Digital Nexus

2 | AI is to be feared... or is it?

Chris Sinclair and Mark Monfort Season 1 Episode 2

🎙️ Our 2nd episode of the "Digital Nexus" podcast and this week it's all about fear (and loathing) in AI🚀

In this episode, we talk about hallucinations and randomness when it comes to generative AI outputs as well as the ways in which we can control this. We also talk about where expertise and creativity come in and highlight some examples of AI tools we've been playing around with this week (including SIKE and Claude 3.5 Sonnet).  

Don't forget to like, subscribe, and stay tuned for more episodes exploring the frontiers of technology!


Timestamps:
🔹0:00 - Intro - what Chris and Mark have been up to during the week
🔹9:00 - AI outputs and how to control the hallucinations and randomness 
🔹21:15 - The need (or want) for experts in the AI space and the double edged sword of thinking AI is only for mundane tasks 
🔹25:49 - AI does not do math - but good at anomaly detection 
🔹29:12 - Scrutiny of new technology / rabbit and the hare of AI (as well as a comparison to blockchain tech)
🔹Show and Tell Time
37:47 - SIKE's Dreams are and example of mimicking writing styles 
46:21 - Claude 3.5 Sonnet for creating charts and dashboards

🔹51:54 - What do developers do next in light of AI advancements - the need to raise the bar in a world of AI
🔹1:01:09 - NVIDIA and their support for startups

- Fuzzy Sequence - https://fuzzysequence.com/
- SIKE - https://sike.ai/
- Claude - https://claude.ai/

#DigitalNexus #Podcast #AI #Blockchain #Startups #TechInnovation #GenerativeAI #ChatGPT #DigitalDisruption #DigitalVillage #NotCentralised #SIKE #TechTrends

Other Links
🎙️our podcast links here: https://digitalnexuspodcast.com/
👤Chris on LinkedIn - https://www.linkedin.com/in/pcsinclair/
👤Mark on LinkedIn - https://www.linkedin.com/in/markmonfort/
👤 Mark on Twitter - https://twitter.com/captdefi

SHOWNOTE LINKS
🔗 SIKE - https://sike.ai/
🌐Digital Village - https://digitalvillage.network/
🌐NotCentralised - https://www.notcentralised.com/

YouTube Channel: https://www.youtube.com/@DigitalNexusPodcast
X (twitter): @DigitalNexus

