51,640 I guess we'd need to keep on the AI. 354 00:21:51,640 --> 00:21:52,640 I can't help. 355 00:21:52,640 --> 00:21:56,640 I'm a bit of a VR nerd as well as an AI nerd, so I can't help but get pretty excited about 356 00:21:56,640 --> 00:21:57,640 that stuff. 357 00:21:57,640 --> 00:21:58,640 I did see the... 358 00:21:58,640 --> 00:22:06,960 In fact, I think we might touch on it later, but Zuckerberg's email response basically 359 00:22:06,960 --> 00:22:09,360 going, nah, they're not doing anything that special. 360 00:22:09,360 --> 00:22:10,360 It's fine. 361 00:22:10,360 --> 00:22:13,480 I'd be interested to see how that plays out. 362 00:22:13,480 --> 00:22:21,280 I think there'll be the look and click mechanism of use from the few people who've been using 363 00:22:21,280 --> 00:22:23,200 it. 364 00:22:23,200 --> 00:22:27,600 I get the impression is really very intuitive in terms of you. 365 00:22:27,600 --> 00:22:28,600 Very much so. 366 00:22:28,600 --> 00:22:29,600 Yeah. 367 00:22:29,600 --> 00:22:32,020 You look at something, the eye tracking picks up what you're looking at. 368 00:22:32,020 --> 00:22:36,800 You just touch your fingers together as a click and how intuitive. 369 00:22:36,800 --> 00:22:42,080 Look, one of the things here that we don't know how it's going to play out necessarily 370 00:22:42,080 --> 00:22:46,320 is will this transform a category that struggled to get going? 371 00:22:46,320 --> 00:22:51,840 As marketers, if this starts to become a dominating way of how people interact with content and 372 00:22:51,840 --> 00:22:55,840 information, we're of course going to have to keep that in mind because we're going to 373 00:22:55,840 --> 00:23:01,880 be able to create new experiences, how people interact with their content is going to change. 374 00:23:01,880 --> 00:23:04,040 I think we need to keep it in mind from that perspective. 375 00:23:04,040 --> 00:23:08,300 The key question that it will take probably several years to really unpick is will this 376 00:23:08,300 --> 00:23:14,760 be a leap forward enough to overcome some of the issues of technology until now? 377 00:23:14,760 --> 00:23:18,680 To be honest, seeing someone's eyes through the front of the screen is either going to 378 00:23:18,680 --> 00:23:20,720 work or it's not. 379 00:23:20,720 --> 00:23:27,480 For those that are not aware, the Vision Pro VR headset has a screen on the front of it 380 00:23:27,480 --> 00:23:31,000 in essence that shows the user's eyes under certain conditions. 381 00:23:31,000 --> 00:23:35,320 So that if Martin's using the headset and I walk into the room, it's almost as if an 382 00:23:35,320 --> 00:23:40,120 opaque window on the front of the headset suddenly goes clear and I can see Martin's 383 00:23:40,120 --> 00:23:41,800 eyes as I have a conversation with him. 384 00:23:41,800 --> 00:23:43,560 But of course, I'm not looking at Martin's eyes. 385 00:23:43,560 --> 00:23:46,880 I'm looking at a digital representation of them. 386 00:23:46,880 --> 00:23:50,440 Basically there is a camera on the inside of the headset that's photographing the eyes 387 00:23:50,440 --> 00:23:53,840 so I can see them and I can see where Martin's looking and stuff like that. 388 00:23:53,840 --> 00:23:56,840 I think we're either going to adapt to that type of thing and go, oh yeah, that's kind 389 00:23:56,840 --> 00:23:57,840 of normal. 390 00:23:57,840 --> 00:24:00,480 I like to see where people's eyes are when I have a conversation with them or it's going 391 00:24:00,480 --> 00:24:04,320 to be one of the weirdest experiences we've ever had because we're going to see these 392 00:24:04,320 --> 00:24:06,600 weird floating eyes. 393 00:24:06,600 --> 00:24:12,320 I think it's going to be in that territory of does it fall in the uncanny valley realm 394 00:24:12,320 --> 00:24:15,960 where it's kind of almost right but just a little bit off. 395 00:24:15,960 --> 00:24:21,080 Am I right in saying that it's not quite like a photo of it? 396 00:24:21,080 --> 00:24:23,600 It's almost, it makes a scandia face. 397 00:24:23,600 --> 00:24:25,840 So that's for FaceTime. 398 00:24:25,840 --> 00:24:35,440 So there's like a 3D representation of you for FaceTime that is very uncanny valley. 399 00:24:35,440 --> 00:24:41,480 Not probably better than like a computer game character, but not as good as a CG character 400 00:24:41,480 --> 00:24:45,480 in a movie, probably somewhere in between based on the examples I've seen. 401 00:24:45,480 --> 00:24:50,720 As I understand it, the eyes that you see is a real time video feed of your actual eyes, 402 00:24:50,720 --> 00:24:53,480 but of course they're not your eyes. 403 00:24:53,480 --> 00:24:54,940 It's a display. 404 00:24:54,940 --> 00:24:57,440 It tries to make it look like the device is see through. 405 00:24:57,440 --> 00:24:58,440 It's not see through. 406 00:24:58,440 --> 00:25:02,120 So I think we're either going to adapt to that and it's going to be quite normal and 407 00:25:02,120 --> 00:25:07,800 over time as technology gets thinner and more glass like, probably never glass like, oh 408 00:25:07,800 --> 00:25:10,080 never, 2000 years maybe it's glass like. 409 00:25:10,080 --> 00:25:13,880 I think it's going to be hard to get these things right down with the level of power 410 00:25:13,880 --> 00:25:17,880 and computing that they're trying to do and sensors that they're trying to do into a pair 411 00:25:17,880 --> 00:25:18,880 of glasses. 412 00:25:18,880 --> 00:25:23,040 Although there are AR glasses out there that you can watch TV on that look like normal 413 00:25:23,040 --> 00:25:26,920 glasses and by watch TV I mean a virtual display. 414 00:25:26,920 --> 00:25:32,880 But hey, maybe we'll do an AR VR special, but for now we probably, because I could talk 415 00:25:32,880 --> 00:25:38,520 about this for hours, we probably should get back to the, what were the last bits here 416 00:25:38,520 --> 00:25:44,760 of Apple's developer conference that really stuck out to you? 417 00:25:44,760 --> 00:25:48,200 Anything else that listeners need to know about? 418 00:25:48,200 --> 00:25:50,600 Not a huge amount. 419 00:25:50,600 --> 00:25:55,720 I just think it's interesting that the focus here is on very much on features and applications 420 00:25:55,720 --> 00:25:58,480 of machine learning rather than the tech and the models. 421 00:25:58,480 --> 00:26:04,640 If you contrast that to the recent Google IO event, so the Google Developer Conference 422 00:26:04,640 --> 00:26:11,720 where I think every fourth word was AI, they just didn't talk about that here. 423 00:26:11,720 --> 00:26:21,160 We came away from Google IO hearing about Palm, or should I say Palm 2 and MedPalm and 424 00:26:21,160 --> 00:26:24,760 Music LM and various other models. 425 00:26:24,760 --> 00:26:29,720 Whereas the focus didn't seem like that's what was going on at Apple. 426 00:26:29,720 --> 00:26:35,920 They were more interested in what the features and functionality of their equipment was. 427 00:26:35,920 --> 00:26:40,720 Now obviously they did talk about throughout the conference itself, Core ML and Create 428 00:26:40,720 --> 00:26:46,760 ML, which are models that developers can access to build and extend the functionality of their 429 00:26:46,760 --> 00:26:50,360 apps and they look very cool in and of themselves. 430 00:26:50,360 --> 00:26:55,320 I would actually recommend that people check out the Apple Developer website to see some 431 00:26:55,320 --> 00:26:59,800 of the capabilities because they really are allowing developers quite a lot of power to 432 00:26:59,800 --> 00:27:03,160 extend the functionality of some of their apps. 433 00:27:03,160 --> 00:27:05,680 But yeah, I just thought it was interesting. 434 00:27:05,680 --> 00:27:11,880 They're not talking about, even when they're talking about the transformer model that's 435 00:27:11,880 --> 00:27:16,760 on the device, they're not telling you anything about that model. 436 00:27:16,760 --> 00:27:20,120 They're not telling you if it's an open source one, that they've trained themselves, they're 437 00:27:20,120 --> 00:27:23,520 not making great claims about its capabilities. 438 00:27:23,520 --> 00:27:27,560 They're just saying there's on device machine learning. 439 00:27:27,560 --> 00:27:35,040 Yeah, I mean, again, probably could spend an hour just talking about Apple. 440 00:27:35,040 --> 00:27:40,120 In essence, they've run a closed ship and a closed system and for me there's a reason 441 00:27:40,120 --> 00:27:42,800 they're not using the word AI and that's because everyone else is. 442 00:27:42,800 --> 00:27:45,800 They probably want to see if there's a way they can own machine learning instead. 443 00:27:45,800 --> 00:27:48,280 Do they want to talk about other people's models? 444 00:27:48,280 --> 00:27:51,720 No, I don't want to give credence to other people's models and work. 445 00:27:51,720 --> 00:27:56,160 They probably just want to make everything very Apple centric and hey, they may be in-house 446 00:27:56,160 --> 00:27:59,040 proprietary models that they've built. 447 00:27:59,040 --> 00:28:01,480 But either way, I don't think we'd get to know. 448 00:28:01,480 --> 00:28:05,800 I just want to clarify one thing on that Vision Pro cost that I said, I've just Googled it 449 00:28:05,800 --> 00:28:09,440 and I cannot find any other reference to an additional fee for the battery. 450 00:28:09,440 --> 00:28:14,920 So I think I was led astray with that tweet. 