Paul Avery [00:00:04]:

Welcome to Artificially Intelligent Marketing, a weekly podcast where we stay on top of the latest trends, tips and tools in the world of marketing AI helping you get the best results from your marketing efforts. Now let's join our hosts, paul Avery and Martin Broadhurst. Hello and welcome to Of Artificially Intelligent Marketing. I'm joined, as always, by my good friend Martin Broadhurst. Martin, how are you?

Martin Broadhurst [00:00:34]:

I'm very good. I've been doing some fun little AI experiments with some clients this week, helping them to document some standard operating procedures using content that they already had in existence. A job that was going to take them quite a bit of time to manually do. But just because I was in their offices as part of my retained work with them, overheard a conversation and said, oh, I think we can streamline that process a little bit. And we did. We jumped into it and half an hour later we had a bunch of documents written that would otherwise have taken them several hours.

Paul Avery [00:01:14]:

Amazing stuff. So you've been waving the AI wand this week. Sounds really good. I've been doing lots of interesting things, but among all of those I had a great conversation with Louis Pereira this week from Audio Pen. So that's our interview this week. Audio Pen is a tool that Martin and I love, which allows you to leave rambling voice notes on your phone that then are automatically summarized and turned into text using GPT Four. So we hear from Lewis a bit later about what it took to build that tool and also different ways that you can use it and just his journey using low code and no code software solutions to build things, which is a fascinating conversation that we will get into later. But as always, we're going to start with what is going on in the world of AI this week? That marketers to know. Martin, kick us off.

Martin Broadhurst [00:02:09]:

So this week there was a new tool dropped by Google called Textfx, and it's specifically a tool for creative writing. It's an AI assistant for creative writing that it's not a full tool that's going to replace the likes of Chat GPT, but it's more of a toolbox with a bunch of specific little tools, each helping you to unlock a new angle or find an interesting way of talking about a theme. So it's particularly interesting for artists creatives. In fact, they launched a tool with some wrappers. It was Lupe Fiasco and did a promo video showing how wordsmiths could use the tool. So the tools that it's got inside it, there's ten of them in total. They've got things like Unfold, which slots a word into other words and phrases. So you might give it the word farm and it will say farm yard, but it will find lots of these different word combinations. You've got Similes, which will find metaphors and similes for you. A tool called Explode where it will break words out into similar sounding phrases, and then Chain, which builds a chain of semantically related items, which as they go through, they get progressively further. The start and the finish words seem quite far away from each other, but there's a chain of steps that take you there. I've been playing with the tool. I tried the tool, and they're called Point of View, which evaluates a topic through different points of view. And I gave it a phrase based on one of my clients, which is temporary warehouses. So a nice industrial client of mine, that's one of their main products. And the angle it gave me was, temporary warehouses are the secret weapon of ecommerce businesses. And this was interesting because ecommerce distribution is a market that they go after. But we hadn't really approached it like that. And when we discussed it internally, that became, okay, there's something there. There's more we can lean on here for some blog content. So just in terms of sparking a discussion, that was a really useful exercise. There's another tool called Fuse, and this will bring two concepts together and it will try to find connections. So I put in Easter and Shoes, and it gave me the angle of New Beginnings, which I quite liked. And then what was the other one? I tried oh, yeah, Farm and Halloween, which gave the connection of harvest. Okay, fine, I'll take that pumpkins and what have you. There's a visual theme at least, but it doesn't always get it right. I tried Farm Plus Dinosaurs based on a project that a client is doing, and it said both farms and dinosaurs are relics of the past. And I thought the $4 trillion global agriculture business might feel like it hasn't quite characterized their sector in the way that they would appreciate.

Paul Avery [00:05:32]:

And you know, what drove that? As far as AI is concerned, it's going to replace humans and it doesn't need to eat food. So as far as it's concerned, farms are a thing of the past. So we just got a little insight there into Skynet future plans.

Martin Broadhurst [00:05:46]:

Yeah, it's thrown it forward and gone. Yeah, no, don't need them anymore.

Paul Avery [00:05:53]:

Clems I should have said anything. Ignore that. Ignore that. Actually, farms are brilliant. I've always been a big fan of farms. You should continue to invest in farms. Polly love farms.

Martin Broadhurst [00:06:04]:

Total necessity. So, yeah, Google's Text Effects isn't going to replace your creative director anytime soon, but it is going to be a good tool for brainstorming on your own or with a team just throwing new ideas into the mix. There's a lot of bad ideas in there. The more you play with it, the more you realize there's a lot of junk. But it does it at scale and speed, so you can filter through the bad ideas quickly. If you want to give it a try, head over to Textfx withgooble.com sounds good.

Paul Avery [00:06:35]:

I think I'll have a play with that. Martin I'm always a fan of these tools for sparking additional creative ideas. We've talked about it previously on the podcast and sounds like there could be some fun campaign planning angles to a number of those different tools that you mentioned. When using AI for creative purposes though, we have to keep in mind our next story. So this week, a US federal judge dismissed a lawsuit attempting to secure copyright protection for art that was created solely using AI. And the judge's ruling reinforced the longestablished principle that copyright law requires human involvement, something we've talked about on the podcast previously. Ultimately, the verdict referenced several precedents highlighting the necessity for human creativity and decision making, a crucial distinction as the lines between AI and human creation continue to blur. So for us as marketers, I think this ruling Martin, continues to remind us that AI generated content may not be protected by copyright laws. So if we're using AI produced content in our marketing, it might be that others can freely use it or replicate it to use it as they see fit, potentially diluting the uniqueness of our brand and our campaigns. Also is raising questions about who owns the rights to AI generated content and how can it be used legally and ethically. So, as exciting as mid journey and stability AI and a number of these other tools are, it continues to be a very gray and muddy area as to how well protected we are by copyright. When we produce content using AI as part of our marketing campaigns, people need.

