Episode - 017: - Where we learn from Glenn what puzzle pieces make up finance transformation, what steps finance teams can take to start using AI, the limitations of ChatGPT, Chat GPT-4, the different applications of different Ais, the importance of mindful work, and much much more.

Glenn is a CFO, Director & Author, known for his industry renowned book ‘Deep Finance’ that charts the course for modern CFOs in leading digital transformation using AI.

His most recent move has seen him become Director and CFO for the Eventus Advisory Group, who provide CFO services for SMEs. And, in his spare time, Glenn is a triathlete and marathoner, is in the process of finishing a 200,000 word sci-fi epic novel.

Audio Podcast Links

Spotify

Google Podcasts

Apple Podcasts

Amazon Music

RSS Feed

Where to find out more about Glenn

Tech Mentioned

People Mentioned

[00:01:48] Why Glenn got into AI

[00:11:37] Glenn’s FP&A tool written by ChatGPT

[00:14:02] The puzzle pieces that make up finance transformation

[00:23:03] What’s the entry level for finance teams that want to start using AI?

[00:28:01] Beware the accuracy of AI

[00:32:50] Differences between RPA & AI

[00:37:39] Is prompt engineering actually going to be a thing?

[00:41:16] Virtual AI assistants, and merging man/woman and machine

[00:47:49] GPT is not a math engine, Wolfram Alpha is a better bet

[00:50:32] Different Ais for different purposes + ChatGPT-4

[00:57:30] Moving from mindless work to mindful work

[00:58:49] The tools Glenn can’t live without

[00:00:00] Glenn: First off, I'd be embarrassed to think that somebody would believe that I wrote as badly as ChatGPT. But, I have used it for ideas.

It's all about moving people from mindless work to mindful work. And the question is, at some point, you know, when the AI gets better and better, are you gonna start excluding large swaths of the population that maybe can't operate up in the 98th percentile or whatever that the computer's doing.

It's gonna be amazing to see where this goes because it's moving so quickly where it goes in the next few years. But, I don't think we're right on the doorstep of, our, AI having, sentience and personalities and, and all that

[00:00:38] Adam: Hello and welcome to Tech for Finance where we help Finance Professionals Leverage Technology to level up their lives.

I am your host, Adam Shilton, and in this episode we're chatting with Glenn Hopper, CFO and Finance megastar with over 20 years’ experience leading finance operations.

Glenn is also an Author, known for his industry renowned book ‘Deep Finance’ that charts the course for modern CFOs in leading digital transformation using AI.

After starting his career as a journalist for the Navy, Glenn moved quickly into finance and then CFO positions for several scale-up companies. His most recent move has seen him become Director and CFO for the Eventus Advisory Group, who provide CFO services for SMEs.

In his spare time, Glenn is a triathlete and marathoner, is in the process of finishing a 200,000 word sci-fi epic novel. And, a little known fact about Glenn is he also wrote an indie film called ‘The Hanged Man’ which you can find on Netflix in the US.

Before we start, if you like what you hear today, please make sure to subscribe to tech for finance on your favourite podcast platform and on YouTube.

But, once again, thanks, thanks for joining me today, Glen. It's great to have you on.

[00:01:46] Glenn: Thanks for having me, Adam. I'm glad to be here.

[00:01:48] Adam: Very good. So we, we had a bit of a, a catch up before this session, didn't we? We, we, we had, half an hour a couple of weeks ago and, and we were introduced through our, through our friend Paul Barnhurst, who, who many of you know as the FP&A guy, who runs, FP&A today.

So Glen was recently on the fp and a podcast, I believe released today. I think it's the 14th of March today, just so that people have got that, that timestamp. and what we're gonna try and do today is not cover similar territory from fp and a today. So what I'll do is I'll, I'll link to the fp and a podcast in the show notes for this.

But we'll try and give a, a slightly different perspective so that Glen's not having to say the same thing that he probably says all the time. on his podcast though, . So Glen, it's up to you to, to correct me where you can and make sure that we cover new territory, eh, so, no. All, all good. So, so first, I mean, it'd be great to understand, especially given all of the, the upheaval in the market now around chat, GPT-3 and, and, and all of those sorts of areas for you specifically, who's been working in the AI space, or the finance and ai, AI space for a while, why did you decide to focus so heavily on the, the AI piece as, as part of your journey in finance?

[00:03:01] Glenn: Yeah, good question, Adam. I think it's, for me, the move towards focus on AI has really been just a focus throughout my finance career on whatever the latest technology is and being able to use that to improve finance. And I go back to, before I had my first CFO role when I was working in, in telecommunications in the early two thousands, I had, I, I was a thin ops guy that didn't roll up.

To the CFO, I rolled up to the COO and the, I was basically responsible for getting all the, budget information and the procurement and the purchase information for the COO. But at the time, the controller at the company was not keen to share financial data with me. So we were tracking everything in spreadsheets, so much manual work, and it was very difficult the scale of the purchases we were making and, and what we were trying to keep up with to manage to budget when you couldn't get access, in as close to real time as possible to the financial data.

So when we're keeping two sets of records, and everybody who's been in finance for more than a minute knows that, that that's gonna lead to problems when you don't have a single source of truth. So my, my tech focus came when. Had to fight the battle to get access through an API or, and I don't even remember how we were connected back then, in the early two thousands, but to get access to the financial data so that we could do our job.

And in the course of that, so I had to learn a little bit about kind of what was going on under the hood and how we could, . That was my first ex, experience with databases and me and my senior manager at the time, were using Microsoft Access and we were just pulling the information down and throwing it in a database.

And we went from reporting on lagging information to as soon as a purchase order was entered into the, accounting system. We had immediate access to it. We had as close as, as we could back then to realtime information, and it really changed. The nature of the work we were doing and the quality of work that we were doing.

So I, from my very first finance role, I've seen the importance of being able to get data out of systems to interact with systems and not rely on people emailing spreadsheets and transferring information that way. So that kind of started me. I'd say my, my tech focus, as I've learned more through my career about finance, it's technology has been joined at the hip with that.

And we've, in every company I've been in, I guess after I left Healthcom and went to smaller startup companies, I realized that the way, if you're trying to get the attention of whether it's private equity or, or banks for lending or wherever you're going out to raise money, or if you're looking to do m and a activity, that one of the secrets to that was to look like a bigger company than you are.

And that means more detailed financial reports and more detailed KPIs and really taking ownership of that. And I'm, I'm giving you a really long-winded answer, but it's gonna get to the AI prompt, . so in that, I, I found a way to connect, to our point of sales systems. And at this point, this is in mid two thousands, I finally now have.