451 00:28:14,920 --> 00:28:15,920 There we go. 452 00:28:15,920 --> 00:28:20,520 That's good to know because that would be a significant bump up in the cost to be able 453 00:28:20,520 --> 00:28:24,600 to not be tethered to a plug socket. 454 00:28:24,600 --> 00:28:26,640 Cool wings. 455 00:28:26,640 --> 00:28:34,680 Let's move on to our next story, which is going to be about the significant amount of 456 00:28:34,680 --> 00:28:37,080 generative video and image news this week. 457 00:28:37,080 --> 00:28:40,520 So we're going to crack through a load of it and then we'll unpick it a bit and talk 458 00:28:40,520 --> 00:28:42,040 about what it means for marketers. 459 00:28:42,040 --> 00:28:45,960 So we'll start with the quick one first because we're going to dive a bit deeper into this 460 00:28:45,960 --> 00:28:51,920 in our tool of the week, but RunwayML has made its Gen2 text to video tool more widely 461 00:28:51,920 --> 00:28:52,920 available. 462 00:28:52,920 --> 00:28:55,160 So we've talked a bit about this on the podcast previously. 463 00:28:55,160 --> 00:28:57,040 It was in a closed beta. 464 00:28:57,040 --> 00:29:01,880 Now in effect, anybody with a RunwayML account can use the tool and we've been playing with 465 00:29:01,880 --> 00:29:02,880 it. 466 00:29:02,880 --> 00:29:04,000 So we'll talk about that later. 467 00:29:04,000 --> 00:29:11,360 We had news that ClipDrop, which is the Stability AI image tool that I think Martin provided 468 00:29:11,360 --> 00:29:16,000 as a tool of the week a few episodes ago, have released a new tool within that suite 469 00:29:16,000 --> 00:29:21,120 called Uncrop, which I think is a great name, which is basically a generative fill tool 470 00:29:21,120 --> 00:29:26,840 for expanding images beyond its borders, a little bit like Adobe Firefly can do, which 471 00:29:26,840 --> 00:29:29,760 we've talked about in previous episodes. 472 00:29:29,760 --> 00:29:34,160 Having had a quick play with it, it's really quick and easy to use. 473 00:29:34,160 --> 00:29:39,200 And for the really tough examples I gave it, I think it may have even done a better job 474 00:29:39,200 --> 00:29:41,160 than Adobe Firefly, if I'm honest. 475 00:29:41,160 --> 00:29:43,800 All testing is required, but it's certainly not a junk tool. 476 00:29:43,800 --> 00:29:48,360 It looks like it could probably do some pretty cool stuff. 477 00:29:48,360 --> 00:29:53,800 While we're talking about images, of course, then we had Adobe Express, which now includes 478 00:29:53,800 --> 00:29:55,140 generative AI tools. 479 00:29:55,140 --> 00:29:56,480 So this is a beta. 480 00:29:56,480 --> 00:30:01,040 As the website says, you can create video marketing and social content, edit photos 481 00:30:01,040 --> 00:30:08,960 and PDFs, make it amazing with all Adobe powers all in one app, including generative AI tools 482 00:30:08,960 --> 00:30:15,600 from Adobe Firefly and easy one click tasks like removing backgrounds. 483 00:30:15,600 --> 00:30:20,800 And a lot of this leans into generative AI, so text to image and text to text effect, 484 00:30:20,800 --> 00:30:26,040 but all baked into Adobe Express, which is a really cool tool for quickly knocking up 485 00:30:26,040 --> 00:30:29,920 banner images for social and little animated gifs and stuff like that. 486 00:30:29,920 --> 00:30:31,440 And so that's pretty cool. 487 00:30:31,440 --> 00:30:38,000 And then the final bit of news here is Adobe Firefly for Enterprises arrived. 488 00:30:38,000 --> 00:30:40,320 So that businesses can leverage that. 489 00:30:40,320 --> 00:30:47,260 I think we've talked previously about some of the, what's the word I'm looking for? 490 00:30:47,260 --> 00:30:53,800 Copyright issues related to using generative AI, especially for image generation because 491 00:30:53,800 --> 00:30:58,280 of what were the tools trained on and was it copyrighted images and will those tools 492 00:30:58,280 --> 00:30:59,400 get in trouble later? 493 00:30:59,400 --> 00:31:01,920 Will users of those tools get in trouble later? 494 00:31:01,920 --> 00:31:09,800 Well, now enterprises can lean into generative AI to create images and amend images. 495 00:31:09,800 --> 00:31:13,680 But one of the most interesting things about this was that businesses can train Firefly 496 00:31:13,680 --> 00:31:16,420 on their brand specific assets. 497 00:31:16,420 --> 00:31:21,040 So this might make it easier to really at scale produce lots of content for your business 498 00:31:21,040 --> 00:31:25,480 if you're able to train it on your brand colors, the typical imagery that you use and all that 499 00:31:25,480 --> 00:31:26,480 good stuff. 500 00:31:26,480 --> 00:31:27,480 So that sounds pretty cool. 501 00:31:27,480 --> 00:31:31,080 But what really caught my eye here, and I will take a breath so we can go a bit deeper 502 00:31:31,080 --> 00:31:36,040 into this together in a minute, Martin, is some of the stuff that's now on the Adobe 503 00:31:36,040 --> 00:31:43,400 Firefly landing page, which includes sketched image and sketched vector creation. 504 00:31:43,400 --> 00:31:47,280 The example that we know will be the best possible version of it that plays on the page 505 00:31:47,280 --> 00:31:48,600 is very cool. 506 00:31:48,600 --> 00:31:55,080 A designer has knocked up some sketch examples for some letters for a logo, highlights them 507 00:31:55,080 --> 00:32:00,840 and asks for some vector based variations and then proceeds to edit them in, I guess 508 00:32:00,840 --> 00:32:03,960 what would be perhaps Adobe Illustrator. 509 00:32:03,960 --> 00:32:05,440 That was pretty cool. 510 00:32:05,440 --> 00:32:09,840 There's an example where you can change the mood, atmosphere or even the weather in a 511 00:32:09,840 --> 00:32:12,460 video through a text prompt. 512 00:32:12,460 --> 00:32:16,080 So describe what you want it to look like and it will change the colors and the settings 513 00:32:16,080 --> 00:32:17,080 to match. 514 00:32:17,080 --> 00:32:22,300 And the example there is some sort of wooden shack in a grassy field in summer that they 515 00:32:22,300 --> 00:32:27,460 then change it to being in winter and then they play the video and the snow falls. 516 00:32:27,460 --> 00:32:32,360 That's a pretty insane example if it turns out that that actually is easy to duplicate 517 00:32:32,360 --> 00:32:33,960 when you have to do it yourself. 518 00:32:33,960 --> 00:32:38,240 And then the last example on the page was high quality 3D renders where you can turn 519 00:32:38,240 --> 00:32:42,040 very simple 3D compositions into photorealistic images. 520 00:32:42,040 --> 00:32:47,120 And the example here is a simple render of a watch that then they give a leather strap 521 00:32:47,120 --> 00:32:52,200 to and different treatments of what the watch face might look like if it was made of like 522 00:32:52,200 --> 00:32:55,200 silver or different types of alloys and all this type of stuff. 523 00:32:55,200 --> 00:33:01,440 And again, very impressive and I should state they are on the coming soon section part of 524 00:33:01,440 --> 00:33:02,440 the site. 525 00:33:02,440 --> 00:33:06,320 It's not that you can get these now, but one assumes they wouldn't be showing us these 526 00:33:06,320 --> 00:33:11,280 examples if they weren't relatively close within the next six to 12 months of being 527 00:33:11,280 --> 00:33:12,880 able to launch those. 528 00:33:12,880 --> 00:33:19,520 So some loads going on at the moment in text to image and text to video and actually some 529 00:33:19,520 --> 00:33:24,760 really cool things that even at the beginning of this a few months ago, I wouldn't have 530 00:33:24,760 --> 00:33:28,040 realised we'd have that power to hand so quickly. 531 00:33:28,040 --> 00:33:31,880 So Martin, what do you think about all of this explosion in text to image and text to 532 00:33:31,880 --> 00:33:32,880 video? 533 00:33:32,880 --> 00:33:34,800 Difficult to know where to start. 534 00:33:34,800 --> 00:33:36,840 There's been so much in there. 535 00:33:36,840 --> 00:33:40,900 I'm going to part runway, uh, Gentoo for a minute because we're going to obviously go 536 00:33:40,900 --> 00:33:45,560 through that one as I'll taller the week, but Clip Drop Uncrop it. 537 00:33:45,560 --> 00:33:46,560 You're right. 538 00:33:46,560 --> 00:33:47,560 It is a great name. 539 00:33:47,560 --> 00:33:48,560 I do love that one. 540 00:33:48,560 --> 00:33:51,580 I've been playing around with it and I like it. 541 00:33:51,580 --> 00:33:59,440 So it throws out four different variations of the uncropped image and it does a really 542 00:33:59,440 --> 00:34:00,440 neat job. 543 00:34:00,440 --> 00:34:06,680 Although I am looking at one that I've created here, which has got the most dreadfully ghoulish 544 00:34:06,680 --> 00:34:11,800 looking AI generated human face on it that I've ever seen. 545 00:34:11,800 --> 00:34:17,880 It's quite scary, but no, it does a good job and this is available for free at the moment. 546 00:34:17,880 --> 00:34:22,160 So if anyone wants to go and play with it, um, get stuck in over there, you don't need 547 00:34:22,160 --> 00:34:24,840 the pro account and it is generating pretty quickly as well. 548 00:34:24,840 --> 00:34:28,040 So there's just so much going on in this space. 549 00:34:28,040 --> 00:34:32,200 I think it's really exciting time to be a creative when I'm demoing this to people at 550 00:34:32,200 --> 00:34:33,200 the moment. 