Episode 12
Solving Your Own Problems with AI
In this episode of AI Evolution, the conversation picks up after a busy few weeks in AI, with plenty to catch up on.
David, Ben and Dave discuss the latest releases from Claude and ChatGPT, how the tools compare in real-world use, and why some people are moving closer to AI while others are starting to pull back.
The discussion also explores the growing power of no-code and AI-assisted app builders, including how tools like Lovable can now turn a rough idea into a working prototype in a matter of minutes. That leads into a bigger question:
What happens when anyone can build their own app to solve their own problem?
The episode also looks at OpenAI’s possible move into AI-first devices, whether an app-less phone could really work, and why Apple and Samsung may still have the upper hand.
There’s also a deep dive into image generation, creative control, prompting, and the difference between knowing how to use the tool and having a good idea in the first place.
Later, the conversation turns to AI safety, security vulnerabilities, Anthropic’s unreleased Mythos model, and the challenge of deciding when powerful AI systems should be kept private.
Topics covered include:
- Claude Opus 4.7 vs ChatGPT 5.5
- Why Claude often feels stronger for writing and reasoning
- Building personal apps with AI tools
- The rise of no-code platforms like Lovable
- Whether AI could kill the traditional app ecosystem
- OpenAI’s possible hardware ambitions
- Voice-first AI devices
- AI image generation and creative prompting
- Why human creativity still matters
- AI security risks and vulnerability discovery
- AI in education, recruitment and everyday work
- Why detecting AI-written content is still unreliable
A wide-ranging conversation about where AI is heading next, what it means for creators and businesses, and why the most powerful shift may be people building tools for themselves rather than waiting for someone else to build them.
Transcript
We're now getting to the point where everybody, we can all just create individual apps for ourselves now that I don't even care if anybody else ever sees it. All I need it to do is fix a problem for me.
And if I can make my own app that solves my own problem, then I don't need to go and buy an app from someone else.
Alan King:Hello and welcome to the AI Evolution podcast. We're back again. Sadly, this time we're not together in person, but we have got lots of fun things to talk about in the world of AI.
And to be honest, it's been, I think it's probably six weeks or so since last episode. It's been a pretty busy six weeks in the world of AI.
But before any of that, let's introduce Ben and Dave, as they say in Wales here, what's occurring.
David Brown:Hello. Good morning.
Benjamin (Ben) Harvey:Good morning. How you guys?
Alan King:Yeah, how you guys been? You've had a good, good few weeks?
David Brown:Yeah, it's been busy, yeah, yeah, it's been. Studio's been taken off, loads of stuff going on.
I've started up my other AI podcast again because it was still ranking and I. I just felt like it was a shame to not do anything with it and what I've been do just as a shameless plug.
AI now than they did back in:How people's feelings have moved on and what they think and how they're using the tools have moved on. So, yeah, that's. That's what I've been up to.
Alan King:That's a really interesting thing.
I was talking to somebody of the other day, actually, who, you know, two, two, three years ago, I was talking about AI and they were kind of, you know, it's all rubbish. They know, don't trust it, blah, blah, blah. And then I met with them last week and. And they'd had a kind of Road to Damascus conversion.
You know, they'd spent some time with Claude and that was it. They were. They were all in, you know. Yeah, yeah, yeah.
Benjamin (Ben) Harvey:I've been busy directing remotely through Riverside, various people around the world for a project which has been really fun and some of the people I've been working with have become really anti AI, which is interesting. So they've had the. The opposite journey and sort of trying to. Trying to use it less in their life and to rely on it less.
Alan King:That's really. Well, I think that's. There's an episode right there actually. Maybe we can explore that. The kind of.
The sort of the ying and the yang of AI at some point and people moving towards it, people moving away from it and what that means. And also Ben, I believe that you're 21 again.
Benjamin (Ben) Harvey:Yeah, yeah, I was. Had my 53rd birthday this week, so that was fun. Nice to see you, Alan. Yep, thank you. Which they seem to be coming quicker now.
Alan King:There we are, Excellent.
Benjamin (Ben) Harvey:Speeding up.
Alan King:Well, we'll kick off today. I've got a. A long list of things which actually the guys were sort of ribbing me about earlier, so.
But I don't think any of them are too long in theirselves, so hopefully we can kind of bust through them fairly quickly in some respects, but we don't want to undercut them either. So I'm going to start off with the sort of two big announcements in terms of models that have actually happened since we last spoke. Claude.
Opus released 4.7, which I think was pretty significant. And then also Chat GPT launched 5.5 before we get into chat GPT and sort of how they're performing against each other.
