Transcript

Matt Heine (MH):

Hi and welcome to today's episode. It's fair to say that it's very hard to have a conversation at the moment without talking about or at least referencing ChatGPT or AI. Now, there is no doubt that AI is here and there is no doubt that it's not going away, but it has been around for a long time, and I think it's important that we start to think about and understand what the applications are and what it actually means for us. And to do that, I'm delighted to have Dr. Michael Kollo joining me today.

Dr. Michael Kollo (MK):

Hi Matt. How are you?

MH:

I'm excellent. Thanks so much for joining us today. As I just introduced, it really is the topic of the month or the week. We've seen from some of the stats that ChatGPT has had the highest adoption of any technology. I think almost just above now, the take-up of Threads as the fastest ever downloaded app.

MK:

Oh wow. Yeah, yeah. I think when it was first released in November of last year, I think OpenAI was just trying to test it and they didn't realise it in the first week, I think it had 1.2 million users, in the first month, 100 million people made free accounts and logged in from around the world. And yeah, the take-up has been absolutely enormous.

MH:

And you must be sitting there sort of scratching your head as to why is it now that everyone's talking about it. I think from memory you've been using or at least playing around with AI for some 20 odd years.

MK:

Yeah, look, I mean, my career began in financial services I suppose back in the mid 2000s. I went to university here in Australia, ended up going to do a PhD in finance in London. And so I was working with data and statistics and forecasting markets. And I ran a career in London for about 14 years working in institutional asset management, as something called a quantitative manager. So somebody who uses data to forecast the world using algorithms. And it's always been a very interesting field for me, I really enjoyed it, but I've also enjoyed talking about it. So I was teaching university, I was kind of engaging people.

And it was about 10 or 11 years ago when machine learning first came on the scene and it was a very different kind of animal than what you've seen before. Before, you had the classic economists and econometricians who were thinking about the world, who had philosophies or ideas, hypotheses of what the world might look like and they would use data to try to confirm them or understand them. Whereas you had this new breed of person coming in from a computer science background with machine learning, who was just trying to throw a lot of processing power data and somewhat blindly saying the data will find the way, that as long as you have data, the truth is in the data and everything will be fine.

And there's a whole different way of thinking about the world. And I think finance looked at that and went, "Well, this is not the best." And we weren't very fast on adoption at the time. But there's a little known field that began to emerge called natural language processing, which was using initially data feeds from Twitter and all these other places. And just to give you a sense, the simplicity of it back then, about 11 years ago, I'm talking about a long time ago, 11 years ago, was looking at positive and negative words on a page. So you read a newspaper article and you looked at how many words were great and how many words were poor, and then you counted them up and then the ones with the more number of good words were good articles, and the opposite for bad articles. And that was as far as we understood language.

11 years later, or maybe about 12 years later, we have a system in ChatGPT underneath the umbrella of an area called generative AI, which understands and speaks language better than we do. So the linguistic IQ of the system is around 140. It's not your traditional AI system that we've been looking at for the last decade, which are kind of advanced forms of statistics and mathematics and so on. What we're seeing here is a system that has essentially cracked the use of language, which we thought, up until now, is an innately human thing. And it was an innately imperfect and it was innately representing, for example, from the creative side, our wishes, our aspirations, from the work side, our complexities of our reports and our conversations and so on written down. But actually these systems are brilliant in picking them up and using them.

And I think the big surprise for the whole world was how quickly that emerged and how rapidly suddenly something like ChatGPT came out and said, "We have passed a significant milestone. We think we can understand and comprehend and write language really, really well." And the world went, "Wow, we want some of this," and thus we're talking about AI. But as you say, AI has kind of come and gone in popularity in the press and in conversations over time. But this is the first time that AI can be talked about in the context of language and importantly it can be talked about by anybody. So this is not an engineering conversation. We're not about to go into data sets and talk about Python code or whatever. We're actually just going to be talking about the way the language is used by these algorithms. And I think that is super exciting in terms of where it's going to go.