Access in real time to, I was working for a retail company. I could see sales and SKUs and what was happening in real time, and we built this dashboard for our investors and for the management where you could go in real time and, and, look at all these results. We built a bunch of metrics and, it really, we, this company was built to sell and it was, I'd say a financially led company.

So I just really started diving in deeper and embracing what can we track? And we, we built our own, inventory management system. We're watching chemical levels and monitoring utility use. And, it was really, that, that was probably the coolest, tool that I had part in building was getting all this real-time information and being able to report these metrics and just having it automated because that freed up my time to shift my focus from just gathering the information and creating reports.

To, actually doing the more detailed analysis and being able to, whether we're talking to the banks or to the board or who, whomever, being able to give them really that next level deep, analysis that, that, you know, it, I think it made the company look better. It looked like operational excellence is, is a big theme in my career and something that I'd always have my teams strive for.

And I think with that we were able to show that we had a handle on what was going on with our business and we knew which levers to pull to change it. So that's that technology iteration. And then continued with small companies, small staffs, I'm always looking to automate. and one of my tricks to automating was, taking financial statements and throwing 'em into database tables and being able to do automated analysis on them.

And. Through all that , we're, we're getting close to the AI part. so, I went and got a, a graduate certificate in business analytics back in about 2018. And this was incredibly eye-opening for me because all the, I mean, I've always thought of myself as a great fp and a guy and, I love modeling.

That's my favorite part of the job. And I've built pretty complex models over the years, but, just in, in, in Excel and when I saw what people were doing with these machine learning algorithms at that point and how much better models they were able to build with that, it, I, I just realized, and, and the whole program, there was very little talk about using this in finance.

Most of it was, sales and marketing. It seems like sales and marketing definitely got the, got the jump there. And so I, at that point, made it my mission. It's, I cuz I, I like to think of fp and a. People as we're the original business analysts. I mean, think about all the models we've built over the years and I thought we need to be sure that we know these tools are out there.

And that sort of started my journey down the machine learning path as just if you can. Go from, you know, linear regression or polynomial regression or whatever you're doing to build your models. And now you've got, you're able to build statistical models that have way more features than you would've considered using maybe in your Excel models.

And you have different correlations, how accurate you can get on these forecasts. And so by using machine learning algorithms in finance and, and building out these models and, finding ways to do it, because I'm, I'm not a developer, I basically just know enough to be dangerous. But through all that I've been very, Tied into, what's going on in the AI and machine learning world.

And I took a Andrew s AI class and, and several others. Just again, not that I'm wanting to be an engineer, but I feel like it's incumbent on finance people to really understand the technology that's out there. So that all leads up to, where we are now with ai. And I think, you know, ChatGPT that we've been talking about is the most accessible form, the most, I mean, you know, recommendation engines and stuff that, you know, the net, I'm thinking back to the Netflix contest.

Mm-hmm. , for, you know, who could make the best recommendations or what Amazon uses to recommend similar products. I mean, machine learning has been out there forever, but it's sort of been. Quiet in the background. I mean, I think about the predictive text on your phone and on your, when you're writing an email, you know, that's been out there.

But ChatGPT is, you know, going to a hundred million users in however few days they did. I mean, it's, it, it's the most accessible AI that's out there. And so it just was a natural progression for me. And, I started this, my last project that I did very, you know, within days of, ChatGPT being publicly released, and I wanted to do something different because I knew a lot of people were gonna be, doing in, you know, using APIs and ma doing interfaces with, now they're out there for Google Sheets and for Excel and.

I really wanted to see how it was at coding. So my, my project that I just did with it, I had chat, GPT write all the code for this, application that, that F P N A specialists can use.

[00:11:37] Adam: Hmm mm-hmm. And I, I have, I haven't read the, the full PDF of you building the tool. I've got semi detailed, but again, I like you cause I've been reading your book, you know, you, you started with, you know, HTML and a bit of JavaScript and all that sort of stuff that, that's, that's basically, you know, my grounding in, in coding.

And as I said in a previous podcast, I, I have been plugging ChatGPT into stuff like Google Docs and Excel, but I've just been copying from GitHub. I've just been changing the promise over so that, that's the limit of my development knowledge. but I'll link to that into in the show notes. Cause I think it's very interesting if.

The end result of producing the tool. It's the, it's the methodology that you, you, you applied during that process, you know? So I think some of the ways that you handled, asking the questions and using the correct prompts worded in the right way is something that everybody can learn from.

[00:12:32] Glenn: Yeah, that's a great point.

And there's, I've , if you've been on YouTube in the past two months, there are a million, of course, this is the YouTube algorithm has figured out that I, am interested in this stuff. But, there's a million videos out there that are just training you how to write proper prompts for ChatGPT, which I thought was interesting cuz I remember, I'm old.

So, I remember when, internet search was first out there, you had to, you know, you would use boos or you'd, you'd have to do things like if you wanted it to search for two words, you know, you'd say, financial plus planning plus analysis, or you'd have to put things in quotes. And it was very, you know, you, you had, there was more, science required to get good search results.

And Google and, other search engines have figured out, you know, how to return better search results with all, without all that now. But I think ChatGPT in this nascent early stage, it's similar. You have to know how to, how to prompt it to get the right results and how to get it to, in, in the case of writing code, how to get it to QC the code and fix errors that it originally created.

That was an interesting part of it. And I, I left that out of the paper because it, the paper would've been, you know, three times as long if I put all the error correction in there. So, you know, I don't know, maybe the magic of post-production. I made it look a little easier than it was , but I did, I did not write a single line of code for that chat.

GPT did write all the code. There just was some, some bug fixes that happened in the background that I didn't bring on the paper. Yeah,

[00:14:02] Adam: there's always some background bug fixes, isn't there? No, that's fine. And, and we might get back around to the concept of prompt engineering later in the conversation, even though it's, I dunno whether it's a term that I'm fully accepting yet, because there's a part of me that thinks that.

The role of a prompt engineer shouldn’t exists. These, you know, you, you shouldn't need to think about the questions that you should be asking. You know, the, the AI should get to the point where it can understand context, but either way, we, we can, we can come onto that if, if we need to. But I just wanted to backtrack a little bit and the reason for the next question, and it, it wasn't in the list of topics that I sent you, Glenn, I, I apologize.

But as you know, I do a lot of work with finance teams, and running a podcast like Tech for finance, with the advent of stuff like ChatGPT, people are coming to me, you know, saying, you know, we wanna leverage these technologies. You know, we want a bit of guidance, you know, where, where do we start?