551 00:34:33,200 --> 00:34:39,200 Um, so I did a workshop last week and we were showing people, um, the kind of the similar 552 00:34:39,200 --> 00:34:47,800 capabilities, uh, people really set up and pay attention because things that, for instance, 553 00:34:47,800 --> 00:34:51,760 a workshop that I was doing recently was with some small business owners, uh, kind of business 554 00:34:51,760 --> 00:34:53,400 coaches, consultants, that kind of thing. 555 00:34:53,400 --> 00:34:58,160 They don't necessarily have a team of designers and what on hand all of the time. 556 00:34:58,160 --> 00:35:04,000 So showing them that, Hey, look, you can use this generative fill. 557 00:35:04,000 --> 00:35:06,860 You can expand that image that wasn't quite aligned correctly. 558 00:35:06,860 --> 00:35:08,220 So you couldn't crop it as a square. 559 00:35:08,220 --> 00:35:12,520 Now you can crop it as a square and it works. 560 00:35:12,520 --> 00:35:13,520 It's really useful. 561 00:35:13,520 --> 00:35:18,160 These are just, you don't have to be a designer to figure it out now. 562 00:35:18,160 --> 00:35:20,840 So I love it. 563 00:35:20,840 --> 00:35:27,600 The Firefly examples are really exciting. 564 00:35:27,600 --> 00:35:32,920 The ones that you mentioned there, the vector ones are going to be very, very cool, but 565 00:35:32,920 --> 00:35:36,560 I'm really interested in the enterprise. 566 00:35:36,560 --> 00:35:42,480 That's where things are going to get interesting from a question that always comes back to 567 00:35:42,480 --> 00:35:45,760 me whenever I'm demoing this tech at the moment is can we do it with our brand colors? 568 00:35:45,760 --> 00:35:47,080 Can we do it with our brand assets? 569 00:35:47,080 --> 00:35:52,040 Can we do it with, you know, how much do I need to, can I train it? 570 00:35:52,040 --> 00:35:55,800 This is when people have kind of right at the beginning of their journey. 571 00:35:55,800 --> 00:35:59,360 And now you're going, yeah, yeah, you can. 572 00:35:59,360 --> 00:36:00,360 Okay. 573 00:36:00,360 --> 00:36:01,360 Not immediately. 574 00:36:01,360 --> 00:36:04,440 You've got to sign up and speed to the sales team to get this. 575 00:36:04,440 --> 00:36:11,300 But ultimately this is going to be creating assets that are on brand in seconds with simple 576 00:36:11,300 --> 00:36:13,160 text prompts as it says. 577 00:36:13,160 --> 00:36:14,160 Yeah, I agree. 578 00:36:14,160 --> 00:36:19,280 I think, I mean, there are some tools that promise this, but I've never personally, and 579 00:36:19,280 --> 00:36:23,820 I'm obviously not a designer come across anything that really does it with the type of robustness 580 00:36:23,820 --> 00:36:24,820 that I'm looking for. 581 00:36:24,820 --> 00:36:28,640 But like I create one banner ad or even a print ad. 582 00:36:28,640 --> 00:36:32,600 And then I say, right, I need, you know, 320 by 320. 583 00:36:32,600 --> 00:36:38,840 I need a leaderboard, like the power and the accuracy and the quality of these generated 584 00:36:38,840 --> 00:36:46,860 images at this point, I would trust this tool to go ahead and produce all of the variations 585 00:36:46,860 --> 00:36:52,000 I need to run a campaign across a variety of different platforms, probably at the click 586 00:36:52,000 --> 00:36:53,000 of a button. 587 00:36:53,000 --> 00:36:59,600 And obviously that's not quite here yet, at least not in the Adobe tools as I see them, 588 00:36:59,600 --> 00:37:02,080 but it's can't be far away. 589 00:37:02,080 --> 00:37:06,160 And that just allows you to deploy those campaigns very quickly at scale. 590 00:37:06,160 --> 00:37:12,960 Now plug that into a programmatic ad management tool or Google display network. 591 00:37:12,960 --> 00:37:19,280 Well, now you're now your power to run different messaging variants, subtly different image 592 00:37:19,280 --> 00:37:22,780 variants, but on the same theme, all still in your brand colors. 593 00:37:22,780 --> 00:37:25,880 One assumes you could do even easier at scale. 594 00:37:25,880 --> 00:37:30,680 So there's a load of cool stuff, isn't there, that potentially comes from this in terms 595 00:37:30,680 --> 00:37:34,400 of high quality outputs quickly at scale. 596 00:37:34,400 --> 00:37:42,400 Yeah, and I think that goes to the discussion we were having about Meta's generative AI 597 00:37:42,400 --> 00:37:44,920 tools for advertisers as well. 598 00:37:44,920 --> 00:37:47,440 I think there's going to be quite a lot of overlap, isn't there, with these kinds of 599 00:37:47,440 --> 00:37:48,440 tools. 600 00:37:48,440 --> 00:37:54,560 So yeah, it's just an interesting space to watch at the moment. 601 00:37:54,560 --> 00:38:01,320 I don't do a great deal of design, but I have found that these kinds of functions, particularly 602 00:38:01,320 --> 00:38:04,880 AI, I knock things up in Canva pretty quickly. 603 00:38:04,880 --> 00:38:12,880 But having the magic edit, the magic eraser, having that at your fingertips is, it speeds 604 00:38:12,880 --> 00:38:14,320 up my workflow. 605 00:38:14,320 --> 00:38:18,280 And that's great because quite honestly, I hate any graphic design. 606 00:38:18,280 --> 00:38:25,440 Yeah, I think definitely on the execution front and actually like creating the assets, 607 00:38:25,440 --> 00:38:26,600 I think is a big help. 608 00:38:26,600 --> 00:38:28,160 It has been interesting. 609 00:38:28,160 --> 00:38:33,280 We've been working on a series of creative campaign projects over the last month or two 610 00:38:33,280 --> 00:38:35,360 at Biostrata. 611 00:38:35,360 --> 00:38:40,280 And I personally have been experimenting with some image generation tools to see how they 612 00:38:40,280 --> 00:38:41,520 might support with that. 613 00:38:41,520 --> 00:38:47,080 So Photoshop, generative AI, generative image tools, Mid Journey. 614 00:38:47,080 --> 00:38:53,360 And our team presented some concepts to a client recently in collaboration. 615 00:38:53,360 --> 00:38:58,880 We built a team, a creative team on that, a creative director, scientific copywriter, 616 00:38:58,880 --> 00:39:00,600 brand strategist. 617 00:39:00,600 --> 00:39:07,160 And the ideas that they came up with and the concept, the example concepts, they just blew 618 00:39:07,160 --> 00:39:11,560 away anything I've been able to generate on any tools, honestly. 619 00:39:11,560 --> 00:39:16,240 You get a proper team together who know an audience, know an industry, know how to create 620 00:39:16,240 --> 00:39:18,960 compelling campaign creative. 621 00:39:18,960 --> 00:39:22,960 You have them IDA and then you have them knock something up and then take something through 622 00:39:22,960 --> 00:39:24,560 to final concept. 623 00:39:24,560 --> 00:39:31,440 AI tools are miles away from being able to do that, in my opinion, because what we were 624 00:39:31,440 --> 00:39:35,160 able to generate through that process is just way better. 625 00:39:35,160 --> 00:39:39,760 And in order to try and see what avenues you could go in, I've been playing with Mid Journey 626 00:39:39,760 --> 00:39:44,640 to try and even riff on some of the themes we'd already created, but the ability to really 627 00:39:44,640 --> 00:39:48,760 get what you want out of Mid Journey, and this obviously could be a limitation in our 628 00:39:48,760 --> 00:39:55,040 skill sets and our experience using Mid Journey, it's just you can get cool stuff, but can 629 00:39:55,040 --> 00:40:02,640 you get the thing you want that is very iterative, takes at least in my hands hours and is probably 630 00:40:02,640 --> 00:40:05,360 not the most efficient or effective way to do it. 631 00:40:05,360 --> 00:40:10,000 No, not if you're, you know, if you are a designer, if you're a creative, like I just 632 00:40:10,000 --> 00:40:14,840 don't think like this, but if you are that person, you have something in your mind and 633 00:40:14,840 --> 00:40:21,160 you know what you want to do to get there and you've already got all of those tools 634 00:40:21,160 --> 00:40:25,600 and heuristics that help you get your job done quickly. 635 00:40:25,600 --> 00:40:31,040 You're not sitting there basically at the whim of an AI, typing in a sentence and then 636 00:40:31,040 --> 00:40:33,880 going, let's see what for it comes up with this time. 637 00:40:33,880 --> 00:40:36,920 No, they're all shit. 638 00:40:36,920 --> 00:40:40,040 Like that such an inefficient waste of time. 639 00:40:40,040 --> 00:40:43,080 If you're a creative, you know how to get there. 640 00:40:43,080 --> 00:40:44,080 Agreed. 641 00:40:44,080 --> 00:40:47,920 I think you could probably be inspired by the things that you see, but I don't think 642 00:40:47,920 --> 00:40:51,720 they're going to be the things that you're, that often you're going to produce. 643 00:40:51,720 --> 00:40:52,720 I should caveat. 644 00:40:52,720 --> 00:40:58,080 I've seen some pretty awesome things generated by Mid Journey that are for brands that as 645 00:40:58,080 --> 00:41:01,880 a creative agency, you could use to pitch a brand to get their work. 646 00:41:01,880 --> 00:41:05,240 We've talked about some of them on the, on the podcast previously. 647 00:41:05,240 --> 00:41:10,920 So, so I do think there are certain niche outputs that you're looking for that Mid Journey 648 00:41:10,920 --> 00:41:13,520 is good at that you can use for. 649 00:41:13,520 --> 00:41:19,400 The way a potential game changer might be on this is that sketching tool, right? 650 00:41:19,400 --> 00:41:24,160 Because presumably, I know the example is a logo or, you know, a stylized bee, I think 651 00:41:24,160 --> 00:41:29,840 is one of the examples, but presumably if I have a vision and one of the great creative 652 00:41:29,840 --> 00:41:34,040 directors we've worked with in the past does a lot of sketch work as initial concepts, 653 00:41:34,040 --> 00:41:39,000 but to be able to click a button that then creates a stylized, more polished image based 654 00:41:39,000 --> 00:41:43,600 on that sketch at the click of a button could be really quite interesting. 