MH:

And I think that's a really good point. And like many people, I went into ChatGPT and asked ChatGPT to explain how it works to a five-year-old. And it gave me a beautiful answer actually, which was effectively, imagine you had a magic box with every word that you've ever heard, and you could dip in and out of that box and pick out words to create sentences. Now it's clearly slightly more challenging than that. Do you have a simple way of explaining exactly how the ChatGPT engine does do what it does?

MK:

I can use metaphors and I can certainly use explanations. I think I must profess all of this by saying that any metaphor I give you is going to be imperfect. And so some listeners will shake their head and go, "Come on, Mike, you're missing the more technical picture here." But I think the way to think about this is that language, and I suppose all the sentences and words and conversations that have ever been said, there are patterns inside language. The way that we structure, not just grammatically, but how we string together reasoning systems, how we give examples, how we give point lists and go into them and so on. If you feed enough text into a computer, the computer understands the patterns in language and essentially understands how to complete a language sequence.

So in the simplest ways, that could be like you and I basically having a conversation and wondering what the next part of that conversation is going to be. And these systems give us probabilities of what that could look like in the next part. But I think what is really surprising for people is just how rich language is in terms of the information that's inside of it. I think a lot of the data science profession went into this going, "Well, if we can just understand the next part of language, we can answer really simple questions." These systems are so large in terms of parameters. I think originally ChatGPT was 175 billion parameters. I don't think the next one, ChatGPT-4, hasn't officially been released for that number, but estimates are maybe tenfold bigger. And what that means is essentially the complexity that it's able to find patterns inside our language means that it's able to look like it's reasoning or seem like it understands something and able to kind of transfer that meaning around.

So I think the way I tend to explain ChatGPT to people is to actually personify it, normally something that you don't want to do very much with algorithms, but to say it's got a linguistic IQ of 140, it's like a really proficient literature student that you have at your disposal who understands context, understands how to write and rewrite, but their knowledge of the world is a bit patchy. And you probably don't want to ask them too many factual information at this point, but what they can do is they can transform and they can engage with you using language really well.

MH:

And presumably then when you using something like Dall-E, which is the image generator of some of the other sort of ChatGPT based illustrators, it's again the same thing, where it's looking for patterns, colours, brush strokes in what some would've thought before was a very unstructured environment.

MK:

So that's a great point. So things like Midjourney, Dall-E, they are a form of generative AI that takes text and creates images. So rather than text to text transformation, it's text to image transformation. And again, you're talking about a really atomized idea here, and there's a particular process by which that actually happens, whereby essentially the algorithm builds up an image which is ever more probabilistically like what you have said. But of course initially it feeds it with tremendous amount of images that are tagged in great deal of detail in terms of what kind of ideas and thoughts and notions that they reference. But the way that these algorithms understand pictures is on a pixelated level, a very atomic scale. And the way that it understands language in a way is through something called tokens. So again, it breaks down words, not quite into letters, a little bit more than letters, but not quite to something that you and I would recognise.

So the transformation element is the most important, text to text or text to pictures. And again, these all comes from a discovery called transformers, which I first started reading about in 2015, which was originally done for the language translation, and it was the foundation of which we got really good at translating language. So transformer is like a mathematical representation of a sequence of words or thoughts or whatever, and it's just a very long vector of numbers essentially, super simplistically. And the idea that we can have mathematical representations of the world that we can then transform into other things like images, like voice, like music and so on, is really where the entire generative AI space is sitting.

Listen to more Between Meetings podcasts

In this podcast series Matt Heine, Joint Managing Director of Netwealth, chats to industry professionals and thought leaders on what opportunities and challenges they see for financial advisers and the wealth industry as a whole.

