Demystifying AI

In this episode, Tara interviewed Tim Martin, an expert in delivering Generative AI training for businesses.

Tim shared tips on:

• Where to start with Generative AI for novices  
• How AI can be incorporated into businesses
• Watch out and pitfalls
• The future of AI in business

Transcript

Tara: [00:00:00] Hi everyone, thanks for joining us today. This is the Innov8 in Medtech podcast, a show for medtech and life sciences professionals that aims to help you get ahead in your career. I'm your host Tara Sharma, ex medtech professional turned executive search headhunter. Um, and owner of recruitment agency, innovate search for the medtech pharma and biotech industries with over 20 years of experience in the healthcare field for more tips and tricks on how to get ahead in your career.

Tara: Follow me on LinkedIn. Let's dive into this week's episode.

Tara: Hey, Tim, it's so exciting to have you on the podcast today. How are you? 

Tim: Very well. Thank you. 

Tara: Good, thank you for being here with us. I know there's a lot of people in our network that would love to know [00:01:00] more about AI and may need some of the concepts broken down into some simpler format for them. So I'm really excited to dig into some of these questions with you, Tim, that I have no doubt have been generated by AI.

Tara: So. 

Tim: You could be right. 

Tara: So Tim, just for the benefit of my audience, could you introduce yourself and tell everybody a little bit about what you do and why you have an interest in AI? 

Tim: Oh, sure. Look, I've been a business communicator, I think for maybe the last 15, 20 years, mostly in the tech space, the early days of digital.

Tim: Websites and getting traffic to websites and then social media and so forth. But I guess the thing is somebody needs to be able to bridge the gap between technically what's happening and, the business implication of things. So I wear two hats. I'm not really a technical person, but I do have a business background and that's probably the journey I've been on.

Tim: And the latest topic obviously is AI, but what I do really as a business hasn't changed much in 20 years. 

Tara: Okay, thank you for [00:02:00] explaining that. So look, could you explain what generative AI is and how it's distinct from other forms of artificial intelligence? 

Tim: So look, AI as a concept has been around for quite a few years.

Tim: I mean, it goes all the way back to the 1940s. Term itself was coined in the 1950s, and there's been lots of attempts to make AI work in different contexts. But in a real sense, we've been using the classical sense of it for the last several years. Google Maps is AI driven, Netflix recommendations of what you see or don't see on your social media newsfeed.

Tim: That's all AI. So a lot of people weren't excited about it back then. It was just a tool that we were using, but that's what we call discriminative AI. takes vast amounts of data and then applies confidence scores and makes decisions based on the levels of confidence around categories. A little bit technical, but it doesn't really matter how it works.

Tim: That's there. Generative AI is relatively new on the scene. That's just in the last, you know, year and a half that the public have had access to [00:03:00] a tool that doesn't try and understand data in terms of making decisions, but actually generates something. As a result, and AI. So texting, you use chat GPT for example, input is text and the output is text or now with multimodal large language models, text or image to text or text to audio or music.

Tim: So that's why we call it generative AI. It generates something. And it's also sometimes known as synthetic media because it's creating content that up until you're giving it a command or a prompt, it didn't exist. It's just. Basically made out of nothing. 

Tara: Okay. Okay. Great. That helps to really clarify the difference and where we've seen it before.

Tara: From what you're seeing, we're certainly using generative AI in our processes and it's made a massive difference to our efficiency and speed, certainly in recruitment. But what are some of the promising uses you're seeing of generative AI or general AI that we might [00:04:00] encounter? In our workplace soon.

Tim: Yeah, so I think people need to appreciate that what we're they're seeing at the moment is exciting as it is and it does have workplace application now and there are efficiency gains to be had. It's nothing compared to what's coming down the pipe, so you using it to maybe look at some documents and reframe, adjust the language or summarize the key points and so forth.

