1 00:00:00,000 --> 00:00:09,960 Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the 2 00:00:09,960 --> 00:00:15,700 latest trends, tips, and tools in the world of marketing AI, helping you get the best 3 00:00:15,700 --> 00:00:18,540 results from your marketing efforts. 4 00:00:18,540 --> 00:00:23,520 Now let's join our hosts, Paul Avery and Martin Broadhurst. 5 00:00:23,520 --> 00:00:28,280 Welcome to Artificially Intelligent Marketing, Episode 27. 6 00:00:28,280 --> 00:00:32,520 My name's Martin Broadhurst and I am your host today. 7 00:00:32,520 --> 00:00:34,640 That's right, I'm flying solo. 8 00:00:34,640 --> 00:00:37,120 Paul is unable to join us this week. 9 00:00:37,120 --> 00:00:44,360 That means it's just me delivering the latest in AI technology developments for marketing 10 00:00:44,360 --> 00:00:45,360 professionals. 11 00:00:45,360 --> 00:00:52,640 And while we may lack a Paul, we do not lack content, not by any stretch of the imagination. 12 00:00:52,640 --> 00:00:58,600 This week has been a hot week for product updates from all of the big players. 13 00:00:58,600 --> 00:01:02,560 So let's get straight into the stories and start off with my favorite, which is that 14 00:01:02,560 --> 00:01:07,840 ChatGPT has now got multi-sensory capabilities. 15 00:01:07,840 --> 00:01:15,120 That's right, OpenAI has announced a significant upgrade to ChatGPT, bringing both voice and 16 00:01:15,120 --> 00:01:18,560 image functionality to the platform. 17 00:01:18,560 --> 00:01:26,160 This brings a shift in the way that the ChatGPT product works, making it more interactive, 18 00:01:26,160 --> 00:01:30,000 more multimodal for the users. 19 00:01:30,000 --> 00:01:36,460 We saw when they announced GPT-4 earlier this year that they were bringing the ability to 20 00:01:36,460 --> 00:01:40,640 input images as a prompt. 21 00:01:40,640 --> 00:01:48,080 We saw them doing things like the creating a website based on a, what was little more 22 00:01:48,080 --> 00:01:50,280 than a napkin sketch. 23 00:01:50,280 --> 00:01:58,800 It was a very low fidelity drawing and GPT-4 created a functioning website with HTML and 24 00:01:58,800 --> 00:01:59,800 JavaScript. 25 00:01:59,800 --> 00:02:04,720 Well, that capability has now been launched. 26 00:02:04,720 --> 00:02:12,240 As well as that, they've announced a new voice feature where we can turn the ChatGPT interface 27 00:02:12,240 --> 00:02:18,860 into a back and forth real conversational tool. 28 00:02:18,860 --> 00:02:22,600 It will respond with a human-like voice. 29 00:02:22,600 --> 00:02:29,400 So rather than seeing text on screen as the response, it will speak back to you and you 30 00:02:29,400 --> 00:02:34,480 can have a back and forth dialogue with ChatGPT. 31 00:02:34,480 --> 00:02:42,240 At the moment, this is available on iOS and Android, exclusively for the time being at 32 00:02:42,240 --> 00:02:51,160 least to ChatGPT Plus subscribers and ChatGPT Enterprise subscribers. 33 00:02:51,160 --> 00:02:56,960 This is expected to be rolled out to a wider user base in the coming weeks and months, 34 00:02:56,960 --> 00:02:59,800 but for now, this is what we are working with. 35 00:02:59,800 --> 00:03:02,640 It's just the people with a paid plan. 36 00:03:02,640 --> 00:03:09,680 While these new capabilities can serve creative and accessibility focused applications, OpenAI 37 00:03:09,680 --> 00:03:16,480 remains cautious, acknowledging the potential risks such as impersonation and data misuse. 38 00:03:16,480 --> 00:03:20,320 My early impressions of these tools are that they are pretty good. 39 00:03:20,320 --> 00:03:27,320 Now I should say at this point, I actually only have access to the voice version of this. 40 00:03:27,320 --> 00:03:35,560 At the moment, the image inputs haven't quite made it to me and a very small number of users 41 00:03:35,560 --> 00:03:37,480 appear to have access to that. 42 00:03:37,480 --> 00:03:44,680 Although you can do this using Bing AI and we'll talk about that in a moment. 43 00:03:44,680 --> 00:03:51,200 But in terms of the actual Chat interface and being able to speak back and forth with 44 00:03:51,200 --> 00:03:59,320 OpenAI and ChatGPT in the app, I found it to be a somewhat frustrating experience. 45 00:03:59,320 --> 00:04:01,800 Let me just demonstrate why. 46 00:04:01,800 --> 00:04:08,440 This has been my experience for the past 36 hours and I'm not alone in this. 47 00:04:08,440 --> 00:04:13,080 So let's try and have a real conversation with ChatGPT right now. 48 00:04:13,080 --> 00:04:14,840 Hi, ChatGPT. 49 00:04:14,840 --> 00:04:18,320 Welcome to the Artificially Intelligent Marketing podcast. 50 00:04:18,320 --> 00:04:20,360 You recently got a voice. 