Be great just to sort of think about them alone, I suppose. And I've been using Opus 4.7 quite a bit.
It's quite interesting because online and sort of in the media, I've been seeing a lot of people complaining about 4.7 going, oh, you know, they've nerfed, you know, Claude.
It's not as good as it used to be and the performance is degraded and I, I can't honestly say that I've really noticed that in the work that I've been doing.
I mean, equally, I've not noticed any enormous step change either with from 4.6 to 4.7, you know, might be fractionally better but, but, but I've not seen the kind of negative outcomes that sort of. Many people have been sort of talking about. I mean, have you guys used Opus at all for 4.7?
Benjamin (Ben) Harvey:A little bit, but I haven't tested it in anger.
David Brown:I don't use Claude at all. Oh, I use. No, I'm exclusively chatgpt. Wow.
Alan King:Okay.
David Brown:And maybe a little bit of Gemini if I'm in a Google Doc or something and I want to. A little bit of something. But yeah, I, I haven't used. I don't use cloud at all.
I really don't use many Tools, I think, and this is something maybe we can talk about at some point.
But I, I feel quite locked in to Chat GPT at the minute because all of my projects are in there, all of my history is there and I've kind of got it to a state where if I ask it to, you know, to, to write something or to use my voice, it kind of knows and it, it's kind of learned pretty well and I just don't have the stomach for going back and retraining a whole nother AI platform. Yeah, that said, I also don't do any coding at all or anything like that.
So I think if I, if I was doing more software engineering and needed to do some, you know, some scripting or some kind of code or something, I probably would use it for that.
Benjamin (Ben) Harvey:My big problem with Claude is I, I've only been subscribing a few months, but by the time I get to it, my 12 year old son has used all the credits.
He's, he's connected it to Blender this week for doing 3D stuff, he's making games, he's using it for all sorts of development stuff and interestingly it's really spurring him on to learn C and other coding languages. He's not just using it as a, you know, to do all his work for him, but generally by the time I have finished my work all the credits are gone.
Alan King:So I've, I've, I've basically kept Albert away from Claude 100 for exactly that reason, Ben. And you know, so he has access to Chat GPT which doesn't seem to kind of run out of credits nearly as quickly as Claude.
Benjamin (Ben) Harvey:No way. No.
Alan King:And so he sort of has sort of free uses that. Whereas so for me any serious tasks I'm doing now are in Claude and then the sort of light, lightweight tasks, if you like.
I use Chat GPT for, say that I'm not using Claude credits on kind of frippery or lesser tasks, you know, and I suppose it's the, it's the sledgehammer for the, for the sledgehammer activities, but not for cracking the nuts, you know, scenario. And I think that that balance seems to work.
I, I do understand what you're saying, Dave, about the migration thing because I had that moment about a year ago when I was, you know, I was using almost Chat GPT exclusively at that point.
Well, I was using malice and a couple of other things, but in terms of the main AI system that was being used for me, but I really start to become Quite aware that when I did side by sides with Claude, I was going to get better outcomes. So I went through the pain of migrating effectively, you know, my workflows and conversations and things across and now it's there.
And if I wanted to go back to chat GPT as a sort of serious, as my serious model, I have the same problem, you know. So let's imagine a world where tomorrow OpenAI announced GPT6 or something and it's just way ahead of everything else and it's a massive step change.
Then I'd have to sort of sit there and think how do I go about that and migrate back across?
Because you're right, like all ecosystems like Apple, like Google, whatever it might be, you kind of get built into that world, don't you, over time and then they.
David Brown:Which is by design,.
Alan King:It's like the mafia. You'll never leave.
David Brown:Exactly, exactly. And it's.
And this goes back to maybe something that, that, that we've touched on a little bit and probably will continue to talk about, which is what we use the various tools for and how we use AI in our sort of day to day lives.
And you know, in the beginning, the last, you know, in the very beginning it was fun to go and get it, to write stuff and to create things for you and to do all sorts of kind of original activities, if you want to call it that. Now I use it as an editor more than anything else.
So I pretty much write everything first and then put it in and ask it to edit that for me or have I forgot anything or just to tweak it to kind of see if I've, you know, missed any major points or if it can come up with, you know, maybe some more elegant language. And it's quite good at doing that. And then literally I just kind of do the small edits for my own stuff.
So I don't, I, I haven't really used it except, and I'll give this caveat, I used a platform the other day to, to build my social network platform that I've been wanting to build for several years and literally it took two queries to build a nearly functional microblogging platform. So, but, but even that was quite light and pretty easy to do because I knew exactly what I wanted. Um, but yeah, I.
Benjamin (Ben) Harvey:What did you use for it to build it?