Listen to the episodes

 

MH:

And that's probably a good segue. I think before we jumped on the podcast, we were talking about the typical questions that you get asked, one being, are robots going to take over the world? So I think from our perspective, and something that we've been talking around for a number of years, is it's really man with machine versus man without. And I know that you've been doing a lot of work in the industry, particularly with financial planners, to get to the point where people shouldn't be afraid by what's coming, but looking at the opportunities to actually harness the power of all of the tools that are increasingly becoming available to us, to run better businesses, to drive engagement. But I'd love to hear from you, I guess, where you see it all heading and some of the best use cases that you are seeing.

MK:

Oh, look, absolutely, and I think I want to say at the beginning that when it comes to AI in Australia specifically, we as a country have a pretty good understanding relative to other countries about the risks. They did a survey, 17,000 people, 17 different countries, they showed that Australia was somewhere in the middle of our awareness of risks of AI, the dangers of AI, but we were right at the bottom at understanding the benefits of AI and what are the positives. And so I think for a lot of people, the average person on the street, when you say AI, they don't get an image of like, wow, yes, I can't wait to be part of that world. Or if you say, "Look, imagine an Australia that is using AI and deploying AI, AI is everywhere. Now picture that image." Most people go, "Skynet," or they go, "Robots on a beach," or something like that.

We need to change that. I think collectively we need to create more thoughts and images around how there could be a much better world with AI for all of us. Now in terms of the industry adoption, financial advisors, that's actually been remarkably positive. We run these masterclasses for them and people have turned up from 25 to 75 and they've come curious and they've come with open minds and questions and they've wanted to understand the strengths and weaknesses. They have a very pragmatic approach. Many of these people have businesses that they've run or are running, and so they're thinking about these as tools that they can deploy and utilise. And so the kinds of use cases, which I think are some of the most powerful and what surprises people about how useful it is, is ability to handle complex information.

So financial reports or financial market rundowns are often not in the easiest language. They talk about interest rates and macro and so on. And a lot of financial advisors need to translate that to clients. Clients who have very different ideas about the world, that they care about their savings account, maybe they care about family and holiday and basic everyday kinds of things. And so far they are the translation vessel for that. When they start to see use cases where complex information is taken and essentially translated to elements that people find relatable or interesting or maybe even entertaining, and suddenly all of these passages open up the financial advisors to say, "I can engage with my clients much better. I can make this experience much more interesting. I can get rid of the benign work." I think the amount of time you can generally save on a, let's say a written piece of report or a blog or something is by 35 to 37%.

One of the important elements, of course, all of this stuff is genuine writing or the authenticity element. People worry about the fact that if we're just going to hand it over to AI, where's the authenticity? And I think authenticity is wonderful except when you're writing benign and emails and whatever, and if you count up the number of hours you do in a day, you end up doing a lot more of that. So I feel like authenticity is important. You have to put your name behind these things, but ultimately AI is a tool that helps you reduce, radically reduce not just the time it takes, but the quality of output that you have and the range of outputs you can have, as well as somebody working in this field.

So again, financial information is inherently complex and subject domain experience, it handles that very well and it manipulates and transforms and works with it very well. And I think that was something of a surprise for most people who had kind of used it for more benign use cases, like write a poem for my wife about how much I love her or something like that.

MH:

Yeah, I think that point about keeping it human is absolutely critical and certainly I'm guilty of using ChatGPT way too often for many of the things that I do, but the reality is that it'll often get me 70% of the way there, and then you can build on it and tweak it and make it your own.

MH:

We often hear, or we're certainly warned regularly internally, that we shouldn't be putting any personal information or anything sensitive into the ChatGPT engine. Do you want to just talk about why that is such a risk and why particularly financial advisors with so much access to so much information really need to be mindful?

MK:

Yeah, I think the data privacy thing often comes up. So there's a couple of things here. I think there's a couple of misconceptions and then there's a couple of reasonable things as well. I think the reasonable things are that whenever you have personal information of people, you have to be very mindful of the fact that that's an asset or that's something to protect and be very careful with. So generally speaking, you don't want to have it lying around, you don't want to email it around. You want to keep it protected, cybersecurity becomes an important consideration for the protection of that kind of data. And therefore anytime you're sharing it with a system, an external system, just be mindful that you are doing so. Of course, we do that almost every day. If we use Microsoft 365, if we use Slack, if we use all of these other, Jira, et cetera, we are putting information into a system that is aligned to us, but ultimately that system is still away from us, like it's in the cloud or something else.