51 00:04:20,360 --> 00:04:22,120 How does that make you feel? 52 00:04:22,120 --> 00:04:25,200 Sorry, an unknown error occurred. 53 00:04:25,200 --> 00:04:26,800 Please try again later. 54 00:04:26,800 --> 00:04:31,920 And that listeners has been the experience that I have had for the past 36 hours. 55 00:04:31,920 --> 00:04:38,360 And unless I clear my cache in my phone, the app refuses to let me speak to it back and 56 00:04:38,360 --> 00:04:39,360 forth. 57 00:04:39,360 --> 00:04:44,560 And a quick Google search has found that this is the experience of quite a few users at 58 00:04:44,560 --> 00:04:45,560 the moment. 59 00:04:45,560 --> 00:04:52,720 And videos like this particular feature is not quite bug free. 60 00:04:52,720 --> 00:04:59,120 In fact, it's quite broken for quite a few people, which is a shame because in the first 61 00:04:59,120 --> 00:05:07,440 24 hours while using the app, I found it to be incredibly fun and helpful. 62 00:05:07,440 --> 00:05:15,720 And I could imagine being on a long drive, having ChatGPT open in the car and just speaking 63 00:05:15,720 --> 00:05:18,380 to it back and forth. 64 00:05:18,380 --> 00:05:24,160 Having it as a real time assistant, completing small tasks for me as I go. 65 00:05:24,160 --> 00:05:28,540 Ideal because it's completely hands free and it would be like speaking to a passenger in 66 00:05:28,540 --> 00:05:31,060 your car. 67 00:05:31,060 --> 00:05:35,680 The scenario I had in mind was coming away from a client meeting, being able to summarize 68 00:05:35,680 --> 00:05:39,880 notes, maybe draft a follow up email or even draft an email to a colleague, giving them 69 00:05:39,880 --> 00:05:43,640 the main points from the session itself. 70 00:05:43,640 --> 00:05:45,240 All of that would be really helpful. 71 00:05:45,240 --> 00:05:50,640 Again, just extending the capabilities of this assistant in your pocket. 72 00:05:50,640 --> 00:05:56,480 Unfortunately, that reality is not now for me and a few other users because it just refuses 73 00:05:56,480 --> 00:05:57,480 to work. 74 00:05:57,480 --> 00:06:04,840 Okay, I mentioned that we've got the image input capabilities as well. 75 00:06:04,840 --> 00:06:06,440 This has been a lot of fun. 76 00:06:06,440 --> 00:06:14,480 I've been playing with it to just test the functionality out and I was doing it using 77 00:06:14,480 --> 00:06:18,080 the image input in the Bing AI chat. 78 00:06:18,080 --> 00:06:24,760 So if you go to chat.bing.com, you can actually access this feature early. 79 00:06:24,760 --> 00:06:31,360 Is it as fully featured as the image input on ChatGPT? 80 00:06:31,360 --> 00:06:35,800 I don't know because I haven't been able to test the ChatGPT version yet, but it's still 81 00:06:35,800 --> 00:06:40,160 pretty good and if you want to give it a try, you can throw in an image and ask it to describe 82 00:06:40,160 --> 00:06:46,080 things for you and get it to tell you something interesting about the image or give you alternatives 83 00:06:46,080 --> 00:06:48,160 or what have you. 84 00:06:48,160 --> 00:06:56,680 The ChatGPT vision discussion online has been interesting to see though because people have 85 00:06:56,680 --> 00:06:59,480 found some really advanced capabilities. 86 00:06:59,480 --> 00:07:02,720 And there have been some examples that we've seen already. 87 00:07:02,720 --> 00:07:09,280 For example, people creating websites and Figma files based on nothing but a pencil 88 00:07:09,280 --> 00:07:10,280 sketch. 89 00:07:10,280 --> 00:07:16,840 McKay Wrigley writes, ChatGPT vision breaks down Christopher Nolan's early diagram for 90 00:07:16,840 --> 00:07:17,840 inception. 91 00:07:17,840 --> 00:07:24,520 The best part, the diagram doesn't mention the word inception once. 92 00:07:24,520 --> 00:07:36,240 When you look at the sketch, it's little more than a series of lines showing you the linear 93 00:07:36,240 --> 00:07:43,700 progression of multiple timelines from the film Inception. 94 00:07:43,700 --> 00:07:53,760 There is the heist and then the escape at the end and various rough annotations on this 95 00:07:53,760 --> 00:07:54,760 sketch. 96 00:07:54,760 --> 00:08:00,720 And the response from ChatGPT is quite something because it's very detailed. 97 00:08:00,720 --> 00:08:05,560 It actually starts off saying the diagram appears to be a representation of the dream 98 00:08:05,560 --> 00:08:09,280 levels and the progression of the events in the movie Inception. 