<b>[MUSIC] Kickstart a problem, kickstart a problem.</b><b>Yeah, exactly, exactly.</b><b>So, shout out to the Speed Shorts.</b><b>How's your week been?</b><b>What have you been up to?</b><b>This week has been an interesting week.</b><b>I've been testing out a lot of a couple of tools on the sales front for AI.</b><b>Sales front.</b><b>Yeah, there's things that allow you to, I</b><b>guess it is short cutting, you know, the outreach</b><b>stuff you might usually do on things like LinkedIn and email, etc.</b><b>Like automating and stuff?</b><b>Just automating that entire process.</b><b>So, that's been pretty cool.</b><b>There's a tool called Fuzzy Sequence, which</b><b>a friend of mine, or now a friend of mine,</b><b>has been working on quite a lot in the background over around Singapore.</b><b>So, it's a pretty cool tool.</b><b>And then the rest of the week is being</b><b>sort of getting excited and prepped for this.</b><b>Oh, nice.</b><b>Yeah.</b><b>I created a brand.</b><b>We've got the website up or landing page.</b><b>The podcast is out.</b><b>First episode out.</b><b>First episode out.</b><b>You go to, what's the website again?</b><b>Plug for the...</b><b>DigitalNexusPodcast.com.</b><b>Dot com.</b><b>There you go.</b><b>And then you can email us at digitalnexuspodcast.gmail.com.</b><b>Dot com.</b><b>That's right.</b><b>We could set up a proper domain exchange for that.</b><b>Yeah.</b><b>We'll see how popular this gets, if at all.</b><b>No, I don't think anyone's watching this.</b><b>But it's funny.</b><b>Speaking of like Gmail, it's like people get</b><b>fooled by like those, "Oh my God, BlackRock</b><b>emailed me," and it's like blackrock at gmail.com.</b><b>Yeah.</b><b>Anyone.</b><b>You can have any Gmail folks.</b><b>Be careful about that.</b><b>BlackRock Gmail 3425 at gmail.com.</b><b>Yeah.</b><b>The...</b><b>Oh, it's the ATO at gmail.com.</b><b>ATO19465 at gmail.com.</b><b>Tell you what, I do appreciate when they</b><b>try at least, you know, the scammers they try.</b><b>When they just have like some fake ass</b><b>email that doesn't even make sense, anything to</b><b>do with the thing.</b><b>They're trying to scam you on.</b><b>Then it's like, "Come on, just give me a little bit of effort."</b><b>I'm going to say no, you know, but like try me.</b><b>What baffles me about the whole scammer</b><b>world, and I'm probably giving them insight now on</b><b>how to get better, but AI exists.</b><b>Why are their emails still so shoddy?</b><b>Like I'm almost offended at the little effort</b><b>that they've put into some of the emails that</b><b>get sent to me.</b><b>Yeah, exactly.</b><b>Spelling mistakes, no like paragraph errors, just like the designs.</b><b>That stuff's automated now.</b><b>What are you doing?</b><b>Exactly.</b><b>Like you are a real...</b><b>Do better.</b><b>You are a real bank employee, email so and so.</b><b>But it does mean that, you know, there's</b><b>obviously the need to have defenses up and stuff like</b><b>that, and there are still ways to tell.</b><b>The one way that, you know, you always</b><b>used to look at the email or whatever, and the</b><b>email comes from a domain service provider, like a...</b><b>So you have your DNS provider basically.</b><b>And I saw one, the scam where the DNS got hacked.</b><b>So the owner of that domain, they got hacked.</b><b>So it was real emails coming from that</b><b>place, technically real, but from a scammer that</b><b>had hacked it.</b><b>That's super hard to tell then.</b><b>And now there's ways that I think that</b><b>that could be stopped because with blockchain,</b><b>you can have ways that only when you sign up, it's called a ZK proof, and you would</b><b>have a way that only that person and that</b><b>company had that relationship at the start,</b><b>maybe in person or maybe through a signup process.</b><b>So any later transactions would need to have</b><b>that extra signature on there that can only</b><b>be proven.</b><b>It doesn't even matter if it's from the same domain, etc.</b><b>You just can't.</b><b>So anyway, we digress.</b><b>There's some people online who like some of</b><b>my favorite YouTubers, I don't know, scammer</b><b>payback and stuff like that.</b><b>Coffeezilla.</b><b>They hack the hackers.</b><b>Oh, nice.</b><b>I love watching them.</b><b>And they hack the cameras and they're looking at the call center.</b><b>That's exactly it.</b><b>It's fantastic.</b><b>Watch out folks.</b><b>Shout out to those guys.</b><b>Once we get big, that shout out will actually mean something.</b><b>And the one that you mentioned before, the fuzzy, what are they called?</b><b>Fuzzy Sprout?</b><b>Fuzzy sequence.</b><b>Fuzzy sequence.</b><b>Shout out to them.</b><b>Yeah.</b><b>So we'll do a bit of this today, some tools that we use day to day.</b><b>One we made and then another one that, you</b><b>know, we've been playing around with in Claude.</b><b>So we'll show some videos on here.</b><b>So exciting.</b><b>This is leveling up and we might even go live with some of those stuff next week.</b><b>But I wonder how we're going to do it.</b><b>If like stuff is here, we might have to have</b><b>lapel mic or something or maybe a mic stand.</b><b>Yeah, we can start.</b><b>I mean, we've got, we've got the stuff.</b><b>We can bring it over.</b><b>We've got the equipment.</b><b>We can start having mics.</b><b>We have the technology folks.</b><b>We can bring directly into the, into the recording.</b><b>Wow.</b><b>We could even start live streaming this if we wanted to.</b><b>Could we then also bring the crowd in here?</b><b>Could we have a live audience?</b><b>I mean, it might be a bit tight.</b><b>Yeah.</b><b>The room might be a bit warm.</b><b>I might get a bit.</b><b>Yeah.</b><b>All the heavy breathing of.</b><b>Yeah, we'll see.</b><b>In the future.</b><b>In the future.</b><b>Well, I mean, your other podcast.</b><b>Yeah.</b><b>We've got plans to make that.</b><b>Yeah.</b><b>Go live.</b><b>For some of the events.</b><b>In recordings.</b><b>Not for everything because it's a bit much to kind of manage.</b><b>But like every now and then I think we're going to go like live with that to select</b><b>few down at crypto.</b><b>Depending on the topic will be about down at</b><b>crypto Lulu or Bar Lulu as it's more publicly</b><b>known as you typed in crypto Lulu on Gmail on Gmail.</b><b>Don't type it on Gmail folks.</b><b>It's not going to get you anywhere.</b><b>Don't do Google search on Facebook or TikTok.</b><b>Anyway, but like, yeah, we're going to do some more stuff there.</b><b>The event that we did this week and I'll put</b><b>it out in founders journey will be all about</b><b>and maybe I'll even put this video.</b><b>There we go.</b><b>One thing I'm thinking about.</b><b>So for folks that don't know, I write the I</b><b>was writing and I still am for chain reaction,</b><b>which is Oz defies newsletter out each Monday at AM.</b><b>And then I was doing data science in Australia as the president.</b><b>I was writing the newsletter for over a year.</b><b>I've stopped that I do it for not centralized because we talk about AI there.</b><b>But what I was thinking was a few months ago, I had people that weren't like that.</b><b>They wanted to like catch up all the time and stuff.</b><b>And it's great, but they just wanted the updates.</b><b>I'm like, well, if I publish the updates, maybe the catch ups would be more related</b><b>to work stuff.</b><b>I don't mind, but I was just having so many catch ups taking me away from work.</b><b>So that's why I wrote founders journey.</b><b>And now I'm even thinking that given that this is like a bit of a recap from what's</b><b>going on in the week and the stuff that we're</b><b>seeing cross emerging tech, I might have even</b><b>less to write, which is great.</b><b>No, just send them send them this.</b><b>Yeah, exactly.</b><b>Exactly.</b><b>Or I'll just include it.</b><b>I'm just going to put it on the distribution</b><b>channel because all of our PR marketing SEO,</b><b>all that stuff.</b><b>Anyway, should we do a couple of shout outs to first stone and chalk?</b><b>This is where we are.</b><b>They're a big sponsor for the work that we do.</b><b>We work here.</b><b>Absolutely.</b><b>We do events here.</b><b>We do events here.</b><b>So we're here on one of the levels where they've got these high chairs.</b><b>It's kind of like the podcast room.</b><b>Bring your own equipment.</b><b>Yeah.</b><b>Awesome stuff.</b><b>It's really great stuff that the guys do here.</b><b>This building is just full of like startup things.</b><b>Such a good ecosystem.</b><b>Level one startup hub for free.</b><b>Level two and three fish burners doing great</b><b>stuff in various events and like they've got</b><b>great event space.</b><b>Level four and five is stone and chalk.</b><b>Then we got six, which is what's the other one?</b><b>Labs.</b><b>Yeah.</b><b>Yeah.</b><b>Tankstream.</b><b>Tankstream.</b><b>That's the one.</b><b>Yep.</b><b>And then there are a few.</b><b>It's a great, it's a great little ecosystem.</b><b>It's ridiculous.</b><b>Like governments that support this kind of</b><b>thing, wherever you are watching from amazing.</b><b>Do it because this is what fosters</b><b>innovation and you see these startups grow scale and</b><b>you know, call back to where it all began.</b><b>Exactly.</b><b>Just like this podcast.</b><b>Got quite a couple of friends who have like</b><b>built very successful business and they all</b><b>started downstairs in the old hub.</b><b>It's fantastic.</b><b>It's amazing.</b><b>To be in the home in Sydney, the most expensive place to live in the world.</b><b>But at least we have this.</b><b>Exactly.</b><b>It's one thing.</b><b>We got one good thing.</b><b>Let's go.</b><b>We'll never afford a house.</b><b>It's for many people.</b><b>Some have, which is great, you know, but at least we have this.</b><b>Yes.</b><b>I could spiral ages into the housing crisis.</b><b>Let's avoid that.</b><b>Let's talk about tech.</b><b>Let's talk about tech.</b><b>No, final shout out.</b><b>Oh yeah.</b><b>No centralized, obviously big sponsor of this and Digital Village also.</b><b>Absolutely.</b><b>And we have leaders in tech innovation,</b><b>AI, all things fun and exciting in the world.</b><b>So the way I say it, every village needs a</b><b>mark or an idiot or a mark or a marking idiot.</b><b>Yeah.</b><b>Yeah.</b><b>It's very, that's what I'm here for.</b><b>You're very marketable.</b><b>Um, yeah, quite a remarkable comment.</b><b>So yeah, the puns will just continue.</b><b>We'll have an episode on puns.</b><b>Yeah, let's do it.</b><b>Next one.</b><b>Episode three.</b><b>It's the AI pun.</b><b>We could probably ask the AI what are some great puns.</b><b>It would be horrible.</b><b>Hilarious.</b><b>We're still more creative.</b><b>Let's jump in and let's get into the crux of what we hear about.