So I think different companies have different levels of risk tolerance for that kind of thing. As a general point of view, you do have to be mindful. I think there's some misconceptions about ChatGPT and information, and I think the misconceptions are that if you feed something to ChatGPT, then it gets stored into a database somewhere and then somebody looks at that database and then they put that database into the language model just to make it bigger. That's not currently how it works, my understanding is. Again, I'm relying upon my knowledge of the models and the way that OpenAI operates and has said that they operate and they are a very significant company with a huge investment for Microsoft. So they're not a kind of small little garage shop.

What they do, especially for the paid accounts, but even for the non-paid accounts, is you don't have to save any of your conversations. So that means that none of the information that you provide it gets saved anywhere, and you don't have to provide feedback to the model, which means even if you say that was a good piece of feedback or bad piece of feedback, that essentially goes back into the training process. What's important is if I, let's say put in a question into ChatGPT, that's very sensitive, I say, "Should I make a takeover bid for my rival?" That information, that text is reduced into these tokens, as I mentioned before, which are atomized versions, and it kind of filtered through the model and out comes the response and that's it. It's not a piece of text that is put into a database and then used to reference something else. It literally flows through the model and out comes the response to it.

That way of understanding models is new. For most people, they think about data and databases and that's what they think about. Think about these AI algorithms as filters that you pour your information into, and then you have a look at what you get out the other end of it. And at that point, that's it. So your information's gone through, it's not retained, it's not captured, and so on and so on. So again, I don't think ChatGPT is uniquely data stealing or anything like that. I think the reason that many banks and other institutions have shut down access early on is because they saw rampant usage across their staff, especially their younger staff. They didn't really understand the system, but they could see that people were all too gung-ho about putting anything on there, including things that may be more sensitive. And so they were like, "Okay, this is not okay. It's a new system, we don't quite know exactly what it's doing. Until we do, let's just shut it down. We haven't authorised this. We haven't told people to go ahead and use this."

I think as businesses start to intertwine AI into their business strategy and to start to think about themselves as AI enabled businesses, the next generation of financial advisor that puts AI at the heart or as a pillar of the offering that they have, and that's a good thing to do, and not just efficiency, but also from a service perspective, I think they'll basically be much more proactive about understanding these systems and going, "Yeah, absolutely, use the next thing, use the next thing, use the next thing that comes along." So again, data privacy, as a general point, absolutely big deal. Be mindful of people data, be protective, be respectful, and so on. Specifically for ChatGPT, be careful as you would be any system, but you can easily anonymize. If you don't want to send in Mrs Smith's details in that email that you want ChatGPT to write to her, call her something else and then get ChatGPT to do it and then send a letter to Mrs. Smith, or re-personalise it afterwards. There's lots of ways around this.

MH:

Since we've gotten to the nuts and bolts fairly quickly, maybe a more detailed question, something I learned about last week as a paid user of ChatGPT's plugins. So there's now a huge marketplace of plugins that extend the capability of ChatGPT and what it can do as far as scanning and summarising PDFs, pulling on financial data. What are some of the best plugins that you've seen? And maybe if you can just talk a little bit about how they work or extend what ChatGPT is already doing.

MK:

Yeah, absolutely. So one of the observations early on was that ChatGPT has been trained to data from 2021, to the end of 2021. And essentially that means that if you want find that knowledge, like what's the NBA scores or how's the cricket going or something, it won't know. It will have only information up until a certain point. That doesn't matter for the purposes of what we talked about in terms of language, but it does matter for the purposes of using it as an information source, like knowledge. Almost like a quasi search engine, the way that you would use Bing for example, or Bard, which is Google's offering. So what ChatGPT came out with is these plugins, and this is the paid version on the GPT-4, which is the more advanced model. They've taught it to use plugins. In order to use plugins, it has to understand that you're asking for something when the plugin becomes relevant.