Martin Broadhurst [00:08:17]:

To start really thinking about how they document the creation process, because the ruling makes very clear that human input is really important. So if you're using AI as part of a wider process to get to an output, you will likely be able to copyright the end result, but something that is wholly created by the computer, as it was in this case, Baylor's example, it will be rejected. So I think it's important to document which bits are human inputs. And basically, if you can say, look, the majority of the work that we created here was done by humans and we've used AI in these specific areas, then I think you'll be fine.

Paul Avery [00:09:03]:

Yeah, I think that's really good advice and maybe advice to take into our next story about GoDaddy's work with videos. Mine? Yeah.

Martin Broadhurst [00:09:11]:

This is very much a story aimed more at small businesses, solo entrepreneurs and people looking to create content for their businesses where they're not necessarily a skilled videographer or graphic designer. GoDaddy, the web hosting platform and domain management platform, has launched GoDaddy Studio, which is a tool to help people create content using AI and templates. It's designed to basically help people capitalize on social media networks where video content is prioritized by the algorithms. They've explicitly said this in their press release. So you can see what they're doing here. They're trying to put a suite of tools together that enable entrepreneurs who aren't talented creators to create videos easily and it is quite fun. I've downloaded it, I've had a play with it and it has some good outputs. Some of the features in the app include things like background removal, which we've seen through stable diffusion, and we can do this in Canva or on Adobe Firefly. It also has an instant video feature, which is pretty neat. You can put in some photos, a little bit of video footage that you've taken and it will then create a little video. What was quite interesting with that particular feature is, without writing anything, I just put in some images. In fact, I took some video of my boy on his little balance bike riding around outside our house, and I loaded that into the app, created the video, and it cut it quite nicely, but it put a text. Slogan on the start of it and gave me suggestions for ten or 15 other slogans over it, all of which were related to exploring the world on your bike. So it understand and interpret what the video was about, which was pretty neat. I haven't seen anything doing that before. On the whole, I would describe it as being kind of Canva light it's 999 a month or 89 99 a year. That's British pounds.

Paul Avery [00:11:35]:

It does a pretty good job.

Martin Broadhurst [00:11:36]:

But I would say if you're looking for video content with AI editing and content generation, there are other tools out there that personally, I think they do it slightly better. They're desktop based. Things like PICTORY, AI or Wave Video I think do it better and have more comprehensive templates. However, they don't have mobile apps and this is an iOS only app at the moment. So if you're a content creator or looking to create content for your business, not a bad place to go.

Paul Avery [00:12:13]:

Interesting. I think I'll have to go and have a play with that of mine. Thank you for sharing it with us. We are shifting gears slightly for our next story, which is a recent report from LinkedIn which reveals that AI centric terms like chat, GPT and prompt engineer are now 15 times more prevalent in user profiles. At least they were in June compared to January. So it's quite interesting because there's like a rising tide here of people positioning themselves as AI enabled talent. But it does sort of ask the question like how valuable are these skills and how valuable is it, for example, to be an AI engineer? I would argue very valuable versus a prompt engineer, perhaps less so. I think when you look at and go a bit deeper into LinkedIn's report, it becomes even more interesting because as much as that was going on, on the one hand, on the other side, the hottest in demand skills, according to the research, were not actually AI centric at all. So when polling people, the in demand skills are actually flexibility. Where demand surged by 158%, professional ethics, which rose by 120%, social perceptiveness, which increased by 118%, and self management, which had an 83% height. So overwhelmingly, top tier execs are championing people skills as being indispensable. And there was even some stats in there from the report that said that 92% of top tier execs really are looking for those people skills. So you got a whole bunch of people adding in. I'm a prompt engineer at the same time as execs are like, wow, I really need people with uniquely human skills and reliability. So what does this mean for us as marketers? Well, while people rush to make themselves seem like AI experts after they've played with Chat GPT prompts for a day worth remembering that us as marketers, our true skill set will always be much more than this, right? It's our ability to think strategically. It's our experience working on marketing programs and knowing what works well and what doesn't and what gets resolved. It's our ability to adapt and actually leverage these tools as part of our wider marketing experience, knowledge and skill set. So, sure, absolutely learn to get to grips with these tools. We've talked about it a lot on the podcast in the past, but it's combining them with all the other skills that you've developed over the years as a marketer that will create the true magic, not just positioning yourself as someone who knows how to get half decent results out of Chat GPT.

Martin Broadhurst [00:14:47]:

That very much echoes another report that came out this week from IBM. IBM spoke with 3000 executives and 21,000 workers to understand AI's footprint in our workplaces, and they found that about 40% of the global workforce might need to reskill within three years. Thanks to AI's expanding role in the workforce. 87% of the executives thought that AI isn't about replacing us, but about enhancing our roles. The most interesting point of this report very much aligned to the LinkedIn survey findings. It basically said that human skills, people skills, are the most important thing in the workforce right now. This was shown with one particular graph that said that in 2016, the most in demand skill was proficiency with Stem. So science, technology, engineering, maths. Now that was the least prioritized skill with skills like communication, time management, teamworking being much more in demand. The report goes on to say that AI won't replace people, but people who use AI will replace people who don't. Something that we've heard in various different guises from various sources in recent years. So, yeah, very much reiterating what LinkedIn says people skills are going to be very important. Those soft skills, those abilities to the ability to interact with colleagues, to self manage time management, things like that, they're going to be very important. But to reiterate AI is going to be important. So learn these tools and introduce them into your workflows.

