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 --> 00:00:22,280 results from your marketing efforts. Now let's join our hosts, Paul Avery and Martin Broadhurst. 4 00:00:22,280 --> 00:00:28,880 Hello everyone. Welcome to a special edition midweek episode of Artificially Intelligent 5 00:00:28,880 --> 00:00:34,920 Marketing. Is it episode 32? Is it episode 31A? No one knows. Let's go with episode 6 00:00:34,920 --> 00:00:39,360 32. The reason that I decided to record a quick episode, because I know Martin can't 7 00:00:39,360 --> 00:00:45,280 join me today, is just to get out a quick bit of info for you about OpenAI's developer 8 00:00:45,280 --> 00:00:51,280 conference yesterday, because they shared really loads of information that I think marketers 9 00:00:51,280 --> 00:00:56,680 will hopefully want to know about. The first thing is that they announced the launch of 10 00:00:56,680 --> 00:01:03,440 GPT-4 Turbo, something that we predicted in the run up to the event. But there was some 11 00:01:03,440 --> 00:01:07,320 additional information in here that we didn't predict that's really quite exciting. So, 12 00:01:07,320 --> 00:01:13,600 for example, one of those is 128k context. So for the listeners of the podcast, you'll 13 00:01:13,600 --> 00:01:19,120 know. One of the reasons that we love Claude is because it's got 100k of context, which 14 00:01:19,120 --> 00:01:24,400 means you can feed it many documents or really large PDFs with hundreds of pages and then 15 00:01:24,400 --> 00:01:30,800 interrogate that content, ask for summaries, etc. Something that you can't do with chat 16 00:01:30,800 --> 00:01:35,800 GPT out of the box because it's only got an 8k context. And even some of the tools do 17 00:01:35,800 --> 00:01:41,840 have access to 32k context for GPT-4. But that's actually quite rare because it's been 18 00:01:41,840 --> 00:01:48,640 quite expensive. So for developers now, they're going to be able to access GPT-4 128k. Expect 19 00:01:48,640 --> 00:01:53,440 to see that baked into some of the tools that you might use. I'm hoping to see it in Magi, 20 00:01:53,440 --> 00:01:59,720 which is one of my favourite tools. And that context is really now very powerful against 21 00:01:59,720 --> 00:02:06,240 the likes of Claude. They've also promised to reduce the price of using GPT-4 for developers. 22 00:02:06,240 --> 00:02:10,320 So hopefully we're going to get access to this without any increases in price for us 23 00:02:10,320 --> 00:02:16,240 as software users. Although even though there's a reduction in price for input and output 24 00:02:16,240 --> 00:02:21,040 tokens, obviously, when we've got that much more context window, there's going to be a 25 00:02:21,040 --> 00:02:25,360 lot more tokens involved. So exactly how much cheaper it turns out for the providers of 26 00:02:25,360 --> 00:02:31,960 tools and therefore us as marketers that subscribe to those tools, we will wait to see. It's 27 00:02:31,960 --> 00:02:40,600 also likely, in fact, a dead cert that GPT-4 Turbo will become the underlying model for 28 00:02:40,600 --> 00:02:47,400 chat GPT-plus users. So we're going to have that massive context window to be able to 29 00:02:47,400 --> 00:02:55,000 work with when we're using GPT-4 within chat GPT-plus. So again, if you're used to pasting 30 00:02:55,000 --> 00:03:01,160 in materials into chat GPT and getting some sort of error message about the information 31 00:03:01,160 --> 00:03:05,800 you've provided is too long and breaks the context length, that is about to disappear 32 00:03:05,800 --> 00:03:12,680 with the introduction of 128k context windows. Something that will be interesting as we see 33 00:03:12,680 --> 00:03:21,160 this GPT-4 Turbo 128 roll out is going to be how good is the model at delivering the 34 00:03:21,160 --> 00:03:24,920 outputs that we've become used to. So you might remember some months ago, there was 35 00:03:24,920 --> 00:03:31,680 quite a bit of debate about whether GPT-4 was getting less smart over time and producing 36 00:03:31,680 --> 00:03:39,060 worse outputs. People were comparing it against GPT-3 Turbo or 3.5 Turbo, which they actually 37 00:03:39,060 --> 00:03:44,800 felt