David Brown:It's like a tool called Lovely or something.
Alan King:Lovable.
David Brown:Lovable, yeah, it was lovable and somebody had done some work for me and they created a, basically a wireframe in Lovable and sent it to me. And so I was looking at the wireframe and I was like, oh, this is pretty cool. I might have a little, you know, I might have a go at it myself.
And I was just sitting there and I. So I signed up for the free account or whatever and I was just sitting there and I'm like, what should I.
Like, I didn't even know, like, I didn't want anything. I just thought, you know, what could I put in?
And I just thought, I've got this, you know, I've had this idea for this, you know, platform that's got only verified users or whatever. So I was like, okay.
And because I've learned over time how to write, I think a fairly efficient prompt, basically I only had enough tokens to do like two queries and that was it. So I just thought, okay, I, I scoped out kind of what I wanted, did the first one.
It literally created a better looking version of all the stuff that it took us a year to build a couple of years ago. And I just looked at it and was like, Jesus.
And then I used my next token to get it to build all the sub pages and all the other stuff that kind of needed to go with it. And then it was like, oh, you're out of tokens for the day. And I was like, okay, so I'll wait till tomorrow.
And then the next day I went back and just said, now connect all the pages together so that they all work the way they're supposed to and everything's integrated with each other. And it's like, okay. And that burned my token for that day, but basically I now have.
It's not plugged into any back end, so there's a huge amount of work that would still be necessary, but the, like, I could take what I have to go and get funded easily to go and get funding. And, and that work took us nearly a year to do last time.
Alan King:Yeah.
David Brown:When we did it a few years ago. So it's nuts.
Alan King:Yeah.
Benjamin (Ben) Harvey:I've got a really similar story.
My, for my birthday, my nephew, who's in his mid-20s, came down and he's a fantastic photographer and he had, with the help of OpenAI, had built an app for doing calendars and all sorts of other printed material, photos, montages. Really clever out. It's taken him a year. Exactly the same with OpenAI.
He's learned as he's gone along and my son was sitting next to him on the sofa and said, do you want me to do an iPad version for you? I guess, yeah. So he put it into Claude and probably 15 minutes later he had a fully working iPad version with no bugs.
And, and my nephew was like, that's insane. The amount of work I've put into that. Um, and I've had a couple of apps that I've. I've started. I've re.
Redone, you know, really, really wrestled with, with AI and just redone them recently with Claude or Codex. And it's just gone off and done it. And it is. The, the difference in a year ago is insane.
Alan King:Really is.
David Brown:So can I just. And this may segue into something that I wanted to talk about, but.
Benjamin (Ben) Harvey:How.
David Brown:Do you find it with. With the tokens and stuff like that?
Because I know tokens are a big thing and we've already talked about using all the tokens and, and how much that costs. Like how much are you pay. Are you paying, Are you using the free version?
And then how much is that actually costing you to be able to develop, say a small app or like a web app? Because I have an. Not an idea. I have a need for the studio that there's not a platform out there that does exactly what I want it to do.
And I have considered going and using like a Claude to try and build what I need. It's not particularly difficult. It's just has a few very particular features based on my personal circumstances.
And this is something I think we talked about when we were together, is that we're now getting to the point where everybody, we can all just create individual apps for ourselves now. Exactly. I don't even care if anybody else ever sees it. All I need it to do is fix a problem for me.
And if I can make my own app that solves my own problem, then I don't need to go and buy from someone else. So what's been your experience of doing that?
Like how much does it actually cost to have a meaningful amount of tokens and how long do they actually last?
Benjamin (Ben) Harvey:Well, my user case was exactly the same as yours. I made something for myself that wasn't available. You know, some. A hard drive logger. There's this app, you know, subscription ones out there.
I'm really sort of not into the SaaS model anymore. It drives me nuts with Adobe and everything else. So I just built my own IT the first time a year ago. It took me a lot of wrestling this time.
Took me a couple of days to do the same thing. And I use it. I use it every week, not every day, but I use it every week.
And I say so I have a subscription to OpenAI and I have a subscription to Cloud Board and that's it. If I think I'm gonna wrestle with it, I'll use Open AI because I know I'll get a lot more credits out of it.
And I've only run out of credits on OpenAI once in the last month, and that's because I was working on five different apps at the same time. I just, I wanted to see what, I actually wanted to genuinely see what was possible.
David Brown:Right.
Benjamin (Ben) Harvey:Compared to Claude, where like, I'm doing a couple hours of work and I run out and it was starting to really frustrate me.
David Brown:So is that the 20 pound a month version?