Paul Avery [00:16:51]:

In our next story, Meta has announced Seamless M four T, which they claim is the first all in one multilingual, multimodal AI designed for translation and transcription. It supports an impressive range of functions from speech to text, speech to speech, text to speech and text to text. Try saying that faster than that and you might fucking fall over. And it can do this across up to 100 languages. If you visit the press release, the blog post, et cetera, that announces the tool, there's lots of videos showing how it works. And it's really rather impressive in its ability to take spoken audio and then translate it into spoken audio, but in a different language. And all those text and speech variations that I mentioned, it's not available yet. So it's more about, hey, this is what we developed and this is what's coming down the line. But what this could mean for marketers is that in our globalized business landscape that we operate in, being able to communicate effectively across languages is crucial. And with tools like this, we have the potential to create marketing materials in a range of languages, but all at the click of a button. It could also make it easier for businesses to collaborate, even if they don't share the same language, by ushering in like an era of a general translator, like we see in Star Trek. Because one of the things that they claim is that the lag between the input and the output is actually quite low. So you could imagine speaking in English automatically being translated into the language of choice with the person that you're trying to converse with. And if you were doing this with earbuds in, it's almost like you'd only be hearing them in your language when they were speaking to you and vice versa. And so that could be pretty powerful. Might change schooling a bit. If we end up maybe one day we won't teach languages anymore. I think that would be a real shame if that ever became the case. But, yeah, a step closer to a universal general translator, Martin, who doesn't want that?

Martin Broadhurst [00:18:54]:

The stuff of science fiction coming to real life. That is really exciting. When I saw that story, I thought, yeah, I'm looking forward to that. It's that reduction in the latency, being able to do it at speed, this almost real time translation. That's the bit where I think where I read the research thought, oh, yeah, cool.

Paul Avery [00:19:14]:

Well, I was in Istanbul recently, as you know, and the family we were staying with, most of them couldn't speak English, so we were trying to communicate through Google's current Translate app, which actually did a reasonably good job, especially when we'd lubricated ourselves with a bit of alcohol to remove the sort of social inhibitions of playing with something like that. But it was slow. So I think the key thing here is the ability to really do this at speed, like you said, Martin, and enable us in our business meetings to be able to easily communicate across languages. But I also love the idea of potentially localizing campaigns very quickly at the click of a button. That would be pretty cool.

 Louis Pereira [00:19:58]:

Yes.

Martin Broadhurst [00:19:59]:

Still have to watch out for those hallucinations though. Hey?

Paul Avery [00:20:02]:

Indeed.

Martin Broadhurst [00:20:03]:

So in other news related to AI generated content, there was a story in the Times today which reported that students are using chat GPT for coursework and then trying to beat AI detectors, which universities are said to be using by using rewriting tools such as Quillbot. And this rendered detectors ineffective, only catching AI content between two and 5% of the time. However, anyone that's been paying attention to this space for a while knows that the detectors are absolutely trash. None of them work. They're completely unreliable. The idea that universities are using these to try to detect Aigenerated content in students coursework and assignments is laughable. They are wildly inaccurate with too many false positives, false negatives. It's a joke. And this has actually been studied recently. There were some findings of a study that looked at exams and coursework from TEFL students. So teaching English as a foreign language students taken by nonnative English speakers, widely used detectors consistently mislabeled human written essays as AI generated over 50% of the time. The detectors perform well on essays written by native English 8th grade students. This was at GPT-3, I believe, but it suggests that there's significant bias against non native writing patterns at GPT Four level. The detectors are grocious. Why do I bring this up? Marketers that I speak to are always concerned about this. We've mentioned it in previous episodes specifically focused around search engine optimization. People are saying, is Google going to detect AI generated content and label it as AI generated content, only for it to then not rank well? I think this shows us that the detectors just aren't there at the moment. They're very easily beaten if they can detect an AI generated content. And they have significant biases against certain communities, in this case, non native English speakers. So the idea that Google would deploy them in the real world doesn't seem right. And ultimately what we have to remember with Google is that what they want is quality content. They say in their guidelines, they don't care if content is AI generated. It's got to be quality and it's got to be saying something new in the subject matter that you're speaking about.

Paul Avery [00:23:05]:

Yeah, I think we talked about this last week a bit, didn't we, Martin? In terms of AI enabling your writing, GPT four is great at certain aspects of writing and it can do creative writing that, as a trained writer, I can't do. Honestly, asking it to create Shakespeare style wraps about shoes at Easter, I think it would do a better job than me. But a lot of the content that we need to produce as businesses, it turns into a slightly essayish style. Or if you try and give it prompts to go down a particular style, it goes overboard. It loves hyperbolic statements, as we've talked about previously. So the human editor in the loop is still critical for what I would consider really high quality business content. And as you said, it's got to be novel because if you're just regurgitating information that's already available on the internet, you're not really helping users find out new information and so you're not going to rank that well. So it's almost like we're looking at AI's role in SEO from slightly the wrong angle, which is that it makes it too easy to just pump out crap content. And we should still be focusing our time and energy on coming up with interesting takes our own perspectives, combining information that's out there with our own experiences to offer audiences something new and interesting. Good story, right? Thanks, Martin. Next, we're going to move on to a story about midjourney, one of our favorite AI image generators. And what midjourney has done has introduced a new feature called Inpating. So here's how it works. You produce your image in midjourney. You can choose a part of that image and specify what you want in its place. And just like that, you get your edited image. So it's kind of like adaptive's, generative fill that we've talked about previously. But you're working in Mid journey with midjourney's AI generated images. So for the marketers and designers listening, this might be exactly what you need to really get a bit more flexibility in the outputs that you get from Mid journey. Because I think one of the things that and we've talked about this with all the AI generated tools for images on the podcast, Martin, is it's so iterative trying to get what you want. And ultimately, in many cases, especially in Mid journey, you have to start from scratch. Like every prompt generates a new image. So finally being able to select areas of an image and update your prompt to change just that section of the image gives all of us as designers, as marketers, much more control over the final image output. Some of the examples that were given in the original press release around this or the blog post around this included swapping glasses on an animal character from different colors and doing that very easily without having to change any of the rest of the image. So if you've been playing with mid journey and maybe you've given up because you've been frustrated by there have been certain bits of images that you didn't want, or you had to port them into another tool like photoshop. Go back, have a play with this new in painting feature and see if it makes it easier for you to be able to get the final output that you're hoping for.