was better than the original GPT-3.5. So then the question remains is in our hands, 38 00:03:44,800 --> 00:03:51,720 how accurate and useful will GPT-4 Turbo be? Will it perform better on a number of benchmarking 39 00:03:51,720 --> 00:03:56,880 tests than the original GPT-4? Will it deliver on a number of the marketing applications 40 00:03:56,880 --> 00:04:02,360 that people have started using chat GPT-plus for? Will it be worse or better? We don't 41 00:04:02,360 --> 00:04:07,560 know until we get the opportunity to play with it. My gut says it will probably be slightly 42 00:04:07,560 --> 00:04:13,520 better, but certainly the 128k context window could really be a game changer for a lot of 43 00:04:13,520 --> 00:04:19,080 people's applications. Then there's a couple of things that are probably more important 44 00:04:19,080 --> 00:04:25,120 for developers, but that could be important for us as users, which is OpenAI has changed 45 00:04:25,120 --> 00:04:30,680 some of the way that developers will interact with the model. So now there are function 46 00:04:30,680 --> 00:04:35,320 calling updates whereby a developer can now call multiple functions. So this is going 47 00:04:35,320 --> 00:04:40,960 to make it much easier for them to create systematic agent-like experiences within the 48 00:04:40,960 --> 00:04:49,120 API. There's also a new JSON mode, which is going to make it easier for developers again 49 00:04:49,120 --> 00:04:55,360 to get the types of information out of the API that they would want to make it easier 50 00:04:55,360 --> 00:05:05,160 to move information between different tools. The other thing to note is that GPT-3.5 Turbo 51 00:05:05,160 --> 00:05:09,120 there's going to be a new version of that that has 16k context window by default. So 52 00:05:09,120 --> 00:05:13,200 those of you that are using the free version of GPT, you're now going to have access to 53 00:05:13,200 --> 00:05:18,800 this slightly improved GPT-3.5 and for developers whose tools rely on it, it's going to be even 54 00:05:18,800 --> 00:05:26,640 cheaper than it was before. Another big news item announced by the OpenAI team was the 55 00:05:26,640 --> 00:05:32,560 assistance API. So this is the first step towards helping developers build agent-like 56 00:05:32,560 --> 00:05:37,360 experiences within their own applications. An assistant is kind of like a purpose-built 57 00:05:37,360 --> 00:05:44,480 AI that has specific instructions and leverages extra knowledge and can call models and tools 58 00:05:44,480 --> 00:05:49,360 to perform different tasks. And I think what's really interesting about this is for developers, 59 00:05:49,360 --> 00:05:56,940 this is going to allow them to build assistance on top of chat GPT that can also leverage 60 00:05:56,940 --> 00:06:05,160 tools like Code Interpreter and DORLY 3 and GPT-4 Vision to create really interesting 61 00:06:05,160 --> 00:06:11,120 applications, the likes of which we just haven't been able to build so far, which again is 62 00:06:11,120 --> 00:06:16,320 going to translate into the different types of tools that marketers get to play with and 63 00:06:16,320 --> 00:06:23,960 get to access. OpenAI have also built in a new retrieval-based system. This has been 64 00:06:23,960 --> 00:06:27,560 one of the limitations that's been really quite hard and made it a bit more challenging 65 00:06:27,560 --> 00:06:34,280 to develop on top of OpenAI's platform is having the system remember the information 66 00:06:34,280 --> 00:06:39,280 that the tool that you're using or you as a user didn't put it into that tool that 67 00:06:39,280 --> 00:06:44,200 had been sent to OpenAI. And then all of that had to keep being packaged in the information 68 00:06:44,200 --> 00:06:50,440 that was sent in the context window. So expect your tools to remember much more about instructions 69 00:06:50,440 --> 00:06:58,400 that you've given them in the past, et cetera. And I think why is this really cool? I think 70 00:06:58,400 --> 00:07:04,840 it's nicely summed up by the way OpenAI summarized it on the blog post on their website is that 71 00:07:04,840 --> 00:07:09,840 there's so many potential use cases for this. And