1 00:00:00,000 --> 00:00:10,080 Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the 2 00:00:10,080 --> 00:00:15,820 latest trends, tips and tools in the world of marketing AI, helping you get the best 3 00:00:15,820 --> 00:00:22,480 results from your marketing efforts. Now let's join our hosts, Paul Avery and Martin Broadhurst. 4 00:00:22,480 --> 00:00:29,220 Hello everyone, welcome to Artificially Intelligent Marketing, a podcast for marketers looking 5 00:00:29,220 --> 00:00:33,840 to use AI to help them be more efficient, more effective as marketers really. I'm joined 6 00:00:33,840 --> 00:00:37,000 today by my very good friend Martin. Martin, how are you? 7 00:00:37,000 --> 00:00:42,600 I'm very well, just trying to keep my head above water with all of the AI news going 8 00:00:42,600 --> 00:00:46,600 on in the world at the moment. I feel like I am drowning in it. 9 00:00:46,600 --> 00:00:51,400 I think we broke WhatsApp this week, or at least you broke my fingers trying to reply 10 00:00:51,400 --> 00:00:56,280 to your WhatsApps. Yeah, crazy deluge. Some of the stories that we're going to be featuring 11 00:00:56,280 --> 00:01:01,000 today were released a handful of hours ago, and it feels like at this point you'd need 12 00:01:01,000 --> 00:01:05,920 to do a new podcast episode every day to try and keep up with all of these bits and pieces. 13 00:01:05,920 --> 00:01:07,880 But you've had a good week overall. 14 00:01:07,880 --> 00:01:14,840 Yeah, definitely. Excited for tomorrow actually. I'm doing my first AI marketing presentation 15 00:01:14,840 --> 00:01:19,880 seminar, call it what you will, conference, first one of 2023. So I'm looking forward 16 00:01:19,880 --> 00:01:24,880 to that. That's at Thrive Park in Derby, a digital marketing conference and expo. 17 00:01:24,880 --> 00:01:28,720 Yeah, looking forward to it. Good stuff. Sounds exciting. First for me this week as well, 18 00:01:28,720 --> 00:01:33,680 not work related, I took my little puppy to puppy training for the first time today. Little 19 00:01:33,680 --> 00:01:41,960 Pippin was learning to sit and down and stand and recall and mostly did a real good job. 20 00:01:41,960 --> 00:01:45,440 He's a whippet, so he's a little bit scared of all the other dogs at the moment, but he's 21 00:01:45,440 --> 00:01:47,640 very, very sweet and it was lots and lots of fun. 22 00:01:47,640 --> 00:01:50,720 So Pippin, lovely name as well. 23 00:01:50,720 --> 00:01:55,280 Pippin, he's had four or five names during his life, none of which he responded to. He 24 00:01:55,280 --> 00:01:59,280 was like, I'm not interested in that. Pippin, he's quite happy with. I love the Lord of 25 00:01:59,280 --> 00:02:02,880 the Rings, so I'm quite happy to go down that route. And Scotty Pippin was one of my favourite 26 00:02:02,880 --> 00:02:08,040 basketball players from back in the day when I was young. So it all just fit together. 27 00:02:08,040 --> 00:02:12,800 Right, let's get through this inane-ness and into the stuff that all of you lovely people 28 00:02:12,800 --> 00:02:16,520 are here to hear about. Let's talk about what's been happening in the world of marketing and 29 00:02:16,520 --> 00:02:18,760 AI and this week we got a lot. 30 00:02:18,760 --> 00:02:23,520 As you can see, GPT-4, as predicted last week, did come out. It came out even earlier than 31 00:02:23,520 --> 00:02:27,760 we thought it was going to. And so we've got some great info to go into on that. 32 00:02:27,760 --> 00:02:33,440 As well as that, Google provided an announcement of what they are planning to do by baking 33 00:02:33,440 --> 00:02:41,600 AI into all of their work-based tools. And then Microsoft as well coming in with Copilot 34 00:02:41,600 --> 00:02:46,660 a few hours ago. So we're going to go into those. We're also going to talk about a couple 35 00:02:46,660 --> 00:02:52,320 of tools we've been playing with. One called DID, which is quite fun and maybe could show 36 00:02:52,320 --> 00:02:56,280 the future of website chat. We'll look at that later. And then of course we'll have 37 00:02:56,280 --> 00:03:00,600 our tool of the week, which we're going to not tell you what it is until we get to it 38 00:03:00,600 --> 00:03:04,520 because we're that excited to show you some of the stuff that we've been playing with. 39 00:03:04,520 --> 00:03:11,480 So let's go into story one. GPT-4, it's alive. It's alive. 40 00:03:11,480 --> 00:03:16,800 Right. You've been playing, Martin. Tell us, tell us, tell everyone what's going on first. 41 00:03:16,800 --> 00:03:23,160 Why, why GPT-4, 3, 3, 4, 5, track GPT? Tell us what GPT-4 is. 42 00:03:23,160 --> 00:03:33,840 The latest iteration of the general processing transformer, large language model of open 43 00:03:33,840 --> 00:03:38,800 AI. Have I got GPT? Have I got the words right there? I don't know. Anyway, it's the new 44 00:03:38,800 --> 00:03:48,920 iteration of the engine really that powers the interfaces like chat GPT. So GPT-4 widely 45 00:03:48,920 --> 00:03:54,360 anticipated. People have been talking about the number of parameters, basically how powerful 46 00:03:54,360 --> 00:04:02,680 it is going from 175 billion to a trillion, a hundred trillion. And it's come out now 47 00:04:02,680 --> 00:04:07,560 and open AI. I haven't told us how big it is. So that was a really interesting observation 48 00:04:07,560 --> 00:04:13,720 when they announced it straight off the bat. It was a very different tone in their announcement. 49 00:04:13,720 --> 00:04:17,880 It was much more, we've done this cool thing and this is what we've done to it and this 50 00:04:17,880 --> 00:04:21,640 is what it can do, but we're not going to tell you how we've done it. 51 00:04:21,640 --> 00:04:26,800 No secrets, no secrets revealed. And there's been a little bit of angst on the Twitter 52 00:04:26,800 --> 00:04:33,800 spheres about that. Yeah. Open AI not being quite as open perhaps as they used to be. 53 00:04:33,800 --> 00:04:41,600 Really not. It's like Google's don't be evil motto. All of a sudden, they just pivot away 54 00:04:41,600 --> 00:04:49,400 from that. Previously we were an open AI research company. We're going to share. We want artificial 55 00:04:49,400 --> 00:04:53,480 general intelligence. And then all of a sudden, oh, actually no, sorry. We're going to be 56 00:04:53,480 --> 00:04:58,920 a profit based company, but we're going to put a profit cap on it. It's a hundred times. 57 00:04:58,920 --> 00:05:02,400 So if we invest a billion, can we make a hundred billion? 58 00:05:02,400 --> 00:05:04,800 That's it. But we're just going to cap it there. 59 00:05:04,800 --> 00:05:10,480 Just that's it. That's enough. We know our limits. 60 00:05:10,480 --> 00:05:14,400 Just the GDP of a small country. We just cap it there though. And then we'll go buy a country. 61 00:05:14,400 --> 00:05:15,400 It's all good. 62 00:05:15,400 --> 00:05:19,800 We use AI to take over. And I think actually that's something that they could probably 63 00:05:19,800 --> 00:05:25,680 do. GPT-4 is pretty powerful. It's quite impressive. The announcement, one of the first things 64 00:05:25,680 --> 00:05:31,760 that they said is that it's multimodal. So we discussed this last week. What does that 65 00:05:31,760 --> 00:05:36,720 actually mean? Is it going to input? Are you going to be able to input video and text and 66 00:05:36,720 --> 00:05:44,880 it output video and text? Well, it's, it's an input multimodal. Input multimodal. You 67 00:05:44,880 --> 00:05:51,960 can use this for you to say. You can input images along with the text based prompt. And 68 00:05:51,960 --> 00:05:58,280 what's incredibly powerful about this is that it has some incredible capabilities to understand 69 00:05:58,280 --> 00:06:04,440 those images that you put in. The outputs are all text based. So there's still no integration 70 00:06:04,440 --> 00:06:10,280 with the likes of Dali or any video capabilities or anything like that. Sure. That will come 71 00:06:10,280 --> 00:06:18,000 in GPT-5 or something like that. But for now, text based. And that's really quite powerful. 