Martin Broadhurst [00:26:21]:

Yeah, I love this update and I love that Mid journey are constantly evolving. I mean, not that we think they're not going to. Everything that they've done in the past twelve months is push that model forwards. While we're talking about Mid Journey, another thing that I think people that are using it will want to know about is a new product that was released this week on Product Hunt, which is Playbook by Mid Journey or sorry, Playbook and Mid Journey. Playbook is a cloud based storage system specifically designed for creatives to organize and version control different assets. It's really cool. When they launched last year, they gave out huge amounts of storage for free, which was brilliant. But now you can sign up and connect it to your Discord account where you've got your Mid Journey login and it will automatically import all of your Mid Journey creations, keeping all of your iterations and your prompts effortlessly organized. So I would highly recommend playbook X mid Journey.

Paul Avery [00:27:26]:

Yeah, that sounds really powerful. And ultimately we're still waiting for a really usable web based front end for midjourney. Right? It's still a bit of a pain getting Discord set up and getting subscribed to Mid Journey from within Discord like Troy Mike. If you go to the Mid Journey website and try and subscribe there, you will fight a losing battle because you have to get into Discord and it's tough. Rumors are they are working on that web front end and I think when they do, their user base will explode even further. And with tools like you're describing, it's going to get even easier for power users to really organize themselves and keep track of all their best prompts. Right, next story.

Martin Broadhurst [00:28:07]:

Martin fine tuning Chat GPT has finally arrived. I say finally arrived as if OpenAI have been sat on their hands doing nothing for months on end. But no, we can now fine tune. Chat GPT 3.5 is the model that we can fine tune and fine tuning allows us to basically train models on our own company's data. So owner voice specific factual information, they're great for things like question and answering systems. We can now do that and deploy it using the Chat GP 3.5 model. While prompt engineering prompting as we know it day to day is solid for general tasks, it isn't great at giving responses that are relevant to our own businesses. And that's exactly why we try to use fine tuning. So it could be used for customer support, it could be used for an internal knowledge base for your company and for your employees, all sorts of different use cases. However, there are a few things to note about this update. One, it is a little bit pricey, so it's eight times more expensive than it would be if you were to just use the base model. The other thing to note is that it is bit of a technical process. This isn't something that your average marketing exec is going to knock up and create your business's fine tuned model in an afternoon. It requires some technical knowledge in the first place. And more importantly, it requires data not just any old data unformatted, unstructured data. It requires high quality, well labeled data. Which is why OpenAI also announced this week that they are partnering on this particular project with Scale AI. They're a US based startup that works exclusively on data labeling. So if you need your data labeling, you can work with Scale AI and then start to deploy that across your fine tuned chat. GPT 3.5 model. For marketers, this is a pretty big deal. Personalized AI outputs means more brand centric content, more tailored messages. So if you're looking to create a real, state of the art chat bot that integrates with your company data, this is the way to go.

Paul Avery [00:30:48]:

Yeah, it's really interesting stuff. I think we've talked about it on the podcast previously. Training these bots on your own data and having them speak in your voice and all those other things I think is really interesting. But the need to have structured data to be able to feed the tools and we've talked. Should we get should you get into this now when you need fairly well structured data to be able to train the models effectively? Or should you wait until AI is sophisticated enough to basically figure out how to label and analyze your unstructured data itself? And that's a question mark. I'd be really interested, actually. Any listeners in the podcast. Hit us up on the Twitter's or the LinkedIns if you know more about this, because I think that's going to be a key question for businesses. But more and more options are popping up for you to easily train models on your own data, which is great, because it's something that we get asked about a lot, isn't it? Mine by folks thinking, this would be great if it knew more about this topic that we know about as a business, but GPT Four doesn't seem to be strong enough in and of course, this is a mechanism to enable that right into our next story. There was a story this week about how major brands like Nestle and Unilever are using generative AI to trim costs, amp up their efficiency, and boost their creativity. There was a number of examples in the article that we read this week, including WPP, the well known advertising agency working with Mondalay on an AI crafted Diwali campaign. Using the tech, they had a Bollywood star, Shah Rukh Kang, enticing folks to drop by local shops. Basically, small businesses could tailor the ad for their own storefronts using a dedicated microsite powered by AI. And they got loads of unique ads. People created 130,000 versions of this ad customized for themselves, which themselves garnered 94 million views on platforms like YouTube and Facebook. Meanwhile, over at Nestle, they've been using Dorley Two to craft imaginative extension to Vermeer's, the Milkmaid painting. They had 1000 AI rendered scenes created with a media value bump of more than $766,000. And then there's Inner Lever, which has many, many brands, many of which will know. Like Dove soap. And Ben and Jerry's. And they have been using AI for creating product descriptions and churning out visual content. They haven't told us results they've got from this, but they are using it. So it's really another clear indication of how AI is really starting to reshape how we advertise, both in terms of creating content at scale and helping us be more efficient and effective. But I would say the best examples so far are where ad agencies and brands are figuring out how to use AI to enable customers and audiences to be part of the creative process. Right. Customize a key asset to your local storefront in one of the examples that we gave. And I think when you can bring your customers and your audiences into that process, it just makes it much more easy to go viral. So I think as a marketer, I'd be interested, I'd be thinking, how can I bring users into this process so that they create sort of AI generated derivatives of the work that we produce and maybe help our campaigns go viral? So some of the big brands are already doing this, but the great thing about AI is it's cheap and it's easy to use, and these days it's pretty much ubiquitous. So how could you do this for your brand?