99 00:08:09,280 --> 00:08:15,360 The movie involves navigating through multiple levels of dreams, each deeper and more time 100 00:08:15,360 --> 00:08:17,840 dilated than the previous one. 101 00:08:17,840 --> 00:08:23,520 Let's break down the diagram and then it goes into line by line detail breaking it down. 102 00:08:23,520 --> 00:08:27,520 The conversation of the back of this tweet was really interesting as well as people were 103 00:08:27,520 --> 00:08:33,200 saying well, this diagram must have been used in the training data and characters names 104 00:08:33,200 --> 00:08:38,100 are mentioned on the diagram, which has led other people to take this even further, doing 105 00:08:38,100 --> 00:08:44,200 an even more crude sketch, removing all of the characters names, in fact, changing all 106 00:08:44,200 --> 00:08:48,440 of the characters names and then repeating the process. 107 00:08:48,440 --> 00:08:52,400 And ChatGPT successfully completed the task. 108 00:08:52,400 --> 00:08:54,760 So it's really quite something. 109 00:08:54,760 --> 00:08:59,220 We can see that there's lots of capabilities with ChatGPT Vision. 110 00:08:59,220 --> 00:09:00,600 This has only been out a week. 111 00:09:00,600 --> 00:09:06,560 I expect many more interesting and exciting use cases to come to the fore. 112 00:09:06,560 --> 00:09:13,720 But for now, we can say that this is an exciting development for the ChatGPT subscriber user 113 00:09:13,720 --> 00:09:16,280 base. 114 00:09:16,280 --> 00:09:19,920 Next up, we have Dali 3. 115 00:09:19,920 --> 00:09:27,440 So sticking with OpenAI, this was a big announcement that they made last week in the text to image 116 00:09:27,440 --> 00:09:29,360 generation field. 117 00:09:29,360 --> 00:09:35,520 Dali 3 is set to transform text to image generation with even more nuanced and accurate image 118 00:09:35,520 --> 00:09:37,800 outputs. 119 00:09:37,800 --> 00:09:43,960 Slated for full release in early October, it promises enhanced capabilities, especially 120 00:09:43,960 --> 00:09:47,840 for GPT plus and enterprise subscribers. 121 00:09:47,840 --> 00:09:53,760 Again, this is an additional feature that is coming to ChatGPT. 122 00:09:53,760 --> 00:10:00,160 Unlike previous iterations where having a full grasp of prompt engineering was really 123 00:10:00,160 --> 00:10:07,480 necessary to get the most out of the Dali system, Dali 3 generates images that closely 124 00:10:07,480 --> 00:10:10,640 adhere to the user provided text. 125 00:10:10,640 --> 00:10:15,200 And it's built natively into ChatGPT. 126 00:10:15,200 --> 00:10:22,600 So the system itself can refine your prompts to produce detailed and very accurate images. 127 00:10:22,600 --> 00:10:31,120 Safety has also been prioritized with features to limit violent, adult, or harmful content. 128 00:10:31,120 --> 00:10:37,280 Dali 3 won't generate images in the style of living artists, a move likely aimed at 129 00:10:37,280 --> 00:10:42,240 respecting creative control and respecting copyright. 130 00:10:42,240 --> 00:10:48,920 OpenAI is researching ways to help users identify AI generated images and is developing a provenance 131 00:10:48,920 --> 00:10:52,160 classifier tool for this specific purpose. 132 00:10:52,160 --> 00:10:55,600 How that actually manifests remains to be seen. 133 00:10:55,600 --> 00:11:01,320 But it's an interesting point that they're looking to roll that out to users as well. 134 00:11:01,320 --> 00:11:07,360 So Dali 3 presents an absolute treasure trove of opportunities, particularly for creative 135 00:11:07,360 --> 00:11:11,040 ad campaigns, content marketing, and brand storytelling. 136 00:11:11,040 --> 00:11:15,280 In the early demonstrations that we've seen of it, there seems to be things like consistency 137 00:11:15,280 --> 00:11:20,680 of character, which is something that's been really hard to get previously. 138 00:11:20,680 --> 00:11:27,640 In the example that they put in the demo video, there is a hedgehog character that seems to 139 00:11:27,640 --> 00:11:34,760 stay pretty accurate and close to the original version that it creates, despite several different 140 00:11:34,760 --> 00:11:38,980 iterations of images being created after the original. 141 00:11:38,980 --> 00:11:42,960 This is something that's really exciting for producers because, quite frankly, it's the 142 00:11:42,960 --> 00:11:45,360 bit of the puzzle that has been missing. 143 00:11:45,360 --> 00:11:56,040 How do we create images that maintain a consistent style and character going forward? 144 00:11:56,040 --> 00:12:02,640 This appears to be available already in chat.bing.com, so if you want to give it a go, you can head 145 00:12:02,640 --> 00:12:03,640 over there. 