they give examples like a data analysis 72 00:07:09,840 --> 00:07:16,720 app, a coding assistant, an AI powered vacation planner, a voice controlled DJ, a smart visual 73 00:07:16,720 --> 00:07:24,320 canvas. It really now is down to the developer's imagination to really be able to call on so 74 00:07:24,320 --> 00:07:32,260 many of those different aspects of how GPT and OpenAI's tools work. If you want to play 75 00:07:32,260 --> 00:07:38,200 with Assistant's API and you've got access to the playground, it's currently in beta. 76 00:07:38,200 --> 00:07:42,320 So you can go and actually have a little play with what this looks like. It's also worth 77 00:07:42,320 --> 00:07:51,640 noting that I don't have access to GPT-4 Turbo in my chat GPT instance yet. And it's currently 78 00:07:51,640 --> 00:07:56,100 in preview access as well, but they expect to roll it out more widely before the end 79 00:07:56,100 --> 00:08:01,920 of the year. The other thing that's coming to GPT-4 that we also talked a bit about on 80 00:08:01,920 --> 00:08:08,480 the podcast previously is you no longer have to in GPT-4 Plus, chat GPT Plus, I should 81 00:08:08,480 --> 00:08:15,160 say you no longer have to select the modality that you want to use. So up until now, if 82 00:08:15,160 --> 00:08:19,600 you wanted to use code interpreter now called advanced data analysis, you had to go ahead 83 00:08:19,600 --> 00:08:23,740 and select that in a dropdown. And then you could interact with that tool. If you wanted 84 00:08:23,740 --> 00:08:28,840 to create images, you had to select Dali 3 in the dropdown. And it was just a bit of 85 00:08:28,840 --> 00:08:33,720 a pain in the bottom to use, to be honest. Now that's all been consolidated. So there's 86 00:08:33,720 --> 00:08:39,560 no dropdowns anymore. You just make a request of chat GPT like you would normally. And then 87 00:08:39,560 --> 00:08:43,400 it figures out what tools it needs to draw on in the back end to deliver on your request. 88 00:08:43,400 --> 00:08:47,560 So if you ask it to create an image, it will know it needs to use Dali 3. If you ask it 89 00:08:47,560 --> 00:08:52,440 to perform some sort of data analysis or some action that requires the ability to create 90 00:08:52,440 --> 00:08:57,600 Python code and run that code, then it will use advanced data analysis. So that's going 91 00:08:57,600 --> 00:09:02,920 to make it all easier for us to be able to just use the tool, which is quite nice. The 92 00:09:02,920 --> 00:09:08,520 other thing is all this stuff, as I mentioned, is being made available via the API. So you 93 00:09:08,520 --> 00:09:15,360 may use a tool now that is a wrapper or a skin of chat GPT like we do with Magi, but 94 00:09:15,360 --> 00:09:20,120 expect to be able to do stuff beyond just text generation, because we have access to 95 00:09:20,120 --> 00:09:24,560 the Dali 3 API, for example, you're going to be able to create images in your application 96 00:09:24,560 --> 00:09:31,600 of choice, providing they open up access to Dali 3 within that tool. And it will create 97 00:09:31,600 --> 00:09:36,720 images for you using Dali 3. So a fairly obvious use case of this, I can imagine will spring 98 00:09:36,720 --> 00:09:43,560 up fairly quickly is if you use a tool to write like writer or Jasper or Magi, you'll 99 00:09:43,560 --> 00:09:49,440 also be able to create images for those pieces, let's say blog posts using Dali 3. And one 100 00:09:49,440 --> 00:09:53,760 assumes they'll be able to use the text of whatever thing you've written, social posts, 101 00:09:53,760 --> 00:09:58,800 whatever, to influence the prompt that then creates the image alongside it. So an example 102 00:09:58,800 --> 00:10:07,240 would be write me a blog post about the five most exciting things that OpenAI have released 103 00:10:07,240 --> 00:10:11,000 and create me an image to go with that blog post should be something that is all very 104 00:10:11,000 --> 00:10:16,000 easy to do, certainly within chat GPT, but also other tools. The other thing that they've 105 00:10:16,000 --> 00:10:22,280 made available via API is text to speech. So many of you listen to the podcast will know 106 00:10:22,280 --> 