Tim: And that's super useful, but where it really will come into its own is not so much. Create a new content like give me an idea for an essay or that sort of thing or reframing content take this and shorten it by 50 percent, but it's understanding. Data so large language models why they called language models allows people to use natural language to perform an action.

Tim: So if you had a large data set, for example, it could be a thousand job applications or membership names or thousand of anything. And to be able to go in and say, [00:05:00] yeah, do you think you could pull out the data? It's around what could be interesting with business development in that zone, but I'm really interested just in the last maybe three, four weeks, that natural language, me saying that the way that you talk to another human being is the way to be able to talk to data and vast waves of data.

Tim: So all of a sudden, your ability to pull insights from big data sets, big Excel spreadsheets or databases containing any sort of data by talking to it, that is a game changer. That will revolutionize the way that people act at work because their ability to get insights has accelerated a hundredfold. Wow, that's incredible.

Tara: Okay, my mind's tickled. Yeah, that really is, that's making me very excited for everybody's business. 

Tim: What's cool about it, it's not pie in the sky, futuristic flying car stuff. This is now, so anyone that wants to go and [00:06:00] access Gemini's latest model, 1. 5 pro, which is an open beta, just Google it and you can end up there.

Tim: Can upload, they've got a token window of a million tokens, which means represents data in different ways. So that's equivalent to about 700, 000 words or an hour of video, 11 hours audio. You can actually start playing with this now. You can upload some vast amount of data and just start talking to it. And tease insights out and that's ready to go today. 

Tara: Wow. Okay, that's great. Thank you for those insights. And, and so then what do you come across? You talk to a lot of people about this topic and about the knowledge that you can share. What are some of the common misunderstandings that you see about AI and how can they be effectively addressed?

Tim: Look, I think there's a lot of confusion around that acronym. I artificial intelligence, what that actually means. I mean, firstly, that's not you, as I said before, it's been kicking around for quite a few years, secondly, that you've been using it. So don't be afraid of it because you use it every day, whether you [00:07:00] know it or not.

Tim: But I think the big one is the misunderstanding around the impact that it's going to have. And there should be a term for what I'm about to explain, and maybe I should be the one to go here to make it up. But it's a concept where somebody comes across a technology in its very early days. Right. 

Tara: Yeah, 

Tim: they play around with a little bit and the case of chat GPT, for example, they ask it a question about something that gets it wrong, or they do a best man speech.

Tim: And it's like, Oh, that's pretty good. But they sort of write it off as not really being something that needs to be considered as how it's going to change things in their workspace or in their personal lives or career even more importantly, and they write it off. That was a mistake. The thing that you were looking at was in its formative stage.

Tim: It may not have performed the way that you thought it would. The magic of AI was going to perform, but don't write it off because whatever you were seeing then is not the way it's going to be two months from now, or 12 months from now, 24 months from now, it will [00:08:00] rapidly change. So for all of those folk out there that might be listening that have played around a little bit with large language model, like ChatGPT, the only game in town, by the way, there are other brands, go back and revisit it because it can do some things now that it could not do a couple of weeks ago.

Tara: Okay. Definitely. Even just I think about using it even 6 12 months ago and the sorts of prompts you had to give it versus how much the newer models can learn and you can create the avatars within them and then you're not teaching them every time like it, it's evolved a lot and I'm just talking about track GPT in that instance.

Tara: Yeah. 

Tim: It has. And the whole idea of prompting, I still see people struggling with their prompts because they're using a keyword mindset, but it's really the case of a lot of people disappointed with an output they're getting because the prompt wasn't right because. It was keyword jogging as opposed to a detailed brief that you would give another human being.

Tara: Yes, yes, you can talk to it, you can't really write to it, but you're, I've [00:09:00] been treating mine, not the beginning, I think I was following what I was seeing about prompt engineering, but now in some ways. Instead, I give it a brief as if I was briefing someone else in my team and it's so much better. 

Tim: That's right. So it's not like saying, Oh, talk to it like a human being. It's like you must talk to it like a human being because it's trained on human being language, large model data set was pulled off the web, which is a function of what people humans are posted there. So for the patents to kick in and really make it work, you have to talk human, just talk natural.