</b><b>Okay.</b><b>Go ahead.</b><b>Um, thoughts on AI.</b><b>It's the leading title.</b><b>No, we can't, we are still here.</b><b>We had a lot of great discussions last week,</b><b>but obviously one of our key things was, you</b><b>know, giving the audience a chance to know who we are.</b><b>Yeah.</b><b>A bit of introductions.</b><b>And then we talked to a couple of topics around AI uses in business.</b><b>And we wanted to sort of elaborate on a few</b><b>of those things and expand on them a bit as</b><b>well.</b><b>Absolutely.</b><b>Um, so I guess, yeah, kicking off on those</b><b>thoughts of AI and looking at our notes here,</b><b>we've got some stuff around programmatic</b><b>prompting and that led off from a discussion we were</b><b>having, which is, um, when you are being too</b><b>simplistic in the language and questions you're</b><b>putting into potentially your LLMs and the</b><b>outputs that you then get, um, which tends</b><b>to turn away a lot of businesses because they realize that they're giving, getting</b><b>bad responses, incorrect responses, or they're</b><b>getting, um, well, they call it hallucinations,</b><b>hallucinations, or whether it's hallucinations</b><b>or even, um, references to things that aren't</b><b>relevant to their business, their query.</b><b>I checked the link and the link doesn't exist.</b><b>Exactly right.</b><b>Oh, frustration and stuff.</b><b>But, um, yeah, it's funny because, you know,</b><b>starting off one thing is that the term that</b><b>AI hallucinates technically because it's a</b><b>probability model and maybe the data scientists</b><b>and sad decisions will appreciate this, but it hallucinates all the time.</b><b>Like everything that you are coming back</b><b>with is technically a hallucination because it's</b><b>a probability model going, well, what's the next letter feeds that back in?</b><b>What's the next letter?</b><b>It's this iterative process back and forth to figure out the next letter and sure.</b><b>It looks like sentences, paragraph stories,</b><b>really long outputs, but at the end of the</b><b>day, it's just adding a next letter and then</b><b>feeding that, um, output back into the model</b><b>to go, okay, now that you've got this output, what is the next letter beyond that?</b><b>It's so simplistic when you break it down and lift the veil on top of it.</b><b>It's like when AI before generative AI was</b><b>a whole bunch of like the joke was that if</b><b>you lifted the hood, like, um, Fred in</b><b>Scooby-Doo would lift the hood on the ghost, that meme.</b><b>Um, and it's just nothing but if then else</b><b>statements and that was your AI and it's not</b><b>wrong and it's the same thing for this kind</b><b>of stuff, but at the same time, whilst yeah,</b><b>it sounds simplistic, it's actually quite</b><b>powerful because what we've realized is that</b><b>there's patterns in our language.</b><b>This is why I can sound coherent.</b><b>And the thing is if you read something and</b><b>you go, actually, that's a really good output</b><b>in your line of work, your profession, your</b><b>field, then it's technically not a hallucination</b><b>that was subjectively not a hallucination because it's you, the subject that as the</b><b>one to judge.</b><b>Now we can go deeper down into the field of,</b><b>um, in future, whether AI is judging itself</b><b>and how good that is or not.</b><b>But I think just like right now, um, we</b><b>are prone to going, well, AI hallucinates and</b><b>then we just turn away from it.</b><b>And there is nuance to that.</b><b>There's a spectrum.</b><b>The, the easiest example, the simplest</b><b>example, which I think you hinted at is, uh, so if you</b><b>ever type, and this is, this is mainly for</b><b>LMMs, particularly if you go into chat to</b><b>between you say how many hours are in</b><b>strawberry, it will nine times out of 10, get it wrong.</b><b>It'll say two or one.</b><b>Wait, there's five, right?</b><b>That's what it could be.</b><b>Sometimes I might say five and you could either go, no, it's not to try again.</b><b>Yeah.</b><b>No, it's not one.</b><b>Try again.</b><b>No, there is more than two.</b><b>Yeah.</b><b>Try again.</b><b>And it will never get it right.</b><b>But if you say to it, how many letters are in</b><b>strawberry, please count every single individual</b><b>letter.</b><b>Put those letters aside and then give me the number of hours on that.</b><b>It will get it right.</b><b>And as an example of, because you've given it</b><b>such a simple statement, it's not a mathematical</b><b>tool.</b><b>It's a language model.</b><b>It's looking for patterns.</b><b>It's not looking for one plus one equals two.</b><b>It doesn't know how to do that without</b><b>generating Python script and all type of stuff, which</b><b>it spiraled into a different conversation.</b><b>Exactly.</b><b>Oops.</b><b>But if you tell it to go through every</b><b>single division, a visual letter, it starts to go,</b><b>okay, cool.</b><b>There's a pattern here.</b><b>I need this particular letter and therefore it will give you the correct output.</b><b>There's a perfect example of where, of why that starts to occur.</b><b>If you're not descriptive, you're not going to get the right responses.</b><b>No, you're precisely right.</b><b>And what you just described there is one of</b><b>the two elements I think that are important.</b><b>And you know, Arturo and I who've been</b><b>working on our tool psych, this is true for any AI</b><b>tool, but it's especially so given that</b><b>we've made it easier to tackle this problem.</b><b>And it's actually two problems.</b><b>Like I mentioned hallucinations is one, but</b><b>it's also to this point, even if something</b><b>doesn't hallucinate as in it gets this, the details, right.</b><b>You're happy with the details, the content.</b><b>It's still this randomness of the response</b><b>that it can look like paragraphs to you dot</b><b>points to me, super summarized to someone else and really long to another person.</b><b>And what you described there with</b><b>strawberries analogy was that you were basically telling</b><b>it exactly what steps to take.</b><b>So if we think about two things, one is context.</b><b>The context is what you want it to look up.</b><b>The second is the execution steps, what</b><b>steps you want it to take to get to an answer.</b><b>And sure, there's a bit of back and forth</b><b>you might have with AI, but just looking at</b><b>that first one context, what it means is that</b><b>when you just ask chat GPT, and this is a very</b><b>simple example, like it will probably do better</b><b>at this, but just simplistically, if you asked</b><b>it about Apple, you're not tell and you don't</b><b>tell it, I mean, Apple computer, Apple phones,</b><b>I mean, Apple, the fruit, it's going to have to guess what the context is.</b><b>But rag models, which is stands for retrieval</b><b>augmented generation is a process, especially</b><b>when you upload documents or you're doing,</b><b>you see AI doing Google search, it is looking</b><b>up something first and looking for relevant chunks of the query.</b><b>So if you ask about Apple and you're pointing</b><b>it to a set of documents that are only about</b><b>Apple computers, then it's going to look</b><b>inside those documents for relevant sections of the</b><b>wording.</b><b>And then when it what it does next is it</b><b>combines that it goes and when you say tell me about</b><b>Apple, and it's looked up Apple and it's</b><b>got the chunks, it'll only answer from those.</b><b>So that's the context side of things.</b><b>And there's ways that you can be specific</b><b>or even just a bit better than being broad</b><b>with the context and execution steps is</b><b>exactly what you did there where you said, I want</b><b>you to count each letters out.</b><b>First step is this break it down.</b><b>Next step is like you count the number of hours.</b><b>The third step is like return, you know,</b><b>that that answer being specific with the steps</b><b>that you want it to take.</b><b>Those sorts of things are part of the way</b><b>that you reduce hallucinations and you make</b><b>things like more, more accurate or at least less random.</b><b>That's what we've been saying.</b><b>Yeah, it's very true.</b><b>And if you once you start to realize this and</b><b>you start to understand how you can articulate,</b><b>describe, explain, tell it what to do in a</b><b>specific way, it unlocks so much power and</b><b>potential with what you can get out of the system.</b><b>Like one of the fun things I had doing even</b><b>actually making this podcast was a great example</b><b>is you can start to get to a point where you</b><b>can ask it to debate with itself and you can</b><b>actually see it in real time.</b><b>As you put in a query that goes, we have a particular customer doing XYZ.</b><b>You are a person that does this, this, and this.</b><b>I want you to find a solution of this, but</b><b>then I need you to debate it with these type</b><b>of personas.</b><b>And suddenly you're creating this group of</b><b>smaller LLM models that are starting to argue</b><b>with each other about the response to then get a final query or final answer that.</b><b>And that look, that answer might not be</b><b>always still the best answer, but it forces it to</b><b>double check itself and then be a little bit more creative with that.</b><b>I guess with what the output is.</b><b>Yeah.</b><b>So much fun.</b><b>Like you can actually have this have it make discussions with itself.</b><b>You'd be flying the wall.</b><b>Yeah, exactly.</b><b>Just sit back.</b><b>Like one's Einstein, the other one's Oppenheimer and other ones, blah, blah, blah.</b><b>But like they call that mixture of experts.</b><b>So you're getting it to go back and forth.</b><b>The first one I saw when people were doing</b><b>this, it wasn't from a research paper that</b><b>we watch a lot of like YouTube podcasts on AI.</b><b>And one of them was showing that there was a ball in the cup problem.</b><b>Like I've got a ball in a cup here and then I go outside to another room.</b><b>I turned the cup upside down and then I come back into this main room.</b><b>Where is the ball?</b><b>And when they were doing chat GPT 3.5,</b><b>that version couldn't just straight shot, just</b><b>asking it once and not getting it to think</b><b>it would get it wrong or like half the time</b><b>it was right.</b><b>But when they did Mitch mixture of experts</b><b>on GPT 3.5, they were able to get a higher</b><b>increase in, sorry, a better level of accuracy.</b><b>But then GPT for just had better reasoning</b><b>when it first came out and then to you, but</b><b>the reason and Omni now it's just inbuilt that it's got better reasoning.