And it's very difficult because, you know, you and I sort of breath this tech on a day-to-day basis, but finance teams that may be working a slightly dated way, they're still quite a big jump to go from legacy applications. To an advanced business that is making strategic decisions with AI that's plugged into to their data.

And I just wanna go back to a point quickly from, from your book, and you must have people quoting from your book all the time, so sorry, sorry. But, and, and I won't read the, the whole quote, but there's a section that says that there's four stages of digital transformation. So you've got ai, but then you've got robotic process automation and machine learning, you know, within that as well.

But they're all small pieces in a larger puzzle of digital transformation. And like any good puzzle, it must be approached intentionally and strategically. So what I'm saying at the moment is you can't just plug an advanced AI platform into a dated process. Where either the people or the foundations aren't producing accurate data, cuz, because it's all, it's all about the data when you look at it.

Right. And I'll let you respond in a second, but before I forget about it, the point that you made there about, you know, you used the, the point of sale project as an example. So you had to get data from point of sale systems. And I think I read in the book, whether it's that project or another one, the, you, I dunno whether you got fed up, but your main priority is I just need the data.

You know, I need the data to, to, to make these decisions. So I don't care who the software provider is, you know, providing the, the point of self front end. I just want somebody to give me access to the data. And I think that's the right approach to take process data first and then accuracy before we can start getting into these advanced tools.

So I dunno whether you can speak around that a little.

[00:16:43] Glenn: I can, and if, if I start rambling again, just wave your hand and say, Sean, you're fine. Glenn. So, okay. So let me back it up, because this has been a pivotal part of my finance career and it's probably the weirdest thing for a CFO to do when he comes in on day one.

But , when I show up at a company, the first thing I want to do is, and I call it an ISO9000 audit, and it, it was based on that. And again, I'm showing my age, with, with talking about I s o I don't know if anybody even talks about that anymore. But the purpose of this audit is not to, get ISO certified.

That's not the businesses I'm in, that. Wouldn't matter for anything, but it is a nice framework and really all that I'm doing, and I'm just taking components of it, but I wanna look at the processes where there are gaps, where there are manual entry, where, you know, potential error points, what people are involved.

And what I want to do when I come in as the CFO is understand what information we have, what we're collecting, what we're not collecting. And this goes all the way back to from prospect to lead to wine, to bringing them into the system and putting 'em in, whatever, a, a provisioning system, a CRM, well, CRM may be early, but, a, you know, project management tool, the accounting system, all, everything that's going in.

I wanna understand that data flow and see where, maybe we're, we've got customer information in the CRM and then we're. Redoing all that, you know, manually entering it into the accounting system where somebody could fat finger an email address or mess up the billing contact, or you don't have the billing contact in one system and you, you do in another.

And the reason that I want to do all this is because, When I start thinking about when I'm gonna have to report to investors, to boards, to the rest of the management team, I wanna know what that world of data that I have is, and then where there's any gaps. And it would be really nice, you know, for our customer acquisition cost, if we could tie back and, figure out how much we spent, you know, on the front end on advertising or whatever the case is.

Or if, I have some information about the customer that's in one system and not in another. and I can't see if there's a customer that is late on paying their bills. And I can't see in the c r M system that they've actually had a lot of support issues with us. I mean, it, it, it just, it all feeds together whether I'm reporting on it or acting on it.

So step one is understand that process and then, because if you just throw, and whether you're doing an ERP implementation or trying to bring in new technology like AI or, or machine learning into your tech stack, It's garbage and garbage out, right? So, it's first understanding the data you have, understanding the process flows, understanding where, how you can integrate that data across all the systems.

And then you can start looking at, and I don't mean pick a software platform and just throw that at it and say, this is gonna fix everything it is. Figure out the requirements first, and then have the software or whatever you're adding to your tech stack, make sure that it is meeting the needs that you need.

And you're not just jumping on the latest bandwagon, or you got a really good salesperson for some off the shelf, accounting software that, you know, did a good sales job. Just make sure you understand at the fundamental level what it is you're trying to do and what you have currently.

[00:20:23] Adam: Mm-hmm.

And I think it's, it's, it's so important to make sure that there's, there's that initial benchmark to say, This is where we are. And then as you say, instead of just lumping a load of, you know, stuff that you can justify business case with, you know, all this AI tool is gonna enable us to save, you know, 20% of our day, or, you know, 70% of our day, the, the is the questions first, right?

It's, you know, we know what our baseline is, we've got that benchmark. What questions do we need to be asking to produce the results that we need? And then fi fill in the, fill in the gaps from there. Because wh when you start getting into RPA territory, and I, and, and I suppose it's wrong to say that you can't use AI full stop unless you get your process sorted because there are some tools.

There are there to improve accuracy of data input, for example. You know, so, so even basic systems now you can set up with a validation that says if somebody tries to overlay this dimension against this GL code warning, you're not, you're not meant to do that. So I don't wanna misspeak and say that tech technology can't support this sort of stuff, but it'll only work if you first define the process and the end goal for what needs to come out the other end.

Right.

[00:21:35] Glenn: A absolutely. Absolutely. And it's just, it's amazing to me how, I mean, it's for, for me, I'm enough of a geek that getting into that process level and doing your workflows and, and, diagrams and, you know, showing how data is passing through, I love that stuff. I, I don't know that I'm normal in that, but.

You can't, you can't make decisions without the, the full picture of the landscape that you're, that you're doing it in. So, yeah, and, and, and to your point, there are systems out there that are kind of made, and I think as AI improves, you know, maybe systems will make imple, you know, implementation of these systems will be become easier because the tool that is that we're buying to fix this problem may be able to be more intuitive and help us figure out what we, we don't know.

But for, for now, and to me for the foreseeable future, it's gotta start with, your foundation of processes and understanding what data you have and, you know, you could have the same data and, and multiple systems. So which one is the source of truth? And when you're feeding it in, that's something very important to keep up, to keep track of.

And especially if you've got systems going back and forth talking to each other. And like you said, there are tools that can keep you from overriding one, but if you don't give that thought before you put the tools in place, you could have the wrong system, you know, keeping that piece of information and overriding the good information with bad data.

Mm-hmm. .

[00:23:03] Adam: So if, if we imagine that, that people have got the, the process piece sorted and we've got a tick box that says we are pretty confident that the data input's accurate, we know what a master source of data is, we know what the relationship is between our sy between our systems, you know, and whether that's a, an end-to-end ERP or whether it's a couple of applications that we do have completely integrated.