655 00:41:43,600 --> 00:41:44,600 Right. 656 00:41:44,600 --> 00:41:46,760 Let's jump into story three, Martin. 657 00:41:46,760 --> 00:41:54,620 This is about DeepMind announcing generalized AI to optimize computing performance. 658 00:41:54,620 --> 00:41:55,880 Tell us a bit more about this. 659 00:41:55,880 --> 00:41:56,880 Yeah. 660 00:41:56,880 --> 00:42:03,720 At first glance, you could be forgiven for thinking that this is a bit of a yawn story, 661 00:42:03,720 --> 00:42:05,880 but actually it's, it's quite big. 662 00:42:05,880 --> 00:42:13,440 And I think marketers and AI enthusiasts, generally speaking, should be very interested 663 00:42:13,440 --> 00:42:14,440 in it. 664 00:42:14,440 --> 00:42:21,560 So DeepMind has announced some new applications of its AI models. 665 00:42:21,560 --> 00:42:26,300 So AlphaZero, MuZero and AlphaDev. 666 00:42:26,300 --> 00:42:30,960 And what they've found is that they're able to take these, what are basically general 667 00:42:30,960 --> 00:42:33,520 purpose AI models. 668 00:42:33,520 --> 00:42:41,280 So you might remember AlphaZero was trained to play games and you had AlphaGo and it learned 669 00:42:41,280 --> 00:42:49,040 the game Go and became a world champion and beat the world champion back in 2016. 670 00:42:49,040 --> 00:42:55,600 And that kind of blew away people's kind of expectations for what we can think of the 671 00:42:55,600 --> 00:42:57,400 capabilities of AI. 672 00:42:57,400 --> 00:43:05,040 Now these models have now been put into real world applications and to try to optimize 673 00:43:05,040 --> 00:43:07,360 different computer systems. 674 00:43:07,360 --> 00:43:09,960 And what they've done is basically three big things. 675 00:43:09,960 --> 00:43:17,140 One that they've, they've driven particularly good advances in data center optimization. 676 00:43:17,140 --> 00:43:21,520 So optimizing the hardware within data centers. 677 00:43:21,520 --> 00:43:27,960 Video compression, they've been working on that, which is for a company like Google, 678 00:43:27,960 --> 00:43:33,480 which has YouTube, that's quite impactful. 679 00:43:33,480 --> 00:43:41,440 And they've also discovered faster algorithms for sorting and searching. 680 00:43:41,440 --> 00:43:45,680 So it can seem a bit dry, but please bear with me. 681 00:43:45,680 --> 00:43:55,880 AlphaZero has been used to reduce underutilized hardware at Google's dentists, dentists centers. 682 00:43:55,880 --> 00:43:56,880 I can't speak today. 683 00:43:56,880 --> 00:44:00,600 Dental centers at the dentist, at the dental centers. 684 00:44:00,600 --> 00:44:01,600 You sure that's the tooth? 685 00:44:01,600 --> 00:44:04,200 Is that the whole tooth and nothing but the tooth? 686 00:44:04,200 --> 00:44:08,680 They're getting the drills and saying, yep, this drill isn't used enough. 687 00:44:08,680 --> 00:44:10,640 We're going to make sure it's properly utilized. 688 00:44:10,640 --> 00:44:12,520 Anyway, right back to the real world. 689 00:44:12,520 --> 00:44:14,600 We might get complaints this week, Martin. 690 00:44:14,600 --> 00:44:17,040 We don't actually get complaints. 691 00:44:17,040 --> 00:44:18,720 I might complain. 692 00:44:18,720 --> 00:44:19,880 It's warranted, isn't it? 693 00:44:19,880 --> 00:44:22,480 Bless you, dear listener. 694 00:44:22,480 --> 00:44:23,680 Google's data centers. 695 00:44:23,680 --> 00:44:30,840 So they've been able to reduce underutilized hardware by 19%, which is massive. 696 00:44:30,840 --> 00:44:38,000 So it's improving resource utilization by recognizing tasks as they come into the hardware 697 00:44:38,000 --> 00:44:43,680 and allowing it to make better decisions about what hardware should be used to do a task. 698 00:44:43,680 --> 00:44:44,680 You can... 699 00:44:44,680 --> 00:44:48,920 The kind of visual example I saw of it made me think of... 700 00:44:48,920 --> 00:44:52,680 You remember when you used to have to defrag your hard drive all the time. 701 00:44:52,680 --> 00:44:55,880 All my PC is full, I have to defrag it. 702 00:44:55,880 --> 00:44:57,520 I need to do that with my brain. 703 00:44:57,520 --> 00:45:01,240 But yeah, I do also remember that on the hard drives. 704 00:45:01,240 --> 00:45:04,360 Well, that seems like what this is doing, basically. 705 00:45:04,360 --> 00:45:10,120 It's real time defragging of the data centers is the best way I could describe it. 706 00:45:10,120 --> 00:45:15,920 Then you've got Mu Zero, which DeepMind have been working in collaboration with YouTube. 707 00:45:15,920 --> 00:45:21,680 And this doesn't sound like much, but when you think about this at scale, it's massive. 708 00:45:21,680 --> 00:45:29,640 They've been able to reduce the video bitrate by 4% without impacting visual quality. 709 00:45:29,640 --> 00:45:34,740 So optimizes individual frame compression and how the frames are then grouped. 710 00:45:34,740 --> 00:45:42,200 And it suggests that there is much broader potential applications for video compression. 711 00:45:42,200 --> 00:45:49,960 Video makes up the vast majority of data that we send globally around the internet. 712 00:45:49,960 --> 00:45:57,680 So if you can shave off, if you can compress videos by 4% without losing any issue, well, 713 00:45:57,680 --> 00:46:00,760 without degrading the video quality, that's enormous. 714 00:46:00,760 --> 00:46:06,320 That's a really significant saving on bandwidth and energy. 715 00:46:06,320 --> 00:46:11,560 And then finally, AlphaDev, a version of AlphaZero, has discovered faster sorting and hashing 716 00:46:11,560 --> 00:46:20,900 algorithms enhancing efficiency in sorting short sequences by 70% and longer sequences 717 00:46:20,900 --> 00:46:23,720 by 1.7%. 718 00:46:23,720 --> 00:46:31,520 So hashing efficiency has been approved by 30%, all of which basically says that data 719 00:46:31,520 --> 00:46:37,460 retrieval and searching and sorting of data is going to become more efficient, which, 720 00:46:37,460 --> 00:46:39,680 like I say, bear with me, this is a bit dry. 721 00:46:39,680 --> 00:46:46,300 But it just means that if this gets applied to, if you're an e-commerce store, right, 722 00:46:46,300 --> 00:46:52,720 and you can tap into these models for your store's search engine, you've got maybe hundreds 723 00:46:52,720 --> 00:46:56,000 of thousands or millions of SKUs. 724 00:46:56,000 --> 00:47:01,120 The retrieval is going to increase by, again, it's tiny amounts, it's milliseconds, they're 725 00:47:01,120 --> 00:47:06,180 shaving milliseconds off, but milliseconds count for user experience. 726 00:47:06,180 --> 00:47:09,960 It's going to be using less energy, dramatic reduction in energy. 727 00:47:09,960 --> 00:47:16,680 I saw one industry commentator saying it's going to take like 30% of the amount of energy 728 00:47:16,680 --> 00:47:22,880 to do searching and sorting compared to what it is currently if you were to apply this 729 00:47:22,880 --> 00:47:23,880 mechanism. 730 00:47:23,880 --> 00:47:33,360 So it's something of a dry topic, but I think when we listen to publications from DeepMind, 731 00:47:33,360 --> 00:47:38,180 quite often the application seems somewhat abstract, with the exception of something 732 00:47:38,180 --> 00:47:42,880 like the protein folding, which was obviously pretty massive. 733 00:47:42,880 --> 00:47:48,760 But when it's like, oh, we can train a model to be anyone at chess or at Go, or look at 734 00:47:48,760 --> 00:47:55,080 them, they've applied AlphaGo to this Atari video game and it's just completed it quicker 735 00:47:55,080 --> 00:47:57,680 than any human could ever. 736 00:47:57,680 --> 00:47:59,000 You go, that's super cool. 737 00:47:59,000 --> 00:48:02,760 But now we're starting to see this come into, there's the same technology, the technology 738 00:48:02,760 --> 00:48:07,040 that started off just playing Atari and playing Go, we're now starting to see this come into 739 00:48:07,040 --> 00:48:14,000 real world applications where at a global scale, it can actually make a difference on 740 00:48:14,000 --> 00:48:20,400 both energy efficiency and user experience and more. 741 00:48:20,400 --> 00:48:21,400 Yeah. 742 00:48:21,400 --> 00:48:24,760 Do you know what I thought was cool about this story? 743 00:48:24,760 --> 00:48:31,080 And this is me speaking as a novice in a lot of these topics, is if I understood correctly, 744 00:48:31,080 --> 00:48:37,920 what actually happened was DeepMind asked AI tools to try and solve computer and software 745 00:48:37,920 --> 00:48:42,480 problems that humans had run out of ideas for solving. 746 00:48:42,480 --> 00:48:50,360 And then the algorithms, AlphaDev, was able to actually find new ways of compressing information 747 00:48:50,360 --> 00:48:56,960 or achieving and overcoming these technical issues in ways that humans hadn't thought 748 00:48:56,960 --> 00:48:57,960 of. 749 00:48:57,960 --> 00:49:04,400 So therefore, what other domains are we going to be able to apply AI? 750 00:49:04,400 --> 00:49:06,200 We've run out of ideas to solve this problem. 751 00:49:06,200 --> 00:49:08,240 AI, what would you do? 752 00:49:08,240 --> 00:49:12,280 Because this is the first evidence that I know of, of AI being given a tough problem 753 00:49:12,280 --> 00:49:14,160 where humans are gone that was stumped now. 754 00:49:14,160 --> 00:49:16,880 We've got all the optimization out of this that we can. 755 00:49:16,880 --> 00:49:17,880 What do you think? 