72 00:06:18,000 --> 00:06:22,600 Some great examples. I'm sure you've seen a few of them. The chicken nugget tray was 73 00:06:22,600 --> 00:06:23,600 a brilliant one. 74 00:06:23,600 --> 00:06:26,320 Didn't you see that? Tell us that. I didn't see that. 75 00:06:26,320 --> 00:06:31,720 So the chicken nugget tray was a baking tray with chicken nuggets laid out in a kind of 76 00:06:31,720 --> 00:06:38,000 map of the world. Chicken nuggets, you've got the making of North America and South 77 00:06:38,000 --> 00:06:45,680 America and Antarctica and all the rest of it. And it's asked to explain why this, basically 78 00:06:45,680 --> 00:06:52,960 explain this mean photo. And it does. It explains it and says, this is funny because it's, oh, 79 00:06:52,960 --> 00:07:01,120 I think the mean picture across the top said the beauty of Earth from space. Looking at 80 00:07:01,120 --> 00:07:05,760 the magnificent sense of Earth from space. But it explains it in really great detail. 81 00:07:05,760 --> 00:07:09,360 It says, you know, it sets you at the bottom of the, in the description, it says, it sets 82 00:07:09,360 --> 00:07:13,040 you up thinking that it's going to be this great, profound, beautiful image. And actually 83 00:07:13,040 --> 00:07:18,640 it's just something mundane and chicken nuggets. So it knows, it gets context. It has much 84 00:07:18,640 --> 00:07:24,400 more relevancy and an appreciation of the world. Another example, a photo of someone 85 00:07:24,400 --> 00:07:30,680 holding balloons on helium and helium filled balloons on string. And the prompt said, what 86 00:07:30,680 --> 00:07:36,560 would happen if the string was cut? And it said the balloons would fly away. 87 00:07:36,560 --> 00:07:41,640 So not just able to interpret accurately what the image is, but also to bring logic into 88 00:07:41,640 --> 00:07:47,400 inferring next stages in a story, for example. That's quite impressive. I think my favourite 89 00:07:47,400 --> 00:07:52,560 was the one, there's been riffing on this, but my favourite was probably the one that 90 00:07:52,560 --> 00:07:58,400 was in the demo, the development demo that was shown on YouTube shortly after the news, 91 00:07:58,400 --> 00:08:04,560 where the person presenting had drawn a sketch for a simple website. They wanted to build 92 00:08:04,560 --> 00:08:09,740 a joke website, took a picture of the sketch, fed it into GPT for, and said, give me the 93 00:08:09,740 --> 00:08:16,460 code that I would need to actually code this website. GPT spits out the HTML. You render 94 00:08:16,460 --> 00:08:22,640 the HTML and you get this website with this joke website and even has clickable buttons 95 00:08:22,640 --> 00:08:28,160 to reveal, you know, JavaScript clickable buttons to reveal the punchlines to the jokes. 96 00:08:28,160 --> 00:08:32,280 The hand drawn script, I mean, it was better than I could do, but it wasn't great, right? 97 00:08:32,280 --> 00:08:39,160 And somehow it was the people talk about low fidelity wireframes. It was the lowest of low 98 00:08:39,160 --> 00:08:44,480 fidelity wireframes. It was a box with, so in the text, you could barely, the text was 99 00:08:44,480 --> 00:08:50,280 barely legible. It had, what I thought was really impressive as well actually was, hint, 100 00:08:50,280 --> 00:08:54,200 text, the text that he'd written, he put in brackets where that was kind of placeholder 101 00:08:54,200 --> 00:08:58,040 text and he had understood that was placeholder text and inserted it and replaced it with 102 00:08:58,040 --> 00:09:02,640 real text. It was, yeah, that was... 103 00:09:02,640 --> 00:09:07,200 Assuming that's all real, right? And it's not been manufactured. That was pretty cool, 104 00:09:07,200 --> 00:09:08,200 right? 105 00:09:08,200 --> 00:09:13,920 Yeah, totally. That was like all of a sudden you went, holy, holy cow, that is quite something. 106 00:09:13,920 --> 00:09:19,840 So what else does it have? It's multimodal. It can, it has this enhanced reasoning that 107 00:09:19,840 --> 00:09:26,360 like we said a moment ago with the example with the balloons. If you can actually access 108 00:09:26,360 --> 00:09:32,560 this model through ChatGPT plus at the moment. So I've been playing around with it and it's 109 00:09:32,560 --> 00:09:39,400 quite interesting the kind of subtleties, because it's also billed as being more concise 110 00:09:39,400 --> 00:09:44,200 and having played with it quite a bit, it will output quite a lot, but it is more concise. 111 00:09:44,200 --> 00:09:49,160 It gets to the point in each section. So if you ask it to write a blog, it's less, and 112 00:09:49,160 --> 00:09:56,040 not that GPT-3 was particularly waffly, but the writing's tighter. Yeah, it's more succinct, 113 00:09:56,040 --> 00:10:01,240 it's more to the point. It's more on point as well actually. I think it's, yeah, it's 114 00:10:01,240 --> 00:10:06,240 really quite surprising that. They talk about it having improved safety and alignment, but 115 00:10:06,240 --> 00:10:12,800 don't really go much more in into how that is. And that's a point of contention for some 116 00:10:12,800 --> 00:10:19,040 people. Again, private company, why do they get to set the rules on what is allowed? What 117 00:10:19,040 --> 00:10:24,240 is harmful content? What is not harmful content? I'm sure that will all come out of the wash 118 00:10:24,240 --> 00:10:30,280 now in the wash over the next few months, then years as more people get involved in 119 00:10:30,280 --> 00:10:36,040 that discourse. I'm not too concerned personally there. I think OpenAI tended to be quite, 120 00:10:36,040 --> 00:10:40,960 responsible in the way that they've put the models out there and they've been quite quick 121 00:10:40,960 --> 00:10:45,240 to adapt. I think we saw that with ChatGPT when it was released. But one of the most 122 00:10:45,240 --> 00:10:51,280 interesting elements of this, other than the multimodal, for me is the longer input text. 123 00:10:51,280 --> 00:10:58,200 We've taken it from being able to handle with ChatGPT about 3000 words, and it's suddenly 124 00:10:58,200 --> 00:11:08,760 gone up to 25,000. So the amount of data you can add into a prompt has just shot through 125 00:11:08,760 --> 00:11:15,280 the roof. It's really quite significant. And that opens up the doors for much more, in 126 00:11:15,280 --> 00:11:20,680 terms of practical applications as marketers or content marketers, you have being able 127 00:11:20,680 --> 00:11:26,240 to take transcripts of podcasts like ours and just stick them in and get your show summaries 128 00:11:26,240 --> 00:11:30,760 done in a click. You're not having to break it up into smaller pieces. If you're doing 129 00:11:30,760 --> 00:11:35,960 research that is going to enable you to put whole documents in there and find key takeaways. 130 00:11:35,960 --> 00:11:41,040 There's a myriad of use cases for this, but I think that one is really, really quite exciting. 131 00:11:41,040 --> 00:11:48,520 Yeah, I'd agree. I've been having quite a play as well. I got a really good book outline 132 00:11:48,520 --> 00:11:52,520 out of it. I'm not going to share what the book's about because I actually think I might 133 00:11:52,520 --> 00:11:57,720 see if I can produce it really quickly as an example and maybe talk about it next week's 134 00:11:57,720 --> 00:12:06,800 podcast or the podcast after. But I gave it a really challenging book concept, without 135 00:12:06,800 --> 00:12:11,440 giving too much away, based on things like the hero's journey. And it did a really, really 136 00:12:11,440 --> 00:12:17,240 good job of taking something super abstract and somehow squeezing that into a hero's journey 137 00:12:17,240 --> 00:12:22,280 character arc, which I was really impressed about. I started to get it to write it. And 138 00:12:22,280 --> 00:12:26,440 by chapter four, it had kind of forgotten what it had written about in chapter one. 139 00:12:26,440 --> 00:12:31,200 So it knew the characters still, but it wasn't consistent with some of the things that had 140 00:12:31,200 --> 00:12:35,160 happened to the character in chapter one. And the chapters, I call them chapters, they 141 00:12:35,160 --> 00:12:40,400 were like 800 words max, right? Like this was not a book. So I think the strategy you'd 142 00:12:40,400 --> 00:12:44,680 have to go through and I saw someone, I can't remember who it was, had written a whole book 143 00:12:44,680 --> 00:12:49,840 of 250 pages and it required 2000 prompts. And that doesn't surprise me because I think 144 00:12:49,840 --> 00:12:53,960 if you really want to produce that type of long form content