Yeah. So we know that we are working from a single source of the truth. What would be a good starting point, and again, taking the ChatGPT hat off for a second, for people that are wanting to get a, a feel for how potentially machine learning or, or some sort of predictive tool can help start. Allowing them to spot stuff that they wouldn't otherwise know.

So I guess the short version of the question is to people that have got the data sorted, what's the entry level step with more advanced AI tech?

[00:24:02] Glenn: Yeah, so the entry level step is basic statistics. I mean, and that's what's so crazy, and we, we don't think about this, but even, even GPT is a statistical model.

It is a, it is predicting what the most likely next word is based on a series of words. And, you know, trained on billions of parameters. It, it gets pretty smart at this, but I think before you start diving into doing support vector machines or, or more complex machine learning algorithms, understanding basic statistical models that would let you.

You take a, a list of features. You don't know if, if there's correlation between all these, variables and what you're trying to measure, but basic statistics lets you figure out if there are correlations. And then, and I'm thinking of just building a, a, a typical statistical model. You find, out of your, I don't know, a hundred, 200, however many features you have, and, which features are correlated to what you're trying to measure.

And then you find out which are most correlated. And then you build a model, a statistical model that makes predictions based on what you have. And I think without an understanding of that, it's, you know, once you understand that, then you can look at, I mean, and there's all kinds of drag and drop and drag and drop.

I don't mean that. Even me necessarily who lives and breathe this stuff could go in and just easily do it. But I'm thinking of tools like TensorFlow and Keras, Pie Torch and the, tools that have made these machine learning algorithms much more accessible to people who know what they're doing. it's easier than it used to be, and obviously with the compute power being what it is now, that makes that easier.

But I think before you can even look at, before you even need to look at more complex, machine learning models, you need to understand the basic statistical models. And I don't know what, you know, what your audience demographic is, but for the world that I live in, small and mid-size enterprises typically don't have enough internal data that would even justify doing more than a basic statistical model.

But if you do, if you're, you know, an IOT company that has all kinds of data coming in from your devices out there, I mean there are, you know, there's all kinds of examples where you would, but. To think, especially if you're someone who hasn't been following it, to think, oh, we're just gonna throw some, we're gonna throw AI at this, which people would normally mean just machine learning algorithms and, and try to figure it out.

That's not the place to start. it is a basic understanding and I think there's tools out there, and I love this tool. I had access to it when I was a student. It is very expensive for an individual , but DataRobot is an amazing tool that lets you dump your data in, it'll run several different machine learning algorithms on it to make predictions.

You can pick the model that you want to use and, and, you know, use your data, that way without having to be a true data engineer. But the danger is, it's a, it. If you're just throwing this stuff into a black box, and I know neural networks are always kind of in just one type of, machine learning, but, you know, neural networks are always a black box where you don't, you can understand the process that's going on, but it's, it's always kind of magic what happens in those, in those middle layers in a neural network.

But if you're just throwing stuff in a model and then presenting that as if it's fact, it's like, it's like believing when ChatGPT has a hallucination and tells you something that's wrong. I mean, the machine will kick out with seemingly great certitude. What what it wants you to, you know, what, what it's trying to tell you.

But if you don't understand what's happening in the, in the model that you've picked, how can you be, how can you trust the results that come out of that?

[00:28:01] Adam: So, and, and you're dead. Right? And, and it's, it is a bit of a tangent. So I was, I was listening to. I can't remember who the, the, the dude was. But, they were talking about some of the early stage, Meta AIs that they were working on, which again, they, they were, they were language models similar to, similar to ChatGPT, just not quite as advanced.

and they used the example of the AI producing a justification for how eating glass was good for you and, and intuitively that we, as we as humans know, you shouldn't eat glass . Right? I mean, but, but to your point there, it, it included citations, you know, it was that inclusive and convincing in its response that a first glance, you'd think, There's a point there.

You see what I mean? And, and that's, that's again, and we're going on to a different topic here, which is obviously how much can you trust the machine at this moment in time? But that's why whenever I'm speaking to people, it, it comes down to that validation. Validation, you know, check, check, check. Because just because something's giving you a convincing response doesn't necessarily mean that it's right.

And it's the same with people as well. People can give you a convincing response based on their credibility, but they might not be right either. Yeah. So again, it's I guess something to, to watch out for. But, but you're dead, right? You know, applying a machine learning algorithm to accurate data is one thing as an entry level, you know, providing, as you say, you've got the, the volume of data there, but then applying something generative, you know, with the advent of GPT-4 now that's now got access to live data, you know, there's a shed load of information on the internet.

It's just damn right. Wrong.

[00:29:50] Glenn: Yeah, absolutely. And that's, and I'm, I am. an evangelist for all this stuff. I, I love it and I'm so excited about chat. I, I keep a, a ChatGPT tab, opening my browser all the time. I'm just always testing it. . Same. Good. That said, I have, never put any actual production data like any, any, you know, proprietary information or anything into chat.

GPT I always, you know, I've, like when I did that project, I would dump my, CSV files or I would just copy and paste the table in and see what ChatGPT could answer on that. But it is, it is a novelty at this point to me. I mean, I don't think that, The publicly available interface on the web that you use is a novelty.

Nobody, and I think, the companies, who have said, do not use ChatGPT for company information. I think they are right. But, I think it's a novelty right now. And it is a, we can call it a proof of concept to say this is the power of what this can do. But until you have, and there, there's already companies out there that are doing this.

And I think that, that over the next 12 months is gonna be the very exciting next phase of what generative AI is doing. There are companies who are taking the GPT base and applying it to different industries, different professions, and making it experts at whether it's law or medicine or FP&A.

And, so then you could, and if you could, you know, figure the, the, whoever is the startups that are applying GPT in this way. Figure out a way to protect your proprietary information, give you this interface that's a chat bot that you can ask. This is my dream. And I think it's getting closer to reality every day, where any stakeholder can, you know, the same way we have dashboards, visual dashboards right now, you know, dashboards are great.

They give you all this information at a glance, but drilling down sometimes, you know, in Tableau or whatever, you can have one level deep that somebody can click on and see. But to really answer questions, you have to go to an analyst. but I think in the very near future, in a situation where you have this chatbot that can operate.