756 00:49:17,880 --> 00:49:20,640 Secretly probably thinking, humans have done a good job of this. 757 00:49:20,640 --> 00:49:22,520 There's not going to be any opportunities in there. 758 00:49:22,520 --> 00:49:24,260 AI is not going to find anything. 759 00:49:24,260 --> 00:49:25,880 And then look at the amazing stuff it found. 760 00:49:25,880 --> 00:49:29,720 So I think that will be really interesting to see how that plays out. 761 00:49:29,720 --> 00:49:30,840 Yep. 762 00:49:30,840 --> 00:49:33,720 And I think that's where DeepMind are heading. 763 00:49:33,720 --> 00:49:41,080 And that's why training this, the approach that they've taken with AlphaZero is so interesting 764 00:49:41,080 --> 00:49:46,800 because they've just trained this machine to just, without telling it the rules of the 765 00:49:46,800 --> 00:49:53,800 game, but telling it the objective and then letting it kind of run with that and play 766 00:49:53,800 --> 00:50:00,320 against itself and basically millions and millions of playthroughs of a scenario. 767 00:50:00,320 --> 00:50:03,840 It just finds better strategies. 768 00:50:03,840 --> 00:50:07,840 Really interesting, really powerful approach to problem solving. 769 00:50:07,840 --> 00:50:10,800 Yeah, absolutely mind blowing. 770 00:50:10,800 --> 00:50:16,880 Right for the final story of today, we're going to talk a little bit about Meta's roadmap. 771 00:50:16,880 --> 00:50:23,040 So Mark Zuckerberg has outlined Meta's plans for the future in an all hands meeting this 772 00:50:23,040 --> 00:50:30,480 week, highlighting how AI and the metaverse will align with the company's vision. 773 00:50:30,480 --> 00:50:36,040 Actually they've gone so all in on the metaverse, Martin, but they've been really good at AI. 774 00:50:36,040 --> 00:50:38,320 They've got loads of smart AI people doing loads of great work. 775 00:50:38,320 --> 00:50:42,200 They must've been like, ah, crumbs, we went all in on the metaverse and it was AI that 776 00:50:42,200 --> 00:50:44,320 exploded and we're really good at that. 777 00:50:44,320 --> 00:50:49,080 Yeah and to their credit, they've been really big contributors to the open source community 778 00:50:49,080 --> 00:50:50,480 in AI as well. 779 00:50:50,480 --> 00:50:54,280 So that must be somewhat frustrating. 780 00:50:54,280 --> 00:50:57,800 Yeah, so he's obviously like, right, we're really good at AI. 781 00:50:57,800 --> 00:50:58,800 We're not going to give up the metaverse. 782 00:50:58,800 --> 00:51:03,840 We're going to bring those things together and Meta's going to continue to be a really 783 00:51:03,840 --> 00:51:05,480 successful company. 784 00:51:05,480 --> 00:51:10,680 And then in the in the all hands, they introduced plans for their AI powered assistance across 785 00:51:10,680 --> 00:51:15,400 all meta apps, trying to make these technologies more accessible to users. 786 00:51:15,400 --> 00:51:21,280 And they also revealed that they were working on Project 92, which sounds like an order 787 00:51:21,280 --> 00:51:25,720 to kill all Jedi from the Star Wars universe, to be honest, but there you go, which is a 788 00:51:25,720 --> 00:51:28,400 social app similar to Twitter. 789 00:51:28,400 --> 00:51:33,560 And they've also got plans to improve Instagram's reels to better compete with TikTok. 790 00:51:33,560 --> 00:51:39,940 So it's sort of a galvanizing, we're good at AI, we're still believe in the metaverse. 791 00:51:39,940 --> 00:51:43,320 We realize our competitors are doing all these cool things. 792 00:51:43,320 --> 00:51:46,900 We're doing cool things and we're going to be successful with our cool things. 793 00:51:46,900 --> 00:51:50,200 What was your take on this new story, Mide? 794 00:51:50,200 --> 00:51:57,480 They are bringing the chat GPT Bard, call it what you may like experience to to make 795 00:51:57,480 --> 00:51:58,640 messenger on WhatsApp. 796 00:51:58,640 --> 00:52:04,520 Now this for this to be functional, I'm sure they'll do their own kind of chat GPT Bard 797 00:52:04,520 --> 00:52:08,520 version that you can chat with and they'll call it God knows what they'll call it. 798 00:52:08,520 --> 00:52:15,920 But where I think this will be a kind of game changer for them is by extending this into 799 00:52:15,920 --> 00:52:18,440 businesses. 800 00:52:18,440 --> 00:52:26,120 One of the things that separates Facebook or meta from all of the other tech companies 801 00:52:26,120 --> 00:52:32,340 is the user base for their big chat apps. 802 00:52:32,340 --> 00:52:40,840 Messenger and WhatsApp are massive, the global adoption of these two chat tools is huge for 803 00:52:40,840 --> 00:52:44,000 both consumers and businesses. 804 00:52:44,000 --> 00:52:49,220 Now in certainly in Asia, I know in India, WhatsApp is one of the primary channels of 805 00:52:49,220 --> 00:52:53,680 communicating with companies these days. 806 00:52:53,680 --> 00:52:58,140 We're seeing more and more customer support queries being dealt with on Facebook Messenger 807 00:52:58,140 --> 00:53:03,820 around the world, we're seeing marketing, direct marketing being done through messenger 808 00:53:03,820 --> 00:53:07,040 and through through WhatsApp. 809 00:53:07,040 --> 00:53:16,680 If Facebook or meta can get their language models plugged in to business data, I think 810 00:53:16,680 --> 00:53:21,240 this is going to be an absolute showstopper. 811 00:53:21,240 --> 00:53:25,840 We keep talking about, we've mentioned a few times in recent weeks that increasingly the 812 00:53:25,840 --> 00:53:29,360 differentiator is the UX. 813 00:53:29,360 --> 00:53:35,480 Make it easy for people to use the AI without necessarily having to have, you know, any 814 00:53:35,480 --> 00:53:38,360 technical knowledge or you just just make it easy for me to do it. 815 00:53:38,360 --> 00:53:44,360 That's why chat GPT was an overnight success because it made using AI as simple as using 816 00:53:44,360 --> 00:53:45,880 WhatsApp. 817 00:53:45,880 --> 00:53:54,440 If Facebook can do that with, let's say, enabling companies to connect their customer data, 818 00:53:54,440 --> 00:54:03,920 their CRM, whatever it may be, to Facebook Business Suite, for example, and then given 819 00:54:03,920 --> 00:54:09,280 that everybody can log into apps and services these days with Facebook, you know, sign in 820 00:54:09,280 --> 00:54:14,000 with Facebook, log in with Facebook, Facebook Authenticator, all of that kind of stuff. 821 00:54:14,000 --> 00:54:21,280 If I can log in and then can just fire off a message to a customer support AI in Messenger 822 00:54:21,280 --> 00:54:25,880 and it knows who I am because of all the authentication and it's pulling in data in real time because 823 00:54:25,880 --> 00:54:31,280 it's connected to the relevant database, like chat spot, you know, like we're seeing with 824 00:54:31,280 --> 00:54:36,120 or expecting to see people be able to do with HubSpot's chat spot. 825 00:54:36,120 --> 00:54:39,200 I think that's going to be immensely powerful. 826 00:54:39,200 --> 00:54:43,640 Now, if that's not where they're planning on taking it, Mark, please have the idea and 827 00:54:43,640 --> 00:54:45,160 run with it. 828 00:54:45,160 --> 00:54:52,560 My only concern there would be that over the recent years, every time I try to use Meta's 829 00:54:52,560 --> 00:54:58,900 Business Suite, I find it more and more infuriating every time I have to log in and do even the 830 00:54:58,900 --> 00:55:00,240 most simple task. 831 00:55:00,240 --> 00:55:07,440 So whilst they might want to do this, I think they'll still make it incredibly hard for 832 00:55:07,440 --> 00:55:12,680 businesses to actually be able to implement it. 833 00:55:12,680 --> 00:55:13,680 Usability is key, right? 834 00:55:13,680 --> 00:55:18,960 We talk a lot about UX on the podcast and how the magic at the moment, I think there 835 00:55:18,960 --> 00:55:23,880 is absolute technical wizardry, but that seems to be something that lots of companies have 836 00:55:23,880 --> 00:55:25,280 access to. 837 00:55:25,280 --> 00:55:32,960 It's how wonderfully easy and perhaps almost magic in the background that you can actually 838 00:55:32,960 --> 00:55:36,960 make your tool and that will be what helps a company win. 839 00:55:36,960 --> 00:55:37,960 And I agree. 840 00:55:37,960 --> 00:55:44,720 On the WhatsApp front, there's a number of companies, we talked about the tool I was 841 00:55:44,720 --> 00:55:49,480 trialing where it's basically like an advisor on anything you want, like mindfulness. 842 00:55:49,480 --> 00:55:53,120 It's like, oh, I want to, how can I be calmer in these situations? 843 00:55:53,120 --> 00:55:57,560 You can get advice from this app and here's a business decision I need to make and you 844 00:55:57,560 --> 00:55:58,720 can get advice from this app. 845 00:55:58,720 --> 00:56:02,040 So I think third parties are already leaning into WhatsApp to do that. 846 00:56:02,040 --> 00:56:06,480 And if Facebook can basically gobble up a lot of that before other apps can get in, 847 00:56:06,480 --> 00:56:07,480 it makes sense. 848 00:56:07,480 --> 00:56:11,520 We talked previously on the podcast about where the big companies have their motes. 849 00:56:11,520 --> 00:56:15,080 Well, it may not be the large language models because there's loads of them, there's open 850 00:56:15,080 --> 00:56:17,680 source ones and they're all pretty good now. 851 00:56:17,680 --> 00:56:22,120 But I was doing some Googling while you were talking and according to some research from 852 00:56:22,120 --> 00:56:29,200 last year, about this time last year, WhatsApp had around 2.25 billion monthly users. 