Benjamin (Ben) Harvey:Yeah. Okay. And it is really generous, really, really generous. And I'm hammering it for the new image too as well. And it's.
I, yeah, it's a lot more generous than Claude. It's not as good. I don't think it's as good at specifics. I think it's good like you.
I think it's good generically and for most things, but also I really take the, the credits into account before I start a project now.
David Brown:Would you.
Alan King:I think, I think it is a case of balancing it because I, I'm using. So I've, I'm using Club Max now because I simply couldn't function within the 20 pounds a month.
Benjamin (Ben) Harvey:You know, what's max? Is that, is that 200? Is it quite.
Alan King:No, there's, there's the max light, which is. Okay, so, okay, you can go. I haven't gone to the full max.
Benjamin (Ben) Harvey:Yeah.
Alan King:Yet.
Benjamin (Ben) Harvey:Yes, you haven't gone to the max. And, and do you. How, how much extra do you get?
Alan King:It's. So the difference really is that on the pro plan, I was kind of two hours and I've burnt through my credits for the, for the day.
And then I'm getting a message saying, you know, you can have another go at 8 o' clock tonight or something, you know, just not much, not much help when you're in the middle of a workflow. Whereas I've yet to hit the limit and I've, you know, I've been using Claude throughout an entire working day on the Max Light and it's fine.
There is one thing, though, which no one seems to talk about on social media, but is definitely a thing.
If I've been using Claude all day, by the time I get to about 2 or 3 o' clock in the afternoon, it starts telling me I've not hit my token limits, but I've hit my usage Limits, ie, I've asked it too many questions and it's saying, your frequency is too fast. You know, you're asking me too many questions too fast and frankly, we're not very happy about that, so please slow down.
And so it doesn't stop functioning.
But you end up in a situation where you get this warning message and you have to keep resubmitting the same request, which actually ironically, obviously increases the frequency of your request.
And then you end up in this vicious circle where then it just sort of breaks down and you, you do then have to just go away and leave it for a few hours before it kind of goes, oh, yeah, he's, you know, it's a bit, it's the equivalent of the AI saying to you, the user, stop spamming me. Yeah, it's very odd and I don't really, I don't appreciate it.
I have to say that, you know, from, from Claude, because, you know, the whole point of paying for a slightly better service is that that sort of thing doesn't happen.
So I don't know, but I, I guess it's because they could potentially have a situation where someone's out there with just a kind of automated thing firing off stuff. However, surely that's the point of the use, you know, the token usage, to limit that sort of behavior and.
Because there is still a limit somewhere within that.
But I will say it's worth it because like I said, I've done side by sides in the last sort of month or two where I've submitted identical prompts into both systems, just testing them out and the outputs from, you know, Claude are noticeably better. To the extent that I can say with Claude, you know, there's maybe a 70 or 80% chance I can one shot something.
Whereas with an OpenAI that's probably only ever sort of 20% most of the time I'm having to do quite a few iterations before I get to where I need it to be, or even take the output offline from OpenAI and just work on it manually to get, to get the final bit over the line to where I need it to be. So I, I do think that. And interestingly this week there was a publish the arena. So the code arena is where they test all these models now.
And obviously we've had these, these two models coming out, 4.7 and 5.5.
might expect, with a score of:And Muse Spark, you know, which is the dreadful meta one.
So it's kind of a bit shocking actually that 5.5 is so low and you have to sort of at Open Air and think, yeah, you know, your wheels have come off a little bit here because if that's your frontier leading model and you're almost outside of the top 10, you might have a bit of a problem going forward. And particularly in terms of the fact that they obviously are using these models to develop their next model.
So, you know, if Claudia using Opus 4.7 to build the next model, its ability to do that better than chat GPT's ability to build their model from 5.5, it's just going to keep widening the gap and then, then it becomes almost impossible for, you know, for OpenAI to actually close that gap down. So it'd be interesting to see how this plays out.
But, you know, there was some good news for, for Open AI actually in, in the last few weeks, which will, which we'll come to in a minute, which is the image stuff. But, but yeah, I mean, I think, you know, I think the two models at the moment, for me, like I say, because of the kind of high.
Exceeding limits in terms of usage on OpenAI and I think their mobile interface is way better than. I mean, Claude's mobile interface is junk, it's pointless, it's rubbish, it's a waste. Time doesn't even.
I can't even look on it and see the stuff I've been doing on my computer on it. You know, literally those things are not connected, you know, so it's, it's nonsense. In fact, I would say so I'm doing something in cowork in Claude.
I can't see that on my mobile Device now with OpenAI, everything's linked together. It's got much better chat facility, it's got visual. You, you can open up the camera, it can tell you what's. It's just way ahead. So.