</b><b>And so that's definitely, you know,</b><b>something that you can take advantage of because to</b><b>your point as to how you put it together,</b><b>this mixture of experts, you can get it to</b><b>do things like that.</b><b>People even ask us, and this is a common</b><b>question we get, I don't know what prompt to ask.</b><b>Okay.</b><b>Well, what are you after?</b><b>I'm after this.</b><b>Well, putting what you're after into the AI,</b><b>whether it's a whole tool or others, and then</b><b>they please help me craft a better prompt.</b><b>Yeah.</b><b>And even if you don't know, well, like,</b><b>cause I say to it when I'm not sure, cause even</b><b>we, I get unsure cause I don't know exactly what the features are.</b><b>And I'll say, please craft a better prompt</b><b>to help me get a more rounded, reasonable</b><b>and well thought out response.</b><b>If you don't remember what say that just</b><b>put out your prompt and go, what is blockchain</b><b>and go make this better, make this prompt better.</b><b>That's it.</b><b>You can be as simple as that and you can learn</b><b>along the way and you can even do other things</b><b>like when you give the output, tell me what else I should be thinking about.</b><b>So when it gives you that output, it even</b><b>gives you what next to think about and you</b><b>can choose whether you want that as the next prompt.</b><b>It's crazy what you can do.</b><b>And like, there's the thing that then people</b><b>ask you is like, why does this, why is this</b><b>applicable to me?</b><b>Right?</b><b>So if I'm learning all these different</b><b>responses, like why would I want to have the tool argue</b><b>with itself or why would I want to have learned</b><b>about these different ways of asking questions</b><b>to get responses in certain ways?</b><b>And again, it comes back to applicability,</b><b>but like the education side of it, right?</b><b>If you're able to start getting the tool to</b><b>answer in a very specific way that provides</b><b>a more formal education response, huge bonus</b><b>rather than someone going in there just going,</b><b>Hey, can you give me this essay that does</b><b>this, this and this and that is getting the</b><b>answer straight away.</b><b>You can actually tell it provided in a</b><b>meaningful way, probably far superior than sadly, a lot</b><b>of the education system is currently doing to people.</b><b>But then, absolutely.</b><b>And on the consulting side of things, or</b><b>even the research side of things, you can start</b><b>to validate things in a way that is more</b><b>powerful and you're not just getting straight answers</b><b>and then thinking, okay, that's the correct response.</b><b>You're able to start getting it to challenge</b><b>you challenge what it is doing, which is the</b><b>whole meaning research.</b><b>You're just challenging, challenging what you're learning and what you're seeing.</b><b>Yeah, absolutely.</b><b>It's not quite an adversarial kind of</b><b>model that is there to kind of red team as that</b><b>as you do in software engineering, like your</b><b>red team to like really test and hone where</b><b>can you break things in the AI space you do it through prompt engineering.</b><b>You prompt inject attacks and see if it will</b><b>give out the instructions that it's not supposed</b><b>to or give details of documents and not supposed to, et cetera.</b><b>But what's interesting as well with all of</b><b>that is that these are discoveries that people</b><b>are making over time.</b><b>They're playing around with things in real</b><b>time and we're seeing it shared globally.</b><b>This is an industry that I feel like is</b><b>moving at a really fast pace compared to like other</b><b>innovations that don't have this level of hype.</b><b>So the hype can be bad for, you know,</b><b>valuations going crazy and stuff and people losing money</b><b>because they invest in something that really wasn't valuable.</b><b>But it's also good when it comes to maybe,</b><b>maybe, um, but it's come to the, you know,</b><b>you always got to take everything with a</b><b>grain of salt, like whether it's humans doing it</b><b>or it's AI doing it, you've got to like validate these things.</b><b>And, um, AI is not just going to be this</b><b>like quick fix or this silver bullet or the</b><b>AI told me this.</b><b>So therefore I'm going to take this medication.</b><b>No, there's going to be a doctor or a medic,</b><b>you know, someone, an expert in, um, the process,</b><b>like human in the loop.</b><b>But I think, um, you know, one thing that I find quite interesting with all of this</b><b>is that people are looking at, uh, maybe</b><b>this is like the next topic, but people looking</b><b>at, uh, and I'm talking about experts here.</b><b>We see these things initially, like it's</b><b>like a wave where initially you, you had people</b><b>saying that AI is going to take jobs.</b><b>Oh, okay.</b><b>Let's all be scared.</b><b>And you know, um, I think it was right to be scared, but at the same time, the fear</b><b>mongering, the fear mongering is it, it's, it sells, right?</b><b>And so it makes the news media and it sells headlines and it sells YouTube clicks.</b><b>And um, we are going to try not to do that,</b><b>obviously, but at the same time, it led to</b><b>this quelling of the fears, a laying, a</b><b>swaging, all these words that I learned with AI and</b><b>my old, I did English at uni, um, or a high</b><b>school, like we all did, but advanced English.</b><b>But anyway, I digress ADD.</b><b>Uh, so the thing was is that made to, to</b><b>call those fees, you, people said that tasks are</b><b>what is going to like be fixed with AI.</b><b>Don't worry.</b><b>It's not going to take a job.</b><b>It's just going to do tasks, mundane tasks</b><b>even, but then experts, and I have discussions</b><b>with, you know, experts in different fields,</b><b>some of them look at AI and go, I don't need</b><b>it because that's a mundane task.</b><b>So it's had this double edged sword effect that experts think they don't need it.</b><b>I think they do.</b><b>It's debatable as to whether they need it.</b><b>Interesting word.</b><b>Yes.</b><b>Yeah.</b><b>I think to some extent they do, but let's</b><b>just, I'll take that, I'll take that side,</b><b>that extreme.</b><b>So I'm not an extreme guy, you know,</b><b>otherwise there'd be a fire truck like that'd go after</b><b>me.</b><b>Just a big siren happening outside.</b><b>You know, and only, what is it?</b><b>Only the Sith deal in absolutes.</b><b>Sorry, George.</b><b>Um, but you know, if I take that extreme</b><b>absolute that experts need AI, what do you have to</b><b>say that?</b><b>Uh, I've, I just got lost in the AI in the quote that you just made around this.</b><b>And now I've, I've literally got that whole scene playing out of my head.</b><b>I have the high ground.</b><b>Everyone can OB and, uh, pulling out his legs.</b><b>Oh, this is absolutely.</b><b>I love that.</b><b>Um, uh, fantastic.</b><b>Oh, look, I was, I was, um, honing in on</b><b>the, that the word need and the ultimate thing</b><b>around it, particularly my role and the work</b><b>that I do in this space is around experience</b><b>and the, I think what Lofton gets a lost between,</b><b>um, especially with, especially with specialists,</b><b>experienced users, you know, people who have</b><b>specific fields, qualifications, et cetera,</b><b>et cetera.</b><b>And they approach AI with this fear of what</b><b>it does, but then they also approach it with</b><b>this, Oh, it's being forced upon me.</b><b>There's a need for it.</b><b>It's not answering the questions I wanted to answer.</b><b>Yeah.</b><b>I don't want to use this challenging what I do.</b><b>It's like, it's going to take over my job.</b><b>You know, the old South Park is doing a deck of germs.</b><b>It's the people from the future and AI is going for the future.</b><b>Whereas if you, if you approach it in a way</b><b>of how can this enhance the experience I'm</b><b>having with the experience that I'm having</b><b>with the work that I do with the, um, with</b><b>the tools that I currently use, it changes your mindset around the potential of AI.</b><b>And so if you can bring it into your, to</b><b>the stuff that you do and utilize it in a way</b><b>that enhances the things that you're doing,</b><b>it can allow you to focus on the other stuff</b><b>that could probably be more meaningful for</b><b>your work and probably generate you more revenue</b><b>in the longterm.</b><b>You know, you know, do I spend so much time</b><b>on sales or having these discussions or doing</b><b>all the minutia pieces within your research</b><b>or generating lengthy documents when you've</b><b>already got all the key points mapped out on a, in some other format or whatever.</b><b>Absolutely.</b><b>It's like, if you approach it with my</b><b>experience, it really changes the perspective that you</b><b>have on, on AI and the work that you do.</b><b>And that's actually a really good point.</b><b>So I thought you might be getting it.</b><b>Like obviously, uh, I am not a Sith.</b><b>Obviously.</b><b>Questionable.</b><b>Although funnily enough, if folks watch the</b><b>acolyte, there is a Filipino slash Sith connection.</b><b>That guy is amazing.</b><b>We probably shouldn't be talking about it.</b><b>People haven't seen this.</b><b>No, well, we won't say anything further.</b><b>Probably give it too much away.</b><b>So if you have started, ignore anything that I say.</b><b>Well, we won't release this until probably next week.</b><b>So hopefully people have seen it all.</b><b>Yeah, yeah.</b><b>We'll see a bit, but anyway, um, I thought</b><b>that maybe you might get into the, the whole,</b><b>you know, taking away my extreme hat on.</b><b>I am very pragmatic is what I'm saying.</b><b>And, um, I don't think AI is there as the be all and end all.</b><b>Certainly not.</b><b>And even then we spoke about like the strawberry thing.</b><b>AI does not do math.</b><b>Like so that current AI models, unless you</b><b>are thinking that GPT five is going to be</b><b>Q star and Q star is a way that language models are supposed to be able to do math.</b><b>That's a whole nother thing.</b><b>Cause right now what they do, if you lift</b><b>the veil, they're actually just taking your</b><b>query.</b><b>They think instructing.</b><b>Yeah.</b><b>Yeah.</b><b>Yeah.</b><b>And instructing, um, Python are SQL code,</b><b>um, to do something that then does the actual</b><b>math.</b><b>So that's like, um, it's not probabilistic</b><b>now it's deterministic, which means it's just</b><b>black and white.</b><b>It's just one answer, right?</b><b>Um, but AI is not good.</b><b>Like, sorry, current AI models are not good at that.