and have it be coherent, 145 00:12:53,960 --> 00:12:59,560 you're almost, say you want to produce a book chapter, you're really iteratively prompting 146 00:12:59,560 --> 00:13:05,400 every section. And maybe even, I think, giving it, either it should remember giving it a 147 00:13:05,400 --> 00:13:10,160 bit of context for each section or for each chapter so it doesn't forget what it's supposed 148 00:13:10,160 --> 00:13:14,880 to be writing about. But I was pretty impressed compared to when I ran the same test with 149 00:13:14,880 --> 00:13:22,720 GPT-3, like GPT-3's, ChatGPT, the original version, the book concept didn't work. It 150 00:13:22,720 --> 00:13:31,240 just couldn't interpret what I meant. I also really loved, and I thought was a great demonstration, 151 00:13:31,240 --> 00:13:39,120 Paul Reutzer from the Marketing AI Institute showed a post on LinkedIn where he asked GPT-3 152 00:13:39,120 --> 00:13:46,720 driven ChatGPT or GPT-3.5 to give him an out of office, a creative out of office, like a 153 00:13:46,720 --> 00:13:51,840 humorous out of office. And then he asked GPT-4 to do it and the difference is night 154 00:13:51,840 --> 00:13:58,640 and day. And I found this myself even when prompting ChatGPT to be like really creative. 155 00:13:58,640 --> 00:14:03,320 I had it right in social media posts where the prompt was, be so creative and persuasive 156 00:14:03,320 --> 00:14:10,200 you offend me. Like I was asking you to be provocative. I wanted it to say things like, 157 00:14:10,200 --> 00:14:14,240 you're a shit marketer and you don't know what you're doing. And if you don't read this 158 00:14:14,240 --> 00:14:18,880 blog post, your career will be over forever. And I could never get it to get anywhere near 159 00:14:18,880 --> 00:14:24,720 that provocative or creative. But this, and I should say Paul Reutzer is not angry and 160 00:14:24,720 --> 00:14:31,680 aggressive, but it is a great example of the difference because whether you like the output 161 00:14:31,680 --> 00:14:37,200 from GPT-4 or not, it is incredibly creative. It's like, hi Earthlings, I'm away speaking 162 00:14:37,200 --> 00:14:42,480 with robots at AI conferences. Like it is really, really, really creative, which I just 163 00:14:42,480 --> 00:14:49,920 couldn't get out of GPT-3. So yeah, as content marketers, this is a massive upgrade, I think 164 00:14:49,920 --> 00:14:54,000 for a variety of different reasons in terms of its reasoning ability. It's almost like 165 00:14:54,000 --> 00:14:59,240 knowledge of topics and ability to speak about them. The fact it's more concise, the fact 166 00:14:59,240 --> 00:15:05,080 that it really understands the prompts that you give it better than the old model did. 167 00:15:05,080 --> 00:15:12,280 So much cool stuff there. If you're a marketer or you work in content marketing and you've 168 00:15:12,280 --> 00:15:16,680 played with chat GPT, you need to play with chat GPT-4 because I think you will see a 169 00:15:16,680 --> 00:15:22,520 big difference. And if you're a content marketer and you haven't played with either yet, you're 170 00:15:22,520 --> 00:15:28,440 going to be obsolete pretty quick if you don't start to learn how to manipulate these tools 171 00:15:28,440 --> 00:15:34,360 to get a decent first draft out. Don't get me wrong, I still think heavy editing is required. 172 00:15:34,360 --> 00:15:38,760 I still think if you want things with genuine novel insight, you as the human have to bring 173 00:15:38,760 --> 00:15:44,920 the insight to the party. But yes, these things are getting crazy powerful very quickly. 174 00:15:46,360 --> 00:15:52,280 And that's coming through in the benchmarks that they're showing. The capabilities of 175 00:15:52,280 --> 00:16:02,360 chat GPT or GPT-3.5, I think it could, in terms of one of the big domain areas was law. And in 176 00:16:02,360 --> 00:16:07,880 America, passing the bar is the kind of big example. Everyone's just passing the bar. Can 177 00:16:07,880 --> 00:16:15,640 it pass the bar? And I believe chat GPT or GPT-3 had about a 10% pass rate. 178 00:16:15,640 --> 00:16:24,200 I think it was able to, against humans, it was in the bottom 10% of scores. 179 00:16:24,200 --> 00:16:29,880 Oh, she was right. Yeah, I wasn't sure of the... Whereas this is now in the top 90%. 180 00:16:29,880 --> 00:16:34,440 Yeah, it's scores in the top 90%. 181 00:16:35,880 --> 00:16:42,680 Which is, you know, that's quite incredible. And it doesn't say so many domains as well. 182 00:16:42,680 --> 00:16:50,920 I saw some in the life sciences. One of the examples that somebody put on Twitter was about, 183 00:16:50,920 --> 00:16:55,800 we were talking about bio GPT in last week's episode, talking about the drug discovery 184 00:16:55,800 --> 00:17:04,600 capabilities or the drug interaction capabilities of GPT-4 that people have been playing with. I was 185 00:17:04,600 --> 00:17:10,360 like, wow, that's, you know, it's not trained on this, but it's capable of doing it. Well done. 186 00:17:10,360 --> 00:17:18,520 Yeah. I mean, it must be sucking that out of some training material somewhere. I think the exam 187 00:17:18,520 --> 00:17:24,360 thing is interesting. And I honestly haven't probed how they do it deep enough to say, 188 00:17:24,360 --> 00:17:32,280 but one criticism would be training something to take a specific test is potentially quite easy. 189 00:17:32,280 --> 00:17:39,240 Right. Even humans know how to gain certain tests to get a good mark in, you know, for some tests. 190 00:17:39,240 --> 00:17:47,160 Right. So in some areas, I'm not super surprised, but it was the massive jump for some of the, 191 00:17:47,160 --> 00:17:53,080 like you were saying, like for the bar exam, GPT-3 sucked at some, like it was at 0% for a 192 00:17:53,080 --> 00:17:57,560 couple of the tests. And if you look out there, lovely listeners, you'll find there's a bar chart 193 00:17:57,560 --> 00:18:03,080 floating around the interwebs that shows, I don't know, 20 odd tests where GPT-3 scores, 194 00:18:03,080 --> 00:18:09,480 where GPT-4 scores. And it really is those ones where GPT-3 sucked, but GPT-4 is like way better 195 00:18:10,200 --> 00:18:17,000 than most humans, not just better than GPT-3. I actually had an issue this week, a consumer 196 00:18:17,000 --> 00:18:25,800 rights issue, shall we say. And I thought, I don't know how to tackle this one. And I just thought, 197 00:18:25,800 --> 00:18:29,640 you know what, I'm going to throw it into GPT-4 and see what it says. And I kind of set it with 198 00:18:29,640 --> 00:18:36,120 that you are an expert in UK consumer law with domain expertise in this particular area. 199 00:18:37,000 --> 00:18:42,840 And I kind of explained the situation, bullet pointed key points, tried to keep it as 200 00:18:43,480 --> 00:18:50,520 fact-based as possible and ask it to give me advice and then write a letter to the company 201 00:18:50,520 --> 00:18:59,080 that I was engaging with and then provide me with kind of next steps strategy. That was fascinating. 202 00:18:59,080 --> 00:19:06,040 First off, the letter that it wrote, it had the address of the company on the letter. I told it 203 00:19:06,040 --> 00:19:12,520 the name of the company and that it was in Derby and it put the company's full address, their real 204 00:19:12,520 --> 00:19:21,000 address on there. So that's a thing that it can do. Because it can't access the internet still, 205 00:19:21,000 --> 00:19:28,360 right? No, but it's obviously got within its corpus of training data, lots of business directories 206 00:19:28,360 --> 00:19:35,160 and things like that. So it'd be worth sticking Biosrata in and seeing. I'm about to ask it 207 00:19:35,160 --> 00:19:40,040 what the address is. Yeah. And then, so in the strategy actually, in the kind of like, 208 00:19:40,040 --> 00:19:49,400 what should I do? It gave six potential kind of next steps, but in each one, it gave like how 209 00:19:49,400 --> 00:19:56,920 likely that is to get me the resolution that I was after. So it was said, you know, there's this 210 00:19:56,920 --> 00:20:03,320 breach of contract, there's miss selling, you could raise a complaint with the FCA, you could 211 00:20:03,320 --> 00:20:09,000 do this, you could do that. But it gave a detailed kind of explanation. The last one, for instance, 212 00:20:09,000 --> 00:20:15,080 