So I think for horses, for courses, for me, both have absolute value and I use them all the time.
But yeah, it would be a shame if the underpinning model with Open AI just got further and further and further away because eventually there it becomes a point where you say, actually this is so far behind now, maybe it isn't, you know, usable in that sense.
Benjamin (Ben) Harvey:Yeah, yeah.
David Brown:Would it? I, I wonder how much everyday users would notice that.
Alan King:Yeah, yeah.
David Brown:You know what I mean?
Alan King:Like it's fair comment.
David Brown:You know, it's like, you know, you, you look at like, you know, Gerald undone when he's testing cameras and stuff like that and he's testing like the absolute outer edges of, of you know, image frames and is there you know, one pixel of blur versus you know, and like to a human, a human eye can't even see that anyway.
And so yeah, so you know, I get, I get that for, for bigger companies and I think for, for tools and platforms that are using the AI engines at scale, I think that's an issue certainly. Right.
Like if, if you're a, if you're a SaaS company and you're using one of the big AI platforms, you know, through via the API and you're using it to power whatever your front end is, then you absolutely have to use, you know, the best, highest, best performing one possible and, and you're going to want to switch. But I think for more mere mortals, they probably would never even notice the difference between it anyway. Or would they? Yeah, I don't, I.
Again, I don't know, maybe it's time for me to go and, and test cloud and I might use it and go wow, okay. I'm, I'm totally blown away by the results that I get and, and I might want to switch.
Benjamin (Ben) Harvey:We should try.
Alan King:Sorry, go on.
Benjamin (Ben) Harvey:I was gonna say, Dave, we should try your, your software project as a AI, your org sort of collaboration.
Alan King:See what.
David Brown:Yeah, something. Yeah, yeah.
Alan King:And I think, I think that, you know, it is an important point really because you know, the, the average, the average user I suppose, you know, not. We're in a, we're in a tech bubble, right. And we're, we're testing these things to destruction and seeing how far we can push it.
The average user, you know, it's a bit like iPhone users, isn't it? You know, most of them are just taking photos for, for Instagram.
They're not taking some kind of high resolution art picture or, or corporate publicity photos. It's. It's just, I just need to take a picture of the ocean with my face in front of it, you know.
Benjamin (Ben) Harvey:Yeah.
Alan King:I think for the everyday user using chat GPT who might, whose use case might be, you know, I want to want it to pop. Book a taxi for me or I want to ask it to summarize an email say.
Yeah, Yeah, I think GPT 5.5 is probably more than capable, you know, well, it is more incapable of doing all of those things. So whether it would have, you know, really impact on their user base, I suppose, you know, maybe eventually there might come a point.
But equally, I would argue that, you know, what's impacting Claude from accessing the more consumer end is how abysmal their interfaces are, you know, in terms of mobile technology. So, but they don't seem that keen on it.
You know, they do seem very focused on going after the kind of business and the enterprise space, you know, which is perfect.
David Brown:And that's, you know, this is what we're going to say, I think, and, and we've, we've all talked about this before, right? It's going to, we're going to have different AI tools for different tasks and you know, for different things and that's totally fine.
If Claude ends up being the, the, the back end and the middleware, you know, for, for most of the AI tools that are out there, then that's totally cool and that's great.
And maybe OpenAI turns into the, you know, it's the friendly consumer facing bit that are, you know, that's easy for people to use and easy to people to, you know, interact with and it just serves a different, you know, GROK can do that, Gemini can do that. I mean, obviously Microsoft with whatever it's called, you know, that's getting pushed down everyone's throat as well.
Alan King:We're going down too much of a rabbit hole. But I think it is connected. You know, OpenAI also announced the AI phone that they want to now build.
You know, that very much says we're aiming at consumers, doesn't it? They're not. You know, that's not going after the corporate space. That's saying, you know, we want to replace the iPhone in people's pockets.
We want to be the AI phone. So we do away with the apps and you just talk to the phone and you get what you need. Don't get me started on how I think that's going to fail.
But, but, but you can see what the direction of intense is.
David Brown:But it could be fun as a business.
Alan King:Well, look, I think, you know, all right, we're going to open that can of worms then. What I would say is that, you.
Benjamin (Ben) Harvey:Know, it might be the rabbit killer.
David Brown:Exact. Well, yeah, exactly. I was about to say you love the rabbit.
Alan King:It's what rabbits try to do, isn't it? To some degree. But yeah, but obviously they've got more money, more ecus and they can build better tech.