146 00:12:03,640 --> 00:12:09,600 I've been playing with it already, and a few noticeable things to mention that it can create 147 00:12:09,600 --> 00:12:10,600 text. 148 00:12:10,600 --> 00:12:15,520 However, it's inconsistent, and if you're looking for it to create lots of text, it 149 00:12:15,520 --> 00:12:18,880 will soon fall down. 150 00:12:18,880 --> 00:12:26,280 The image quality is much higher though, and it does stay much closer to the prompt, far 151 00:12:26,280 --> 00:12:32,100 more than any of the other language models and text models that we've seen, such as mid-journey 152 00:12:32,100 --> 00:12:33,800 or stable diffusion. 153 00:12:33,800 --> 00:12:42,920 There are just subtle nuances that it understands and can bring into the image better than those 154 00:12:42,920 --> 00:12:43,920 other models. 155 00:12:43,920 --> 00:12:49,280 I think this is one that's going to be very useful for creatives. 156 00:12:49,280 --> 00:12:56,920 The power of this plugged into something like the Adobe Suite will be very welcome for lots 157 00:12:56,920 --> 00:13:03,280 of people, and I think this is the kind of model that creatives will want Adobe Generative 158 00:13:03,280 --> 00:13:11,560 Fill and Firefly to reach the same levels as very soon. 159 00:13:11,560 --> 00:13:16,520 Moving away from new models, now we're going to look at new partnerships. 160 00:13:16,520 --> 00:13:21,320 This is a big one announced by Amazon and Anthropic. 161 00:13:21,320 --> 00:13:28,040 They've announced that they're tying the knot to supercharge the generative AI capabilities 162 00:13:28,040 --> 00:13:32,400 on Amazon Web Services AWS. 163 00:13:32,400 --> 00:13:37,000 Amazon is not just backing Anthropic with a token investment. 164 00:13:37,000 --> 00:13:47,640 It's going all in with a staggering $4 billion investment, taking a minority stake in Anthropic. 165 00:13:47,640 --> 00:13:56,760 It will leverage Amazon's AWS Tranium and Inferentia chips, solidifying AWS as the preferred 166 00:13:56,760 --> 00:13:59,120 cloud partner. 167 00:13:59,120 --> 00:14:05,400 Anthropic models are set to feature on Amazon Bedrock, which is Amazon's AI platform, providing 168 00:14:05,400 --> 00:14:10,840 AWS customers with the inside track on model customization. 169 00:14:10,840 --> 00:14:17,240 In the press release announcing this new partnership, there were some interesting details from 170 00:14:17,240 --> 00:14:19,600 customers of Anthropic and AWS. 171 00:14:19,600 --> 00:14:23,560 So it says the partnership is already making an impact. 172 00:14:23,560 --> 00:14:30,360 Chris Wide, SVP of Engineering and Data Science at Lonely Planet, stated that by building 173 00:14:30,360 --> 00:14:38,840 with Claude 2 on Amazon Bedrock, they reduced itinerary generation costs by nearly 80%. 174 00:14:38,840 --> 00:14:45,880 Bridgewater's co-CIO, Greg Jensen, commented, we are using Amazon Bedrock and Claude's model 175 00:14:45,880 --> 00:14:53,080 to create a secure large language model powered investment analyst assistant. 176 00:14:53,080 --> 00:14:57,160 It's a bit of a mouthful, but they've created an investment analyst assistant powered by 177 00:14:57,160 --> 00:14:58,160 Claude. 178 00:14:58,160 --> 00:15:03,560 If companies like Lonely Planet and Bridgewater can dramatically reduce operational costs 179 00:15:03,560 --> 00:15:07,360 and streamline their research, it makes you wonder what the rest of us could do when we 180 00:15:07,360 --> 00:15:12,920 can tap into this cloud environment. 181 00:15:12,920 --> 00:15:14,400 So this was quite a big one for me. 182 00:15:14,400 --> 00:15:18,920 I was surprised to see this announcement. 183 00:15:18,920 --> 00:15:25,680 Google had previously announced that they had made a significant investment into Anthropic. 184 00:15:25,680 --> 00:15:31,920 And it seemed like the Google Cloud Platform was going to be the cloud platform of choice 185 00:15:31,920 --> 00:15:33,000 for Anthropic. 186 00:15:33,000 --> 00:15:36,760 But obviously things have taken a bit of a turn. 187 00:15:36,760 --> 00:15:43,720 So now we've got Amazon investing $4 billion in Anthropic, Microsoft owning 49% of OpenAI 188 00:15:43,720 --> 00:15:54,480 with their $11 billion investment into the AI leader, which leaves Google out there focusing 189 00:15:54,480 --> 00:15:57,960 on its own in-house product development. 190 00:15:57,960 --> 00:16:00,680 So they're not looking to snap up anybody. 191 00:16:00,680 --> 00:16:04,560 There doesn't look to be any acquisition strategy right now. 192 00:16:04,560 --> 00:16:08,360 Maybe they might move from one of the smaller players, but I think that they're putting 193 00:16:08,360 --> 00:16:14,080 all of their eggs into their own basket and backing themselves to create a great model 194 00:16:14,080 --> 00:16:22,000 using something like the upcoming Gemini, which is said to be as powerful as GPT-4 with 195 00:16:22,000 --> 00:16:25,740 some multimodal capabilities as well. 196 00:16:25,740 --> 00:16:32,560 If the rumors are true, we'll find out more about that in Q4 of this year. 197 00:16:32,560 --> 00:16:33,760 And so watch this space. 