Tim: Um, but you've got to be very clear. Because, uh, the, the model will only ask or respond based on what you've asked it to do. So it won't volunteer information. And a lot of people like are waiting for something else to come out, but the machine won't do the other thing unless you ask it to. 

Tara: You need to ask it to expand. Yeah, for sure. Okay, that's good. 

Tim: You have to have a conversation with it. 

Tara: And same as if you, Went to a [00:10:00] copywriter and they sent you something back and then you'd give them feedback. It's the same thing. Right? 

Tim: Exactly, right? And occasionally someone might fluke an output based on a prompt the very first time and go excellent That's exactly what I wanted.

Tim: That is so rare My experience is you start the process you get the feedback in terms of its output You feedback, it's feedback, and this is why they call chatbots or chatterbots. You're having a conversation and the two of you having a conversation eventually work it out and get the result that might take a number of iterations and X number of minutes, not quick in and out the way you would a Google search.

Tara: Yeah, for sure. Absolutely. And so look, these changes have already come and they're happening. How can individuals and the businesses we work with prepare for some of these changes that are being brought about by AI? And what, at this moment, and for the next sort of six to 12 months, what skills or knowledge should they be prioritizing?

Tim: Yeah. Okay. So, [00:11:00] and I think this has been true beyond AI. Any new technology, if you want to know how it works, you play with it. Right? You get your hands dirty. It's what I call constructive tinkering. If it's a personal use case, for example, you're planning a trip overseas, use a large language model to try and understand what it can do and not do, what you like about it.

Tim: And then you start to join the dots in terms of, well, if it worked in that context, You know, what are the implications or how could I extrapolate to make it work in a business or work context? I think you won't be able to join the dots without the dots. You've got to play with it, be prepared to go on that journey.

Tim: You can't say, Oh, I'm going to wait for the book to be published or I'm going to go to that conference at the end of the year. That's way too late. It's not too late. That's not what we want. You've got to get on the tools and play with it. In terms of the business side of implication, the thing that we can't have happen is.

Tim: Any business sector to be in denial that something's happening. It's like, okay, we've seen tech [00:12:00] ways before. We've a little bit different. We're insulated from that rubbish. You're not. So that state of denial is still a very real thing. And I think a lot of people are writing that off as, Hey, it's high, it's AI.

Tim: We've seen this all before. But the conversations that you and I have had a couple of times, just our own personal experiences of how we use it. And it's like, once you've been introduced to the superpower, you're never going to get it back. 

Tara: Oh, absolutely not. No. 

Tim: And that's tip of the iceberg. What about the people that don't have the superpower?

Tim: It's like, so this person is using it. This person over here is not. There's going to be this yawning gulf of efficiency between those two parties, just because one adopted and the other one didn't. 

Tara: Yeah, and that's it. And the kind of competitive advantage when used correctly to the people that are using it is going to be huge.

Tara: Just going back to that example of somebody maybe who hasn't ever used it. And it really is an absolute beginner to this. [00:13:00] What would be the first step for them? Could you just give us kind of a step that they can do first, then maybe a question that they could ask so that they can start to tinker and play with it.

Tim: Yeah. Okay. So there's a couple of platforms out there. I mean, the obvious one chat should be T and 3. 5 is still a things and that's free, right? 

Tara: !Yeah. 

Tim: Google. Gemini is another option that's free. So those are the two probably go to platforms at the moment. So money's not an issue because I just said they're free.

Tim: That to go in and start, probably the best way to approach it is we were just discussing is to try and probe it, have a conversation and be quite nuanced in what you're asking it to do. So for example, I'm thinking of going to Italy in, in, May. Now that's important because the Large English Models knows what the temperature is in May and it knows how busy it is for tourists, so you have to be quite, thinking of going to Italy in May, I'm not too sure whether I want to see some monuments or might be good to just hang out on the coast.