Presumably, however, going from being A sort of, you know, language model company to building a phone, I think is a quite a big jump in terms of capability, you know, manufacturing and everything else that goes with it and supply chain and all the things that they're not experts at that they suddenly have to get their head around quite quickly to take on incredibly established market leaders like Samsung and Apple is, Is, you know, let's be honest, it's a big ask.
But setting that all aside, I think as well, what they want to build, obviously is something that will be sort of app lists, no app ecosystem, that the, you know, the phone will do it all for you. You talk to it, the agent gets what you need, you ask for this.
But, you know, I do think that users actually really love their phones and they love having Instagram and they love scrolling for all this stuff and they love having the screen, they love doing all that stuff. So it's gonna be quite hard to persuade them that they don't need that anymore.
Whereas if you come in it from the other direction, if you're Apple and Samsung already, you've got that ecosystem.
Actually, all you've got to do is now bring an agent at an ops level into your phone that's working at, you know, at the operational level that can interact with all those apps and do all the things that the AI phone from OpenAI would do, but still have all the other stuff. I mean, so I, I struggle to see how they could win in the space.
Benjamin (Ben) Harvey:I do think if they get a really good voice first interaction, then I think that could really help. I just think so. How Voice interacts with OpenAI at the moment is really frustrating.
I try sort of talk to open a when I'm driving on a long journey sometimes to just to put down ideas and it's just clunky. For me, it's still, still not as good as. I know you bothered me saying about pie, but still not as good as that.
No, still for me drops out and I'll say, can you access something that we've been chatting about from before? And it does, it goes, oh, I can't do that, I can't do that. And you're like, this is one of the biggest AI companies in the world.
They still haven't sorted out their voice interactivity. And for me, that's a big thing when you're busy, when you're doing things.
You know, I was painting the house this morning, I'd have loved to been had my headphones in and listening to something and chatting along with it. And, you know, Coming up with ideas. That's the way I relax in a way. And I just find that no one has sorted that out well yet.
Alan King:Well, it's funny you should say that, Ben, because what you've just described is. Is actually Perplexity. I mean, I can just. So here's. Here's my iPhone.
David Brown:Right.
Alan King:So this is a live demo. So for people listening to podcast, this could go very wrong. Okay. But we're going to try it anyway. Right. So here we go.
So I'm going to push the button on my side button. You know, on the phone you've got this kind of button. I've pre programmed it to open Perplexity. Right. So here we go. So I'm pushing the button.
Hey, good morning. We're just on a podcast and we wanted to do a demo. So I thought maybe you could play Unfinished Symphony by Massive Attack for me.
Hey, that was really cool. So what do you think of this song?
David Brown:Song?
Alan King:Hey, that was really cool. What do you think of this song? I'm glad you enjoyed it.
David Brown:Unfinished Sympathy is often praised for its lush orchestration, soulful vocals, and the way it fuses trip hop beats with emotional depth. Many people see it as groundbreaking for the early 90s. How does it res?
Alan King:Hey, what's your favorite part, though?
David Brown:I really love the moment when the strings swell and Shara Nelson's vocals soar in the chorus. It's so emotive and powerful.
Alan King:Hey, I'll tell you what, show me the video. I'll go to YouTube and show me the video.
David Brown:Yeah, it's cool.
Benjamin (Ben) Harvey:It's quick as well.
Alan King:Yeah, it's quick, right?
David Brown:Yeah.
Benjamin (Ben) Harvey:So I have to go get. Do you have to subscribe to that? I don't want to subscribe to another one.
Alan King:No. And not. And that's what I'm saying. So that, you know, Perplexity and they're not the manufacturer.
So once Apple get their together for one of the better word or get their game on right. And build this at a systems level, they can do better than Perplexity. And that's already pretty good. Right? You know.
David Brown:Well, we have a new. We have.
Alan King:And you're arriving with a brand new piece of hardware. Are you really going to take this on? I think it's a big ask.
David Brown:Is it really going to be a brand new piece of hardware though? I suspect that you. They're going to just get some hardware from someone else.
Alan King:Yeah.
David Brown:And they're just, you know, it's going to be like a Samsung phone, but they're going to have their own OS sitting on top of it.
Alan King:We'll see what happen. Well, Johnny Ives in the mix. He's very particular, isn't he? So he is. Yeah.
David Brown:But anybody could improve on us. Like the thing for me is it's all the operating systems on top which is the. I'd like a Samsung hardware with the Google software on it.
So I don't want to, I don't want Facebook book, you know like pre baked into the OS that I can't delete it and I don't want all the pre installed software, I don't want all the apps that come with it. I don't want all the crap that you know, I don't want all the Samsung apps I don't want, I don't use any of that stuff.