00:10:28,440 OpenAI has a speech to text engine called whisper, so for transcription. But if you 107 00:10:28,440 --> 00:10:34,360 have the mobile app of OpenAI chat GPT on your phone, you can actually speak with it 108 00:10:34,360 --> 00:10:39,360 and it will transcribe what you say, but it will then speak back to you because chat GPT 109 00:10:39,360 --> 00:10:43,960 now has this text to speech function. And it's kind of interesting and fun to go play 110 00:10:43,960 --> 00:10:48,160 with. You're basically just having a conversation with the computer like they might do in Star 111 00:10:48,160 --> 00:10:53,680 Trek. Well, now that capability is being rolled out via the API to developers. So expect to 112 00:10:53,680 --> 00:10:59,520 see more of this speech to text and text to speech type interactions popping up in lots 113 00:10:59,520 --> 00:11:04,760 of different tools, because they're basically democratising and making it much easier for 114 00:11:04,760 --> 00:11:09,120 developers to create tools that are software applications that you speak to and they speak 115 00:11:09,120 --> 00:11:14,520 back to you. So again, that's pretty exciting and could change fundamentally how marketers 116 00:11:14,520 --> 00:11:18,920 interact with lots of different software to get what they want, especially because those 117 00:11:18,920 --> 00:11:25,240 tools are going to have access to advanced data analysis and other aspects of the OpenAI 118 00:11:25,240 --> 00:11:31,080 back end. You can imagine sitting in front of your favourite dashboarding program or 119 00:11:31,080 --> 00:11:39,400 your marketing software suite and asking in natural language to see a graph of contacts 120 00:11:39,400 --> 00:11:45,440 generated by a particular source that say organic search over time, and then ask the 121 00:11:45,440 --> 00:11:49,960 tool to explain the graph to you in speech. So to speak to you basically. So that could 122 00:11:49,960 --> 00:11:56,080 be pretty cool as well. There's a bunch of other stuff there that's probably more important 123 00:11:56,080 --> 00:11:59,680 for developers than it is for marketers at this point, but it's worth going and checking 124 00:11:59,680 --> 00:12:07,400 out the blog post on OpenAI's website. The other thing that OpenAI mentioned is their 125 00:12:07,400 --> 00:12:11,940 new copyright shield. So as you see on their website, OpenAI is committed to protecting 126 00:12:11,940 --> 00:12:16,640 our customers with built-in copyright safeguards in our systems. Today, we're going one step 127 00:12:16,640 --> 00:12:22,140 further and introducing copyright shield. We will now step in and defend our customers 128 00:12:22,140 --> 00:12:27,540 and pay the costs incurred if you face legal claims around copyright infringement. This 129 00:12:27,540 --> 00:12:31,960 applies to generally available features of chat GPT enterprise and our developer platform. 130 00:12:31,960 --> 00:12:36,040 And so this is very much in line with the steps taken by Microsoft and Google so far 131 00:12:36,040 --> 00:12:41,720 to give users of large language models confidence that they're not going to get in trouble 132 00:12:41,720 --> 00:12:48,280 around copyright if they use the outputs of those tools by effectively indemnifying and 133 00:12:48,280 --> 00:12:53,780 protecting them against any legal claims brought against them. How that will actually play 134 00:12:53,780 --> 00:12:57,940 out in the real world is not clear at the moment, but I think all of the providers are 135 00:12:57,940 --> 00:13:02,760 working really, really hard to give especially large enterprises confidence that they can 136 00:13:02,760 --> 00:13:07,720 use these tools without worrying that they're opening themselves up to some sort of copyright 137 00:13:07,720 --> 00:13:13,680 infringement claim and lawsuit in the future. OpenAI are also going to be releasing their 138 00:13:13,680 --> 00:13:19,640 next version of the Whisper model that I mentioned earlier. So this automatic speech to text 139 00:13:19,640 --> 00:13:24,400 model. So now we're up to Whisper version three. Martin and I were talking about this 140 00:13:24,400 --> 00:13:28,600 on the WhatsApps