Martin Broadhurst [00:34:42]:

Right, final story of the week, and it's a battle of humans versus AI. Here, AI bots can now surpass humans at those dreaded capture tests. This was a really interesting study done by University of California and Microsoft that has shown AI bots boast an accuracy of 85% to 100% at solving captures. And they do so in around 1 second compared to a human, which is averaging 15 seconds. So it's vastly quicker and very, very accurate. And human accuracy hovers around 50% to 85%. So they're not just beating us quickly, they're absolutely smashing it out of the water. Google's even more advanced capture tool, which is called Recapture, they launched that in 2019. That isn't even safe either, because bots are completing that task with 85% accuracy in a similar time to humans averaging about 17 seconds, and humans are averaging around 18 seconds. So this rapid AI progress means that the future of capture is likely to be no capture. It's losing its potency. So when we think about all of those forms that we have on our websites with capture detection to stop the bots spamming our web forms, I think as marketers, we might need to reassess what tools we use in the future. I'm sure it won't be long before a new alternative comes to the market, but just know that for the time being, capture is not safe.

Paul Avery [00:36:40]:

Such a funny dichotomy. As a user, I cheer the death of capture and recapture. As a marketer, I cry about the death of recapture because, yeah, it is a pain when you get a load of spam submissions to your forms. But there you go. So that's what's happening in that area of marketing. Speaking of areas of marketing, we finish our podcast this week with an interview and it's Lewis Pereira. Lewis is a offline entrepreneur by day and an online entrepreneur by night, and he takes through how he built his tool, Audio Pen, which you can use to create blog posts, leave yourself voice memos, dictate emails, etc. So thanks to Lewis for his time and we'll go into the interview now. Other than that, thanks for your time this week, as always, Martin, pleasure to be here. We'll catch you next week right into that interview. Today I am joined by Louis Pereira, a business leader and entrepreneur that spans the online and offline worlds because he runs a traditional business in Goa, India during the day and at night he is creating web apps. Lewis, welcome to the podcast.

 Louis Pereira [00:37:52]:

Thanks, Paul. I'm happy to be here.

Paul Avery [00:37:54]:

I'm glad you're here as well, especially because you are the creator of Audio Pen, one of my favorite apps and just a random story for the listeners. The way this podcast interview happened was I was having a problem with the app and I sent a support email thinking it would get answered in, I don't know, a week or one day or two days. And within three minutes, Lewis had already responded, was diagnosing my problem, and had me fix within the next ten minutes, which I thought was pretty awesome for what is still I'm guessing it's just you and the team at the moment, Lewis, is that right?

 Louis Pereira [00:38:28]:

Yep, it's just me.

Paul Avery [00:38:30]:

So that's fairly unbelievable. And then that got me thinking, I love this tool, we should get you on the podcast and here you are.

 Louis Pereira [00:38:36]:

Yeah, I mean, it's nice to have you as a user. Thank you for being like a paid user as well. It's always a pleasure to talk to someone who's paying for something that I built.

Paul Avery [00:38:46]:

Great stuff. Well, you and I know what audio pen is. Perhaps you can give the listeners the simple description of what it is and what it does.

 Louis Pereira [00:38:54]:

Cool. So Audio Pen is a web app that lets you go from messy fuzzy thoughts into clear text that's written in a style of your choice. And it does that by letting you record voice notes on your phone or computer. And then it uses AI to transcribe those notes and then rewrite them in the style that you've chosen, even in the language that you've chosen in a length that you've chosen. So it basically lets you ramble without worrying about saying something incorrectly, using the wrong grammar, doing sort of a double take, et cetera. And it's really useful to sort of it's almost like a replacement for typing. It's a very quick way to get from a messy thought to like a first draft, whether it's a message, an email, an essay, even if it's just book notes that you're saving, you can use it for a wide variety of things.

Paul Avery [00:39:40]:

Yeah, absolutely. In fact, one of my favorite things is trying to figure out different use cases for it while I'm playing with it, to be honest. What motivated you to create it?

 Louis Pereira [00:39:49]:

So audio pen was an accident. I'd love to take credit for the idea, but I can't because it was born during a week where I was experimenting with learning to use a bunch of AI tools or other AI APIs, and OpenAI's Whisper API was one of those. My plan was just let me play around with this stuff and let's see what happens. Can I build something that's useful to people? Or if not, let me at least learn how to use these things. And I did that by just building a bunch of tiny tools on my own personal website. So I didn't set up separate websites and domains, et cetera, because I've done that in the past and it just usually proves to be an optimistic experiment, which usually fails and ends up being expensive. So I said, let me just do this on my own site, not try to charge people for anything, just build like small little experiments and see what happens. And I built about four or five that week, one of which was Audio Pen. It was probably the one I expected least to succeed. So it surprised me the most when I got a lot of positive feedback from people saying, hey, this is useful, I can use this for this, that and the other, because I'm not an Audio Notes person by nature, or at least I wasn't up until now. So then I said, okay, let me sort of drill down and see if I can build something around this. And the following weekend, I built the first version of Audio Pen as a standalone tool on its own domain. And yeah, that was about five and a half months ago now, I think just been building ever since.

Paul Avery [00:41:21]:

Yeah. One of the things I love about Audio Pen, and I don't know how you do it, is the constant updates that you've made to the platform since it launched. It feels like every week or two there's some new cool feature that isn't just something to add for the sake of having it, it's like a genuine thing that makes the tool more powerful. Most recently, you launched the ability to customize how much the AI changes your original transcript, right? Like almost like a temperature setting that can almost leave your transcript as is. Which my best notes, I basically want dictation software, but my worst notes, I need someone to clear them up because I've really been rambling along and so it's very nice to be able to choose, when you save each note, which type of output you're going to get. That's one of my favorite features. I really love the fact you launched. What do you think are the most powerful features of Audio Note in its.