In a closed system that doesn't expose your, and this, this is, leave out the compute part of it that's, for engineers to solve more than me, but . but that you basically can ask just like you would your, Amazon device, you know, ask it a question about your data and get results, whether it's talking to it, typing it in, get a report that may be in a, in a current environment, you have a report request, it goes into the, backlog in the finance team and whoever's doing your reports, it may take 'em a week to get to your request and get it back to you.

this is gonna be real time, good financial information that you will be able to, to trust more than just plugging something into an open, you know, HTTP interface. So, yeah.

[00:32:50] Adam: And, and that's, so if, if I could summarize it at the moment, and may maybe I'm off, I dunno, I'm, I'm by no means the expert in this, but the, the best term for me at the moment is co-pilot.

And I know that's an official term, I think in, in GitHub. So, so I had Adam Sk, who's, who's building, an FP&A tool specifically for SaaS companies. So you can't talking that industry approach that you mentioned there. He's doing that. and he's built ChatGPT into his, similar to Chris Bell who's building Chat to Xero as well.

So, so it's similar concepts. We, we are slowly seeing that emergence as, as you say. but the reason I, I tend co-pilot, and it goes back to our, our previous conversation about predictive text, right? You know, when, when, when you take all the fills away at the moment, it is kind of, we're into that, you know, the AI is just kind of filling in the gaps for you.

So GitHub co-pilot. Obviously it slashes the time it takes him to code, because as he's coding, it's, it's using its intelligence to say, oh, we think that you're doing this, or we think that you need to add in this snippet and we think that you need to do this, that, and the other. I've also referred to it as, like a, an an AI assistant sort of outsourcing to, to an AI assistant.

but I think to, to rely on it at this stage, and again, I welcome, I welcome your perspective here. So going back to your points about the jigsaw puzzle pieces and the concept of, you know, AI doesn't exist in a silo. You know, there's rpa, you know, there's the other fundamentals in terms of automation that, that we can use as part of this wider transformation.

You know, we need to be careful that we don't confuse stuff like generative AI as a, as an R RPA replacement. Because, because they serve two completely different purposes. So RPA is great. If you have a structured do this, then do that, do this, do then do that. If you tried to give that instruction to a generative AI model like ChatGPT-3, you're gonna get wildly different results and it's just not fit for that purpose.

I, I dunno whether you'd agree from, from your experience with that as well.

[00:34:56] Glenn: as of state of tech today, yes, but I did, I was over the last couple days just seeing what we can expect from GPT-4. Yeah, I was, you know, reading some of the Google paper, well, not, not necessarily on GPT-4, but on the sort of the next generation of where GPT is going and the models, whether, you know, whether, Facebooks or Googles, or, or OpenAI, are getting better at making inferences and at understanding jokes, at, at taking things out of context. And you, a lot of the papers would go side by side with, you know, existing GPT-3, the res response you get where GPT-3 says, I don't understand.

And the In GPT-4 or similar, I mean, you're now starting to get to a point where on some of these responses there's that they're already above sort of the midline of, what human responders can answer. Mm-hmm. , and getting close to the, the top of that line. and, , the way that they measured it was, you know, mechanical Turk on, the Amazon where you hire people to go through and, and do jobs for you.

So they, as part of the training of the model, they hired a bunch of mechanical Turk contractors to, answer questions and then rate them how they answered compared to GPT. And so, GPT-3 stayed kind of below the, midline of that. So, you know, like it's writing is more, you know, it, it's very basic elementary writing styles, and in the same way it does everything.

But with these newer iterations, you're gonna see. The responses that come from the computer get closer to the top of that curve. And I guess once it goes past that curve, I don't, we're probably close to the singularity not, not really. Because, and actually I do, I do wanna, I do wanna throw this out. I mean, that will be great when you can get a response from a computer that is maybe better than 98.

You know, it's, it's a mensa level computer. Yeah. but it's very important for all of us to remember this is a large language model doing predictive text. It is nowhere near even on the spectrum of sentient. So, you know, it's . and I think, you know, for, thinking about Ray as well and the singularity, and I think that that is very different than what you can just get from a predictive text model, which they, they're getting.

Really good and they're gonna, it's gonna be amazing to see where this goes because it's moving so quickly where it goes in the next few years. But, I don't think we're right on the doorstep of, our, AI having, sentience and personalities and, and all that .

[00:37:39] Adam: Yeah, Yeah. That's it. And, and, and to be fair, I think going back to the point I made earlier about, you know, this, this whole sort of hype about, you know, the need for prompt engineers and, you know, the, the people that are basically telling other people how, how to ask better questions of ai, you know, is that that's so matter, isn't it?

You know, you've got somebody telling somebody else how to ask better questions to an ai, you know, ask us this 10 years ago. Well, we have, even, we, we wouldn't even thought that we'd be in this situation. Right? But the reason I come back to that, and it's a, it is a point that, Both Chris and Adam on the previous podcast pointed out is the biggest limitation we have with the likes of ChatGPT and whatever the similar models are out there is it's in the user interface at the moment.

You know, it is, it is, as you say, you've got your tab open, you know, but we're, we are copying and pasting text. You, you, you know, that that whole interaction is in itself clunky and doesn't make for the best responses because the minute you've gotta think about how do I ask a question is the minute that you know, you, you're kind of spending time and attention that you shouldn't really be spending on, on this sort of thing.

So as soon as we get to the point where, and I'll use the example of my notion for example. So I use Notion, bang on about it all the time, but my notion is a massive repository of my brain. Essentially, my thoughts go into it. So if chat boutique can go into my notion and it understands the way that I speak in my sentiment, In theory, it shouldn't have to guess as much when I'm asking chap GPT questions because it knows my language and it knows how to infer the meaning behind my words.

So I'm hoping the improvements come there to the point where we don't need prompt engineers because the AI should be intuitive enough to be able to understand the context. But again, I dunno how far we are off that.

[00:39:28] Glenn: Yeah, and really, I mean, if, if you think about the way GPT was trained in a specifically taking it from GPT the language models are currently being, or chat, GPT. Was trained first on, just doing the predictive text, and then the next round of training was human interaction.

And so humans would go and, put in prompts and get the response from, or they'd get, say, three or five responses from GPT, and then the humans would wait, which was the best response. Mm-hmm. , you know, all the way down. And that's how they got it to be much better at interacting with humans and understanding the prompts and what to put out.

So in your example, you could take another layer of fine tuning and say, I'm, a podcaster and an FP&A expert, and these are my notes, and review my notes , and understand the kind of things I'm talking about, the way that I structure sentences or whatever. And then fine tune the model on that. And maybe there's even some example where you and people at your company are using, doing the same kind of training that people did with, ChatGPT, to get that first level of training.