So if you enable the plugin testing and so on, what you're able to do is to grab some externally sourced plugin and say, "Now please look up the scores of the NBA for me." One of the versions, it's just a browser. So it just goes off to the general search rhythm, in this case Bing, and finds some information, brings it back to you. And then there's all these plugins that try and do specific things like draw a little graph, like make a little music and so on, but they're all provided by third parties. And these third parties essentially make a deal with OpenAI that every time that someone asks for at say a piece of content, like draw me a picture about a tree, then if you've selected that plugin in your list, then your request essentially gets sent off to that company, the company does something for you, and then ChatGPT tells you that here's a link to that something that the company's done for you.

That can be useful if you're trying to do multiple things in the same go and you say, "Oh, I can, I have a picture," you click on that button, you go to the company. There's no doubt that at that point your data or your request is moving not just from OpenAI but to that third party as well. I mean, for me, the plugins are still uncertified, it's a little bit like early days of Apple Store. So I'm still of the opinion that at this point I use the plugins that allow me to browse PDFs or the internet. So I can point to a PDF document and say, "Summarise the PDF document." What's really important about this stuff though is in the background, a certain amount of data can be fed into large language models, and that's not infinite. Every model has different capacities. Sometimes you're told about that explicitly.

So if I try and copy and paste a three-page document into ChatGPT, it will tell me, "Fine, I can work with that." If I try and copy and paste a 10-page document or 15, it might tell me, "No, I'm sorry, this is too long," and it errors. If you use a plugin to say, "Go and read this website," which is roughly about the same amount of text, it will just make its way halfway down, which is the maximum it can do, and then a stop reading, but it'll pretend that it's gone all the way. So there are a couple of cautionary tales here. The most probably adventurous plugin, which is coming to a plugin store near you, which is a slightly different category, is what's called an interpreter. So an interpreter is a Python interpreter.

So obviously with ChatGPT, you can ask it to write you code. So you can say, "Please write me a piece of code," then it goes off and calculates, I don't know, some numbers. So here's my client list, for example, or better yet, here's a bunch of stocks with valuation numbers next to them and industries, show me what the average valuation of industries are, for example. And what it will do is it will write a piece of Python code up until now, that you'd have to go and run yourself or your data analyst would have to, or whoever. Now with the interpreter, it's able to run that internally to ChatGPT. So what that means is that if you upload a piece of data, again, being mindful of what you upload, but if you upload an Excel spreadsheet, it can pick that up and it can create data analytics for you, basic data analytics for you, and maybe some exploration as well.

For people that don't know how to code or don't really understand that side of the world, it becomes a really easy thing to talk to a database essentially through ChatGPT to get you to do things. Again, that Python code that it runs cannot access the internet, so they try to make it safe. They were very cautious about the fact that people could ask it to do malicious things, but it is able to explore data. So I suppose as a kind of a plugin, the ones that are most powerful are the ones that allow you to read the internet and the ones that enable you to work with your own data sets, to interrogate it and to ask questions, what's the biggest this? What's the biggest that? What's the average of this? And so on and so on.

The Netwealth Portfolio Construction Podcast

In this podcast series, our investment research team pick the brains of key wealth management professionals to uncover unique insights on the investment areas they are most passionate about.

Listen to the pocasts

 

MH:

You've referenced a number of times what you might ask the engine to do, and I touched on an example earlier on where I asked ChatGPT to explain ChatGPT to a five-year-old. The prompts that you put in or the questions that you ask, will give you very different answers and therefore it becomes very important how much information or what it is that you're actually asking. If we bring it back to financial advice and wealth management again, what are some of the suggestions or what are the great prompts that you've seen that have seen really good results?