Tim: I did a trip a couple of years ago to [00:14:00] Peru and we got a bit mixed up in the museum scene. What do you think? 

Tara: Yeah. 

Tim: That's a prompt. 

Tara: Yeah, 

Tim: that's a way to start and it comes back with something and then you go, Oh, look, I hadn't thought of that. Maybe we could unpack that and it's, you're actually using those words or typing in those words and I think that's a sort of a moment that people get what's happening there is that when you start engaging on a human level, even though it's not human, it's a, Washing machine at the end of the day, just nuts and bolts, but that ability to tease stuff out is really as well.

Tim: This is going to change a lot, but how is that going to change things at work? So doing that at home, the next stage to start seeing things differently as to how it's going to impact workflow, how you're going to assimilate data, draw insights from data. Create content, reframe content, change marketing matrix.

Tim: A lot of things will click in, as long as you've had that hands on experience and realize what it's capable of doing. [00:15:00] 

Tara: What it can do. Yeah, that's good advice. Now, moving on then from someone tinkering and playing with this at home and in more of a personal application, how do you think or have you seen that generative AI can improve decision making process in a business?

Tara: And maybe you could share an example with us. 

Tim: Look, I mean, there's so many, one really powerful one is stress testing and ideas. Sometimes it's called that red teaming. So you've got a business case, a proposal, a tender document or whatever. And we share that with the large English model. That could be an upload.

Tim: It could be as simple as a copy and paste. Ask the model to punch holes in it. So look, this is a line of argument that I'm going to run with the client. The client has been with us for a while. They told us that they're shopping around for some alternatives. This is the sort of money we spend with them.

Tim: These are their options. These are some things we're thinking of saying, ask the machine to feedback what it thinks the holes in your argument, and even to go one [00:16:00] step beyond that, which I do this all the time, try and anticipate objections from the client and tell me some really powerful counter objections to their objections.

Tara: Yeah. 

Tim: Like, wow. 

Tara: That's a great idea. 

Tim: It's like having a business coach that's trying to lovingly punch holes in your argument to give you the tips before you go into the meeting, like, who wouldn't want that? 

Tara: Absolutely, that sounds incredible. Yeah, I have heard about a platform, and we work in devices, medtech, biotech, pharma.

Tara: We work with a lot of sales professionals. And obviously, nowadays, It's a lot of business done based, obviously, particularly in our healthcare industry, but a lot of business is conducted virtually. And I heard about a platform that sits in the side of your virtual meeting. And as you're talking, it's giving you prompts to say, Did you notice this just got said, or that's a buying signal, or perhaps you could ask this.

Tara: Do you know about platforms like that? 

Tim: Yeah, so look, one thing to [00:17:00] appreciate is that a lot of these tools are manifestations of the underlying technology, the large language models, which is just one. Lots of brains, but the underlying technology of an LLM, a large language model, is the ability to take Data information in and run it through a filter like a system prompt or a set of instructions and have an output on the other side.

Tim: So in the instance of what you're talking about, the ability for a large language model to have voice to transcribe that into text to run it through a set of instructions, which is to say isolate each person talking, take a note of each person, analyze for sentiment, pull out key ideas, pull out any commitments.

Tim: Thank you very much. To do actions that we're committed to, and in real time present that up on the screen. That's just really basic stuff. So this, you can see how this is fundamentally going to change contact centers. In fact, it is already. But what you've raised is a really [00:18:00] interesting point for clarification.

Tim: That the tool is the application, but the application is built on a large language model. And this is where it's going to get exciting. It's like, we've got this technology. What will people build over the top of it? 

Tara: What applications will there be? What will get thought of? 

Tim: That's right. So, you know, good analogy would be electricity. Electricity gave rise to the telegraph, which was pretty important. You know, real time. Communication electric signal going down a wire and a lot of people like, well, okay, so that's a electricity is the telegraph and no, that was the first thing that was built on it. Then we've got electric engines and we've got radios and we've got television and we've got the internet.