and we've been really impressed with Whisper version two and can't really 141 00:13:28,600 --> 00:13:33,780 think of much that we'd like Whisper version three to be able to do that's not already 142 00:13:33,780 --> 00:13:38,160 covered by the current version of Whisper. So it'll be interesting to see exactly what 143 00:13:38,160 --> 00:13:43,760 improvements this provides. Those of you that have seen Martin speak or have listened to 144 00:13:43,760 --> 00:13:48,720 the podcast will know he does a really fun bit in a lot of the talks that he gives where 145 00:13:48,720 --> 00:13:55,360 he uses Whisper to automatically transcribe a recording that he does live while he's talking 146 00:13:55,360 --> 00:14:00,160 where he'll interview and ask questions of a few audience members and then has an automated 147 00:14:00,160 --> 00:14:06,160 workflow through Zapier that automatically creates a summary blog of his talk and includes 148 00:14:06,160 --> 00:14:11,520 the quotes of the people in the room published in real time. So at the end of his talk there 149 00:14:11,520 --> 00:14:17,800 they are in the blog post on Medium. Whether or not this new Whisper version will make 150 00:14:17,800 --> 00:14:22,400 that even better is hard to see but it's nice to see that they're continuing to improve 151 00:14:22,400 --> 00:14:28,400 the tools. So you'd be forgiven for thinking that that would be it because there's certainly 152 00:14:28,400 --> 00:14:33,280 a lot of things for us to chew over in there as marketers in terms of the increasing power 153 00:14:33,280 --> 00:14:41,320 and ease of use as it comes to ChatGPT and also the sheer range of options now for developers 154 00:14:41,320 --> 00:14:45,760 to use a lot of the things that we love in ChatGPT in their own applications including 155 00:14:45,760 --> 00:14:52,760 image creation and text to speech creation and advanced data analysis and the ability 156 00:14:52,760 --> 00:14:57,080 to pull information in from lots of other tools which we should probably focus on for 157 00:14:57,080 --> 00:15:03,880 just a second actually because that ability to use data from lots of different tools could 158 00:15:03,880 --> 00:15:09,240 be a real game changer for developers because they can build on top of your existing stack. 159 00:15:09,240 --> 00:15:16,320 So integrating data from your CRM, from your marketing automation platform, from Google 160 00:15:16,320 --> 00:15:21,480 ads, from Google search console and give you something that you can interrogate with a 161 00:15:21,480 --> 00:15:29,080 large language model like ChatGPT, let's call it, I don't know, Marketing Analytics GPT, 162 00:15:29,080 --> 00:15:33,680 that you can then ask questions of that either by speaking to it like I'm speaking to you 163 00:15:33,680 --> 00:15:38,720 now or by typing it in. That is all going to be really cool but they've gone a step 164 00:15:38,720 --> 00:15:44,000 further than that and they've made it so that this type of power and functionality is not 165 00:15:44,000 --> 00:15:48,920 going to be just available to developers and people who know how to code. They have introduced 166 00:15:48,920 --> 00:15:56,600 a new thing that they call GPTs. In essence, users of ChatGPT can now create custom versions 167 00:15:56,600 --> 00:16:03,160 of ChatGPT that combine instructions, extra knowledge and any combination of skills to 168 00:16:03,160 --> 00:16:10,680 deliver a customized chat capability. So in the examples they give on their website, they've 169 00:16:10,680 --> 00:16:16,680 got a creative writing coach, they've got a tech advisor, they've got a puppy trainer. 