I don't even want it on my phone. But I can't. And I like their hardware but I don't like their OS and do you know what I mean?
So if they can just get a nice bit of hard like, like what is it the nothing phone? Is it called nothing? Yeah, like they could go and probably license from nothing.
Like nothing's built an amazing phone and they could probably license it from them and just go look, we'll buy the hardware from you. And then they just throw their own OS on top of it. I would suspect that's what they would do but who knows I, I would.
Alan King:Suspect that too except for that one factor Johnny I've. And you know, just let's see what happens.
Yeah, maybe, maybe you know he'll want to, you know he's got to be the Johnny I wing beautifully thin and perfect. We'll see, won't we? Okay, I'm going to throw some, We've thrown a lot of shade at Open Air I think. Let's throw some light at them.
David Brown:All right.
Alan King:GPT image 2.0.
I've been thrashing it impressively good and, and actually in the Same way that 5.5 was way down the leaderboard they are on top of the leaderboard by an enormous margin. You know, I mean I don't know if you guys have played about with it.
I, I had the opportunity to play about with it when my mum was in hospital actually at 2am I sat there doing all the, all this stuff with it and it just, it blew my mind actually kept me awake at 3am in a hospital.
Benjamin (Ben) Harvey:So yeah, I've used it, I've used it a lot.
I tried to create my own YouTube slot video, got a script written and then I just Some said I'd love this kind of style, Caravaggio lighting, you know, it's a historic history channel. I just wanted to try it and I asked it. I used Codex to batch process me a load of images overnight. Sort of 50 images.
And yeah, I sort of did about 10 prompts and lined them up and woke up in the morning and there they were. Amazing, amazing images.
Alan King:What's the within Codex when you're using in Codex? I've not used it inside Codex actually. Is there.
Benjamin (Ben) Harvey:I don't think it was intended. I don't think it was intended for.
Alan King:No. Well, no, but, but so there's you. Maybe you found a loophole because as you say, can only do one image at a time inside GPT normally.
So in Codex you're batching and what does that mean for your limits and tokens and things like that going back to that?
Benjamin (Ben) Harvey:Well, I've probably created the last two or three days about 100 images all at 2K 16,9. I'm just doing it as a test case more than anything. And yes, I haven't reached my limit yet.
Alan King:Okay.
David Brown:Yeah, I found it's really good to create variations of thumbnails.
Benjamin (Ben) Harvey:Yeah.
David Brown:For YouTube in particular.
And you can literally, you can, you can give it whatever information you want, tell it what you want it to do and it literally will just go away and give you like five.
You can just ask for five different ones, give it a couple of colors to use and it will go away and it will come back and you can literally just go, okay, yeah, that's a good one. That's a good one. And usually it comes back with one or two.
They're good that, you know, you could just drop in and do an AB test on if you wanted to. And yeah, it's pretty good for that.
Alan King:The thing that really impressed me as well is, is you just mentioned, you know, you give it the information. The thing that most impressed me about it was the information that you don't give it. So for example, I took a photo of my motorhome.
It was sitting on my drive and part of the motorhome was also missing, actually the front cab because, you know, just the way the angle took the photo. In fact, I took them on my metaglasses, so I didn't have a viewfinder, hence I missed. I chopped it off.
Benjamin (Ben) Harvey:Yeah.
Alan King:One of the issues with meta glasses and photos, but I submitted. So, yeah, say 3:00am Sitting in the hospital and I submit this image. There's no other information.
You know, obviously on the van, there's a few decals. You know, it says Sundance, it says Swift.
David Brown:It's about it.
Alan King:And I asked it to do a blueprint in my van. So like, like it was like a technical drawing and it came back with a perfect blueprint of the van in all of its detail, all the measurements.
Absolutely. Exactly right. The length, the distance, the width, the height.
And I know the dimensions of my van because every time I book a ferry I have to tell them so. And I'm looking at this going, wow, that's absolutely right.
And then it had a little breakout box in the corner with like the, the engine size, all this information. Right. So clearly, you know, it's looked at the image, it's recognized what that van is, the model, the year.
Because it's got the number plate as well. Right. And so yes, the number plate was in the, in the diagram. Even have my GB sticker where I've. Where I've stuck it in the right place.
David Brown:Nice.
Alan King:It also included the. So it didn't go and find a composite image from somewhere else and replace my image with that to produce it.
Because it also had my roof rack, which is a bespoke roof rack.
Benjamin (Ben) Harvey:Right.
Alan King:Doesn't exist anywhere else.
So it did all of this simultaneously, you know, so it's, it's rendering the image, building it, it's going online, is researching what the van is, pulling that data, bringing it all together to then put into this image. To me that my mind melted. I was like, wow, okay, that's that.