853 00:56:29,200 --> 00:56:32,840 And according to another bit of research that would at the time would have made it the world's 854 00:56:32,840 --> 00:56:34,800 most popular messaging app. 855 00:56:34,800 --> 00:56:43,720 That's a fairly chunky moat in terms of having a, you know, 25% of the world's population 856 00:56:43,720 --> 00:56:51,880 give or take, maybe 30% of the world's population using your tool, you know, monetize that. 857 00:56:51,880 --> 00:56:54,480 And of course, I'm sure they're already thinking all of this. 858 00:56:54,480 --> 00:56:58,440 It'd be brilliant, Martin, if me and you were coming up with ideas that their teams of strategists 859 00:56:58,440 --> 00:57:00,440 haven't thought of, I think it's unlikely. 860 00:57:00,440 --> 00:57:01,440 Wouldn't it just, wouldn't it just? 861 00:57:01,440 --> 00:57:03,040 Have that one for free, Mark. 862 00:57:03,040 --> 00:57:05,240 Yeah, like get in touch with the podcast. 863 00:57:05,240 --> 00:57:06,960 We don't mind having you on. 864 00:57:06,960 --> 00:57:09,360 Yeah, we'll make space. 865 00:57:09,360 --> 00:57:11,240 To be honest, he won't stop messaging me on LinkedIn. 866 00:57:11,240 --> 00:57:14,920 And this might be the bit where I have to finally say, all right, Mark, you can come 867 00:57:14,920 --> 00:57:15,920 on. 868 00:57:15,920 --> 00:57:19,320 You'd be very welcome, but we can only give you 15 minutes. 869 00:57:19,320 --> 00:57:20,320 Yeah. 870 00:57:20,320 --> 00:57:22,120 So, so some interesting stuff there. 871 00:57:22,120 --> 00:57:27,480 And I think we should, as marketers, it's what channels we have to reach our audiences. 872 00:57:27,480 --> 00:57:31,720 As you said, a lot of businesses are already using WhatsApp, but what ways might we make 873 00:57:31,720 --> 00:57:37,800 that even better and even easier for our audiences, leveraging the data we have about our products 874 00:57:37,800 --> 00:57:43,880 and services, plus the data that then Meta has about our customers in order to really 875 00:57:43,880 --> 00:57:48,720 provide personalized customer service and product recommendations and all those great 876 00:57:48,720 --> 00:57:52,480 things at scale using a messenger app like WhatsApp, right? 877 00:57:52,480 --> 00:57:57,080 We've got to be imagining what these things might look like, what type of data and systems 878 00:57:57,080 --> 00:58:01,680 and processes we might need internally to take advantage of those opportunities. 879 00:58:01,680 --> 00:58:08,600 Just on that, it would be nice, and I hope the industry heads this way in the near future, 880 00:58:08,600 --> 00:58:16,560 if we saw some standardizations and standardized protocols across the industry to help companies 881 00:58:16,560 --> 00:58:21,600 make their data more accessible to things like AI chat box. 882 00:58:21,600 --> 00:58:30,280 Now, I know that with chat GPT plugins, the idea was basically you would natural language 883 00:58:30,280 --> 00:58:32,720 the connection to your database. 884 00:58:32,720 --> 00:58:36,080 You could basically say, this is our data, this is what it looks like, and off you go, 885 00:58:36,080 --> 00:58:38,280 and it would figure it out yourself. 886 00:58:38,280 --> 00:58:44,600 Maybe that is the way it's going to go, but something akin to website schema markup that 887 00:58:44,600 --> 00:58:47,400 just says exactly what is in your database. 888 00:58:47,400 --> 00:58:53,680 If there was that, that would help us to standardize a connection between, let's say, OpenAI versus 889 00:58:53,680 --> 00:58:59,760 Messenger and WhatsApp, and then some sort of Google chat interface and make our content 890 00:58:59,760 --> 00:59:04,800 easily searchable, crawlable, and accessible to users. 891 00:59:04,800 --> 00:59:11,760 That would be nice, particularly for not enterprise businesses, but large businesses and SMEs 892 00:59:11,760 --> 00:59:19,000 that want to make their customer data accessible via these kinds of chat-based interfaces. 893 00:59:19,000 --> 00:59:22,320 Yeah, I think it's an interesting one. 894 00:59:22,320 --> 00:59:24,920 When's the best time to plant a tree 10 years ago? 895 00:59:24,920 --> 00:59:28,200 When's the second best time to plant a tree today? 896 00:59:28,200 --> 00:59:33,440 If you're a business that's not thinking about your data as an asset, and more importantly, 897 00:59:33,440 --> 00:59:38,440 how that data is structured so you can leverage it as an asset, now is the day to start thinking 898 00:59:38,440 --> 00:59:39,440 about that. 899 00:59:39,440 --> 00:59:44,400 I think the thing that will be really interesting, because I think you're absolutely right, mine, 900 00:59:44,400 --> 00:59:50,160 even similar fields of data in similar systems work very differently. 901 00:59:50,160 --> 00:59:55,040 Anybody who's tried to do a sync between Salesforce and HubSpot will know all the ways it can break 902 00:59:55,040 --> 01:00:01,360 in completely unexpected ways because of what HubSpot calls a contact and Salesforce calls 903 01:00:01,360 --> 01:00:06,160 a contact and the difference between a lead and even difference between an MQL in one 904 01:00:06,160 --> 01:00:11,820 company versus another, things that seem like they should be pretty standard between systems 905 01:00:11,820 --> 01:00:14,600 and between businesses are not. 906 01:00:14,600 --> 01:00:17,520 I wonder if it's kind of a bet. 907 01:00:17,520 --> 01:00:22,180 Do I bet that those with structured data will be able to move faster, monetize their data 908 01:00:22,180 --> 01:00:26,600 through these tools and win, in which case invest time and energy in structuring the 909 01:00:26,600 --> 01:00:32,220 data I already have, because a lot of companies already have a lot of data, or do I bet that 910 01:00:32,220 --> 01:00:37,100 a lot of these companies that are pumping up, I think is it scale.ai one that springs 911 01:00:37,100 --> 01:00:44,880 to mind to basically come in and help extract the insights from your unstructured data to 912 01:00:44,880 --> 01:00:49,800 be able to then build models and stuff on top of them and as they mature, will they 913 01:00:49,800 --> 01:00:55,840 almost be out of the box, AI driven models themselves that can crawl all your disparate 914 01:00:55,840 --> 01:01:01,480 data and figure out how it should connect to other bits of your data and basically turn 915 01:01:01,480 --> 01:01:07,120 unstructured data, instructure data for you without you having to do anything. 916 01:01:07,120 --> 01:01:08,440 That would be the bet, right? 917 01:01:08,440 --> 01:01:14,440 Because if that appears in the next year or two or less, then any effort you put into 918 01:01:14,440 --> 01:01:17,320 structuring your data might be wasted. 919 01:01:17,320 --> 01:01:19,960 So I think it's a difficult one to know how it's going to play out. 920 01:01:19,960 --> 01:01:25,800 One thing is for certain, structured data about your business, about your customers, 921 01:01:25,800 --> 01:01:30,640 about your market that you can leverage in intelligent ways to provide better customer 922 01:01:30,640 --> 01:01:34,560 service, improve your marketing, improve your products. 923 01:01:34,560 --> 01:01:38,360 People have been saying data is the new oil for years and I think we're going to see how 924 01:01:38,360 --> 01:01:42,800 AI tools are going to help get some of that oil out of the ground. 925 01:01:42,800 --> 01:01:47,760 Yeah, although that analogy always just make me think of how bad things are when there's 926 01:01:47,760 --> 01:01:50,760 a leak. 927 01:01:50,760 --> 01:01:53,760 Absolutely. 928 01:01:53,760 --> 01:02:00,240 So after we were about to put the podcast live, we noticed a story that we thought we 929 01:02:00,240 --> 01:02:01,280 should feature. 930 01:02:01,280 --> 01:02:04,760 So here's a little addendum to the big stories this week. 931 01:02:04,760 --> 01:02:10,720 And this big story is about the CRM company Salesforce, who I'm sure many of you will 932 01:02:10,720 --> 01:02:18,200 know, debuting two new generative AI products this week as part of the company's connections 933 01:02:18,200 --> 01:02:19,200 conference. 934 01:02:19,200 --> 01:02:25,600 And those products are called Marketing GPT and Commerce GPT, plugging into Salesforce's 935 01:02:25,600 --> 01:02:29,040 Marketing Cloud and Commerce Cloud products. 936 01:02:29,040 --> 01:02:32,960 So in essence, why is this important? 937 01:02:32,960 --> 01:02:36,040 Well I guess the first thing is they're rolling this out in phases. 938 01:02:36,040 --> 01:02:41,340 So if you are a Salesforce user, you may not have access to these yet, but you hopefully 939 01:02:41,340 --> 01:02:44,160 will in the future. 940 01:02:44,160 --> 01:02:49,040 And in essence, these are kind of cool because what they're going to allow is a number of 941 01:02:49,040 --> 01:02:50,320 different things. 942 01:02:50,320 --> 01:02:56,640 So Marketing Cloud users should be able to put in natural language prompts to query the 943 01:02:56,640 --> 01:03:05,140 data in their Salesforce implementation and identify new audience segments to target. 944 01:03:05,140 --> 01:03:12,460 They could also ask Einstein GPT to write or modify personalized emails, complete with 945 01:03:12,460 --> 01:03:16,240 subject lines and body content for their campaigns. 