MK:

So as part of the masterclass, I end up teaching prompt design, and I think that's a very important part and people often miss that out. For example, many people listening to this podcast would be like, "Well, I tried it once. I asked it to write a report for me, but it was so bland and generic, I never touched it again." And so generally when you're writing prompts, you can be extremely prescriptive. You can literally write a page of prompts, which would be, this is exactly how I want you to do things. I mean, generally speaking, most prompt design will say to you, there's probably three things that you have to do. The first one is provide context. So for example, pretend that you are an equity analyst and you're looking at this fact sheet that I'm showing you, and you'd like to understand what are the kinds of risks that this fund is taking, for example.

The second part of the prompt would be read the below text and highlight two of the major risks that I highlighted by the manager about the economy. And then the final prompt would be, how would you like the information back from the system, which you could say things like, "And write to me in three bullet points or five bullet points," or whatever else you want. And just that simple structure of thinking about it, where you're not making any assumptions about what the system knows or doesn't know how to do, but you're kind of being almost as prescriptive as you would be to a junior. Imagine sitting there with someone, an intern or whoever, read this, find out for me, tell me the highest risk that I mentioned, and write it to me in the style of Wall Street Journal or the AFR or whatever. And the system will do exactly that. It will take that beautifully and it will ride it out.

So I think for most people, for example, one of the things that we do with the masterclass is we take an investment process and we say, "Right, here's your investment process of your manager. Now let's say we want to do some funky stuff with this." And so the first one is express the investment process as a simple five bullet points. It's quite complex, so often that's kind of a nice one. Express it in terms of a five-year-old, absolutely, it's a great one to get a hold of. But then you can do things like write a multiple question and answer on this that I can give to other people to test them on their knowledge. Or pretend that you are the investment manager that has this, and let me have a conversation with you.

And so one of the key parts of what I just said is when you're prompting the system, people assume that you're going to be asking the questions and the system's going to be giving you the answers, which is reasonable most of the time. But what's wonderful about this system is you can play with it all kinds of different ways. You can get it to ask you questions. You can say to it, for example, slightly tangentially, "Guess the name of my dog, without asking me the name of my dog. Ask me a question and I'll provide a response." And the system very cleverly, works its way through. "Do you like animals? What kind of animals do you like?" "Oh, I like turtles and doves and I like dogs." "Well, dogs, what kind of dogs? Oh, do you have a dog? What kind of breed of a dog?"

And it works the way through. And then eventually you go, "Okay, so what's the answer?" And it'll kind of make a guess. And sometimes it's wrong, sometimes it's right. But the point is that it's trying to engage you. So I think prompt design for the quality of outcome that you get is very important. And the second thing is very important is expecting to iterate with the system. It's not a one shot, "Do this for me. Oh, you failed. Okay, I don't want to talk to you again." It's this, okay, but do it slightly longer. Do it slightly shorter. Emphasise this, don't emphasise that. Take this away, put that back in again. And within three or four of those kinds of prompts, you get to really decent, I think, response.

MH:

So the question that clearly comes up is this how SOAs and ROAs are going to be produced in the future? And I saw a post earlier on LinkedIn regarding what it means for paraplanners. I'd love to get your feedback on that.

MK:

So I think the genuine content part is very interesting. I'll tell you a story, which I probably can declassify now because it's happened maybe more than a decade ago. I used to work at BlackRock at the beginning of my part of my career, which is a wonderful, wonderful shop. And I was working in the risk department and we had about 400 billion worth of mandates that we're looking after, and it was a very high quality risk group, 90 people, PhDs, whatever. And we would talk to unusual behaviours in markets, and it was quite formulaic by the end of it, it felt very formulaic. And so I wrote a little bit of automation, I remember in Excel back in those days, that would just create the comments. And it caused a big upheaval because on the one hand, the comments had become quite formulaic, so the algorithm was correctly representing what we would've said, but the assumption was that you were looking at some information and data and making those comments.