170 00:16:16,680 --> 00:16:23,000 So what this does is it's going to make it possible for anyone to create their own customized 171 00:16:23,000 --> 00:16:30,780 ChatGPT bot trained on information provided by the user. Why is this important and why 172 00:16:30,780 --> 00:16:36,800 is this different to what they've provided to developers? And in essence, this is because 173 00:16:36,800 --> 00:16:44,680 there is going to be a GPT wizard. So you don't need to code. You just open up the wizard 174 00:16:44,680 --> 00:16:51,120 and then you can build your own custom chat bot. So in essence, what this means is by 175 00:16:51,120 --> 00:16:58,580 providing information sources for the bot to use, so by uploading a lot of your company's 176 00:16:58,580 --> 00:17:04,360 documents for example, you could create a customer service bot. The great thing about 177 00:17:04,360 --> 00:17:09,520 the tool is it's built using natural language. So in essence, you have a conversation with 178 00:17:09,520 --> 00:17:16,040 ChatGPT about what you want your custom GPT to do and it will help you build it. It will 179 00:17:16,040 --> 00:17:20,640 build rules in the back end in terms of what questions that GPT will answer and what questions 180 00:17:20,640 --> 00:17:26,000 it won't and how it will answer those questions. In the live demo that they gave during the 181 00:17:26,000 --> 00:17:31,760 developer conference yesterday, Sam Altman showed us an example where he took transcripts 182 00:17:31,760 --> 00:17:36,800 from some previous talks that he'd done during his time at Y Combinator giving advice to 183 00:17:36,800 --> 00:17:44,960 startups to in essence create a startup advice bot based on previous advice that he'd given. 184 00:17:44,960 --> 00:17:50,200 So we've talked on the podcast previously about how organizations should really be looking 185 00:17:50,200 --> 00:17:55,240 at what data they have internally, what information that they've captured, but also what information 186 00:17:55,240 --> 00:18:00,880 could they capture if they more proactively look to record more internal conversations 187 00:18:00,880 --> 00:18:07,120 and calls, record customer service calls, record sales calls and then leverage that 188 00:18:07,120 --> 00:18:13,560 information through something like a custom bot to create customer service bots, internal 189 00:18:13,560 --> 00:18:19,840 knowledge based bots and all of the above. Again, I think the imagination of people is 190 00:18:19,840 --> 00:18:24,800 the limitation here somewhat because I think the tool really offers quite a broad canvas 191 00:18:24,800 --> 00:18:30,800 for people to build quite interesting stuff and OpenAI is certainly betting on this themselves 192 00:18:30,800 --> 00:18:36,520 because they believe that the best GPTs will be invented by the community and so they're 193 00:18:36,520 --> 00:18:41,360 going to make it really easy for anyone to build one and they're going to create a GPT 194 00:18:41,360 --> 00:18:48,080 store where you can share your GPTs publicly. So the GPT store is going to be searchable 195 00:18:48,080 --> 00:18:52,520 and then if a particular GPT becomes popular, then there's going to be like a leaderboard, 196 00:18:52,520 --> 00:19:00,280 a little bit like popular apps on like Google Play Store or the Apple Store. And if people 197 00:19:00,280 --> 00:19:05,080 use your bot, you're going to be able to earn money based on how many people are using it, 198 00:19:05,080 --> 00:19:10,520 which I think is really, really cool. They're also developing them with all of the sort 199 00:19:10,520 --> 00:19:15,520 of safety guards in mind that they've used to develop all the rest of their tools. And 200 00:19:15,520 --> 00:19:19,880 if you are an enterprise customers, you've got chat GPT enterprise, you're going to be 201 00:19:19,880 --> 00:19:26,260 able to create bots that can only be used within your organization. So this is kind 202 00:19:26,260 --> 00:19:31,520 of akin to what we've heard some of the consulting companies like Bain have been doing where 203 00:19:31,520 --> 00:19:37,120 they've created an internal knowledge GPT or bot for their teams to use, trained on 204 00:19:37,120 --> 00:19:42,440 all of their proprietary information. So if you are a chat GPT enterprise customer, you're 205 00:19:42,440 --> 00:19:47,040 going to be able to do that. One example I can think of is if you have a customer service 206 00:19:47,040 --> 00:19:52,960 team, you could empower them to find it easier to answer customer queries, not by searching 207 00:19:52,960 --> 00:19:58,400 through knowledge bases, but to actually just feed the queries that customers provide into 208 00:19:58,400 --> 00:20:04,400 those GPTs to help answer