</b><b>But one thing that they are good at and</b><b>it's worth testing is, okay, don't get it to do</b><b>math and sums and additions and</b><b>multiplication and correlation analysis or anything like</b><b>that.</b><b>But what you could do is you had a time</b><b>series laid out as a, don't get it to read tables</b><b>because then it's actually using Python to read the tables.</b><b>But if you just had something laid out as</b><b>a, a time series history CSV comma delimited</b><b>as in like, here's the names of the fields</b><b>and commas in between them, here is all the</b><b>values commas in between them.</b><b>And if you had that as a time series, you can understand that.</b><b>And what's been interesting is that it's good at anomaly detection.</b><b>If I feed it two time series and I go now it's not, and we can verify that it's not</b><b>using Python because you can see that</b><b>doesn't use function calling like our app psych.</b><b>We don't have function calling in there,</b><b>but using pattern recognition, it's just using</b><b>pattern recognition.</b><b>It's looking at the language.</b><b>It's just seeing that, okay, well, when</b><b>this happened, this is also like something that</b><b>seemed to happen before you can, you can steer it that way.</b><b>But yeah, digressing and coming back to it</b><b>to clarify though, something there is around</b><b>it.</b><b>It's not good at maths, which is true.</b><b>What it is good though is generating code.</b><b>So it's, which is why it uses Python in the first place.</b><b>So it's really good at you instructing it.</b><b>I need to be able to do something like this to this set.</b><b>And it goes, great.</b><b>Here's the code example we've seen do something similar to it.</b><b>It will generate that Python or other</b><b>scripts like all the programs do a lot of different</b><b>scripts now.</b><b>It's pretty amazing.</b><b>And then it uses that script to then, as you said, give you the one.</b><b>So it's not having to do probability to get a result.</b><b>It's actually doing an actually applicable on the spot.</b><b>Exactly.</b><b>Equation to the problem that you're trying</b><b>to solve, which is why AI isn't solving the</b><b>problem.</b><b>It's using AI to generate a code and the code then solves the problem.</b><b>But it's helping you get there.</b><b>Right.</b><b>Like, so I've seen this thing where my thoughts</b><b>on it are that AI is going to be this efficiency</b><b>gain, even if it means that to get a</b><b>really accurate answer, humans have to be in the</b><b>loop.</b><b>And I've seen arguments from some people</b><b>that say that, well, that's, you know, trying to</b><b>have humans in the loop is just inefficient.</b><b>Like, whether it's inefficient or not, A, it's inefficient.</b><b>If you're going from five hours to do</b><b>something down to one hour, I don't care what you say,</b><b>that is efficiency right there, even if humans have to be involved.</b><b>But the second thing is that it doesn't.</b><b>I don't agree that we have to achieve 100% automation from day one, right?</b><b>We're still kind of day one of this</b><b>journey where we're going to be in 100 years from</b><b>now.</b><b>Like, it's crazy when you look at what</b><b>technology was 100 years ago versus what we can do now.</b><b>As long as an EMP doesn't go off from the sun, everything is reset.</b><b>That's a big problem.</b><b>Yeah, you've got a point here around engineers and hurdles.</b><b>And I really like this insight.</b><b>I think we raised this in the last episode,</b><b>if you went back around, yeah, like 100 years</b><b>ago, people did things differently.</b><b>Yeah.</b><b>Even further back then, we had horses and</b><b>carts and technology innovates and gets you</b><b>to different states.</b><b>We start with horses, I think.</b><b>Yeah, but yes, I think so.</b><b>Do they exist?</b><b>I don't know.</b><b>Are they AI?</b><b>I think the environment's killing them off it.</b><b>But at every single point of those</b><b>innovation releases within the world, there was always</b><b>negativity.</b><b>There was always nuances that people had to</b><b>fight to get the technology into the hands</b><b>of users.</b><b>It's what makes people people. Exactly what makes people people.</b><b>And that hurdle, as you've rightly said, does</b><b>create opportunity off the back end, though.</b><b>Whether it's opening up different types of</b><b>jobs, different types of roles, opening up</b><b>different ways of thinking, enhancing the ways that we think about things.</b><b>And yet there's definitely going to be spaces where there will be sadly losses.</b><b>But unfortunately, that is the nature of how the world evolves.</b><b>It's how business evolves.</b><b>A business when it starts off, and it's...</b><b>Look at how open AI started and how it becomes within the years that pass.</b><b>People lose jobs, the business moves on, new</b><b>technology comes out, it changes how it does</b><b>things, which opens up new roles, but unfortunately you lose others.</b><b>Exactly.</b><b>Speaking of roles and stuff there, and even just like open AI, one is that...</b><b>I can't remember the website, we'll have to</b><b>put it in the show notes, but there's a website</b><b>that tracks job movements and stuff, and</b><b>they're seeing a lot of movements from companies like</b><b>Google to open AI, which is fascinating.</b><b>So those losses at certain companies, there's</b><b>a net gain, I guess, because people move into</b><b>the AI space.</b><b>Speaking of open AI, it reminded me of the</b><b>analogy of the turtle and the turtle, the</b><b>tortoise and the hare.</b><b>The tortoise and the hare.</b><b>The turtle and the hare, the turtle and the rabbit.</b><b>Depending on what part of the world you're</b><b>from, I'm sure there is a similar kind of</b><b>story for this, but the turtle in this case</b><b>being anthropic and even others and the rabbit</b><b>being open AI, who's just known as like just rapid deployment of things.</b><b>And what I'm highlighting is that they put</b><b>out that we're going to do Sora as a way to...</b><b>Google came out with, "We've got 1 million</b><b>tokens," and the next day, open AI has overshadowed</b><b>that news with Sora, and it looks amazing</b><b>for this video generation, but then it's not</b><b>released yet, or like few people have access to it.</b><b>And so anthropic, which was just slow and</b><b>steady when they announced something, stuff</b><b>came out, and then they saw it's popular.</b><b>They go, "And we've also got this."</b><b>And when they announced it, it came out.</b><b>And there's other companies like Runway and</b><b>Luma, which does Dream Machine, which we played</b><b>around with turning your static images into</b><b>animations that didn't do it well enough for</b><b>this.</b><b>But I mean, I should have just done it on</b><b>this image alone and getting these guys to</b><b>move, which we'll do for the next episode, even though it takes hours to generate.</b><b>But my point is that slow and steady and</b><b>releasing things as you're making it is definitely back</b><b>in Vogue when there's been a whole lot of hot air and you can see videos of these.</b><b>It's like games, right?</b><b>Where these games, they look so great at E3</b><b>and these other gaming conventions and stuff,</b><b>when it's released, or if it is released, it's not as good.</b><b>Yeah.</b><b>I mean, yeah, there's lots of reasons why</b><b>that occurs from system performance to consoles.</b><b>And so you can just say there's lots of arguments around that, but it's spot on.</b><b>It's very true.</b><b>And it's that MVP release now is his</b><b>comeback in fashion for these big companies, which</b><b>is fantastic.</b><b>It means that you're putting tools into</b><b>people's hands earlier and getting them to test and</b><b>play around and feedback and then move on</b><b>to the next thing a lot faster than if you</b><b>just went, "Hey, here's this thing.</b><b>Oh crap, we've got to spend years trying to</b><b>fix it because it's horrible," which gaming</b><b>companies do a lot of that traditionally.</b><b>Yeah.</b><b>I think it's a tale as old as time, and</b><b>you're going back to tell us all the time what you</b><b>said there about people just fearing these things.</b><b>We saw it with blockchain and crypto because we're working in that space.</b><b>And yeah, many people might think it's</b><b>scams, but you'll not know it, but your payments</b><b>and the ways that we do business will just be</b><b>much better because of blockchain technology</b><b>that is just continuing to be worked on, even</b><b>though all the hypey stuff crashes and burns</b><b>and comes back again.</b><b>But things like digital escrows, like in</b><b>construction, a lot of tradies don't do escrows because</b><b>it's very expensive, and yet with blockchain, we can do it.</b><b>And we did it with the RBA on a project</b><b>there, and we're doing it with other projects we're</b><b>working on.</b><b>But the technology, if it's good, it will prevail.</b><b>And it's the same thing with AI.</b><b>I see the same messages that crypto critics</b><b>or tech critics say, "Oh, it's all the hype</b><b>stuff.</b><b>Where's the real value and stuff?"</b><b>And it's just like it's being built up.</b><b>You're just not hearing about it because the</b><b>really massive things that have fallen over</b><b>or really hypey stuff is what sells.</b><b>It's so true.</b><b>Good NFTs are the perfect example where the</b><b>technology that underpins it is absolutely</b><b>amazing.</b><b>The stuff you can do from ownership of digital</b><b>content to ticketing to transferring of assets,</b><b>et cetera, that stuff is really, really cool.</b><b>And it could have been done straight away</b><b>when it launched, but it wasn't the hype thing.</b><b>Let's do the images and stuff.</b><b>It was purchasing images and all these pictures online.</b><b>And suddenly people latch onto that.</b><b>And then 50 million dollar pictures get sold.</b><b>And then a week later, it's like, "Why did we do this?"</b><b>Of course, why did we do this?</b><b>It's like selling this digital picture.</b><b>It doesn't make any sense.</b><b>The technology was not built for this.</b><b>It was built for something better.</b><b>And I get the trading cards analogy and</b><b>top shot and some of these other things, some</b><b>of the first NFTs that were out there.</b><b>The first NFTs were not that.</b><b>There's some things like rare Pepe's and other...</b><b>I'm joking.</b><b>It's obviously generalizing. You can go down a rabbit hole, obviously generalizing.</b><b>But you're absolutely right that there's</b><b>this really powerful technology and it's even</b><b>being looked at in real smart scenarios such</b><b>as it is a digital ID because it's non-fungible.</b><b>It's unique.</b><b>It is a digital ID for your uni degree.