And, and it's just further fine tuning the model to be specific to what you're doing. So I, I think that's kind of a logical next step, for, what, what's to come with these future iterations and very industry and profession and even potentially person specific, GPT models.

[00:41:16] Adam: Well it come, comes back to that, you know, virtual AI assistant doesn't.

You know, we, we will get to the point and, you know, to come back to your point about the singularity and all of that sort of stuff, I mean, it, it will be become, before there is replacement, there will be augmentation, right? And we've already got, studies of Japanese companies that are getting AI to read human thoughts, you know, and, and all of that sort of stuff.

We we're entering scary territory there. But I always kind of like the, the thought of, I dunno whether you ever watch Dragon Balls Z or anything like that, you know, it's, no love Dragon Balls Z. But anyway, the, the, all the characters in Dragon Balls Z, they've got like a little lens that goes over the right hand side of their face.

It's like a, a heads up display. It is all that. It is, you know, the equivalent in modern day is Google Glass. And, and I think RayBan and. building, Facebook portal or meta portal into their glasses now. and to me, I think it's, it's gonna be an iteration of that. You know, before, before we have any of this, you know, sentence or anything like that, I think it's gonna augment our capability on an individual basis whereby we are getting the best of both worlds.

So we've got our human brain, but then we've also got a model like chat, G B T, that's feeding us with knowledge that is giving us guidance, that is being that co-pilot that we mentioned earlier. So it's gonna be interesting to see the way that that develops, I guess. a little bit now about the pie in the sky stuff, which, which I love talking about and it's nice to, to have an eye for the future, right? But when we now think of practical applications of some, some of these platforms, so we, we've said in earlier parts of the discussion, the, obviously the starting point is foundation's data end goal. Where are you headed?

Then the entry level in terms of AI is to start building some of these machine learning, models into our data set. Whether the end result is, you know, predictions from an fp and a perspective or, you know, spotting trends as maybe not even just a finance team, but as a wider business that's trying to spot trends from data from different departments and that sort of stuff, which is fine.

So, At the moment, I see technologies like GPT potentially having more wins for individuals over businesses and, and coming back to the co-pilot piece, you know, with the softer stuff like, you know, writing job descriptions and the stuff that it actually, the, the text based stuff, that's, that's where I'm seeing the real time saving because it's not relying on pinpoint accuracy.

Yeah. so that's, that's the first piece. I guess in our last conversation we talked about chat, GPT being a really poor mathematician. So, I think you, you can probably speak to the, the Wolfram Alpha piece that we spoke about last time. And then the, the last question that I've got before I ask my standard set of questions, is what you think the shift is gonna be with a multimodal solution like GPT-4 against what we're seeing at the moment.

So the first piece is, Practical applications for GPT as you see it now, and whether my thinking is aligned with yours or more of the softer side of things, and then your perspective on what we're gonna see in terms of shift with the the later generation of GPT.

[00:44:28] Glenn: Yeah, and I think a great example would be, I would use GPT and I, I do similar things with this.

So say I pick up a new client that is an industry that I haven't worked in before, and before I have the introductory call with my new client. I might say, what are some key metrics that we need to look at in, the manufacturing or construction industry, say? Mm-hmm. And it actually can, you know, it will give you some pretty good information.

And then you, the great thing is you get some sort of high level items and you can, with the prompts, you can say, give me some more ideas or drill in on this. And so if you are, I mean, I think about right now, if, without chat, GPT, if I'm trying to figure out key metrics that are important to a client or an industry, or, you know, depending on the scale of a business, the kind of things I'm looking at, I'm, I might still go back to Google and, you know, see what, what other people have written about it.

But, Being able to ask ChatGPT for things like that, it makes it much quicker. Or, like I was setting up a, a new client intake form this morning and I asked, ChatGPT what are, what's some good information? I'd like to just to make sure, to kind of check myself what's some information I'd like to include on a client intake form.

So all that soft skill stuff is, is great and it really does facilitate, or if I'm, you know, I, I contribute a lot to different publications and I would never. First off, I'd be embarrassed to think that somebody would believe that I wrote as badly as ChatGPT. But, but I have used it for ideas. If I have to, you know, do a new Forbes article every month or whatever I'm submitting to, I might say, “Hey, I'm, I want to target, private equity backed businesses. What are some, article ideas you have for me?”

And then I would take the idea and use it. So there are ways that individually, like you said, you could, find to save, save time and, and be more productive. But if I were wanting to run a model, just like I wouldn't just drag and drop a data set without thoroughly understanding the dataset into something like DataRobot, where I don't thoroughly understand, wh what the processing of that data is doing, I would never do.

And then report that to anything that mattered. Like, you know what, actually the, the key part of my job, because we are, we have to be very accurate and precise, but I, so I wouldn't do that today, but I really do think it's gonna be sooner than later that we're, we are doing just that with, with production data, you know, once we're insured and you'll still have to, it's, we're not even to trust but verify, yet we're like, I, I think I started with don't trust

But yeah. But I think, you know, we'll get to it and then there's gonna be a, a long period where, okay, the, the robot says this, but let's understand why it's saying that and check it, check our answers against it. So that'll be a big step when we get, when we do get to trust, but verify where first blush it seems right, but let's figure out how we got there.

So we we're not, we're not close to there yet. Yeah.

[00:47:49] Adam: And the. The, the Wolfram Alpha piece then, we don't have to spend long on this, but of course, you know, Wolfram Alpha is quite a, a, a well known, facility that is, I mean, you are probably better to describe it than I, but there's, there's a theory that says, you know, for what Cha GPT lacks in, you know, mathematical knowledge, potentially a, a means of plug in the gap by connecting it with something like Wolfram Alpha.

[00:48:15] Glenn: Yeah. And that's important because, in GPT-4, again, we're gonna say it will be better at doing, slightly more complex math. But the way, the way that the. The GPT model works is very different than what a math engine would do. It is saying, oh, I've, I'm making inferences based on this number you gave me.

Or, you know, this trying to, you know, maybe it's getting better at reading word problems, for example, and, you know, for simple math like that. But if you're trying to at scale, have a computer system that is generating your key financial metrics, your ratios, the everything that you're tracking, that's like asking the librarian to do your physics somewhere for you.

You know, it's, it, it's probably, she could, she or he could do it, but that's not, you know, if you're gonna really ask someone to do your physics somewhere for you, it'd be great to find a physicist to do it. So I think that the integration between a large language model and a math engine, and if you wanted to make it a, a full purpose, like, you know, this is pie in this sky stuff down, but eh, it's not really, cuz everything's moving so quickly, but you could have the chatbot as the front end. And even I, I'm seeing people already doing this where they're building kind of a decision engine behind the chat prompts that then say, I think I can answer this, but this is a, you know, statistics question.