Tim: And then on top of the internet, we've got websites on top of that. We've got e commerce. So it's all of those layers that are about to be built out that we really don't have much of a glimpse into at the moment. That's the really exciting frontier. 

Tara: That's the exciting thing, isn't it? That's amazing. So obviously people are starting to implement this in their business [00:19:00] and I know we've certainly attended a lot of webinars and things that will show us the watchouts for our business, particularly in terms of protecting people's data and those sorts of things. But what factors should businesses consider when they are incorporating generative AI into their operations to make sure it's a very smooth transition and minimize any disruption?

Tim: Okay, so there's a couple of considerations. One, you know, What are we made 2024 early days. So whatever you see now, the tools that you're using and how you're using them will not be what you'll be using six or 12 months from now. The space is moving very quickly. So take that on board. It's not like, oh, we get Microsoft Excel and we're going to have it for the next 20 years.

Tara: Yeah, 

Tim: I'll be running with Excel for a couple of weeks and there's going to be something else that's going to come down the track. 

Tara: Yeah. 

Tim: So there's certainly that and be prepared to change quickly and adapt because the tools will change and adapt very quickly that the disruption. So. If [00:20:00] I, you know, say, for example, for 25 years, you had been tabulating numbers.

Tim: If you're an accountant, for example, and you've got your calculator there, and you're really fast at doing that with your hands, you've got the pencil thing happening. It's great. And all of a sudden says, look, we've got this. Excel spreadsheet. It's a no brainer that person is going to migrate to that technology, but the transition is going to be disruptive because all of a sudden you've got to change the way you've been doing things.

Tim: You've got to learn how Microsoft works and you know how to perform a sum function at the bottom of a column and it's everything that a three digit spreadsheet. Like you person is capable of learning quickly, but there's going to be a disruption in terms of going back a step to go for three need to appreciate if they want to get the advantages and the efficiency gains that some of these tools are offering, they're going to have to go back a step.

Tim: It's going to be disruptive, but it's worthwhile doing and the sooner you do it, the sooner you'll enjoy the upside of the switch. 

Tara: Yeah, I think that's been the interesting [00:21:00] thing here in my business, you know, there's some technology that people are quite resistant to. Use because it's more work, but when it's come to anything AI related, I mean, they absolutely snap it up because it just helps them, helps them be more efficient.

Tara: And often it's taking away the more time consuming or tricky parts of the job that you have to slow down for. 

Tim: You said time consuming, tricky. I'd call that the crap work. Like seriously, like I've got to write that report and I've got to read someone else's report. I know it has to be done. It's really important.

Tim: You know, devil's in the detail, but I don't need to spend an hour doing that anymore. I'm going to do that in about seven minutes. And then I'm going to work out what I'm going to do with my other 53. 

Tara: Yeah, that's it. And it's not that you don't still interact with that information, check it, all those things. But like you say, it's just that time part of it that's just quicker. 

Tim: Oh, no, [00:22:00] completely. Like, for example, I'm trying to use it at home as much as in a work context. Got a new induction oven and it comes with a 38 page manual. Upload the manual to a large language model and just as I'm looking at different features of the oven, just ask it questions.

Tara: I love that. 

Tim: What's the second knob on the left do? What does that button do? And it just gives me the answer. I don't have to read the manual. 

Tara: Yeah, 

Tim: I have to find it. The page that button is on in the manual. It'll find it for you. It'll find it. So that becomes a data set. Then I ask questions in natural language.

Tim: It's like, oh, do you mean I could use that button to reheat pizza? And it goes, yes, that would be a really good application for that button. 

Tara: And then surely soon It will say that to us, you won't even have to read it, and then it will also do it for us. 

Tim: Exactly. Like the Jetsons. Well, no, I'm still going to put the pizza in the car, at least I'll get carried away.

Tim: But it's down the track, robots that put pizzas in our room for you. But right here, right now, you'll be doing that. 