it said, you could take legal action through a legal professional, but this could get expensive 213 00:20:15,080 --> 00:20:20,520 and so on. I think that was a really obvious kind of answer. But the ones earlier were quite 214 00:20:20,520 --> 00:20:26,760 specific and nuanced, I would say. I guess your only danger there is we know that 215 00:20:27,320 --> 00:20:33,160 Chachi Bti can churn out absolute junk that's completely incorrect. So I can imagine a use case 216 00:20:33,160 --> 00:20:38,840 for this. The danger for you as a consumer is you still need someone with a legal background to 217 00:20:38,840 --> 00:20:42,600 check that and make sure it's not nonsense, I guess. But at least it arms you with the right 218 00:20:42,600 --> 00:20:49,400 things to go speak to about with an expert. Yeah, 100%. And again, it's always that kind of 219 00:20:49,400 --> 00:20:53,960 human in the loop element to it. What I would say is that I did check some of the details, 220 00:20:53,960 --> 00:21:00,760 because I thought, is that right? And it gave reference to Consumer Rights Act 2015 and pulled 221 00:21:00,760 --> 00:21:07,560 out relevant sections of it. And that was quite powerful. And what it gave me was, I would say, 222 00:21:08,760 --> 00:21:16,760 a starting point, a far better jumping off point that I would have got doing my own research. 223 00:21:16,760 --> 00:21:22,280 They saved me a hell of a lot of time. And potentially, I can imagine a scenario where 224 00:21:23,160 --> 00:21:31,320 I go to a solicitor and say, you know, this is my current situation, this is my understanding of it. 225 00:21:31,320 --> 00:21:35,960 And you save yourself a £600 bill just by not having to spend the first hour explaining everything. 226 00:21:37,560 --> 00:21:43,640 Yeah, agreed. Pretty cool. Right. I think we've probably whacked Lirico enough on GPT-4 scene as 227 00:21:43,640 --> 00:21:48,520 we've got a ton more of other stuff to give. Let's talk about Copilot, which came out a handful of 228 00:21:48,520 --> 00:21:56,600 hours ago. So take us through it, Martin, why it's important for marketers. Well, the productivity 229 00:21:56,600 --> 00:22:04,360 suite that we all use, we all know, we all love, we used to have Clippy in Microsoft Word, but now 230 00:22:04,360 --> 00:22:11,080 we have Copilot. And Copilot definitely feels like Clippy on steroids. Clippy has evolved, 231 00:22:11,080 --> 00:22:18,280 taken new form and is now making our lives much easier. Before we get into the detail of it, 232 00:22:18,280 --> 00:22:22,760 I just want to say one thing. In the promotional video, there was a two minute promotional video, 233 00:22:22,760 --> 00:22:27,320 and then there was the longer kind of keynote session. In that session, Jared Spataro, 234 00:22:27,320 --> 00:22:31,320 the corporate vice president, modern work and business applications at Microsoft, 235 00:22:32,040 --> 00:22:39,720 he said at one point, at best work not only feeds our families, it feeds our souls. 236 00:22:39,720 --> 00:22:47,400 And at that point I felt what I thought was my soul leaving my body. It was a little bit of vomit 237 00:22:47,400 --> 00:22:54,920 in my mouth. I just, I can see he's doing it to you now, just remembering the event. 238 00:22:54,920 --> 00:23:01,720 So I couldn't, I couldn't hack it. I said, no, don't say that. That's awful. Anyway, enough of 239 00:23:01,720 --> 00:23:12,120 the cringe. Copilot looks pretty cool. Yeah. So they've baked GPT-4 and large language. It looks 240 00:23:12,120 --> 00:23:16,360 like they might be introducing, I'm guessing some of their own as well. I'm not sure. I've not read 241 00:23:16,920 --> 00:23:22,600 that they have done specifically, but I wouldn't be surprised. Well, this is off the presses as 242 00:23:22,600 --> 00:23:27,640 well. So we'll probably learn more in the next couple of days. Yeah. This came out literally 243 00:23:27,640 --> 00:23:32,280 six, five, five, six hours ago. So trying to get ahead around it all very quickly. 244 00:23:32,280 --> 00:23:40,040 It's baked into the whole ecosystem. So Word, Outlook, Excel, all of it has it baked into it. 245 00:23:40,040 --> 00:23:46,120 And what's interesting is it all works together. I thought the presentation that they did was 246 00:23:46,120 --> 00:23:51,080 incredibly powerful. So in short, they'd say that it works with all of the apps. It turns natural 247 00:23:51,080 --> 00:23:57,080 language prompts into productivity actions, unleashes creativity, unlocks productivity, 248 00:23:57,080 --> 00:24:02,520 and up-level skills. So what can it do? Well, it generates content, automate tasks, and provides 249 00:24:02,520 --> 00:24:10,520 relevant data and insights. It's enterprise ready as well. So they say that it protects the tenant, 250 00:24:10,520 --> 00:24:16,040 the group, and the individual data. So it's very secure. It's designed to be rolled out at enterprise 251 00:24:16,040 --> 00:24:22,680 level. There's not going to be any data leakage. It's all contained within your instance, within 252 00:24:22,680 --> 00:24:29,640 your license. So this is clear, because people have been super worried in larger companies about 253 00:24:29,640 --> 00:24:35,560 putting things into chat GPT. Like let's say I want some help editing a press release for an 254 00:24:35,560 --> 00:24:40,600 upcoming product that's going to change the world. Well, clearly that's important commercially 255 00:24:40,600 --> 00:24:45,480 sensitive news. I can't copy and paste the press release into chat GPT. I don't know where that's 256 00:24:45,480 --> 00:24:52,200 going or who's seeing that. So incredible sensitivity and risk aversion for larger companies. 257 00:24:52,200 --> 00:24:58,360 I've heard stories of chat GPT has even been blocked in a lot of companies because they're 258 00:24:58,360 --> 00:25:02,200 like, we don't know where the data goes. We don't know who can see the information. None of you are 259 00:25:02,200 --> 00:25:07,560 allowed to use it. So the fact that then Microsoft can do this within their privacy frameworks that 260 00:25:07,560 --> 00:25:13,800 they're ideally, and I guess would have to be held accountable to, frameworks that probably most of 261 00:25:13,800 --> 00:25:19,880 us were already signed up to even in large enterprise should open the door for those companies that 262 00:25:19,880 --> 00:25:26,360 didn't feel comfortable having their employees use tools where they didn't know where their data was 263 00:25:26,360 --> 00:25:33,560 going. I guess the only thing would be, you know, you use a Jasper or a copy.ai or a writer, which 264 00:25:33,560 --> 00:25:41,160 all rely on open AI. Microsoft also relies on open AI. So yes, there's all these privacy things in 265 00:25:41,160 --> 00:25:45,160 place, but where does the data really go? Are they still going to use it to train further models and 266 00:25:45,160 --> 00:25:50,200 all this stuff? Well, Microsoft actually have access to the code, don't they? They have the 267 00:25:50,200 --> 00:25:56,200 model themselves. Nobody else has access to that's their $10 billion deal. Right. Everyone else can 268 00:25:56,200 --> 00:26:02,200 access it through the API. Microsoft has the thing itself. So it's enclosed. Right. I didn't realize 269 00:26:02,200 --> 00:26:07,320 that. That helps answer that question. Thank you, Mark. I'm glad you're here. Of my pleasure. 