And by giving it the hands of an algorithm, you're automating that. I think on the one hand, this can be an extremely good efficiency saver, SOAs and so on are tedious and they're quite structured and so on. But one has to be very careful that you are signing off on that. So don't let the fact that you have a piece of automation behind it mean that you pay less attention to what is being done or how it's being done. And I think as soon as you have an external vendor or a company that's providing you an algorithm to do that, it can feel like who owns that piece of content? And again, I think being very clear about saying, "No, no, no, I am the one providing this advice. I'm using the tool to reformat my advice or my information in a different way that is regulatory compliant. And I take the risk that it isn't regulatorily compliant, importantly," because the algorithm can kind of get me the way there, but not all the way there.

So I think it'll kind of pose some really interesting questions, and maybe we've gone a bit too far into that formulaic reporting mindset. And as soon as you start to, I suppose, create these automations, you're going to start to see this question being thrown up. And you'll definitely see companies who are on the wrong side of that. You'll see companies who are throwing out hundreds of SOAs just for a fee point. And I think the regulator will have to be very careful to come back and say, "Okay, but is this really yours or not?"

MH:

Michael, we're running out of time, but I guess before we finish, I'd love to understand your thoughts on perhaps where this is going to have the biggest impact. We can see applications for investment management and modelling. We can see marketing, efficiency, engagement. There's so many areas that this new technology is going to touch upon and enhance. Where do you think it's going to have the biggest impact?

MK:

I think it's very hard to say because a lot of the impact will be about industry adoption. So the system has a lot of capability, as you said. Now capability is transformed into outcomes when industries adopt it and use it. It's up to us essentially, the question, in terms of how much we want to use it. A big topic that's very close to my heart is education. And I think if we deploy these systems as educators, as mentors, I think we could achieve an enormous uplift in our communities, in our families, in our working environments, in our civilizations. Because so many people receive partial or partially informed, I suppose, clues or information, especially about topics like financial literacy or financial markets. And these systems have the capability to put a much more scientific and thoughtful view in front of every person that wants it and needs it, at the level that they are at, at the stage that they're at, in the way that they want.

And whether these are children we're dealing with or adults, I mean, one of my thoughts was we could be at a point where this will be the last generation solely taught by teachers, and in the future we might have a large chunk of material being delivered and tested by AI algorithms. And that's not a scary concept as it might seem because at the moment we've got three or 30 or four kids in high school, primary school classes all around the country. And while a good teacher is definitely not replaceable by ChatGPT, the bar is these overworked poor teachers who are trying to do their best with a class full of people. Similar with financial literacy. I think it's hard to find good consistent and reliable and applicable sources of that information for all different forms in the society. And as we know, financial literacy is one of those key early skills that you want to embed in people so they make good decisions and better decisions throughout their life, not just about superannuation, but everyday spending and so on.

So I think our ability to use this technology and to create that future we want and that we desire is up for grabs, and that's a once in a generational opportunity.

MH:

Michael, what a fantastic place to finish this discussion. I've thoroughly enjoyed it. It's very rare to find someone that can articulate what the technology does so clearly and even rarer to find someone that is in our industry and understands our industry. You're running masterclasses around the country. I believe you're joining us for our national summit on the 7th of September. How else can people that are interested get in touch with you or find out more information?

MK:

Thanks so much. So I mean, I'm the CEO of a company called Evolved Reasoning. So we have a website, evolvedreasoning.ai. We put contents on there regarding AI, generative AI specifically, we run workshops again, for groups, executives. We are going to be doing some online content as well. I'm quite prolific on LinkedIn, but I'm always also responsive to emails as well. So again, I'm deeply passionate about AI adoption and getting people to use these tools and to give it a shot and see what it can bring them. So any way that I can help organisations or individuals achieve that, I'm all ears.

MH:

Fantastic. I can't wait to see where this is all going and really look forward to continuing the conversation. Thanks very much.

MK:

Thank you.