Tara: [00:23:00] That's a shame. And look, in terms of, again, if I think about different departments within, um, The medtech companies we work for, how do you think AI is going to really revolutionize customer service and customer experience in the coming years?

Tim: It's going to be a whole lot better. So for a start, when you make an inbound call to an organization, that will be a large language model, a bot that will take the call. You probably won't know it is actually a bot. You will, because they'll announce it. It'll be part of their AI policy to say, look, you're talking to a bot.

Tim: Talk to human press too and stand in the queue for 38 minutes. But you're talking to a bot, the bot has access to data in real time. So whatever you ask it, it's able to access it out of the database and come back with the precision level, correct answer. So that makes training contacts into personnel different.

Tim: And you'll love it because there won't be A cue to stand in or sit in with a [00:24:00] silly elevator music. The information will be precise that a million people can call in simultaneously and a million bots can take those million calls. Cause it's scales up and down. So that's going to change contact center interaction and for us, the better, but I was demonstrating a piece the other day where I uploaded a form into a large language model, and then I told the model, look, I don't have my glasses.

Tim: Could you walk me through this? And it said, okay, so it looked at all the questions and there was some for my doctor. So it didn't ask me those ones that asked me the questions and in its own time, it said, okay, can you explain this to me? And I said, oh, I've got this and this and it goes, okay, if I had to choose this or this.

Tim: Which would you rather me put in? I go, I'll just go with the first one. So it completed the form and it said, okay, now what I want you to do is you need to go to your doctor and, you know, him or her to fill in their side of things. But the next point up would be the bot would actually have the ability to perform an action, which is to submit that [00:25:00] form to VicRoads or, Whatever department requires it and in terms of customer service and having to go through quite torturous processes and we don't think like that at the moment because it's just what we do but in retrospect you'll realize how clumsy they were that's going to change a lot and it's going to streamline our interactions with organizations because the bot will be the translator in the middle that will just make sure that what we're giving them what they need is aligned and It'll just be so much easier.

Tara: Easier and quicker. What issues do you see? Particularly if people are experimenting with this. Let's take ChatGPT for an example. And maybe they're experimenting, maybe they're even using it for work and that hasn't been properly integrated in a process. What are the watchouts? 

Tim: You're talking about kids running around the house with sharp scissors, right?

Tim: That's pretty close to what it can be. So what we don't see. One is these very powerful [00:26:00] tools and the wrong hands. And when I say that I'm talking about people that don't appreciate that the tools have limitations. There is definitely a risk profile at the moment. We've still got an issue with the concept called hallucinations where the LLM can occasionally because it's all probability next word based occasionally.

Tim: Make something up. 

Tara: Make things up. 

Tim: It's not based or grounded in truth or fact. So we need to be aware of that. We also need to be aware that, at this point at least, the models are not going to do all of our work for us. It's a good way to get something started. It's a good way to get a framework in place.

Tim: It's a good way to review something, reframe, but we still need that subject matter eye over the top of it before we ship. So, Different people in business and industry will use in different ways. Lower level workers will use it more as a knowledge lift to access information quickly and understand how their business or industry works.

Tim: The likes of you or I will use it differently because we're going to leverage our [00:27:00] expertise in terms of what we want the machine to do and what insights we want to draw from the data and importantly, be able to spot that the machine. That's not quite what we wanted. My expert eye has said, that's actually not the way I really want it framed.

Tim: Whereas somebody that's not operating on our level go, wow, that, that reads really well. Let's just send it out. 

Tara: Yeah, definitely. 

Tim: That's dangerous. 

Tara: Yeah, I think so. I've seen that. I see it all the time on LinkedIn and I can see when someone's written something with chat GPT, but they haven't turned the dial down on it.

Tara: Everybody's posts or content are looking the same. That's not to say for everybody, because I think people have realized that now, but if you do your original prompt, it comes out with quite blurry chat GPT language. I now turn around and say to chat GPT. Please don't write this in such a chat GPT way and it knows what I mean.