198 00:16:33,760 --> 00:16:36,600 I'm sure me and Paul will talk about it at great length. 199 00:16:36,600 --> 00:16:46,280 But for now, Google doesn't look to have much in the way of active industry partnerships 200 00:16:46,280 --> 00:16:48,520 going on. 201 00:16:48,520 --> 00:16:53,080 So following on from that story, I just want to take a moment to look at a new article 202 00:16:53,080 --> 00:16:58,760 that was published by Anthropic related to Claude. 203 00:16:58,760 --> 00:17:05,220 So they've published new research on how to improve their AI assistant Claude's memory 204 00:17:05,220 --> 00:17:07,400 over long documents. 205 00:17:07,400 --> 00:17:17,720 So Claude's unique selling point is this 100,000 token character window or context window. 206 00:17:17,720 --> 00:17:22,600 And they tested Claude's ability to answer multiple choice questions about government 207 00:17:22,600 --> 00:17:29,960 meeting transcripts spanning 75 to 90,000 words. 208 00:17:29,960 --> 00:17:35,960 And any prompting modifications, Claude's accuracy dropped substantially on distant 209 00:17:35,960 --> 00:17:36,960 passages. 210 00:17:36,960 --> 00:17:42,400 However, Anthropic found two techniques that together improved performance. 211 00:17:42,400 --> 00:17:51,800 First, if you ask it to extract relevant quote before answering, this boosted scores. 212 00:17:51,800 --> 00:17:58,980 And second, by providing examples of correctly answered questions about other sections, they 213 00:17:58,980 --> 00:18:05,040 also found that this was more effective than unrelated examples. 214 00:18:05,040 --> 00:18:09,080 With both strategies, Claude's error rate declined by over a third. 215 00:18:09,080 --> 00:18:14,180 The newly published Anthropic cookbook, which we will link to in the show notes, also provides 216 00:18:14,180 --> 00:18:18,880 full details and the code to reproduce the experiments. 217 00:18:18,880 --> 00:18:24,860 So this quantitative analysis gives creators concrete guidance on prompting Claude for 218 00:18:24,860 --> 00:18:25,860 optimal recall. 219 00:18:25,860 --> 00:18:32,400 So if you really want to get the most out of that 100,000 token context window, introducing 220 00:18:32,400 --> 00:18:38,460 these techniques into your prompting is going to get you better results than if you don't 221 00:18:38,460 --> 00:18:39,720 introduce them at all. 222 00:18:39,720 --> 00:18:44,960 It's certainly going to be more effective than a zero-shot prompt, such as summarize 223 00:18:44,960 --> 00:18:50,080 this article or tell me 10 relevant things from this article. 224 00:18:50,080 --> 00:18:53,560 And actually most of the content is at the early stages of the document. 225 00:18:53,560 --> 00:18:59,840 So as AI assistants are increasingly tested on understanding and interpreting long form 226 00:18:59,840 --> 00:19:07,520 content, these results offer us as prompters a model for rigorously evaluating different 227 00:19:07,520 --> 00:19:16,520 prompting approaches and maximizing our capabilities of getting real high quality outputs. 228 00:19:16,520 --> 00:19:22,680 Microsoft is steering into the future of AI with the new assistant Microsoft Copilot set 229 00:19:22,680 --> 00:19:26,040 to become an integral part of its product ecosystem. 230 00:19:26,040 --> 00:19:36,640 It's going to be integrated gross Windows 11 and Microsoft Office 365 and Bing and Edge 231 00:19:36,640 --> 00:19:41,040 and a bunch of new Microsoft Surface devices. 232 00:19:41,040 --> 00:19:44,000 It is going to be rolled out everywhere. 233 00:19:44,000 --> 00:19:49,120 So Copilot is designed to offer personalized assistance by understanding user context, 234 00:19:49,120 --> 00:19:51,620 data and activities. 235 00:19:51,620 --> 00:19:58,400 The enterprise version Microsoft Office 365 Copilot will be generally available starting 236 00:19:58,400 --> 00:20:00,640 November 1st. 237 00:20:00,640 --> 00:20:06,280 This work specific assistant aims to streamline operations across office apps, including the 238 00:20:06,280 --> 00:20:09,240 popular apps Outlook and Teams. 239 00:20:09,240 --> 00:20:15,180 Not just an enterprise asset, Bing and Edge are also levering up their AI features such 240 00:20:15,180 --> 00:20:20,960 as personalized search results and AI shopping assistant, which could potentially influence 241 00:20:20,960 --> 00:20:24,440 consumer behavior and e-commerce strategies. 