270 00:26:08,440 --> 00:26:15,000 Come back anytime. So what else do they talk about? It's integrated into all of the apps in a way 271 00:26:15,000 --> 00:26:21,640 that gives this consistent user experience. And they've said that it's built on responsibility 272 00:26:21,640 --> 00:26:28,760 with AI principles and ethical AI standards kind of at the heart of it, which you would expect them 273 00:26:28,760 --> 00:26:34,920 to say. It does look smart. I thought the demo was great. You also saw the same demo as me. There's 274 00:26:34,920 --> 00:26:39,880 a particular example halfway through the presentation where they take you through 275 00:26:39,880 --> 00:26:45,160 the process, your graduation of the presenter's daughter. She's graduating and it takes you 276 00:26:45,160 --> 00:26:53,320 through sending the invites through to the creation of the presentation that you would 277 00:26:53,320 --> 00:26:59,720 have playing at the party with embarrassing photos and all of that kind of thing. I thought that was 278 00:26:59,720 --> 00:27:05,880 incredibly powerful. I need to stop saying incredibly. That's just how I feel about it at 279 00:27:05,880 --> 00:27:12,840 the moment. What was quite smart was in the chat conversation, it understood the context. So you 280 00:27:12,840 --> 00:27:19,480 could say, send the email invite for daughter's 16th birthday party to family and friends. And 281 00:27:19,480 --> 00:27:26,520 it would draft the email invite, but taking in context from other emails like the venue booking 282 00:27:26,520 --> 00:27:33,800 confirmation email. So it would integrate the party time and date and location into the 283 00:27:33,800 --> 00:27:40,760 message. That is so cool. And they give a business example as well, didn't they, where you're 284 00:27:40,760 --> 00:27:46,600 emailing a colleague and assuming you've got all of your documents on your SharePoint, you can 285 00:27:46,600 --> 00:27:52,760 associate a document with that email, but have Copilot automatically pull the context from that 286 00:27:52,760 --> 00:27:57,880 document to fill the email. So it might be that you need to send over the actions from a call, 287 00:27:57,880 --> 00:28:02,360 but that's easily pulled from the transcription in Teams. It's already been done in the 288 00:28:02,360 --> 00:28:07,560 Teams. It's already been saved. And by the way, if you need to flag to them that they should go and 289 00:28:07,560 --> 00:28:13,800 have a look at the PowerPoint decks, describing the value proposition of the new product, 290 00:28:13,800 --> 00:28:19,960 oh, please Copilot, can you pull out the benefits features and proof points out of the deck and 291 00:28:19,960 --> 00:28:24,760 just add them as bullets in the email? Yeah, cool. No problem. It goes into the deck, knows 292 00:28:25,480 --> 00:28:29,720 semantically what important bits it should pull out, pops them in the email as well. 293 00:28:29,720 --> 00:28:36,600 Like it's kind of mad when you think about the sheer range of possibilities for how much faster 294 00:28:36,600 --> 00:28:41,640 you can do things. One of the points that there's a nice little quote I thought that describes 295 00:28:42,680 --> 00:28:48,920 my experience of using PowerPoint. And the presenter, she said, the average user uses 10% 296 00:28:48,920 --> 00:28:55,000 of PowerPoint. Copilot unlocks the other 90%. And I thought, yes, it's the same with all of 297 00:28:55,000 --> 00:29:00,040 them work for a soft suite. There's so much power in it, but I just scratched the surface of it. 298 00:29:00,040 --> 00:29:04,840 And this is the same with nearly everybody that I know. Just, it can do so much, but you just do 299 00:29:04,840 --> 00:29:09,880 the same thing time and again. The presentation example where it pulled in images automatically 300 00:29:09,880 --> 00:29:16,040 from a folder and then created the deck and you edit the text, that it automatically created 301 00:29:16,040 --> 00:29:19,880 a few tweaks and it would make its auto. And it looked beautiful. Let's be honest. It didn't 302 00:29:19,880 --> 00:29:24,520 even look like a junk presentation. Whatever template it pulled from, I don't know if it uses 303 00:29:24,520 --> 00:29:29,000 branded templates on your SharePoint or if they're built into PowerPoint already, but it looked 304 00:29:29,560 --> 00:29:34,920 mustard. It did with, with animations that didn't look cheesy with transitions that didn't look 305 00:29:34,920 --> 00:29:43,320 cheesy. It looked modern contemporary. It looked smart. How much time in agencies and marketing 306 00:29:43,320 --> 00:29:49,560 departments and design departments, how much time is spent putting together the PowerPoint? 307 00:29:50,120 --> 00:29:53,880 I'm working on a deck, right? The content's there. The content's there in somebody's head 308 00:29:53,880 --> 00:29:58,200 in other documents, in other areas, there was meeting notes, there's this, that, the other. 309 00:29:58,200 --> 00:30:03,960 Someone has to go and sit and create that deck. People will spend hours pulling a deck together. 310 00:30:03,960 --> 00:30:10,600 If that can now be done in minutes, you know, relative, relatively speaking, you know, 311 00:30:10,600 --> 00:30:17,720 10, 15 minutes rather than three hours or something, that's a huge time table. It's an instance. 312 00:30:17,720 --> 00:30:25,560 Yeah. In the life science industry, we, because we deal with such technical and complex subject 313 00:30:25,560 --> 00:30:31,240 matter and the products that we're talking about, showing the applications of talking about the 314 00:30:31,240 --> 00:30:36,120 benefits features applications and all that good stuff. It's usually very complicated. And the 315 00:30:36,120 --> 00:30:41,640 subject matter experts are usually working with the PhD scientists, very detail oriented people 316 00:30:41,640 --> 00:30:47,000 who can, who love the wizardry of the technology that they've built or that they're using and go 317 00:30:47,000 --> 00:30:50,840 into a great amount of detail. It'll be like I'm doing now. Can you tell my next scientists as well? 318 00:30:52,600 --> 00:30:58,760 Because of that, their PowerPoints can massively suffer, right? Like 52 bullets on a page, 319 00:30:58,760 --> 00:31:05,560 basically inviting people to read full prose on slides. But now if there's, it's almost like 320 00:31:05,560 --> 00:31:10,520 there's a magic button that you can press that's going to distill all that key information into 321 00:31:10,520 --> 00:31:15,960 something that's going to be much more digestible for people to, you know, to, 322 00:31:15,960 --> 00:31:20,120 to be presenting, they will focus on you, the presenter. 323 00:31:20,680 --> 00:31:26,600 Exactly. But that, that, as you said rightly is at the moment, you know, the domain of the 324 00:31:26,600 --> 00:31:33,480 marketing department, the domain of the agency, there is absolutely a number of important skills 325 00:31:33,480 --> 00:31:39,160 required to go from dense technical output. Because to be honest, there's, there's reasons 326 00:31:39,160 --> 00:31:43,400 the subject matter experts think and act like that. It's because the majority of their job requires 327 00:31:43,400 --> 00:31:48,840 them to be like that. Right? On the, on the few occasions they have to create a PowerPoint, 328 00:31:50,680 --> 00:31:55,240 this is what you get. And you know, they're not, most subject matter experts don't create enough 329 00:31:55,240 --> 00:31:59,800 PowerPoints to even warrant spending an hours and hours and hours of training to try and be a good 330 00:31:59,800 --> 00:32:04,440 graphic designer or content summarizer or value proposition created. It's just not a good use of 331 00:32:04,440 --> 00:32:09,560 their time, but that used to be the marketing department. And now is it potentially a button? 332 00:32:09,560 --> 00:32:15,240 Who knows? I think until we get to play with it and really see how good it is at extracting key 333 00:32:15,240 --> 00:32:20,360 information and associating appropriate visuals and all that stuff, we have to wait and see, 334 00:32:20,360 --> 00:32:25,000 but it's certainly a step in that direction. Yeah. Huge potential that I did like during the 335 00:32:25,000 --> 00:32:32,840 presentation. They were very heavy on, this is a starting point is that it's designed to give you 336 00:32:32,840 --> 00:32:38,520 a starting point. This is about augmenting human work and not replacing human work. 337 00:32:38,520 --> 00:32:42,680 They were very big on that. And I think that really comes across in the way that the UI of the tool 338 00:32:42,680 --> 00:32:48,680 is as well. Well, still very much in control. This is not, you know, going to replace you, 339 00:32:48,680 --> 00:32:55,480 but it's going to allow you to get stuff on a page quicker. Yeah, I agree. Co-pilot's a good name. 340 00:32:56,120 --> 00:33:02,280 Member of our team, Tom, very smart guy. We, when we started talking about AI and supporting tools, 341 00:33:02,280 --> 00:33:07,720 you know, back when we began testing them, he said, we're going to all have to become very 342 00:33:07,720 --> 00:33:12,840 good pilots. And I just really thought that was a great way of phrasing it because you look at a 343 00:33:12,840 --> 00:33:19,320 modern plane, it can do a lot of things automatically, but there's so many edge case scenarios where you 344 00:33:19,320 --> 00:33:24,840 need the human in the loop. And I don't even think the AIs that we're going to be using for a lot of 345 00:33:24,840 --> 00:33:28,760 these creative use cases are anywhere near as good as the automation in aeroplanes, to be honest, 346 00:33:28,760 --> 00:33:33,880 because I think there's way more edge cases in real life and across all these complex domains that 347 00:33:33,880 --> 00:33:38,760 the AIs are trying to help us with, at least in the marketing industry. So co-pilot is a good word 348 00:33:38,760 --> 00:33:43,480 because we are the pilot and we can get help and we could be more efficient, more effective. 