</b><b>And we're seeing other countries do this kind of thing, right?</b><b>Where the NFTs that and then what gets attached to that are your courses.</b><b>And it's an immutable record on a blockchain</b><b>that is hopefully like a good enough blockchain</b><b>that doesn't go down.</b><b>That is just a smarter way to do it rather</b><b>than where's my record, where's the email,</b><b>I have to call the place again.</b><b>Like these things are just better.</b><b>Not even university.</b><b>You've got places like Estonia and other areas</b><b>where they've actually got their whole system</b><b>within the government and everyone's IDs is now managed by on the blockchain.</b><b>It's fantastic.</b><b>So you suddenly have this semi-decentralized</b><b>model, very important thing to put in that,</b><b>semi-decentralized model for you to have control of your data and how it gets used.</b><b>And that is what the technology was built for.</b><b>And that is like a really fantastic</b><b>application of it rather than our traditional sense where</b><b>potentially it gets hacked, potentially it gets stolen.</b><b>I can't control who it gets shared with.</b><b>One government entity has it, the medical</b><b>people have it, and it's just all over the</b><b>place and you have no oversight of that control.</b><b>Exactly.</b><b>Yeah.</b><b>It's a problem.</b><b>It's a massive problem.</b><b>I think that as well on that, the</b><b>semi-decentralized when we're going to the blockchain side of</b><b>things will bring it back to AI soon.</b><b>One thing that people argue for is like, "Oh, it's not decentralized."</b><b>And it's like, that's kind of like subjective.</b><b>You can look at Ethereum and other things and</b><b>go, "Well, that's not really decentralized."</b><b>And other people can look at another way and go, "That's decentralized enough."</b><b>My point is that when it comes to these</b><b>projects, there's a lot of idealistic people that come</b><b>into this space and they'll go, "I want it completely decentralized off the grid."</b><b>And there's benefits to decentralization, but</b><b>then to make it work, there's always elements</b><b>of centralization.</b><b>And so the saying goes that you decentralize until it hurts.</b><b>You decentralize as much as you can, but to make it work, you centralize it.</b><b>So centralize it until it works.</b><b>And so this is balance of the spectrum where you're going.</b><b>The perfect example of that is oversight and security.</b><b>So if something is purely decentralized, if</b><b>it's 100% decentralized, there's no oversight</b><b>and there's no security on that.</b><b>You're relying on the chain and the pass</b><b>through that information, which is, in some instances,</b><b>it is very accurate.</b><b>And many other though, there's a lot of risk there.</b><b>And so once you put oversight on, if you have</b><b>an entity that just makes sure there's management</b><b>and control, you're now detaching it from</b><b>that decentralized decentralization, suddenly</b><b>it's becoming slightly centralized.</b><b>That's one example of many different ways that obviously there is a need for that.</b><b>Absolutely.</b><b>No, agreed.</b><b>Agreed.</b><b>Shall we get some examples?</b><b>Yeah, that's a great idea.</b><b>What are you thinking?</b><b>Should we get examples first or you want to go to some other articles and stuff?</b><b>Maybe we go examples and we jump back to articles.</b><b>I like it.</b><b>I like it.</b><b>I prepared some examples.</b><b>If I may be so bold as to show some examples.</b><b>So I like this.</b><b>So the first one I'll do is all about a</b><b>tool that some people have only just started to</b><b>see that we do.</b><b>Like there are still some people that</b><b>obviously we do a lot of blockchain stuff and we got</b><b>into AI space.</b><b>And I think we did this in the background last week because we were training people</b><b>on how to use chat GBT.</b><b>We were using it ourselves as part of the</b><b>advisory and consulting work and even just</b><b>managing the operations of the various businesses.</b><b>And we saw this need for privacy focused AI tools.</b><b>And so we built psych with Azure services, which means it stays here in Australia.</b><b>Like it doesn't leave the shores, which is great.</b><b>But adding features in there, we've started to realize, oh, hang on a second.</b><b>We need to have a new feature because this is the bug bear that I'm having.</b><b>Like great.</b><b>I'm not going to do AI, but it's not easy to do things like mimicking.</b><b>And this is something that people are asking us more about.</b><b>How do I get an AI that can mimic what it is that I've got without having to always</b><b>feed it my, my style, my notes and stuff.</b><b>And so one thing I want to show on here.</b><b>So what we're seeing on screen and you'll</b><b>see it on the video as well is that here is</b><b>something where we've got this thing called</b><b>dreams where we can set up various things.</b><b>Like in this case, I've got articles,</b><b>things that I've written about, because I wanted</b><b>to be in that kind of style.</b><b>And then another thing that we can do in</b><b>here is like now that it's got my style, I want</b><b>it to do things like I'll call that style.</b><b>I wanted to do things like add an about section about me.</b><b>And so I can do that.</b><b>My thoughts, my feelings, you know, what I think about the world.</b><b>And then let's say there's a news article and here's something from the AFR talking</b><b>about AI that came out the other day.</b><b>I can copy that in and here's this like a news article.</b><b>Now what I can do then is I can talk to it.</b><b>I can go, well, those different notes there</b><b>that I've laid out here, I'm having to remember</b><b>what's a quote that I, and I can do this is great.</b><b>I can refer to each of the notes there, but</b><b>I have to remember what this query is to get</b><b>it to write in the style.</b><b>Like I've given it my style of writing,</b><b>I've given it an about me and I've given it the</b><b>news article.</b><b>And that's great.</b><b>But the problem is like you still, and</b><b>you've got those ingredients there, but you still</b><b>always have to remember how do I actually piece that together to get the same style</b><b>of output?</b><b>So one thing that we've seen is better is</b><b>the first part instead of just throwing all</b><b>your articles in, because that leads to really lengthy prompts.</b><b>What you can do is take a whole bunch of</b><b>articles, preferably ones that relate to the same kind</b><b>of topic, but tell me about the style and</b><b>tone of this author and then give me quote</b><b>examples because if you do that and what we're</b><b>seeing come out on screen here is it's taking</b><b>those articles and it's going Mark is</b><b>informative and analytical and here's examples of that.</b><b>Mark is also optimistic and forward looking tone in his writing.</b><b>Here are examples.</b><b>Cause if you and I are both the same</b><b>description in formal and optimistic, it's going to be</b><b>different for your writing versus mine.</b><b>For the people who are listening in, um,</b><b>what you're, what you're not seeing is a screen</b><b>where you're able to input in one window, your prompts, your information, your, um,</b><b>your articles that you may be referencing.</b><b>And then in the same screen, a secondary</b><b>box appears to give you the suggestive outputs</b><b>that you have some choices around.</b><b>Whereas in the traditional, I mean, we've</b><b>seen chat to teach and a move with the thread</b><b>on the options of what's your best</b><b>response, but at the same time, you're still in this</b><b>entire single parallel view, really long,</b><b>really long scroll back up, talk box by text</b><b>box.</b><b>Whereas here you're able to roll your windows in post-it note format.</b><b>Yeah, very cool.</b><b>And we set it up that way because we were having these really lengthy conversations</b><b>and prompts.</b><b>And so we created this post-it notes thing and you can be specific.</b><b>You, we spoke before at the start about the context.</b><b>The great thing about dreams is that, um,</b><b>having these post-it notes, you can talk specifically</b><b>just to those post-it notes.</b><b>Um, but another thing that, um, we've got</b><b>here is that if I have to remember, even if</b><b>I've got all those, uh, post-it notes, I</b><b>have to remember what's the query I want to ask</b><b>about it.</b><b>I was saving that in Google drive.</b><b>I was writing it down on my hand.</b><b>I was like doing whatever, right?</b><b>But you got to remember the exact query.</b><b>Guess what?</b><b>I can add another post-it note in there,</b><b>which is something that we'll call instructions.</b><b>It could be anything else.</b><b>But if I add that, I can put in step by step what it is that I want it to do.</b><b>So I give it a label.</b><b>That's what I'm doing on screen here.</b><b>And then I'm writing in steps, your mark, refer to the about note, analyze the news</b><b>note, respond in the style note.</b><b>And then I can even edit those steps.</b><b>I said in here, like I want it to give me three key points and a historical quote.</b><b>I'm going to take away the historical quote.</b><b>I just want it to do the three key points.</b><b>And so what it ends up meaning is that he's the final one.</b><b>Now that it's got the instructions as a note</b><b>there, I can just simply say at the top, run</b><b>instructions.</b><b>That's it.</b><b>And anyone who uses this same set of ingredients will get the same kind of output.</b><b>So it's, this comes from like the</b><b>frustrations that we had as a business using these tools</b><b>and from customers having the same kind of thing.</b><b>So if you were to run something like this,</b><b>you would just be able to control it more.</b><b>And all of this, if we go back to how do you control hallucinations?</b><b>How do you control the randomness?</b><b>You control the context.</b><b>So what are in the notes?</b><b>You control the execution steps as well.</b><b>You're setting a command, a set of command</b><b>prompts that you're able to then reference</b><b>whenever a scenario is.</b><b>So you could have a whole bunch of input that you may have created.</b><b>You may even have secure and private documents that the tool is linked into.</b><b>And then all you need to do is input a</b><b>command that you would have defined with a set of</b><b>steps.</b><b>And that will give you the outputs in the consistent format, right?</b><b>That's it.