242 00:20:24,440 --> 00:20:32,000 So we'll have to see how that plays out for marketers looking to use Bing or Edge as a 243 00:20:32,000 --> 00:20:35,120 potential channel. 244 00:20:35,120 --> 00:20:41,960 At a recent event where they went big on describing what Copilot was going to do and how it was 245 00:20:41,960 --> 00:20:47,240 going to be rolled out, they also announced that AI capabilities were coming to some of 246 00:20:47,240 --> 00:20:53,840 the classic Windows apps such as Microsoft Paint and Microsoft Photos. 247 00:20:53,840 --> 00:20:59,680 Now that might come as a disappointment to those of you who, like me, really respect 248 00:20:59,680 --> 00:21:03,920 the simplicity of Microsoft Paint. 249 00:21:03,920 --> 00:21:09,360 Bringing in AI to Paint might ruin the vibe, so to speak. 250 00:21:09,360 --> 00:21:15,080 They're also rolling out AI powered capabilities to Clipcham, which is the video editing platform 251 00:21:15,080 --> 00:21:22,640 built into Microsoft Windows, which was really a replacement for Microsoft Video Editor. 252 00:21:22,640 --> 00:21:27,280 That's the one where I think for SMEs there could be some real value. 253 00:21:27,280 --> 00:21:33,280 If you're a small marketing team in an SME and you want to create some cool video content, 254 00:21:33,280 --> 00:21:39,560 Clipchamp is already a really affordable and very user friendly way of creating videos 255 00:21:39,560 --> 00:21:43,160 for social media and things like that. 256 00:21:43,160 --> 00:21:51,080 Having some AI power through Copilot could level up this capability dramatically. 257 00:21:51,080 --> 00:21:54,880 That's the one where I'm quite interested to see what happens there. 258 00:21:54,880 --> 00:22:00,840 Who knows, maybe we'll all be creating our social media graphics using Microsoft Paint 259 00:22:00,840 --> 00:22:05,720 once it gets AI powered capabilities within it. 260 00:22:05,720 --> 00:22:14,760 Today I opened my Microsoft Surface Pro and found that Copilot is now integrated into 261 00:22:14,760 --> 00:22:15,760 Windows. 262 00:22:15,760 --> 00:22:23,080 I haven't tried it yet and this happened literally about three hours before recording. 263 00:22:23,080 --> 00:22:29,400 I haven't had anywhere like enough time to give you even my first impressions, but it's 264 00:22:29,400 --> 00:22:31,600 definitely something that is here. 265 00:22:31,600 --> 00:22:32,600 It's live. 266 00:22:32,600 --> 00:22:34,400 It's in the real world. 267 00:22:34,400 --> 00:22:40,160 Hopefully we'll have more news to give you on this front next week. 268 00:22:40,160 --> 00:22:48,000 Moving away from models and hardware and software, let's take a look at societal impact. 269 00:22:48,000 --> 00:22:54,320 Generative AI and the future of jobs is a report that was recently published by McKinsey. 270 00:22:54,320 --> 00:23:00,080 They've been looking at how AI will dramatically shape the future of jobs and workflows. 271 00:23:00,080 --> 00:23:05,760 The report finds generative AI could automate nearly 10% of tasks across the US economy, 272 00:23:05,760 --> 00:23:11,840 affecting various roles, especially customer service, food service, production manufacturing 273 00:23:11,840 --> 00:23:13,880 and office support. 274 00:23:13,880 --> 00:23:19,080 Lower wage jobs are 14 times more likely to experience this shift compared to higher wage 275 00:23:19,080 --> 00:23:20,360 roles. 276 00:23:20,360 --> 00:23:24,440 However, the economic outlook is not all doom and gloom. 277 00:23:24,440 --> 00:23:30,720 McKinsey anticipates GDP growth and creation of higher wage jobs, forecasting 12 million 278 00:23:30,720 --> 00:23:34,640 occupational transitions by 2030. 279 00:23:34,640 --> 00:23:36,640 The catch? 280 00:23:36,640 --> 00:23:37,640 Adaptability. 281 00:23:37,640 --> 00:23:44,120 Both individuals and organisational adaptability will be key to weathering these changes. 282 00:23:44,120 --> 00:23:50,440 McKinsey suggests targeted reskilling programmes, skills-based hiring and public-private training 283 00:23:50,440 --> 00:23:56,120 partnerships as pivotal strategies for a successful transition. 284 00:23:56,120 --> 00:24:01,880 Given that we're in an era where customer centric marketing is data driven, the insights 285 00:24:01,880 --> 00:24:08,320 in this report from McKinsey should serve as a warning bell for organisational leaders. 286 00:24:08,320 --> 00:24:10,440 Leaders need to be adaptable. 287 00:24:10,440 --> 00:24:12,100 They need to be planning ahead. 