946 01:03:16,240 --> 01:03:21,800 And they can use typeface within the platform to create contextual visual assets for specific 947 01:03:21,800 --> 01:03:22,800 contacts. 948 01:03:22,800 --> 01:03:26,600 That's pretty powerful and pretty interesting. 949 01:03:26,600 --> 01:03:30,600 In addition to this, there's going to be some other things that can be done that were going 950 01:03:30,600 --> 01:03:36,660 to be valuable for marketers also, including the creation of personalized shopping experiences. 951 01:03:36,660 --> 01:03:43,320 So Marketing GPT uses Data Cloud and Einstein GPT to allow users to not only create dynamic 952 01:03:43,320 --> 01:03:48,240 product descriptions for digital storefronts, but also to have those descriptions translated 953 01:03:48,240 --> 01:03:52,180 into different languages for different target audiences. 954 01:03:52,180 --> 01:03:55,400 So quite interesting stuff actually. 955 01:03:55,400 --> 01:04:00,880 The original article comes to us from VentureBeat, so it's worth Googling. 956 01:04:00,880 --> 01:04:07,400 But yes, Salesforce, maybe I know Einstein has been around for a while, but been a little 957 01:04:07,400 --> 01:04:10,940 bit quiet on the generative AI front. 958 01:04:10,940 --> 01:04:18,560 And it looks like these two products, the Marketing GPT and Einstein GPT could be opening 959 01:04:18,560 --> 01:04:23,780 up some really interesting applications for marketers who use Salesforce that will very 960 01:04:23,780 --> 01:04:31,000 likely make it faster and easier to create content for your customers and prospects, 961 01:04:31,000 --> 01:04:36,080 but also easier to create hyper-personalized content at scale. 962 01:04:36,080 --> 01:04:38,920 So very interesting stuff from Salesforce. 963 01:04:38,920 --> 01:04:43,200 Right, let's get onto the last bit this week because the lovely folks really stuck with 964 01:04:43,200 --> 01:04:44,320 us and we appreciate it. 965 01:04:44,320 --> 01:04:46,400 Let's talk about Gen 2. 966 01:04:46,400 --> 01:04:53,640 So for those that are not aware, there is a really cool web-based tool called RunwayML 967 01:04:53,640 --> 01:04:59,480 that has a whole suite of clever image editing and video editing and image creation and video 968 01:04:59,480 --> 01:05:01,540 creation tools. 969 01:05:01,540 --> 01:05:07,440 And RunwayML made a bit of a splash at the start of the year when they launched Gen 1, 970 01:05:07,440 --> 01:05:13,920 which was a video tool where you could feed an image and it would create a very short 971 01:05:13,920 --> 01:05:16,160 snippet of video based on that image. 972 01:05:16,160 --> 01:05:20,380 Then Gen 2 came along where you could do the same thing, but with a text prompt. 973 01:05:20,380 --> 01:05:24,000 So you just write a simple text prompt, you get, I don't know, what is it, three or four 974 01:05:24,000 --> 01:05:26,920 seconds of video. 975 01:05:26,920 --> 01:05:29,640 So we've been having a play, haven't we Martin? 976 01:05:29,640 --> 01:05:34,680 Because obviously, as marketers, if we can produce high quality video from a simple text 977 01:05:34,680 --> 01:05:40,800 prompt, that opens up a load of creative avenues for us to be able to produce video at scale. 978 01:05:40,800 --> 01:05:45,280 What have your thoughts been with your access to Gen 2? 979 01:05:45,280 --> 01:05:52,120 I've been playing with this for the past day or so, tried out a few different styles. 980 01:05:52,120 --> 01:06:00,240 The first thing that I tried was a prompt which simply said, two British men recording 981 01:06:00,240 --> 01:06:05,560 a podcast in a dimly lit studio with neon lights. 982 01:06:05,560 --> 01:06:12,320 And it's a curious one, I'll be honest, because it can't count as there are three men in the 983 01:06:12,320 --> 01:06:13,320 scene. 984 01:06:13,320 --> 01:06:19,880 It's dimly lit, it does have a kind of aesthetic of a dogly lit studio with some neon lights. 985 01:06:19,880 --> 01:06:23,880 But the three men in the image are all identical. 986 01:06:23,880 --> 01:06:29,400 It's as if three British triplets had recorded this podcast. 987 01:06:29,400 --> 01:06:31,680 That was a bit of a curiosity. 988 01:06:31,680 --> 01:06:38,360 The videos are all, like you say, about four seconds long. 989 01:06:38,360 --> 01:06:47,140 They have some of the strange, I don't even know how you would call it, just objects merging 990 01:06:47,140 --> 01:06:51,640 into one another, like the hand suddenly becomes part of the table and then comes out of the 991 01:06:51,640 --> 01:06:52,640 table again. 992 01:06:52,640 --> 01:07:02,280 What I have found, when I tried more abstract things, so I tried hand drawn animation, so 993 01:07:02,280 --> 01:07:09,320 that was part of the prompt, I said, tractor racing, Grand Prix, line drawn, hand drawn 994 01:07:09,320 --> 01:07:11,200 animation. 995 01:07:11,200 --> 01:07:16,120 And it basically came, just gave me a static image of a tractor, like there was very little 996 01:07:16,120 --> 01:07:18,720 movement in it at all. 997 01:07:18,720 --> 01:07:21,160 Didn't really give me tractor racing. 998 01:07:21,160 --> 01:07:24,400 I was trying something for my toddler, you know. 999 01:07:24,400 --> 01:07:30,920 I found things like animation, it just doesn't really vibe with very well at all. 1000 01:07:30,920 --> 01:07:32,880 I haven't had any good results with that. 1001 01:07:32,880 --> 01:07:35,680 I've tried three or four variants. 1002 01:07:35,680 --> 01:07:42,480 One that I have had a great deal of success with was, so I said a ginger head woman using 1003 01:07:42,480 --> 01:07:51,240 her iPhone, stock video, professional video, and that was it. 1004 01:07:51,240 --> 01:07:54,600 And it's actually given me a decent clip. 1005 01:07:54,600 --> 01:07:56,160 It works quite well. 1006 01:07:56,160 --> 01:08:04,080 There's no immediate things that are ghoulish or the artifacts aren't really bad. 1007 01:08:04,080 --> 01:08:10,500 Now the file size for all of these images, videos is about three to 400 kilobytes. 1008 01:08:10,500 --> 01:08:14,280 So they're really short and basically they're just like little gifs. 1009 01:08:14,280 --> 01:08:18,480 And for that purpose, I think if anywhere that you might stick a little gif, so in a 1010 01:08:18,480 --> 01:08:24,120 blog post, maybe on something on social, they can be quite useful. 1011 01:08:24,120 --> 01:08:27,880 You do have to play around with them a lot and I don't think with the monthly subscription, 1012 01:08:27,880 --> 01:08:31,360 you don't get huge amounts of generation credits. 1013 01:08:31,360 --> 01:08:39,640 But overall, taking a few words and turning it into an image, when you find the styles 1014 01:08:39,640 --> 01:08:45,600 that work, for instance, professional videography, stock video library, and putting that into 1015 01:08:45,600 --> 01:08:48,920 the prompt, you can get something that looks pretty decent. 1016 01:08:48,920 --> 01:08:50,880 Yeah. 1017 01:08:50,880 --> 01:08:52,440 I think my experiences have been the same. 1018 01:08:52,440 --> 01:09:00,160 I asked for two fighter jets fighting alien spacecraft over an alien planet and I got 1019 01:09:00,160 --> 01:09:06,320 multitude of jet light subjects flying in the sky that weren't really jets. 1020 01:09:06,320 --> 01:09:12,200 They were like three jets had all been smashed together to create this strange flying thing. 1021 01:09:12,200 --> 01:09:14,440 It didn't work to be honest. 1022 01:09:14,440 --> 01:09:25,440 I also asked for a video of Tom Cruise playing basketball and for that, I got a couple of 1023 01:09:25,440 --> 01:09:30,680 male figures that look like humans, but none of them look anything like Tom Cruise and 1024 01:09:30,680 --> 01:09:31,680 then it didn't animate at all. 1025 01:09:31,680 --> 01:09:34,760 Yeah, it's the lack of animation that's interesting and I think they're experiencing this quite 1026 01:09:34,760 --> 01:09:39,960 a lot because under each generation, there's a five star rating. 1027 01:09:39,960 --> 01:09:46,320 So if you give it a rating and you give it a low rating, it comes up with, it's got like 1028 01:09:46,320 --> 01:09:51,000 six options to choose from and one of them is just that it doesn't move. 1029 01:09:51,000 --> 01:09:57,200 So it must be one of the common issues that they're facing. 1030 01:09:57,200 --> 01:10:04,360 Yeah, I saw that as well and I think soliciting this type of feedback tells me they know this 1031 01:10:04,360 --> 01:10:11,160 doesn't work very well yet and they want this type of feedback to improve the models. 1032 01:10:11,160 --> 01:10:15,120 My spidey sense tells me that one of the reasons you got better outputs when you tried to talk 1033 01:10:15,120 --> 01:10:20,760 about stock videography is the training data. 1034 01:10:20,760 --> 01:10:26,500 When I look at my Tom Cruise output, some of that is almost entirely going to be driven 1035 01:10:26,500 --> 01:10:27,680 by the training data. 1036 01:10:27,680 --> 01:10:32,360 Like it doesn't have a video or anything close to an image of Tom Cruise playing basketball 1037 01:10:32,360 --> 01:10:35,200 and so it has literally no hope of producing it. 1038 01:10:35,200 --> 01:10:38,520 Whereas I bet if I try to get something that I know there's plenty of stock video out there 1039 01:10:38,520 --> 01:10:45,760 for or plenty of stock imagery for, I reckon it would probably do a better job. 1040 01:10:45,760 --> 01:10:50,080 So I guess for those marketers out there, as Martin summed it up, you might be able 1041 01:10:50,080 --> 01:10:51,760 to get some interesting gifts out of this. 1042 01:10:51,760 --> 01:10:56,880 They're going to be very specific use cases that work right now and loads that don't. 1043 01:10:56,880 --> 01:10:59,780 You're probably going to have to iterate a lot to get close to something that you like 1044 01:10:59,780 --> 01:11:03,320 and then you're probably going to run out of credits. 