349 00:33:43,480 --> 00:33:48,200 I think, again, it would be great for getting things going. And in the example that you described 350 00:33:49,320 --> 00:33:54,920 in that Microsoft presentation, the raw text that's put in that PowerPoint, the lady who 351 00:33:54,920 --> 00:33:58,840 presents, I can't remember her name, she edits it, right? It's not in her voice. She's like, no, 352 00:33:58,840 --> 00:34:03,080 I wouldn't say that like that. So she edits it. And that's exactly how I expect it to be for some 353 00:34:03,080 --> 00:34:08,120 time, if not forever, because at least at the moment, they don't know what's in our heads, 354 00:34:08,120 --> 00:34:13,720 right? Although if our story from last week where you're an MRI and everything is what you see, 355 00:34:13,720 --> 00:34:19,880 and maybe not so, but yeah, co-pilot is a good name. I think that's how we should all be looking 356 00:34:19,880 --> 00:34:26,200 at it for sure. And I can't wait to get my hands on it. So on that note, Google announced their 357 00:34:26,200 --> 00:34:31,480 own generative AI being baked into Google workspace. But the way that I've just described it to you 358 00:34:31,480 --> 00:34:38,200 is Google has announced his own generative AI in the workspace. I haven't said Google copilot 359 00:34:38,200 --> 00:34:45,080 or whatever. Have they missed a trick? 100%. Trying to get your head around exactly what 360 00:34:45,080 --> 00:34:50,280 Google's product AI infusion is going to look like is hard because from the news announcements, 361 00:34:50,280 --> 00:34:56,120 they don't go into any detail. And Microsoft are way ahead. I think this is Google's attempt to go, 362 00:34:56,120 --> 00:35:01,480 we've got one. We're going to do it. We better tell you so that you think we're still cool. 363 00:35:01,480 --> 00:35:07,640 And holy moly, right team, we need to do it on the same day that GPT-4 is announced and we're 364 00:35:07,640 --> 00:35:11,640 going to launch it a few hours early. Quick, get that two minute promo video done. 365 00:35:11,640 --> 00:35:17,480 You make a massively good point because I was talking about this with some of the team today 366 00:35:18,200 --> 00:35:25,160 after copilot was announced. You can't as an enterprise not access the power of this. 367 00:35:25,160 --> 00:35:30,280 Like if you're on Google apps, if Google don't give you this within three months of Microsoft 368 00:35:30,280 --> 00:35:35,320 bringing it out, even if you've got a two or three year subscription to Google still going, 369 00:35:36,040 --> 00:35:41,400 it's worth binning that off and just transitioning everything over to Microsoft because the sheer 370 00:35:41,400 --> 00:35:48,840 time savings and efficiencies and productivity gains you'll get from having AI embedded into all 371 00:35:48,840 --> 00:35:54,200 your tools and connected to all your data. So Google are now, you know, there's a fair amount 372 00:35:54,200 --> 00:35:58,760 of, you know, there's a fuse burning, mission impossible style, bum, bum, bum, bum, right? 373 00:35:58,760 --> 00:36:02,920 And it is getting quite close to the bit where everything goes boom and they really, 374 00:36:02,920 --> 00:36:07,640 really need to get to the party quickly. They do. And it does look like their promo 375 00:36:07,640 --> 00:36:12,120 video addresses much of the similar things that we've talked about. In fact, on the, 376 00:36:12,120 --> 00:36:20,680 on the example that they give, they have AI baked into the Gmail system. You've come to a threaded 377 00:36:20,680 --> 00:36:26,680 email conversation, laid lots of emails in that thread. It then summarises that thread for you. 378 00:36:26,680 --> 00:36:32,760 It turns the summary into a campaign brief. It's all about marketing campaign. It turns it into 379 00:36:32,760 --> 00:36:39,800 a campaign brief in a Google doc, turns the campaign brief into a slide deck in Google slides. 380 00:36:40,440 --> 00:36:45,960 And it's that, you know, look how smooth this whole process is. But all you see is that that 381 00:36:45,960 --> 00:36:52,520 two minute video, the equivalent of the one that, but it didn't get, it didn't get me excited. 382 00:36:53,240 --> 00:36:59,320 And it, and it lacked the polish and it didn't, it didn't have this feel of something 383 00:36:59,320 --> 00:37:04,120 transformational was coming. Now you and I nerd out about this a lot and have done for a long time, 384 00:37:05,000 --> 00:37:11,480 but I thought Microsoft approach really got people excited. And as much as it was a feeds our families 385 00:37:11,480 --> 00:37:16,280 and feeds our souls a little bit cringe worthy at time. And these things made an event of it. 386 00:37:16,280 --> 00:37:22,040 They tried to make it more exciting. So yeah, Google, you need to start exciting us. Yeah, 387 00:37:22,040 --> 00:37:28,040 very much so. Right. We, this is going to be one of our longer episodes, but there's been loads of 388 00:37:28,040 --> 00:37:32,840 good stuff to talk about. Let's look at a couple of quick things. Just that we saw this week and 389 00:37:32,840 --> 00:37:39,800 we didn't want to make it all enterprise level baked in AI tools. Did we mind? So very quickly, 390 00:37:39,800 --> 00:37:43,160 we were playing with a couple of tools. Let's talk about a depth first. There was a, 391 00:37:44,120 --> 00:37:51,240 a company that got $350 million in funding this week called a depth that is a generative AI that 392 00:37:51,240 --> 00:37:55,640 takes natural language input and turns it into action. So this is a bit different from what we've 393 00:37:55,640 --> 00:38:02,920 been talking about so far. It's almost like it is being designed to be able to operate all the tools 394 00:38:02,920 --> 00:38:08,040 you already use for you. So that when you ask things of it in natural language, it can go do 395 00:38:08,040 --> 00:38:13,160 them. So some of the examples that are given on the website are things like, set a reminder for me 396 00:38:13,160 --> 00:38:18,600 to email this person, find me a refrigerator for under a thousand dollars, add a new column to this 397 00:38:18,600 --> 00:38:24,520 spreadsheet show in this, add this person to Salesforce and give me a reminder to follow up 398 00:38:24,520 --> 00:38:30,920 with them with the presentation. So in essence, multimodal, but in a different sense, like 399 00:38:30,920 --> 00:38:37,160 multi ability to go access all the tools you already use, click buttons, take actions, etc, 400 00:38:37,160 --> 00:38:42,680 etc. So obviously that's a fairly big bet by the folks that have just chucked all that cash into 401 00:38:42,680 --> 00:38:52,840 a debt. Yeah. Text to action as a, as a generative AI, pretty cool. So that'll be fun. Yeah. I can 402 00:38:52,840 --> 00:39:00,040 imagine, you know, sending a text message and all of a sudden my Excel reports are done and, you know, 403 00:39:01,560 --> 00:39:05,320 when you start thinking about these things working together, that your mind just kind of blows a 404 00:39:05,320 --> 00:39:12,520 little bit, you know, writing a prompt to adapt that then speaks to copilot, copilot goes off. 405 00:39:13,400 --> 00:39:17,480 Does this thing for you? I don't, I think we'll write in a lot of instances, like if you're in a 406 00:39:17,480 --> 00:39:22,120 busy office, but honestly, we're going to be speaking to these tools. Yeah. That's a fair point 407 00:39:22,120 --> 00:39:30,680 really. It's computer from Star Trek. That's where we're headed. Maybe not with the sort of super 408 00:39:30,680 --> 00:39:37,720 all encompassing powers that are envisageable by where most