the questions by training those GPTs on all of your documentation, 209 00:20:04,400 --> 00:20:12,080 technical manuals, protocols, all of those other things. So there you have it. Some well 210 00:20:12,080 --> 00:20:17,900 known companies have already built these custom GPTs. So far, the one I'm really keen to get 211 00:20:17,900 --> 00:20:26,360 my hands on is the Zapier one. So you can use the Zapier GPT to basically help you plan 212 00:20:26,360 --> 00:20:30,720 automation. So you kind of ask it, hey, I want to achieve this particular thing. How 213 00:20:30,720 --> 00:20:35,880 could I do this in Zapier with automation? And then you can use that the Zapier GPT bot 214 00:20:35,880 --> 00:20:42,320 to help you create those automations. There's also going to be a Canva GPT, where in essence, 215 00:20:42,320 --> 00:20:47,440 you describe the type of thing that you're trying to make. And it gives you initial examples 216 00:20:47,440 --> 00:20:52,160 to work from. And then you have a conversation with Canva GPT going back and forth, refining 217 00:20:52,160 --> 00:20:56,640 the design. And when you're ready, you see something you like, you click on it, and it 218 00:20:56,640 --> 00:21:00,640 takes you into Canva where you can then make any fine tuning edits that you want to do 219 00:21:00,640 --> 00:21:06,200 manually within the Canva interface. So I've been trying, I haven't got access to these 220 00:21:06,200 --> 00:21:10,220 yet. As soon as we do, we'll let you know what our experience has been on one of the 221 00:21:10,220 --> 00:21:17,200 upcoming episodes of our podcast. So as a marketer, obviously Canva and Zapier's GPTs 222 00:21:17,200 --> 00:21:21,280 are going to be useful and expect to see similar tools popping up all over the place for all 223 00:21:21,280 --> 00:21:25,800 the other software tools that you use. So I think that's the first thing. The second 224 00:21:25,800 --> 00:21:31,240 thing is, how can you create GPTs to support your internal organizations? So how does this 225 00:21:31,240 --> 00:21:37,960 change internal comms and how internal folks access information in your organization? And 226 00:21:37,960 --> 00:21:43,880 also how could you create custom GPTs to better serve your customers, leveraging your internal 227 00:21:43,880 --> 00:21:49,760 proprietary information and data to have a GPT customized to the thought, leadership, 228 00:21:49,760 --> 00:21:54,080 knowledge, et cetera, internal within your organization? So this is going to be really, 229 00:21:54,080 --> 00:22:00,680 really interesting to see how companies deploy this almost as products, but certainly as 230 00:22:00,680 --> 00:22:05,280 freemium products as part of their marketing approach to help bring customers in and get 231 00:22:05,280 --> 00:22:11,160 them wedded to the brand. Really, really interesting. If you want to know the details, go and check 232 00:22:11,160 --> 00:22:16,400 out those two blog posts on the OpenAI website. We're going to be fascinated to see how this 233 00:22:16,400 --> 00:22:21,400 plays out and how they roll this out to users and frantically hitting the refresh button 234 00:22:21,400 --> 00:22:27,920 on our chat GPT plus instances to see when we get access to some of these new capabilities. 235 00:22:27,920 --> 00:22:31,560 So I hope you enjoyed that just a little quick interlude because there was just too much 236 00:22:31,560 --> 00:22:35,560 good stuff in there not to share with you all. We'll be back into our regular scheduled 237 00:22:35,560 --> 00:22:41,540 programming with our episode next week. Other than that, I look forward to speaking with 238 00:22:41,540 --> 00:22:46,000 Martin then and to sharing the latest tips and tricks that marketers need to know in 239 00:22:46,000 --> 00:22:51,360 the world of AI to help them get better results. Until next time. Thanks a lot. Bye. 240 00:22:51,360 --> 00:22:57,200 Thank you for listening to Artificially Intelligent Marketing. To stay on top of the latest trends, 241 00:22:57,200 --> 00:23:03,600 tips and tools in the world of marketing AI, be sure to subscribe. We look forward to seeing 242 00:23:03,600 --> 00:23:06,200 you again next week.