288 00:24:12,100 --> 00:24:18,920 Their workforce is not going to be the same as it is today in just a few years time. 289 00:24:18,920 --> 00:24:22,760 And thinking about what that could look like, making sure that the necessary steps are in 290 00:24:22,760 --> 00:24:28,960 place to train and upskill their existing workforce and identify what roles they're 291 00:24:28,960 --> 00:24:35,800 going to need for the future is going to be a massive step in managing this organisational 292 00:24:35,800 --> 00:24:41,720 change that we're going to see throughout the economy over the next few years. 293 00:24:41,720 --> 00:24:48,000 Our advice on the podcast would be to start thinking about improving AI literacy across 294 00:24:48,000 --> 00:24:49,640 your organisation. 295 00:24:49,640 --> 00:24:58,000 How can you upskill and train your team to help them transition into an AI powered world? 296 00:24:58,000 --> 00:25:04,320 This goes beyond just simple prompting techniques, although that is a really key part of it. 297 00:25:04,320 --> 00:25:09,160 This is about getting people into the mindset where they see AI as a valuable tool to be 298 00:25:09,160 --> 00:25:13,900 using in their day-to-day jobs, an assistant that's going to help them be more productive 299 00:25:13,900 --> 00:25:19,640 and more effective in doing what they need to do each day. 300 00:25:19,640 --> 00:25:23,480 If you'd be interested in learning more about where you can do that, well, turns out me 301 00:25:23,480 --> 00:25:27,480 and Paul have been working with clients doing exactly this. 302 00:25:27,480 --> 00:25:32,960 And if you would like to know more about how me and Paul can help your business to improve 303 00:25:32,960 --> 00:25:43,380 AI literacy throughout your organisation, then drop us an email at hello at artificiallyintelligentmarketing.com. 304 00:25:43,380 --> 00:25:48,360 So one more story looking at the societal impact of AI, and that is about the public 305 00:25:48,360 --> 00:25:50,600 perception of AI. 306 00:25:50,600 --> 00:25:58,200 This is a report that's come from the Bentley Gallup Business in Society study, which found 307 00:25:58,200 --> 00:26:03,320 that three out of four Americans believe that artificial intelligence will cut jobs in the 308 00:26:03,320 --> 00:26:05,160 next decade. 309 00:26:05,160 --> 00:26:12,400 Surprisingly, those with less than a bachelor's degree are more convinced of this, 80%, compared 310 00:26:12,400 --> 00:26:15,880 to those with a degree, 68%. 311 00:26:15,880 --> 00:26:22,240 Age demographics also reveal varying attitudes with younger Americans between 18 and 29 less 312 00:26:22,240 --> 00:26:26,200 pessimistic about AI affecting jobs negatively. 313 00:26:26,200 --> 00:26:33,160 The majority of Americans believe AI performs specific tasks as well as or better than humans, 314 00:26:33,160 --> 00:26:37,320 be it content customisation or product recommendations. 315 00:26:37,320 --> 00:26:42,640 But the general sentiment holds that businesses can't be trusted to use AI responsibly. 316 00:26:42,640 --> 00:26:51,960 In fact, a whopping 79% of respondents are sceptical about companies deploying AI ethically. 317 00:26:51,960 --> 00:26:59,280 This data really underlines the pressing need for transparent AI usage in business. 318 00:26:59,280 --> 00:27:04,280 Marketers can capitalise on the positives, like AI's capacity for task automation and 319 00:27:04,280 --> 00:27:09,600 customer personalisation, while addressing the public's concern through clear communication 320 00:27:09,600 --> 00:27:12,760 and ethical AI deployment. 321 00:27:12,760 --> 00:27:17,980 Younger demographics seem more open to AI innovations, perhaps indicating a future 322 00:27:17,980 --> 00:27:22,760 market that's more receptive to AI-powered solutions. 323 00:27:22,760 --> 00:27:27,200 One way that companies can start to think about the ethical deployment of AI in their 324 00:27:27,200 --> 00:27:33,600 workflows is by coming up with an AI charter. 325 00:27:33,600 --> 00:27:35,720 Create an AI council of your workforce. 326 00:27:35,720 --> 00:27:40,920 This should be a cross-functional team, one or two people from different departments in 327 00:27:40,920 --> 00:27:41,920 your organisation. 328 00:27:41,920 --> 00:27:46,040 This team could be a few people big or it could be 30 people big. 329 00:27:46,040 --> 00:27:49,320 It depends on the size of your organisation. 330 00:27:49,320 --> 00:27:53,200 Put something in place to bring these people together and say, how are we going to use 331 00:27:53,200 --> 00:27:55,480 AI in our organisation? 332 00:27:55,480 --> 00:28:01,880 This AI council can be responsible for bringing new learnings to the table and sharing them 333 00:28:01,880 --> 00:28:03,680 with the organisation. 