Let's kick this one over to Wolfram Alpha, or I think I can answer this, but the user is trying to play, Alpha Go or play, go with me . Let's use Alpha Go. I mean, that's just, you know, one example. So it did it. You know, if you had a decision in engine behind that chat interface and it could direct those queries, then you start having someone that, or someone , I'm already personifying it here, , you start having a, a system that can interact with you on all kinds of levels.

And that's where it starts to get really interesting. And then, you know, going back to our singularity reference earlier, may not be sentient, but it's gonna look and feel sentient when it's able to interact in all these levels of your life.

[00:50:32] Adam: So I suppose that leads nicely onto the second part of the question, which is, you know, the potential for, for GPT-4 is the next iteration.

And, and, and I'll go back to your, your book for a second here. so your book touches on the subject of, of narrow AI. Which, which is, which is basically AI that's pointed at a specific purpose, right? You know, and, and maybe to use, an example that more people understand, more and more. Now, when you're using teams and Zoom and all of this sort of stuff, there are AI that sits within them that will record the conversation and then the AI will do the transcription.

And then we're even now getting to the point where the AI won't just do the transcription, it'll pull out the key points, and it'll also do a summary for you as well. So correct me if I'm wrong, but to me that's a narrow AI because it serves a particular FO focus and it's developed for that end in mind.

But it strikes me that chat, GPT potentially expands beyond that, whereby you've got. A type of AI that doesn't just extend to stuff like audio because we know it's multimodal now, which just basically means it's going to be able to accept more inputs or types of different media inputs and be able to output different types of, output that aren't just text.

Right. So what do you see as the key difference between a multimodal AI such as GPT-4 compared to a narrow AI such as the example that I gave just there? because the question that people will be asking themselves is, well, it's, it's the question people always ask themselves. When I get invest in new systems, do I do, do I get one system that does everything?

Or do I, do I have, you know, a pointed solution for each of those different business cases? So I dunno whether you can talk around that a little bit.

[00:52:24] Glenn: Yeah, and it's not, the funny thing is, I think even when I wrote the book naively, maybe I , my perception of general AI was that it was synonymous with sentience and it's, it's not, now, now, you know, the way, the way that I look at it now, but it, a general AI can perform broad range of functions probably, you know, I, I don't know the super technical definition of it, but would it say that this AI can perform every function that, a human could do?

And then you have this one piece of software, bot or whatever it is that can. and do your physics homework. It can, give you recipe ideas, it can transcribe your meetings and also drive your car or, you know, place an order, online for you, for, for delivering. I mean that, cuz you know, cuz self-driving cars, that is a narrow AI that it just is made for driving the car.

And, but the more that you integrate these, and then the other thing I wonder is, so, but we're also talking about linked narrow AI versus one combined system that can do all of this. So there's, there is a difference there. So a large language model is very far from being general AI, but if you can cobble together enough things that can all interact with each other, it's, you know, just like we have different parts of our brain that have different functions, it maybe starts to look and feel like general AI.

But, this single, you know, unified AI that can do everything. And I guess that's really the difference is if you are having to cobble together different machines and, and, and different programs, does that come as general AI? I think no. Even if it does look like it. So you're looking for, you know, that would be the holy grail of this.

AI does everything internally, so, and with, but in a connected world, does it matter? I it's really about what the end user, what the experience is and if it feels like general AI and then…

[00:54:39] Adam: Absolutely, absolutely. And, and again, going back to some of the use cases, I mean, some of the LinkedIn post there's conversations go and say, oh, I'm keen to learn about what AI can support me in developing my presentations.

You know, so, you know, as you say, if you do have, even if it is a combination of narrow AI tools within the tool set, you know, whether. Somebody uses two systems or one, you've achieved the same objective, haven't you? So I've got a tool that helps me visualize my data in for the purpose of presentations.

And then I've also got that same, same tool that also helps me transcribe my meeting notes, you know, and all of that sort of stuff. So, you know, we, we'll have to see whether those different types of input, you know, then add up to ChatGPT learning more and whether it does become a little bit more autonomous in, in its inputs.

But again, it's still, it is still unknown territory, right? Yeah. So, okay. That's, that's fab. So to, to be continued I guess. I mean, Yeah. To me, and again, I don't really wanna make predictions, but we've mentioned the concept of GPT-3 as being kind of like a copilot, you know, an AI assistant, you know, great for potentially more individual use than trying to automate business processes.

Right. You know, I think, I think you and I kind of agreed on that, right? I think with the advent of GPT-4, if it can produce imagery and it can recognize audio and that sort of stuff, we're, we're potentially just seeing that on steroids. There's potentially not wider applications apart from just being able to communicate and producing more outputs.

Yeah.

[00:56:15] Glenn: Yeah, I mean, it's really, this next wave of what the applications are is gonna depend on, very creative engineers and, and founders who are gonna take this to the next level. so, you know, without playing around with it, I've certainly, read a lot on what it's gonna do, but it's, it's going to feel a lot more like, you know, that executive assistant that can do a lot more.

You know, and I could see it being, I mean, we're, most of us are doing typing interfaces with it, but imagine this power just behind the Amazon devices that everybody has, or the Apple home or whatever, you know, Google Home or whatever devices they have. If you've got that level of power, that you can chat with, I mean, it, it just changes.

Again, we're thinking individual here and not. Company. but I, you know, I think you can extrapolate from that how if you had a system like this that could go through and just like dashboards kind of really changed FP&A and and reporting company metrics, this is gonna be the next wave that is even bigger than dashboards because you can ask it anything about your data.

[00:57:30] Adam: Yeah, hundred percent. And, and I mean, whether it is individual or business, I mean, one of the primary focuses of these tools is free and up time, right? Yep. You know, and, and, and, and I'll, I'll come, I'll come back to a quote from your book again. which is…

“Knowledge is power. And by understanding how to use data and automate the most menial data related tasks, you can gain more power and share that power with your team. Invoicing clients, paying invoices and doing reports aren't powerful job responsibilities. Not when you compare them to taking all of that data that you have and combining it with other data in the company and becoming a strategic partner.”

So, so I dunno about you, but I don't care whether I get that time back from having an a personal assistant, or whether I'm running an AI model against my data to help me with my predictions.

I'm still freeing up my time so I can become a better business partner, right?