people think of AGI, but with enough 409 00:39:37,720 --> 00:39:43,480 power, certainly for marketers and sales folk to be instructing their software platforms to go do 410 00:39:43,480 --> 00:39:47,400 things for them that usually would have required loads of manual input. A click, click, click, 411 00:39:47,400 --> 00:39:54,760 scroll, click, copy, paste, click, scroll becomes, Hey, do this please. And it's done. So yeah, 412 00:39:54,760 --> 00:39:59,160 that's pretty cool. The other thing was we were playing with a tool called DID this week. 413 00:39:59,160 --> 00:40:05,640 Imagine like a near human looking avatar with a near human sounding voice that basically acts 414 00:40:05,640 --> 00:40:11,880 like GPT. So when we say near human and near voices and look, it's kind of like uncanny valley style, 415 00:40:11,880 --> 00:40:20,520 like looks human, not quite sounds pretty human, but still a tiny bit robotic. But you can ask it 416 00:40:20,520 --> 00:40:24,520 questions and through, you know, natural language prompt, and rather than spitting out text, it 417 00:40:24,520 --> 00:40:29,480 speaks to you. So you could absolutely as a marketer, imagine that replacing a chatbot on 418 00:40:29,480 --> 00:40:34,600 your website, or even if you're in customer service, you know, a video customer service rep, 419 00:40:34,600 --> 00:40:40,360 even on the website or almost hosting some sort of video call with a customer. But the customer 420 00:40:40,360 --> 00:40:46,680 service rep is this animated avatar that looks kind of human like and speaks like a human. I mean, 421 00:40:46,680 --> 00:40:51,240 that's kind of mind blowing as well. And I know there's, we've talked about it before, there's 422 00:40:51,240 --> 00:40:57,480 quite a few companies out there trying to do similar things, but this was by far the easiest 423 00:40:57,480 --> 00:41:02,680 one I've played with. Like I literally logged in, signed up, started giving it prompts and having 424 00:41:02,680 --> 00:41:11,240 conversation with it. Very different setup to a lot of the other AI avatar tools that I've used or 425 00:41:11,240 --> 00:41:16,200 seen. Synesthesia, I think is one of the big ones in this. Yeah, and I think so, Bill. I haven't 426 00:41:16,200 --> 00:41:23,240 seen that they're integrating anything with chat GPT or any chat based AI, but it's the next logical 427 00:41:23,240 --> 00:41:28,840 step for them. And obviously it's happening within this space. I'm fascinated to know what the 428 00:41:28,840 --> 00:41:34,200 consumer response is going to be. When these are rolled out, how people are quite happy using 429 00:41:34,200 --> 00:41:39,960 chatbots and kind of live chat agents. They're always quite clogged on here, those, but as those 430 00:41:39,960 --> 00:41:44,920 get more powerful with chat GPT, they're going to be quite useful and functional. Slapperface 431 00:41:44,920 --> 00:41:53,320 on it. And how are people going to interact? Are people going to hate to it? Are people going to 432 00:41:53,320 --> 00:41:59,080 have emotional response? No, I have. What do we do when we fall in love? Yeah, I fell in love with 433 00:41:59,080 --> 00:42:06,920 the chat agent on my Eon meter reading website. Oh yeah, I love my latest iPhone. No, I really 434 00:42:06,920 --> 00:42:19,320 love the fire. Right. Moving quickly on. Tool of the week, Stakey, is still GPT-4 with a comfortable 435 00:42:19,320 --> 00:42:25,800 circle. It is GPT-4. So I wanted to play around with this. There's some people that have been 436 00:42:25,800 --> 00:42:36,760 doing some cool projects on there. Someone made a playable version of Pong, the classic video game. 437 00:42:37,560 --> 00:42:44,680 And I thought, well, wouldn't it be good if we could get the Artificially Intelligent Marketing 438 00:42:44,680 --> 00:42:53,800 podcast to maybe have a new look and feel on the website, maybe create a whole website. Now, the 439 00:42:53,800 --> 00:43:01,240 thing I want to stress about this is we started recording this. I think we jumped on the Zoom call 440 00:43:01,240 --> 00:43:10,520 at 8.30 PM. I started this at seven minutes past eight. Okay. So you had 20 odd minutes to get, 441 00:43:11,080 --> 00:43:17,080 to do what you've been doing. 23 minutes to get this to the point that it is at. And I'm just 442 00:43:17,080 --> 00:43:22,520 going to share this screen now. So for those of you listening, you don't get the benefit of this, 443 00:43:22,520 --> 00:43:28,760 but if you can find the video, you'll see it. But this is the first time Paul has seen it. So 444 00:43:32,360 --> 00:43:33,000 there you go. 445 00:43:33,000 --> 00:43:37,800 Ooh. So wait a minute. Take us through the process for how you did this. 446 00:43:38,360 --> 00:43:45,640 So this is a three-page website. This is, and it works. All of these pages are actual pages. You'll 447 00:43:45,640 --> 00:43:49,800 see we've got the home page, the about page, and the contact us page. Contact us page has got a 448 00:43:49,800 --> 00:43:54,440 form on it. At the bottom, we've got a newsletter signup with stay up to date, Lotus, episode, 449 00:43:54,440 --> 00:44:00,200 news, and events. It's got a form with an API connection. This doesn't actually work because 450 00:44:00,200 --> 00:44:05,640 I haven't put the API key for the email marketing platform. But once I drop in the API key, this 451 00:44:05,640 --> 00:44:10,520 will be connected to Send in Bloom email marketing platform. And you can see it's got a home page. 452 00:44:10,520 --> 00:44:15,960 It's written all of the texts. It's got three buttons that link you out to listen on Apple 453 00:44:15,960 --> 00:44:21,080 podcasts, listen on Spotify, listen on Amazon music. It's got copyright notices and privacy 454 00:44:21,080 --> 00:44:21,400 policy. 455 00:44:21,400 --> 00:44:24,440 Whoa, whoa, whoa, whoa. Right. Okay. Tell us what you did. 456 00:44:24,440 --> 00:44:30,600 Yeah. So what I did, the starting point for this was I actually asked it. I said, 457 00:44:30,600 --> 00:44:33,320 you're a world-class prompt engineer. 458 00:44:34,680 --> 00:44:37,880 We're just talking to yourself there, Martin. What did you say to chat GPT? 459 00:44:39,400 --> 00:44:43,000 To chat GPT, you're a world-class prompt engineer. This is GPT for remember. So 460 00:44:43,000 --> 00:44:51,960 world-class prompt engineer and website developer. Write a prompt for chat GPT to write a website. 461 00:44:52,760 --> 00:44:59,160 So I actually ended up writing this very, very detailed brief, a full written brief with, 462 00:45:00,520 --> 00:45:05,080 you know, it was a very comprehensive brief. I thought it was very solid. I then took that 463 00:45:05,080 --> 00:45:14,040 brief and then popped it into a new chat and said, you're a world-class website developer. Your code 464 00:45:14,040 --> 00:45:21,960 is neat, tidy, minimal, et cetera. You've been asked to design and develop a website following 465 00:45:21,960 --> 00:45:27,080 this brief. And then I asked it to create that website. The interesting thing was its first 466 00:45:27,080 --> 00:45:33,640 response was to almost like write an email to me or write a long response to me saying, 467 00:45:33,640 --> 00:45:36,360 I've written your website. I've created your website. This is what I've done. 468 00:45:37,400 --> 00:45:39,240 It was the account manager. I like it. 469 00:45:40,440 --> 00:45:45,720 Absolutely. It was going, it's this secure. I've added this to the page. And I was like, no, no, 470 00:45:45,720 --> 00:45:55,240 no. I need the files to upload to the server. Send me each of the files now. And then it started 471 00:45:55,240 --> 00:46:06,040 writing the files. So what I ended up getting was three HTML files for the pages. I got, 472 00:46:06,040 --> 00:46:15,800 I think it was two JavaScript files and one CSS file. And I've taken those and opened them in the 473 00:46:15,800 --> 00:46:20,680 browser. Really? I mean, it really is that, that's basically one of them. There were some additional 474 00:46:20,680 --> 00:46:27,800 tweaks. So the logo is actually the hosted logo from our website. So I had to go back and tell it 475 00:46:27,800 --> 00:46:33,240 where to source that from. So I've given it the address for the image. 