 Louis Pereira [00:42:17]:

Current guys, I think this one was probably the best received update. Also, not coincidentally, was also very heavily requested. So I had sent out a survey in my last update email to use. I was just asking them what I could do to improve the tool and one of the most common ones was this, where there were people that were using it that did not want too much of polish on what they had already said. They wanted it to feel like it was closer to their original note in their own tone and voice and vocabulary. And then I spent a fair amount of time trying to figure out how to do this because it had to sort of fit in the existing structure of the tool without confusing other users that did not really care for it. And then I built it. And that's definitely been so far, I think, the most powerful update that people have appreciated. Other than that, I think I've just launched like a stream of smaller updates throughout the months prior to that as well, most of which weren't very user driven, but they were more driven by my own gut feel as the apps user. Myself as well as I do have this constant stream of interaction with users on Twitter and via the support email. So although I had not formally asked people what they wanted in the tool, I have this sort of set of ideas or set of requests maybe sitting subconsciously in my brain and combine that with gut feel from what I thought I wanted in the tool. And that's why you see a higher frequency of updates because I'm not following any sort of methodology or any sort of I'm not doing any user research or testing, et cetera. I'm a one person team, I'm a user, I'm a designer, I'm a developer, I'm everything. I'm not a ten on ten in any of those areas, but I'm a seven maybe in each of them. But I have the advantage of.

Paul Avery [00:44:14]:

Being.

 Louis Pereira [00:44:14]:

The only person building this stuff. So speed is something I can do. I can ship stuff very fast. If it goes bad, I can always just reverse it most of the time. That's why you've seen a bunch of updates every week or two.

Paul Avery [00:44:29]:

You definitely ship quickly. I was mentioning this to a friend of mine in the software world and he was really impressed with the level of updates and the speed of shipping them. I think all those sevens that you gave yourself question how accurate they are, but they're certainly adding up to ten out of ten when it's using the app because I find it very easy to use. I want to talk about usability in a moment. One of my favorite features is the ability to, after you've recorded your voice note, choose how the AI is going to summarize or use the content. So is it going to be a casual email is it going to be a formal business memo? Do you want it to be witty and funny? And I've had some quite a lot of fun, actually, with things that would normally have been like an internal business email, running the note through, that sort of wittier option that you can choose and actually getting some quite good stuff out of it, which has been quite good. So I'm a big fan of that feature as well. From your own experience or what you hear from customers are the sort of best use cases and applications of audio pen, the really power uses.

 Louis Pereira [00:45:30]:

So I've actually been struggling with this for the past few weeks to try and pinpoint who the ideal user is or who this tool sort of impacts the most. I'm on the verge of giving up because I feel like everybody has their own use case for it and I don't want to sort of pigeonhole it into one thing or the other. I want to keep the app almost very universal, as universal as possible on the front end with the ability for people to tweak it to their own use case, like you were mentioning earlier. You can tweak it to write like you. You can tweak it to maybe replicate a particular style that you use very often for a particular use case. So to give you a couple of examples that vary, I have users that use it because they have a disability and they can't type, so they prefer just talking and getting something that's more polished than a raw transcript. I also have people who aren't native speakers of a particular language who again, they can type, but they can't maybe frame their thoughts with the right grammar and the right format, et cetera, in a language that they're trying to communicate in. So they can talk in maybe say, broken English and get like a perfect essay or blog post or message or whatever it is in English. I also have people that are using it to translate languages. So for instance, you can talk in English and get it to write for you in French, vice versa. I mean, there are a bunch of other languages as well that it works for. There are lawyers that use it to transcribe stuff that they want written down for work. There are people like you and me who use it for emails a lot. There are people that use it to just write blog posts, people who use it to just type something as simple as a casual WhatsApp message. So yeah, I mean, there's a wide variety of things that you can use it for and I want to let the tool sort of develop organically. I think by trying to I've been almost like trying to figure out what it is and that was almost for like a selfish reason where I want to figure out how to market the tool so I can have a single angle on marketing, but maybe that's a dumb idea. Let the tool just do its thing and it'll market itself in these different use cases in each of these markets, separately, hopefully through word of mouth, if I can keep it the way it is right now.

Paul Avery [00:47:53]:

You talked about keeping it as simple as user as you could, and I think it's got so many powerful use cases, but it is also deceptively simple to use. So what steps have you taken to really try and ensure that you keep that user friendly experience on the app?

 Louis Pereira [00:48:11]:

I don't know if I have like a checklist or anything, it's more like trial and error. And again, I've let it evolve organically. So when it started off, the MVP was a very basic MVP, right? It was just a button. You could talk to it and it would give you like this one type of summary. It's only after that that I introduced all of the added features that I did in the weeks after launching that MVP. But that MVP felt powerful enough, felt like there was some magic to that interface where you could see this big button. It was very obvious that you had to press that button, and once you pressed it, it was again obvious that you had to press it again to stop. And then you'd see the result and be like, oh wow, that's interesting. So I want that to be the core experience for as long as the app is around everything else can live in the periphery around that. And it's okay if people don't find some of the advanced use cases. I don't want the average user to be a power user. I don't expect them to be a power user, but I expect every person that wants to be a power user to be able to find what they need to find while trying to convert into a power user. And that's been the case more often than not. So I'm okay having the more advanced stuff hidden in maybe a smaller icon at the bottom or somewhere inside the settings pop up window because people that want to use it will find it. And also I have the advantage of the free plan being fairly simple. So on the free plan, you don't have too many options to customize stuff, which basically means that the average user will stay a free user maybe. And if they want a little more, they'll be aware that these add on solutions exist in the prime plan. And then of course, when they update, they'll go looking for it. And if you go looking for it, it won't take you more than a couple of seconds to find it.

Paul Avery [00:49:59]:

Yeah, I think it's very intuitive from that perspective. I love the when you're recording, you see a waveform and it's the little things. That's how I know it's working right. One of the reasons we got in contact was because I didn't see a waveform, so I knew it wasn't working, which was much better than leaving a 15 minutes note to find it wasn't working.

 Louis Pereira [00:50:16]:

Right.

Paul Avery [00:50:17]:

So it's those little things, I think, that also make it really easy and fun to use. You talked about power users and it integrates with Zapier, right? So I haven't really explored that too much yet. What would you consider like a proper ninja power user application of this? That's a good question.