</b><b>Rather than you having to think about the</b><b>same command and that same prompt every single</b><b>time and retyping it in or going, Oh, I</b><b>need to remember what steps one through six, or</b><b>you might do something slightly different</b><b>last time and go, why didn't I get a similar</b><b>output?</b><b>And that's what creates that consistency.</b><b>It's what leads to the frustrations that</b><b>ordinary people have in the mainstream using these</b><b>tools.</b><b>It's like, what prompt do I remember?</b><b>And it's like, well, if we just set up the</b><b>tools in better ways, like we have here, you</b><b>can do that.</b><b>There's one way to do it.</b><b>We even have AI agents that just have these instructions.</b><b>I've written these out, but if you've got</b><b>something that you need to repeat over and</b><b>over again, we've got another section where</b><b>there's AI agents that are just instructed</b><b>to do this.</b><b>And it's the same principles.</b><b>And the more powerful, even additional powerful</b><b>number there is telling it where to pull certain</b><b>insights from in a certain way as well.</b><b>So it's not just going step one, do this step to do this.</b><b>It could be doing step three, do this, but</b><b>always reference XYZ when you do this one.</b><b>Step four, always reference these things.</b><b>And therefore you're removing your</b><b>hallucinations in a more consistent manner as well.</b><b>And you can even get them to do things like where they check each other at the end.</b><b>You know, you have these AI agents that they</b><b>do various steps, but then there's a compliance</b><b>agent at the end, or even one where I've seen it for lengthier documents.</b><b>This is where the human expertise comes in.</b><b>But if you get an agent to look up</b><b>something from a thousand page document and then it</b><b>may or may not find all of those</b><b>references, you can get another agent to refer to the</b><b>output of that first agent and then also</b><b>look at the document and go, did that first</b><b>agent miss anything?</b><b>And that's when it really gets more powerful.</b><b>So really interesting stuff there.</b><b>So I thought I'd just highlight that we're</b><b>going to do more, you know, insights for our</b><b>clients and people that want to use like</b><b>and yourselves on how to be, this is all part</b><b>of programmatic.</b><b>We call it semantic programming or, you</b><b>know, I see the language and way we use like AI</b><b>tools.</b><b>So thinking like programmers is just that</b><b>our coding language is English, not English,</b><b>or another language.</b><b>Yeah, but English is what we've got here.</b><b>So it was actually, you know, we, um, we</b><b>run a Friday little networking event where we</b><b>bring some people and we have discussions.</b><b>And one of them was in that exact line, what is the future of coding?</b><b>And it's like, we all got to the conclusion</b><b>that it's just going to be, it'll be all in</b><b>English.</b><b>If we think about how it's gone now with</b><b>react and JavaScript, they're very kind of written</b><b>code languages with like all their reference</b><b>points and they're, I'm using the wrong terminology</b><b>because I'm not a developer, but they've got, you know, databases of insights that</b><b>it pulls all that stuff from.</b><b>Exactly.</b><b>And the future is just going to be us going, do this.</b><b>And boom, that's your coding language.</b><b>If you can think logically and you know how to</b><b>lay out your language, you can win hackathons.</b><b>There was a project manager, not a coder.</b><b>Maybe she knows a little bit of code.</b><b>She understands it at least, but not is the one that is writing the code.</b><b>She runs teams.</b><b>She won a hackathon last year from just using AI tools.</b><b>But something that you mentioned before, and we were talking about programming and</b><b>like the code that comes up in Python scripts.</b><b>Have you seen, have you played around with Claude 3.5 much?</b><b>A very high level, but yeah, it's amazing.</b><b>It's crazy, right?</b><b>So what I wanted to show on screen here were two things.</b><b>A simple one where I've got Claude,</b><b>basically this is on LinkedIn, a guy that I follow,</b><b>shout out to Tommy Ragulli.</b><b>He's got some great charts on LinkedIn.</b><b>This is a chart that he had on Ethereum and</b><b>where all the Ethereum is across smart contracts,</b><b>et cetera.</b><b>And what I did, I took that image and I</b><b>said to Claude, can you recreate this chart?</b><b>But I want you to make it so that I can</b><b>actually play around with the data that's in there.</b><b>I want you to recreate the look and feel of</b><b>it, but like I want to change the numbers.</b><b>I want to change the titles of the chart and stuff.</b><b>Now chat.gbt, you can do this kind of stuff</b><b>and it will give you the code, but then you</b><b>have to apply the code.</b><b>And so what we're seeing on screen here is that I've had that prompt.</b><b>It's now going to write up the code and it will</b><b>basically go, okay, I'll do a react component.</b><b>I'll here's the code.</b><b>I can see it on a side screen, which is great.</b><b>And then what we'll see in a second is that it will actually show you a preview.</b><b>Now it's still limited in terms of what it</b><b>can show, but here's a preview where yeah,</b><b>literally I can change numbers in there.</b><b>I can now have that chart and I could apply that code to a website.</b><b>I could apply that code into an app.</b><b>I used to do a lot of data analytics apps.</b><b>I built ETF tracking tools and stuff.</b><b>I might have a go at doing this kind of</b><b>stuff again, but the cool thing is that you can</b><b>actually do this kind of stuff.</b><b>It's I'm looking away from the camera, but</b><b>I don't know if you can see how big my smile</b><b>is right now.</b><b>Just looking at this.</b><b>I've never seen Chris happier except even on the day of his, no, no, no, no.</b><b>Shout out to Gabby and Lana.</b><b>But the last one I want to show you from Claude is speaking of charts.</b><b>I used to work in data analytics and I built</b><b>stuff with Power BI with ClickView with all</b><b>these other ClickSense and all these other tools.</b><b>I built something called ETF tracker that</b><b>I had been updating for a little while, but</b><b>then I stopped doing that because I've</b><b>just been so busy with all these other things.</b><b>I fed it images of ETF tracker and I said, I want you to recreate this.</b><b>I had a little back and forth because it doesn't always get things right.</b><b>I'm saying here's something in Power BI</b><b>with two images and I just wanted to create the</b><b>chart that sees on screen.</b><b>It's actually making up the values from what it can see.</b><b>But I also said, I've got a menu that pops up.</b><b>Can you recreate the menu?</b><b>That was my first part.</b><b>So we're watching this and it gets it wrong.</b><b>It goes, oh, well, I can't show it in the preview because it's using a library that</b><b>the preview area doesn't like.</b><b>And so I said, well, do it again with supported libraries.</b><b>Now fun fact, this will still get it wrong.</b><b>It eventually gets it right.</b><b>This is the reason I'm showing the video, but you can go back and forth with this.</b><b>So I said, well, can you do it with supported libraries?</b><b>It still did it with non-supported.</b><b>So I said, okay, ignore the bar chart icon</b><b>library that you're trying to use and redo</b><b>it again.</b><b>And it uses now a library that it's better</b><b>at and it's not as good looking, no doubt,</b><b>but this is the way that a non-react kind</b><b>of programmer can take something off the web,</b><b>like a website and try to get it to recreate.</b><b>And here's the pop-up.</b><b>It even does the navigation.</b><b>Yes.</b><b>That's crazy.</b><b>So you would then take that code and iterate on it.</b><b>Now it's limited in terms of claw 3.5 free, right?</b><b>You can't do as much as that, or you would</b><b>hit limits there in terms of the context window</b><b>size in terms of the conversation.</b><b>But if you pay your $22 US a month, you unlock more context windows.</b><b>So I'm still exploring that dude, fascinating stuff.</b><b>Well, imagine if you put in a designed</b><b>website, which is the other examples that I've played</b><b>with is you get.</b><b>So you put in like a fully designed kind of website?</b><b>Not fully, fully designed, but like pulled</b><b>out an image or two that I've shaped in Fig</b><b>Mode, some of my UX projects and dropped it in</b><b>there and it's created a like mostly interactive</b><b>page, like almost to perfection.</b><b>It's crazy.</b><b>It's absolutely insane.</b><b>You know, there's still a lot of things that</b><b>gets wrong when you know, particularly when</b><b>it comes to scalability across devices and stuff like that.</b><b>But the fact that you are skyrocketing from a</b><b>picture into a landing page that is actually</b><b>quite functional is like, it just blows my mind.</b><b>And I saw someone like doing, they put in</b><b>Spotify and it recreate Spotify without all</b><b>the images, but it's still recreated.</b><b>It's still recreated.</b><b>That's exactly right.</b><b>And like for me, that's the power of that is I do a lot of prototyping with clients</b><b>and customers, right?</b><b>And so I can elevate that ability to prototype now.</b><b>And Fig Mode does has a few features where you can do things like that.</b><b>But nowhere near as powerful as what we're just witnessing there in Claude.</b><b>That's crazy.</b><b>And so I could create these semi</b><b>prototypes that are clickable, presentable to people</b><b>just by going, Hey, can you please recreate</b><b>this image as a website, like, or an application,</b><b>like just done.</b><b>And what happens in a world where, I mean,</b><b>initially there's only a few people that are</b><b>really looking at, I mean, not a few, there's</b><b>a lot of people looking at this, but I think</b><b>experts will be the ones that just because</b><b>it gets stuck, it doesn't just do it off a</b><b>simple prompt.</b><b>It doesn't just go, here's a website, here's a few images, make this interactive.</b><b>It won't do that.</b><b>I have to go back and forth.</b><b>I have to tell it that library doesn't work.</b><b>Do this expertise is still kind of needed.</b><b>But in a world where we get to just one</b><b>shot, you know, I don't have to like go back and</b><b>forth with it too much.</b><b>And anyone can almost do that.</b><b>What do experts then do?</b><b>That's an excellent question.</b><b>That was one of the questions that we had</b><b>raised in our session we had a few weeks ago.</b><b>Where is the relevant, I guess, in both instances</b><b>from an expert's through to what are the developers</b><b>going to do?