476 00:46:34,200 --> 00:46:40,520 Couple of other little things that I, what else did I need to change? I tried to get it to put some 477 00:46:41,560 --> 00:46:46,760 animation into the text and stuff. Couldn't quite get that working within 20 minutes. 478 00:46:46,760 --> 00:46:53,240 Oh, well that's highly disappointing. What a waste of 20 minutes that you only built a website from 479 00:46:53,240 --> 00:47:01,480 scratch. Code and design. I mean, that's, I mean, yeah, for 20 minutes work, it wrote all, 480 00:47:01,480 --> 00:47:06,280 yeah, that's it. It wrote all of the text. What? That's not even pulled off our existing website. 481 00:47:06,280 --> 00:47:12,280 No, this was, this was text I have to write all of the text. Everything was created by it. 482 00:47:12,280 --> 00:47:18,440 I didn't do anything other than tell it the name was it. It does make me sound way better than I 483 00:47:18,440 --> 00:47:22,840 really am. I just saw that. So yeah, now, now I understand. That's just a really good example 484 00:47:22,840 --> 00:47:27,320 of its ability to produce junk. Yeah, exactly. I mean, this says that you serve as the chief 485 00:47:27,320 --> 00:47:32,920 marketing for a leading tech company. I do. Artificially intelligent marketing. Currently 486 00:47:32,920 --> 00:47:38,040 a podcast. Soon to be a tech company. We're all tech companies these days. Isn't that right? 487 00:47:38,040 --> 00:47:41,880 Well, I think we have to be. Yeah, that's what I thought. I would play around with the tour 488 00:47:41,880 --> 00:47:50,120 this week. It's, what's interesting actually is I stuck the same prompt into GPT, into chat GPT 489 00:47:50,760 --> 00:47:59,560 with GPT 3.5. Yeah. And it was just like old model. It just went, nope. It said, as a, you know, 490 00:48:00,600 --> 00:48:06,760 AI large language model, I cannot create this, but I can tell you the steps that you would need to do. 491 00:48:06,760 --> 00:48:12,920 Now I know it can create that or it can create web pages because at my presentation tomorrow, 492 00:48:12,920 --> 00:48:21,320 I'm showing an interactive functional game of Noughts and Crosses that chat GPT made for me 493 00:48:22,040 --> 00:48:28,600 and did a pretty good job of. Using the old model. Using the old model. Yeah, correct. And it did 494 00:48:28,600 --> 00:48:32,600 that and I managed to edit it to get a logo and a button added in there with some different 495 00:48:32,600 --> 00:48:41,720 and states on the bottom and stuff. But yeah, this is a level beyond that. This is multiple pages 496 00:48:41,720 --> 00:48:48,360 created. This is design text. It's kind of crazy. Right, Mark, as we said at the beginning, if you're 497 00:48:48,360 --> 00:48:55,800 not playing with these tools, get playing. GPT 4, get playing because, and be creative, start 498 00:48:55,800 --> 00:48:58,840 Googling around, get on the Twitters and figure out who's playing with this stuff. Cause you'll 499 00:48:58,840 --> 00:49:04,920 be inspired by the examples that you see. Not only can it generate text outputs, it can generate 500 00:49:04,920 --> 00:49:12,680 code. It can do sums. I saw in the demo, its ability to reason through someone's tax return. 501 00:49:15,000 --> 00:49:19,880 Get creative. So an example was really smart. Oh yeah. It was mind blowing. Like do my tax return 502 00:49:19,880 --> 00:49:24,920 for me and break out the logic that you needed to go through. And it even like sucked in some important 503 00:49:24,920 --> 00:49:31,720 and ledger changes that were important for a particular time period. And anyway, search that 504 00:49:31,720 --> 00:49:37,880 as well. But yes, play with it, challenge it, challenge yourself. Look for examples online 505 00:49:37,880 --> 00:49:41,560 of where people are doing really out there things with this because you probably do pretty cool 506 00:49:41,560 --> 00:49:44,920 things. I mean, people are using it to create apps at the moment. There's been a lot of that. 507 00:49:44,920 --> 00:49:49,960 Somebody created an app that suggests five new movies to you every day that you should watch. 508 00:49:49,960 --> 00:49:55,080 And they did that in like a half an hour or something crazy. Like if in terms of unleashing 509 00:49:55,080 --> 00:50:00,440 creativity and allowing you to finally come up with ideas for things that you wanted to do, 510 00:50:00,440 --> 00:50:06,120 but lacked technical skills in specific domains, this opens up a lot of doors. We should say mine, 511 00:50:06,760 --> 00:50:12,920 to go from raw HTML and JavaScript to what you then put together. It's not super challenging, 512 00:50:12,920 --> 00:50:19,080 but you need to know what you're doing, right? You can't just like paste HTML into your 513 00:50:19,080 --> 00:50:24,040 web browser bar at the top and now you've got a web page, right? There's still things that you 514 00:50:24,040 --> 00:50:28,680 need to know how to do, but they're not the hard things that quite easily learn. 515 00:50:29,880 --> 00:50:38,360 Yeah. And if you know your way, a little bit of knowledge, you can get a long way. And actually, 516 00:50:38,360 --> 00:50:42,440 if you don't know, you can ask it to explain it to you. That's the thing I think, honestly, 517 00:50:42,440 --> 00:50:47,240 that's the thing that I think is really remarkable because if you come at it completely, 518 00:50:47,240 --> 00:50:51,320 admittedly, I couldn't do it in 21 minutes like I did with this or 23 minutes, whatever it was, 519 00:50:51,320 --> 00:50:57,800 to create this website, but I'm very confident now I could create a Chrome extension from scratch. 520 00:50:57,800 --> 00:51:03,240 If you were to ask me right now to create a Chrome extension, I wouldn't have a clue, 521 00:51:03,240 --> 00:51:07,880 wouldn't know. I'm not a coder, not a developer. I wouldn't know where to start, but I reckon that 522 00:51:07,880 --> 00:51:12,200 within a few hours with GPT-4, I could have a fully functional one up and running. 523 00:51:12,200 --> 00:51:17,560 Well, you know what you've just done there. Martin, will you please create a Chrome extension for us 524 00:51:18,200 --> 00:51:20,360 so that we can show the folks next week? 525 00:51:21,560 --> 00:51:23,640 That's a challenge accepted. 526 00:51:23,640 --> 00:51:25,400 Martin's like crap. But yes. 527 00:51:25,400 --> 00:51:29,160 As if I've not got busy enough with the toddler. 528 00:51:29,160 --> 00:51:35,560 Just don't go watch Davie County. Don't goof yourself anyway. We've already talked about this 529 00:51:35,560 --> 00:51:40,760 before. So there's like three or four hours backing up. Oh, I'm going to have to do this. 530 00:51:40,760 --> 00:51:44,840 Three or four hours backing up. Or take your laptop and just, that would be even better. 531 00:51:44,840 --> 00:51:47,560 I was watching Davie County. They were three-nil down again. 532 00:51:48,360 --> 00:51:52,680 So I created this Chrome extension and accidentally commercialized it over the 533 00:51:52,680 --> 00:51:55,880 next two weeks and made $2 million. That would be pretty cool, wouldn't it? 534 00:51:55,880 --> 00:52:00,360 We are playing Fleetwood Town this weekend. So being three-nil down could happen. 535 00:52:00,360 --> 00:52:02,360 It would be better off listening to Fleetwood Mac. 536 00:52:03,720 --> 00:52:07,800 And with that, I think we'll end the podcast. I think we've probably talked 537 00:52:07,800 --> 00:52:13,080 quite enough nonsense for one week. Thank you so much everybody for joining us again. 538 00:52:13,080 --> 00:52:16,600 We really, really appreciate it. Hopefully you found this valuable. If you did, subscribe. 539 00:52:16,600 --> 00:52:20,760 And if you did share it with your pals in marketing land. We're getting loads of downloads. 540 00:52:20,760 --> 00:52:25,160 We're loving it, but we'd love more. So please do share this far in mind if you think it's valuable. 541 00:52:25,160 --> 00:52:30,440 Other than that, we will see you all next week. Martin, thank you for your time as always. 542 00:52:30,440 --> 00:52:33,160 I'll see you next week, my friend. See you next week. 543 00:52:33,160 --> 00:52:34,200 Cheers. Bye. 544 00:52:34,200 --> 00:52:37,560 Thank you for listening to Artificially Intelligent Marketing. 545 00:52:37,560 --> 00:52:43,240 To stay on top of the latest trends, tips, and tools in the world of marketing AI, 546 00:52:43,240 --> 00:53:04,600 be sure to subscribe. We look forward to seeing you again next week.