 Louis Pereira [00:50:38]:

Yeah, I think blogging is probably up there blogging because it involves two separate or three separate things that you can use it for. One is capturing ideas. So if you're a writer who blogs, or just general writer on the Internet, one of the things you always do is capture your ideas before you write them. So this becomes a very friction free way to just capture stuff on the go. Say you read a book, you see something interesting in a book, you talk about it to your phone. It sort of summarizes it in a particular style when you capture maybe five or six notes or maybe ten notes across a bunch of interrelated topics from different sources. A power user move would probably be to use the Super Summary feature, which is a feature that allows you to combine multiple transcripts so you could select, say, six to seven transcripts about different topics and then get Audio Pen to write a summary or an article based on all of those notes in a style that you've told it to write in. So you could even give it like a sample of your own writing for it to replicate that style. It could then create that super summary for you in that style, which you could then maybe use Zapier to plug into your blogging software. Yeah, that would be pretty cool.

Paul Avery [00:51:56]:

That would be really cool. I haven't played much with the super summarizing feature and I almost assumed you could only summarize two. But what you're saying is if I had a bunch of ideas, I could select them all and it would crank them out. Is there any word limit on the sort of label?

 Louis Pereira [00:52:14]:

So the APIs I use have token limits, of course, but I've set it up in a way where it kind of cycles through the content a few times if you cross that limit, so it can go for much longer. Transcripts as well. However, I think it's definitely most effective or most useful for stuff that's, let's say, sub 3000 words in the transcript, which is a fair amount of stuff if you're recording live voice notes. Of course, it does run into issues for people that are uploading files to the software. Then in some cases the quality of the output might not be as good because for each of these cycles it loses a little bit of context from the previous one because you mentioned whisper.

Paul Avery [00:53:02]:

So you're using OpenAI's Whisper to do the transcription. What are you using to do the.

 Louis Pereira [00:53:07]:

Text generation GPD Four at the moment.

Paul Avery [00:53:09]:

So I'm guessing once we get 32K Token access to that, you're going to have a little bit more to play with.

 Louis Pereira [00:53:14]:

It's still just like 8000 and something Tokens, I think 8192, I'm not wrong.

Paul Avery [00:53:19]:

As we've sort of talked a little bit about the future there with the expansion of Tokens, what are your future plans for Audio Pen? Any features or improvements that your users can look forward to?

 Louis Pereira [00:53:29]:

Yeah, I think one that I want to get done with this week is just like a better organizational structure for notes because the tool has been around for a few months now and there are users with like hundreds of notes already and they don't want them to sort of all live on the same screen. So some sort of a folder structure that I'm working on right now just run into a bunch of issues with it design wise today, but hopefully it get them sorted out by the end of maybe tonight or tomorrow, hopefully, I don't know. So that's one that's on the horizon, on the immediate horizon at least. Beyond that, I do want to experiment with sharing Notes publicly because some of the reasons one of the ways I like to use it is to just capture like a fleeting thought and have it write it for me nicely. I wonder if there's a nice way for me to share that, like a very lightweight way for me to share that directly from within Audio Pen rather than as like a permanent sort of artifact on my, let's say, Substac or Twitter, et cetera. Something that just feels lighter because it's a fleeting thought that I've just sort of spoken to my phone about. But let's say that's an idea at the moment may or may not build it, depending on whether it feels right at that point.

Paul Avery [00:54:42]:

Let's talk very briefly about building Audipen. You mentioned at the beginning how you went about doing it and how iterative it was and how you were like the first user. What was the sort of the biggest challenge that you faced whilst developing the tool, do you think?

 Louis Pereira [00:54:59]:

I don't think I've had a big problem at the start in terms of the core of the technical style side of things. I don't think I've had an issue yet. I build using a tool called Bubble, and I love building in Bubble. Of course, I've had a bunch of small issues here and there, but so far so good. I mean, I've managed to solve most of them. I did, however, have a challenge quite recently with something that it wasn't technical, it wasn't my fault either. It was just an unfortunate event that happened where Audio Pen just got like a bunch of traffic from a country in the Middle East, hundred x more than it would normally get from that country on a day or two. And I don't know if it was bought traffic or what, I have no idea. But it basically ended up with it being blacklisted by a bunch of antivirus companies, which was scary because I was like, okay, this is happening. I don't know why it's happening. There's nothing I can do to as far as I know, nothing I can do to solve it. And I spent the better part of two weeks just emailing these guys, like, all of these antivirus companies, saying, hey, this was a false positive. But it was like, there's no other way to do it. You've got to email these guys one by one, go to these sites which aggregate the reports, find out who's still blacklisting you. Some sites wouldn't respond, so I had to find them on Facebook, find them on, I don't know, Twitter, et cetera, et cetera. DM them, get them to respond one by one. I get on a call with a couple because they took too long. Then Twitter blacklisted the link because they use one of these databases from one of these antivirus guys. But just yes. No, day before yesterday. I think it got cleaned from Twitter as well. That was like a pain in the butt. It took a month for Twitter and it took about two weeks for everyone else, but two weeks of basically constant emails, high stress people being unable to use it because their antivirus software would just block the site. You won't be able to see it for no fault of mine. Just some random traffic from somewhere. Yeah, that was a big challenge. And there was a point in there where I was just like, I don't know if I'll ever fix this. But, yeah, for anyone out there who's listening, who has the same problem, like, the secret is just persistence through email, which is just quite silly, but it's true.

Paul Avery [00:57:08]:

Well, I'm glad that you did it. I can imagine, certainly when there's only one of you in the team, that was pretty stressful and required a lot of work. So I think we all thank you for your persistence in getting people up and using the tool and finding a way through that last bit. Let's just switch to talking just in general about AI, right? Because you got into this, you were playing with the tools, you wanted to see how they work, so you got into Whisper and you play them with GPT Four. But I suspect that you've probably seen quite a lot looking under the curtain at how these tools work and what's possible and what's not. So how do you see AI influencing how we work in the future?