334 00:28:03,680 --> 00:28:10,280 They can be responsible for drafting company policies about how you are going to use AI 335 00:28:10,280 --> 00:28:11,800 responsibly and ethically. 336 00:28:11,800 --> 00:28:17,720 They can be responsible for checking that AI systems that are in place are being used 337 00:28:17,720 --> 00:28:20,640 ethically and responsibly. 338 00:28:20,640 --> 00:28:25,920 Once that's been done, meet regularly, monthly, quarterly. 339 00:28:25,920 --> 00:28:30,240 Monthly is probably better given the pace of change in this world. 340 00:28:30,240 --> 00:28:34,720 But just make sure that there is something taking place where the AI council have a way 341 00:28:34,720 --> 00:28:41,760 to feed in to the responsible use of AI in your company going forward. 342 00:28:41,760 --> 00:28:48,200 Now our final story of the day is for the Salesforce subscribers that are listening. 343 00:28:48,200 --> 00:28:52,300 Salesforce has unveiled Einstein Copilot, a conversational AI assistant designed to 344 00:28:52,300 --> 00:28:55,880 boost productivity across various business functions. 345 00:28:55,880 --> 00:29:00,400 It's seamlessly integrated into the Salesforce app, offering personalised and context aware 346 00:29:00,400 --> 00:29:04,360 answers driven by Salesforce data cloud. 347 00:29:04,360 --> 00:29:12,360 So for businesses seeking more tailored solutions, Einstein Copilot Studio enables the customisation 348 00:29:12,360 --> 00:29:17,320 of the assistant with unique prompts, skills and AI models. 349 00:29:17,320 --> 00:29:22,960 This suite of features can be applied across multiple CRM workflows and even consumer facing 350 00:29:22,960 --> 00:29:24,720 channels such as chatbots. 351 00:29:24,720 --> 00:29:31,040 So effectively what they're offering here is a fine tuning studio where you can customise 352 00:29:31,040 --> 00:29:34,160 the tools to your requirements. 353 00:29:34,160 --> 00:29:37,800 Key to this launch is the Einstein Trust Layer. 354 00:29:37,800 --> 00:29:38,800 This is interesting. 355 00:29:38,800 --> 00:29:42,080 It's an architecture that safeguards data privacy. 356 00:29:42,080 --> 00:29:45,460 At the enterprise level, this is critical. 357 00:29:45,460 --> 00:29:50,560 This assures that data is not stored by third parties and masks personally identifiable 358 00:29:50,560 --> 00:29:56,320 information, all while keeping an eye out for toxic content. 359 00:29:56,320 --> 00:30:01,680 These are big considerations whenever speaking to enterprises about deploying large language 360 00:30:01,680 --> 00:30:02,680 models. 361 00:30:02,680 --> 00:30:07,560 So it's interesting to see exactly how this is going to work. 362 00:30:07,560 --> 00:30:12,360 I'm going to be keeping a close eye on this for sure. 363 00:30:12,360 --> 00:30:19,520 But yeah, Einstein and Copilot brings together AI with data driven decision making and customisation, 364 00:30:19,520 --> 00:30:26,600 which is relevant for marketing and operations teams to get what they need from their own 365 00:30:26,600 --> 00:30:30,920 data hooked up to language models. 366 00:30:30,920 --> 00:30:36,040 The ability to customise the AI for specific tasks in sales and marketing funnel will mean 367 00:30:36,040 --> 00:30:40,360 more efficient workflows, more personalisation. 368 00:30:40,360 --> 00:30:46,280 And with that extra layer of the trust layer, this addresses those growing concerns about 369 00:30:46,280 --> 00:30:53,040 data privacy in an era of AI powered tools. 370 00:30:53,040 --> 00:30:59,440 It sounds to me like Salesforce understand their audience very well, no surprise there, 371 00:30:59,440 --> 00:31:04,520 and that they are setting the new standard for data privacy in the use of large language 372 00:31:04,520 --> 00:31:09,400 models and AI for enterprise businesses. 373 00:31:09,400 --> 00:31:14,160 How this plays out remains to be seen, but I for one, I'm looking forward to seeing exactly 374 00:31:14,160 --> 00:31:19,160 what happens when businesses start to adopt this new technology. 375 00:31:19,160 --> 00:31:23,960 That's all we have time for in this week's episode. 376 00:31:23,960 --> 00:31:30,400 We'll be back next week with more news from across the artificially intelligent marketing 377 00:31:30,400 --> 00:31:32,000 landscape. 378 00:31:32,000 --> 00:31:33,000 Thank you for listening. 379 00:31:33,000 --> 00:31:36,460 I'll speak to you next week. 380 00:31:36,460 --> 00:31:42,080 Thank you for listening to Artificially Intelligent Marketing to stay on top of the latest trends, 381 00:31:42,080 --> 00:31:45,960 tips and tools in the world of marketing AI. 382 00:31:45,960 --> 00:31:47,720 Be sure to subscribe. 383 00:31:47,720 --> 00:32:15,360 We look forward to seeing you again next week.