Tim: That's right. So reference points are really good. [00:28:00] I'm not telling people how to fudge their job applications or anything, but give a reference point. So I want you to write this in the style of an opinion piece from the New York Times. 

Tara: Yes, or in this, yeah, or in the style of this top copywriter or whatever it might be.

Tim: Yeah, exactly right. So it's interesting. I said before, you know, that report, for example, used to take one hour and now we've done it in seven minutes. Maybe I misled you a little bit. What I'm finding is that the report in seven minutes is just as good as the one I used to get done the old way in one hour.

Tim: But what I'm doing at the moment is still spending one hour doing it. I'm getting. The first block done in seven minutes, and then I'm refining it and polishing it for the remaining 53. So it's still the same amount of time I'm spending, but the quality output has kicked up to the next level. 

Tara: Yeah, yeah, that's true 

Tim: and an analogy I use a lot because I travel a lot, you know, people on the travelators and airports. 

Tim: Yeah. 

Tara: [00:29:00] Yeah. 

Tim: I walk on them. Right. But I'm standing on the travel at it. And I'm like, dude, walk the same speed you were walking before you got on the travelator. And the travelator and you'll get to the other end quicker.

Tim: And I think that's sort of a mindset. Do I want to use these large language models to be a little bit lazier, still get the same output? Or do I want to use them as an accelerant to actually get a higher level of quantity and or quality done? 

Tara: Quality, yeah. 

Tim: By walking. So I think it's, a lot of people are sort of using them in a lazy sense. It's cutting some time and I think that's really what's going to make it work. 

Tara: No. No, I completely agree. And then what, in terms of that data protection side, what do we need to be watching out there? 

Tim: Yes. So look, this is a conversation that goes back a decade or so around cloud computing. You know, who's got control of the data, where the servers are sitting, what legal domain physically are they sitting.

Tim: It's a similar sort of setup. I think the slight difference [00:30:00] with generative AI and the large language models is the idea that what you put into a model could potentially leak out. Yeah. Of that model. Now, in a sense, that's not really a thing because GPT, the P means pre trained. So whatever you put into it doesn't come out of that particular model.

Tim: It's pre trained. It's done and dusted. There's a knowledge cut off, but a lot of the models take inputs and they hold on to the content as part of the data training set for the next iteration of the model. And there's a concern that my data could somehow end up in the next model and somehow leak out.

Tim: Now, the chances are very small, but organizations that are sensitive around their data, who's holding it and potential for it to spill out somewhere down the track, that's valid. Absolutely. Now, I've got nothing in my business that I would be concerned about anyone in the universe knowing. Like master, you know, Facebook campaign.

Tim: You've got it. That's fine. I don't care. But some other organizations, government organizations, health, for example, health [00:31:00] records, super sensitive. That's why what will happen is that we won't use a public platform like ChatGPT. We'll have our own in house. Version of ChatGPT, our own large language model that will have security and control over where the data is sitting.

Tim: That it won't be a part of someone else's data training set. And that's why everybody will have their own large language model because of different concerns around the data. And also what you want those models to do. You should be able to customize your model to be a little bit better at that. And we don't really care about that feature or skill set.

Tim: At the moment we're just given something very simple. Jack of all trades, generic, I was common denominator. A year from now, it'll become a lot more nuanced. The model that you use, Tara, will be more calibrated to your industry space. 

Tara: Yeah, 

Tim: your personal needs or business needs, my model, the same underlying technology, but it will be different because we've got different businesses.

Tara: Yeah, absolutely. Cool. [00:32:00] What kind of collaboration do you see? We've had a lot of people initially being quite fearful of this, AI is going to take our jobs, but I actually don't see it like that. I see it as efficiencies and like you say, putting it on turbo charge, but what kind of collaborations do you foresee between human employees and, AI systems in the workplace and how can businesses go about encouraging these partnerships?