 Louis Pereira [00:57:49]:

That's a deep question. I don't know if I have a very optimistic answer to it. Of course, it'll replace a bunch of tasks that we do. Hopefully it'll just replace the ones that we wanted to replace and not the ones that it wants to do for us. I think it definitely has tremendous upside when it comes to work, but the downside risk is not zero. So given a long enough time scale, I do think things are not going to end very well because regulators aren't going to be able to keep track of this stuff. They usually move much slower than they need to and I doubt it's unlikely that this will be any different. Having said that, I feel like right now AI is in a great spot because the tools aren't good enough to replace us yet, but they're fun enough for us to sort of play with them. So, yeah, I'm having a blast, but, yeah, it hasn't impacted my day job yet, which is like a very offline job, but with the stuff I do at night, yeah, it takes away a lot of the painful stuff, like GBD Four, whether it's trying to write a few lines of code, which I can't do very well. I'm not a developer, I use Bubble to make this stuff, but whenever I need code, I just talk to AI and it freaking does it for me, which is ridiculous. I can even paste bunch of code, like a bunch of code in it and ask it what's wrong with it and ask it to help me do something different, which is just, again, crazy. So I do think, like, a bunch of use cases like that, AI will eat will sort of eat up jobs in the middle layer, not at the top and at the bottom. So the average developer or the average writer will be ousted by an artificial writer or developer, but like, the guys at the top will still survive. They'll probably use AI to their advantage. They may not survive forever, but in at least this initial phase of disruption, I think they will. Maybe in ten years, down the line, we're all done for at least as far as creativity is concerned, I don't know. But for the foreseeable future, I'm just doing what I can to not stay in the middle and try and be closer to the top or on the board, whichever.

Paul Avery [00:59:58]:

Well, there's a couple of things to unpack there. First, the fact that you built this in Bubble, right, a low code stroke, no code solution. That when you use the web app, and for those of you who are listening on the podcast, there is a free version. Go and have a go with it. The paid version is extremely reasonably priced and is well worth upgrading, at least in my experience. But it's polished. When we started this conversation, I was not expecting it to be a low code solution, right, because it's a very polished web app on the front end and works really well. So I think that's very interesting stuff. I think we all need to be keeping on top of the developments in AI to stay up in the top slice that you described, right, because getting better at what we. Need to get better at and using AI to augment us, to be at the top of our game where we can. One of the big things about the podcast is trying to be a gateway to enable people to do that, right? Speaking with people like you, opening people's eyes to tools like Audio Pen, but also the process of developing a tool, right? Like what are you trying to solve in your business? Because we're not far away from AI, helping you develop bits of code that you may not have been able to write otherwise, and low code solutions for you to build tools that help solve your biggest problems, like you did with Audio Pen. And now we all have this awesome tool to use. So with that last thought in mind, do you have aspirations to build anything else with AI that's not Audio Pen? Any other ideas and itches to scratch?

 Louis Pereira [01:01:26]:

Yeah, I mean, I always have ideas. It's been five and a half months now since Audio Pen was born and I've sort of jumped from idea to one idea to another before that for a few years. And now that I've got one thing that sort of solved a nice problem and is doing quite well, I'm trying to just focus and sort of fight my instincts to just keep building stuff. But of course, man ideas are a dime a dozen. And with AI, it was already cheap to experiment with new ideas because of solutions like bubble and other sort of low code, no code tools. It is cheap and it was fast and you can do it alone. With AI, it gets even cheaper and even faster to do that at scale as well. So the way I sort of approach life now is I don't think I'm very good at figuring out what to build from scratch. However, I think I enjoy the process of building and AI, and no code has made that process extremely fast and cheap. So what I can do and what I've done and what's led to Audio Pen was I've just built, like volume. Like I've built a bunch of stuff, seen what sticks, thrown a bunch of darts at the wall and doubled on the one that sort of worked. So I would say whether you're trying to build something new for fun, whether you're trying to improve something within your own business, just play with it. It'll cost you, what, less than $100 a month at best to build something, of course, excluding the value of your time, but it's quite fun to build. So you could just look at it as an experience or a fun thing for you to do on the side. And if you enjoy it, then your time is well spent. Money wise. It's fifty dollars to one hundred dollars tops for you to use this stuff and keep trying stuff until something works. And when it does, just double down on it. I mean, it's not very complicated. You don't have to be a Steve Jobs to kind of figure out what people want. You can just give them everything and see what they buy.

Paul Avery [01:03:20]:

Well, I think it's a very humble description of how you got to where you are, Lewis, because I love the tool. I think it's really powerful. You've inspired me in our conversation, both what we recorded and what we didn't. To go and try some of the power use cases of audio pen that I haven't even begun to tap into yet, but also to think about what low code, no code solutions I might get excited about and ideas that I might want to explore. You've inspired me to go and do those as well. So thanks so much for your time today. Really appreciated it. I'm sure the listeners will love it. What's the Audio Pen website for people to go check it out?

 Louis Pereira [01:03:54]:

Just audiopen. AI. You can just Google Audio Pen. It'll be probably the first result. Click on it. It's an orange website. You can't miss it.

Paul Avery [01:04:02]:

Good stuff. And if anybody wanted to get in touch with you, to speak to you directly more about the product or about your own journey.

 Louis Pereira [01:04:09]:

I am most active on Twitter, so just my full name, louis Pereira on Twitter. That's X.com. Now. That's where I'm at. If not, just download the Audio Pen app and you'll find the feedback form there. Just write to me and I'll respond.

Paul Avery [01:04:24]:

Great stuff. Thanks again, Lewis.

 Louis Pereira [01:04:25]:

Cool. Cheers. Nice talking to you.

Paul Avery [01:04:27]:

Thank you for listening to Artificially Intelligent Marketing to stay on top of the latest trends, tips and tools in the world of marketing AI. Be sure to subscribe. We look forward to seeing you again next week.