1045 01:11:03,320 --> 01:11:08,600 So it's probably not quite there yet, but like all of these things, we do recommend 1046 01:11:08,600 --> 01:11:12,920 you have a play because it's really important to just be informed about what tools are out 1047 01:11:12,920 --> 01:11:17,720 there and how they're emerging and how they're developing and which ones are good for what. 1048 01:11:17,720 --> 01:11:21,960 Because this space is going to continue to move quickly and I think the best thing marketers 1049 01:11:21,960 --> 01:11:27,680 can do is make sure they know how to augment themselves to get more done in less time, 1050 01:11:27,680 --> 01:11:30,040 be more creative and all that good stuff. 1051 01:11:30,040 --> 01:11:33,680 I have a question for you on this. 1052 01:11:33,680 --> 01:11:41,760 Tools like Runway and then companies like Runway, do you think that they are, how do 1053 01:11:41,760 --> 01:11:47,400 you think they're going to stack up when Adobe get into this game or Canva or any of the 1054 01:11:47,400 --> 01:11:51,240 big players in this space? 1055 01:11:51,240 --> 01:11:52,840 What do I think is going to happen? 1056 01:11:52,840 --> 01:11:58,400 Do you think they'll just get crushed or what? 1057 01:11:58,400 --> 01:12:04,560 Are they just a curiosity at the moment or where's their space in the market? 1058 01:12:04,560 --> 01:12:08,240 I think it comes back to users and motes. 1059 01:12:08,240 --> 01:12:14,200 So many creatives and even non-creatives have Photoshop installed already, want to do cool 1060 01:12:14,200 --> 01:12:15,200 stuff. 1061 01:12:15,200 --> 01:12:19,080 The fact that I could just download the beta quite quickly and start playing and it was 1062 01:12:19,080 --> 01:12:23,400 good at some things that a lot of other tools are not, made it very accessible. 1063 01:12:23,400 --> 01:12:25,720 I want to love Runway. 1064 01:12:25,720 --> 01:12:29,040 I love some of the things it can do. 1065 01:12:29,040 --> 01:12:33,120 We talked about in the first or second episode of the podcast, we tried to edit the video 1066 01:12:33,120 --> 01:12:36,280 for the podcast and the transcript to produce was brilliant. 1067 01:12:36,280 --> 01:12:40,240 There was hardly any errors in it, but goodness me, editing the errors out because it was 1068 01:12:40,240 --> 01:12:43,920 a web-based app was so slow. 1069 01:12:43,920 --> 01:12:46,200 I was like, I don't care if there are errors now. 1070 01:12:46,200 --> 01:12:47,640 It's just too painful. 1071 01:12:47,640 --> 01:12:52,400 Whereas if I was in Premiere Pro and Adobe, I could make edits very quickly because there's 1072 01:12:52,400 --> 01:13:00,520 no real lag in terms of how the tool works, but I made loads more errors in the transcript. 1073 01:13:00,520 --> 01:13:12,360 So personally, I think they're going to struggle to maintain applications and tools that they 1074 01:13:12,360 --> 01:13:21,160 can keep unique and unless their user experience is awesome and the editing the video user 1075 01:13:21,160 --> 01:13:25,560 experience was not awesome and that's why I've been there, they're going to struggle 1076 01:13:25,560 --> 01:13:31,000 because they are the ones that have to get us to jump ship. 1077 01:13:31,000 --> 01:13:36,360 They have to give us a compelling reason to make this the place where we get that thing 1078 01:13:36,360 --> 01:13:41,200 that we need done to this image or this video. 1079 01:13:41,200 --> 01:13:44,200 Like I said, unless they can come up with something that other tools can't do and I 1080 01:13:44,200 --> 01:13:49,200 think the evidence so far is that the major companies are not struggling to replicate 1081 01:13:49,200 --> 01:13:51,400 what other people can build. 1082 01:13:51,400 --> 01:13:52,960 That's how it is so far. 1083 01:13:52,960 --> 01:13:58,720 Then they have to be absolutely incredible in terms of usability and let's not forget, 1084 01:13:58,720 --> 01:14:05,160 Adobe's had decades to improve the usability of its tools and in essence, it's trained 1085 01:14:05,160 --> 01:14:10,760 people to use its tools by having people just have to use the left side, but where do I 1086 01:14:10,760 --> 01:14:11,760 go for the lasso? 1087 01:14:11,760 --> 01:14:13,760 It's on the left side where it's a fourth down. 1088 01:14:13,760 --> 01:14:17,440 We know how all of that works. 1089 01:14:17,440 --> 01:14:21,880 So personally, I probably wouldn't be, and this is not investment advice because we are 1090 01:14:21,880 --> 01:14:27,640 not in that business, but personally, I probably wouldn't be investing in them because I think 1091 01:14:27,640 --> 01:14:29,440 it's going to be hard for them. 1092 01:14:29,440 --> 01:14:31,560 Yeah, I do at the moment. 1093 01:14:31,560 --> 01:14:39,200 I find it a curiosity and something that's fun to watch develop, but so far just seeing 1094 01:14:39,200 --> 01:14:41,320 where the... because they are pushing boundaries. 1095 01:14:41,320 --> 01:14:44,000 That's the interesting thing with them and that's what I like about them. 1096 01:14:44,000 --> 01:14:47,960 I really respect them from that perspective, is the usability and again, it just comes 1097 01:14:47,960 --> 01:14:53,720 back to every third or fourth time we've mentioned it today, is having the right UX and the right 1098 01:14:53,720 --> 01:14:59,760 usability that makes people go, yes, this is so seamless and frictionless and really 1099 01:14:59,760 --> 01:15:01,040 makes my life a breeze. 1100 01:15:01,040 --> 01:15:04,840 At the moment, it's like, oh, this is an interesting curiosity. 1101 01:15:04,840 --> 01:15:09,400 I can make something that isn't quite a very convincing stock image. 1102 01:15:09,400 --> 01:15:14,480 See, you raised such a really important point though, because I think the ecosystem as it 1103 01:15:14,480 --> 01:15:19,280 stands right now is critical because I do think that Runway do loads of the cool stuff 1104 01:15:19,280 --> 01:15:21,200 and they launch it really early. 1105 01:15:21,200 --> 01:15:27,720 You could argue before ChatGPT went nuclear, that was kind of a... we're a small... at 1106 01:15:27,720 --> 01:15:31,400 the time, in terms of the mindset, we're an agile, small brand. 1107 01:15:31,400 --> 01:15:36,360 We Google, we're like, we don't want to launch anything or release anything before it's ready, 1108 01:15:36,360 --> 01:15:40,920 which how much longer would we have waited for these tools if ChatGPT hadn't come out 1109 01:15:40,920 --> 01:15:43,080 and open out and just gone, look, let's just get it out. 1110 01:15:43,080 --> 01:15:44,520 Let's just ship. 1111 01:15:44,520 --> 01:15:48,920 Runway's attitude to its tools is ship, ship, ship, ship, ship and we get to play with innovative 1112 01:15:48,920 --> 01:15:49,920 stuff. 1113 01:15:49,920 --> 01:15:53,400 Therefore, we shouldn't probably be too down or critical of things when like Gen 2 and 1114 01:15:53,400 --> 01:15:56,680 they're not really working as well as we'd all like, but it's because they're getting 1115 01:15:56,680 --> 01:15:58,200 out there so we can play with them. 1116 01:15:58,200 --> 01:16:04,640 So if the ecosystem is boiled down to Adobe, Microsoft and Google, that is not going to 1117 01:16:04,640 --> 01:16:10,000 be good, I would argue, for the actual innovation and the emergence of awesome new tools for 1118 01:16:10,000 --> 01:16:11,000 us to play with. 1119 01:16:11,000 --> 01:16:14,560 So let's hope that they do do reasonably well because we need them around. 1120 01:16:14,560 --> 01:16:20,800 Yeah, I do want to see them thrive, but I just want to see the use cases beyond interest 1121 01:16:20,800 --> 01:16:25,400 and curiosity at the moment, which is currently where I feel like it is. 1122 01:16:25,400 --> 01:16:26,400 Agreed. 1123 01:16:26,400 --> 01:16:27,400 Right. 1124 01:16:27,400 --> 01:16:29,020 So that will sign off. 1125 01:16:29,020 --> 01:16:33,040 Remember to follow us on the Twitter's where the handle is Martin. 1126 01:16:33,040 --> 01:16:34,040 A iMarketingPod. 1127 01:16:34,040 --> 01:16:35,040 Woo, that's us. 1128 01:16:35,040 --> 01:16:40,920 Visit us at ArtificiallyIntelligentMarketing.com where you can subscribe to get blog updates 1129 01:16:40,920 --> 01:16:44,080 and you'll find us on all your favourite podcasting platforms. 1130 01:16:44,080 --> 01:16:47,640 And if you love this and this is your first time, please subscribe. 1131 01:16:47,640 --> 01:16:51,400 And if you know someone who works in marketing that might benefit from us doing all the hard 1132 01:16:51,400 --> 01:16:54,760 work of keeping up to date on the latest news that you need to know so that you don't have 1133 01:16:54,760 --> 01:16:57,040 to, then please do share it. 1134 01:16:57,040 --> 01:16:58,040 Right. 1135 01:16:58,040 --> 01:16:59,040 Thanks, Martin. 1136 01:16:59,040 --> 01:17:00,040 It's for bye. 1137 01:17:00,040 --> 01:17:01,040 Bye. 1138 01:17:01,040 --> 01:17:05,600 Thank you for listening to Artificially Intelligent Marketing. 1139 01:17:05,600 --> 01:17:11,640 To stay on top of the latest trends, tips and tools in the world of marketing AI, be 1140 01:17:11,640 --> 01:17:13,400 sure to subscribe. 1141 01:17:13,400 --> 01:17:27,600 We look forward to seeing you again next week.