[00:58:20] Glenn: Yeah. And it's all about moving people from mindless work to mindful work. And the question is, at some point, you know, when the AI gets better and better, it's like, how mindful do you, are you gonna start excluding large swaths of the population that maybe can't operate up at the you know, in the 98th percentile or whatever that the computer's doing.

So it's, I mean, you know, we don't need to get into a whole universal basic income conversation. stuff like this makes it, you know, makes it seem more like, somebody might need to start planning for that.

[00:58:49] Adam: Yeah. I'm not gonna think too heavily about that. I don't, yeah, no, appreciate that. Again. So, so I'll ask you the, the questions that I, I always ask, and then you can tell people where to find more, more, more about you.

you, the response to this answer can't be your smartphone. Yeah. So I, I get that response. I won't say it's a cop out because smartphones are useful. but the question is, you know, what? Apple gadget in either your personal or professional life, Could you not live without?

[00:59:25] Glenn: So, if people, if people are watching the video, they'll see that my video looks different now than it did when we started, had some technical issues.

The, internet crashed in the middle of the recording. but so my response would've been because I've spent the last week setting it up. I've got a new office, I've got this 4K camera, I've got, my, podcast, microphone, my headphones and everything set up. And it was, I. Really, two days ago I thought, this is perfect.

I've got my huge monitors. I can, you know, be doing everything at once, . So that was gonna be my answer, was this, great, new setup that I have with my, computer, but since it's crashed, I've gotta, I've gotta kick that to the curb. This is not my favorite technology , in my personal life, I, I'm a runner and a biker, and so, activity trackers are, I rely on those a lot, but it's, I've been doing this experiment and it's driving me crazy and I think you'll be able to relate to this as a numbers guy.

Yeah, so I've got my phone, my Garmin watch, and my Apple watch. And I've got certain routes that I run, you know, multiple times a week. And on every one of those routes, every one of those devices gives me a different answer to how far it is . And I know, so if I use my phone, it actually measures that it a further distance.

So I like to use my phone because then it makes it look like I'm running faster. I'm . Yeah. So, so even my favorite devices, clearly I'm, I'm fighting with technology every day, which is, kind of the story of being a tech evangelist, right?

[01:00:57] Adam: No, activity trackers are great. You know, I'm, I'm a big, I'm a big a advocate of, you know, you've gotta get the health right before you can get anything else.

Right? Right. You know, it's, there's no quote, oh, what was the quote? It was an amazing quote. somebody that's healthy, it's something like somebody that's healthy. There's, you know, endless possibilities or they can think about everything. But for somebody who's unhealthy, they can only think about one thing.

Well, that one, that one thing is getting healthy. Right. You know? So unless you get that right, you, you can't. So I'll add a part two to that question, which is, is there anything in tracking your metrics that you learn, that you've been able to improve from, that you wouldn't have had if you were just running without a tracker?

[01:01:45] Glenn: Yes. So when I first started running a million years ago when I, back when I was racing dinosaurs, it , I would just, my workouts would be, I don't know, I'm just gonna go run as long as I can or run as fast as I can. I might mix it up with some speed work or whatever in the, in the middle of it, but didn't think anything about heart rates.

Mm-hmm. So now through watching my heart rate and my level of exertion, I can tailor my training for, okay, this is my long run. I need my heart rate to be in this band when I'm doing speed work. This is my max heart rate. This is where I need to be if I'm doing a hit workout or, or whatever it is, or , you know, if I just wanna see how much work I did pedaling up my bike up a hill.

It's pretty amazing to see. the, you know, because you can actually, with these trackers, you can see your change in altitudes and you can know you climbed at this angle, this was your heart rate, this was your speed, and you can start to sort of factor in and you can really fine tune your, what you're training for.

So, yeah, I mean, it's, you know, and as a data geek, I'm constantly, you know, trying to figure out what, what I can surmise from looking at my heart rate and distance and speed and weather conditions. You know, they're all features that you can use in a model. however, I will say , and I don't know if this is, you know, Human versus, I don't know all, I've not come up with a good, and maybe this becomes a product if I do, but I've not been able to come up with a good predictive tool other than just looking at pace for knowing what my 5k time is gonna be or whatever.

It's like, okay, my training, you know, I did this many miles, this many weeks before and all that, and it's, I haven't come up with the, the model for it, but the data's there so I can keep playing around with it and see if, I can use that same predictive, tool that I use for, for business, for my personal life.

So I'll, I'll keep you posted on that.

[01:03:37] Adam: Please do. No, that, that's great. And, and before I get, we, we mentioned the other book that you're writing, that's the, the, the, the sci-fi one, right? Do, do we have a rough date for that? We, we, getting close.

[01:03:48] Glenn: So that thing is, is 98% done and it's been 98% done for a couple of years.

And meanwhile sitting on my, hard drive gathering digital dust . so I don't know. I think, With everything going on. right now, it's gonna be a while before I can dust it off, but maybe it's, maybe it'll be a good summer project to get that thing, wrapped up and see what happens with it.

[01:04:11] Adam: There we go.

Let me know. I'll read it. Yeah, I'll definitely read it. Cool.

[01:04:15] Glenn: Yeah, maybe I'll let ChatGPT finish it..

[01:04:17] Adam: But it can't, it won't be able process the text , which it, it's too long. Right? You know, finish this book for me. Oh, sorry. I can only, I can only, recognize the last page of text because my character limits up..

No. All good. So, so where, where can people find out more about you, Glenn?

[01:04:39] Glenn: I probably LinkedIn. I told someone the other day that, LinkedIn's the only social media, platform I'm on because, well, I think I'm on Twitter, but I don't really ever do anything over there. it's the only social media platform where I don't log in and immediately wanna argue with someone.

So , so LinkedIn, it is for me, so,

[01:05:00] Adam: Yeah, fine. Okay. So we'll, we'll, we'll put LinkedIn in the, in the show notes. I'll, I'll link to your book as well. so pe people can check, check you out there. But, no, once again, I thank, thank you to you Glen, and you say, you know, you've got a lot going on at the moment, so you know, to talk about this sort of stuff I absolutely love.

So, you know, it's, it's been an absolute pleasure.

[01:05:19] Glenn: Likewise, likewise. And I apologize again for the technical difficulties, but I'm, I'm relying on your post-production capabilities to make this happen.

[01:05:25] Adam: We, we, we have the technology and we have the degree in music production, so hopefully we'll get there.

Alright Glenn, see you later. Cheers.

[01:05:33] Glenn: Thanks Adam.