</b><b>Like people who are currently building these</b><b>things from scratch using their own libraries</b><b>or the React and all that, et cetera.</b><b>Like their jobs are going to narrow down in terms of what they can do.</b><b>For me, it's going to be a while, I think,</b><b>till AI gets to a point where it is going</b><b>to be 100% able to give you the correct, like the output that you need.</b><b>Agreed.</b><b>And like I'm talking like five, 10 years, maybe more.</b><b>And there's going to be a need for that one.</b><b>This is one of the things that I firmly</b><b>believe will always be the case is it's not, it's</b><b>going to have that creativity level.</b><b>You're still going to need to have that human</b><b>creative element that AI can't achieve because</b><b>it is not, it is machine learning.</b><b>It is literally just gathering a whole</b><b>bunch of insights on everything that everyone's</b><b>done and giving an estimated output.</b><b>Right.</b><b>So for a long time, it's not going to be acting like a human and be creative.</b><b>So it's going to be that creative sprinkle</b><b>that specialists or developers and designers,</b><b>et cetera, always going to be wanting to put</b><b>on top of things, which I think will accelerate</b><b>innovation.</b><b>But we'll see what happens there.</b><b>The other part of it is when it comes to development</b><b>is less of the creativity, but also the validation,</b><b>the checks, making sure, adding that extra</b><b>touches at the AI just can't do straight out</b><b>of the box.</b><b>Right.</b><b>So it will, they'll still be in need for</b><b>those roles and those people who have previously</b><b>had to do a lot of development.</b><b>Sadly, I think a lot of those skills will</b><b>diminish, but they will become specialized</b><b>in how do we get it from what the AI has</b><b>done to this hyped up amazing experience off the</b><b>back end.</b><b>Agreed.</b><b>The, what will twist everything that we do</b><b>in the future is that I think the standards</b><b>of what people are expecting in the market are going to accelerate upwards.</b><b>And we're already at a pretty high level</b><b>of, you know, what businesses are expected to</b><b>deliver.</b><b>But now with AI doing majority of the work,</b><b>there's going to be a higher expectation than</b><b>ever with the stuff that gets from applications to products that have developed.</b><b>And that is going to be solved by human elements in the short term.</b><b>And us going, what is the standard that we</b><b>have to meet and how do I adjust and make</b><b>that better?</b><b>You touched on a point there that I was going</b><b>to talk about something else, but like, I'll</b><b>just touch on this, but rising standards, I see it with folks like Ethan Molek, you</b><b>know, on LinkedIn.</b><b>He's an educator.</b><b>I can't remember which uni it is.</b><b>Sorry, Chad, that's Ethan.</b><b>Has great content and research and wrote a really</b><b>good book on this AI space called Co-Intelligence,</b><b>augmented between man and machine.</b><b>And I like his views and stuff, basically.</b><b>And one thing that he said is with his classes,</b><b>it's no longer this thing where you just expect</b><b>the same outputs that people had before.</b><b>But now with AI, you actually have to increase and raise the bar so much more.</b><b>So it has to be much higher in terms of expectation.</b><b>But what this leads to, and just like what</b><b>you said, like consumer expectations need</b><b>to be higher.</b><b>Educators and the outputs from education</b><b>need to be higher and stuff, given that there's</b><b>AI, that's the way that we adjust rather than banning it.</b><b>Let's look the standards, right?</b><b>But what happens when, you know, let's</b><b>just say that there are some in the university</b><b>and look, I get it that we all have to try</b><b>to do this, but there may be as the transition</b><b>happens, some students that are really</b><b>embracing it and they just embracing the higher level</b><b>challenges and some that go, well, look, if</b><b>I'm able to prove that I don't use AI, I just</b><b>want to show that I use my brain and stuff for this.</b><b>Now it is like maybe it's a mixture of different types of tests.</b><b>You have to show stuff where it's not AI</b><b>and it's more oral exams and stuff like that.</b><b>And you're just like speaking, presenting and doing that kind of stuff, as well as,</b><b>you know, where let loose kind of like</b><b>Peter Thiel of whatever, where he's from saying</b><b>that we should have an Olympics where everyone just takes drugs.</b><b>Everyone can just use AI, go ham, see what you can create.</b><b>Like, what do you think about that?</b><b>Like as we transition the two worlds, the artisanal, I'm not using AI, I'm using my</b><b>hands of my brain versus I'm augmented, I'm using AI.</b><b>I think that is going to be the case.</b><b>Like especially, it's already happening now, right?</b><b>There are specific, even art competitions now where there's an AI related only art</b><b>competition.</b><b>And, you know, just like when they had the</b><b>real art competition, someone submitted an</b><b>AI and one and the recent one, someone</b><b>submitted a hand draw or a custom one that they had</b><b>built themselves or designed themselves and they won.</b><b>And everyone's like, oh, why can you submit this as an AI comp?</b><b>It's like, you did the same thing like months ago.</b><b>But that is definitely, I reckon that will</b><b>definitely happen, that shift of, I mean,</b><b>think about, there's lots of stuff in the world now where we do that.</b><b>It's like we're, you know, racing now, we're</b><b>starting electric cars versus the petrol cars.</b><b>Yeah, yeah, exactly.</b><b>It's all those types of things.</b><b>There's always going to be a division between,</b><b>you know, what is manual versus what is automated</b><b>and then how do we create a world or a</b><b>competition around the automation versus around the human</b><b>element, which is, you know, that's the creative side of it.</b><b>You don't want to get rid of creative.</b><b>No.</b><b>You don't want to get rid of how a person</b><b>who is athletic is pushing themselves as much</b><b>as they can without the use of drugs.</b><b>Of course, those people need to compete with each other.</b><b>There should always be a space for that</b><b>separate from, you know, the people who want to enhance</b><b>themselves in certain ways.</b><b>It'd be really funny if, you know, right</b><b>now, and we call the Paralympics and some call</b><b>it like the Special Olympics and stuff,</b><b>especially in the States, but it'd be really funny in</b><b>the future world where AI and like enhanced humans become a thing.</b><b>And that's the norm.</b><b>And the non-enhanced human stuff in our heads, like it's already human testing.</b><b>You're a link, but maybe because you're a</b><b>smarter athlete now in team sports or even</b><b>individual sports like tennis.</b><b>But like imagine if it was like you become</b><b>this, you know, the norm is enhanced athletes</b><b>and the non-norm, the abnormal of what we</b><b>consider today normal, like the normal athletes</b><b>that run and jump and kick and all that, that</b><b>becomes Special Olympics now, like the roles</b><b>reverse.</b><b>I mean, look at the cinema world, right?</b><b>I think we, technology has moved so far now</b><b>that there's so many movies and Marvel, all</b><b>those, everything has, has a digital enhancement to it.</b><b>And then you get someone like Christopher Nolan, who releases the movie, he doesn't</b><b>use any, any sort of tech, he does it all</b><b>by hand and it's like, oh wow, he's done this</b><b>thing that we used to do for the last 100 years of cinema.</b><b>He actually went to a black hole for the interstellar.</b><b>Like it's crazy.</b><b>It's crazy.</b><b>But it's just like, it, there will always, there will be that separation.</b><b>And I think exactly what you said, it will become the unique thing to do.</b><b>Like all that person does it by hand or who does that anymore?</b><b>Can I just say something on the, the expertise stuff?</b><b>Oh no, go ahead.</b><b>When was the last time you drove a manual car?</b><b>I so just to let folks know I steer my vehicle.</b><b>I know I don't drive.</b><b>But that's it.</b><b>Like anyone who drives a manual car now,</b><b>especially in Australia, probably a bit of America.</b><b>I think there's still parts of Europe that</b><b>I was, you know, I don't get why they call</b><b>it automatic.</b><b>I still have to steer.</b><b>Well, I mean, if you have a test, but the,</b><b>but it's, there are so many, especially the</b><b>generation coming up, the idea that they have to change gear blows their mind.</b><b>Like it's, it's now, it's now a skill.</b><b>It's like in Back to the Future, they go to</b><b>the future and Marty's like shooting at the</b><b>arcade machine and the kid next to him and goes, you have to use your hand.</b><b>It's like a baby's toy.</b><b>But I just want to say on the expertise thing, creativity, sorry, creativity.</b><b>One thing, and this is my argument that I think experts are absolutely needed.</b><b>So these things are by very nature.</b><b>They're not creative.</b><b>They might look like they reason, but the</b><b>only way they can do that is that they have</b><b>something in their training set, whether</b><b>it is the raw AI model, or you've introduced</b><b>new knowledge into it via rag by adding documents, et cetera.</b><b>But either way, its ability to reason is limited.</b><b>Yes, it's gotten better, but it's still</b><b>limited compared to like human experts and stuff.</b><b>And there's a lot of tests and I'm sure there will be breakthroughs.</b><b>What I'm saying is that where we're at</b><b>now, which is the worst AI is ever going to be</b><b>that Claude stuff.</b><b>That's the worst it's going to be because it's always getting better.</b><b>But when you think about experts where they</b><b>come in is that the reason I think experts</b><b>are needed in this is they know the special knowledge.</b><b>They know the special source.</b><b>If you add your special, if you add your</b><b>expertise into this, sorry, this is not PG folks, but</b><b>if you add your expertise into this, you</b><b>will get, that's where the expertise is really</b><b>powerful because stuff that the model</b><b>doesn't know, bring in digital village expertise,</b><b>bringing in psych and not centralized expertise.</b><b>And our friends at Friday share that to them.</b>

People on this episode

Podcasts we love

Check out these other fine podcasts recommended by us, not an algorithm.

The Digital Village Show Artwork

The Digital Village Show

Digital Village