Tim: Yeah. So. Collaboration is a good word to use because that's exactly what's going to make it work is the machine is very good at doing some things that humans are not so good at doing or can't do as fast and vice versa. So put the two together and you've got one plus one equals three, a beautiful synergy happening.

Tim: To have that mindset that this is a tool that's going to help me but not necessarily replace me. Now the people that should be shaking in their little booties maybe a little bit are the ones that don't want to understand how the tools work and [00:33:00] how to operate them in a work context because if you don't know how to do that I'm going to get someone that does.

Tim: So it would be like someone coming into the workplace saying, you know, the internet thing and email and that Microsoft Excel, not really my thing, but I'm still really keen on the job. So yeah, sorry. That's just, 

Tara: no, you've got to be tech savvy. Yeah. 

Tim: You've got to be tech savvy. So I think the onus is on firstly people.

Tim: Lifting themselves up a little bit by actively playing with the tools to understand what it is and a little bit of the language around them and some of the features and how they work. But businesses are going to have to formally upskill their workforce. They're going to have to bring people in or run courses or do just do more, but it can't be ad hoc.

Tim: There has to be some formal embedding and introduction of the technology. And a policy is really important and understanding of the risks and leading people. Learn how to use the tools properly and correctly and efficiently without [00:34:00] causing damage. 

Tara: Yeah, absolutely. And I think the other one that you said that was, is really important and I think about it now for my business is, like you said, there's new things coming out all the time.

Tara: There's something we've started using in the last week that really is our superpower at the moment. But you've got to be careful financially too, right? Because like you say, in three months or two weeks, there could be something better. So looking out on what the contracts are , and lock ins and all of those sorts of things too.

Tim: Yeah, that's right. I mean, to access the LLMs directly. Uh, I mean, I know a lot of people that are starting to use particular tools and there's payment required for those, but at this point, at least, and it's going to be true for me, I'm sure for at least another six months, I just go directly to the source.

Tim: I'm getting the LLM to perform the tasks and the actions for me rather than using a third party tool that's built on the top of an LLM. 

Tara: Okay. 

Tim: So in that respect, my price to play is 30 a month. The risk for 30 a month is [00:35:00] zero. The risk of wasting some time possibly could factor that in, but even then I'm still learning some things.

Tara: Yeah, exactly. And final question for you, Tim, where do you think this is going to go? 

Tim: Flying cars, Tara. One day you will get in a car and it will fly. Imagine that. It 

Tara: will be just like Back to the Future 2. 

Tim: Yeah, exactly right. Still waiting on those flying cars. So where's it going to go? Look, what I think is going to happen is that it's a new tool that it will become business as usual, much the same way you've got electricity, powering lights, and you've got computers and access to the internet in the workplace, and we use it every day to the point that it becomes invisible, and we don't think about it twice.

Tim: Yeah, that's where I see it's going and everyone will have access to the same technology will be different calibrations depending on tasks and industry sector and so forth, but it'll just be the regard like we won't even talk about it or think about [00:36:00] it. What there is a bit of confusion around is a concept called general artificial intelligence, which is not a thing, right?

Tim: It's the idea that it's not a large English model that can do some interesting things. It's actually operating at our level. That are quite profound and some would say scary and that knowledge workers would be replaced instantly. We wouldn't need you. We would have a 24 machine running that can do everything that you can do plus a little bit more.

Tim: And we can have thousands of them because it will scale. That would transform society. Let's not worry about that because it may never happen. Or if we're having a three o'clock this afternoon, I don't know. But in terms of the LLMs, the technology you and I have been talking about for the last 30 minutes or so, that have just become business as usual.

Tim: That's what will happen. 

Tara: Yeah. Thank you so much, Tim. I know you would have given a lot of insights to people. There's a lot of people out there already using this technology. There's a lot of people who just would really appreciate the 101. So thank you ever so much for your time today. 

Tim: [00:37:00] Oh, you're welcome.

Tara: Bye. 

Tim: See ya.