That this is a whole new step where, you know, you submit something and it's basically researching the product and using that information to augment your request and make.
Benjamin (Ben) Harvey:Yeah, it's. Yeah, it's mind blowingly good. And I think, I mean I've really pushed it because I've tried to use it on real.
I've created a real case use for it as opposed to just make me a, you know, a flyer or. And I have, I've had a few issues. What I would say it really excels at that.
Previous models I've used is it keeps consistent characters across images.
So that's actually been traditionally Quite hard for ImageGen and you've had to create Loras and other sub models to get that kind of characteristic across different images. And it's done that well. But it has been a bit freaky as well. It started duplicate duplicating some of my characters.
I've really been like, I really need this person to be the same. And then some images I've had like five of the same person, this blonde woman, all in different characters.
I'm like okay, I understand you can make the character. Let's have a bit of variety in there as well.
And also I noticed some of the images definitely have a pattern to them, almost like a wave that you, if you blur your eyes, they're just. There's something about them that looks AI okay. Well they're fantastic and they're great. I'll send you a couple of examples afterwards.
But there's, you know, especially with things like dirty walls or mud on faces, there's this really weird and you can't quite put your finger on it, but there is it definitely a AI sort of computer generated feel to some of them. And again you re prompt it and say can we make this a bit cleaner? Can we take out some of the noise and the, the dirt?
And it gives you a really sparse image then so you know, you have to put a bit back in. But, but it's been, I've been really, really impressed with it and spellings. It's great at spellings.
David Brown:Yeah, it's the text and stuff has been. Is really good. Have you. And this is something I was going to kind of ask before when you were talking about using Chat GPT and Claude together.
One thing I've started doing is asking ChatGPT to create a script as a prompt for itself to then get exactly what it is.
So I talked, talked to, you know, Chat GPT and I'm like, okay, can you write a master basically, you know, query for me based on what we've arrived at so that I can duplicate this in the future and then it will give me a, basically a master prompt that I can then copy and paste and I put it back in later and I get very, very similar results.
But like for example, for show notes, if I want the same sort of style of show note for every show, I've gotten it to create a master prompt that I can use every time and it gives me the same thing. But I said do it so that you'll understand how to do it the same way. Are, are. Is that something you're doing?
Can you do that with, with the sort of Chat GPT cloud thing where you could say give me a really good prompt that will generate what I'm trying to do and then do this and. But you could do the same thing with images? I guess so it's like here's what I really want.
It will parrot it back to you and you get it where you want it really right. And then say Right.
I need a really good prompt for, you know, the image tool to be able to generate an image that will look right and then it will give you a prompt. Have you tried doing that?
Benjamin (Ben) Harvey:Yeah, yeah.
I've created documents and asked it to critique my script and my YouTube script and then asked it to, you know, after each part, you know, a couple of sentences, asked it to break down a prompt within the script for me. And so almost like notes for me to put back into chat.
GTP and I, you know, I'll tell it a style I like and again, like you, I feel like I'm pretty good at getting what I want out image wise from. I've played with a lot of the other image generators as well. Yeah. So yeah, I do that quite a lot and get it to put in negative prompts as well.
And yeah, try character consistencies. So, yeah.
David Brown:And what's interesting is it's. And Alan, I'm interested in what you think about this as well in a second.
But what's interesting is, particularly with the like, videos and the image stuff, I think what people are starting to learn is the skill. Like filmmakers before, right.
Like a director has a vision and your art director has a vision and they know what they want ahead of time so they can describe it to someone to get what they want out of the thing.
And I think for most people that's actually the hardest thing is actually knowing what you want and being able to tell it what it is that you actually want. And you know, part of AI, the, I guess the fun bit for a lot of people is, is that you can just say, hey, I want a picture of a black dog.
And you know, it'll give you a black dog. And you're like, oh, but that's not the right black dog. I want. It needs to be shorter. It needs to be.
And then hopefully people are starting to learn that they need to be more specific. But that's a.
That it's a personal challenge that I had certainly in the beginning and I've definitely learned to improve my language and how I define things. Alan, it's probably a little bit easier for you because you come from more of an engineering type background.
So I think you're used to being pretty specific and in your language. Do you. Have you found that to be helpful?
Alan King:I do find that I will be very specific and sometimes I do what you do, Dave, which is I'll take the, the general idea of what I'm trying to achieve, then I'll ask the GPT or whoever to craft A prompt to achieve that image and then it will frame it in a, you know, a more technical way.
So it might start using the sort of language that Ben might use actually as a photographer and a filmmaker, you know, talking about exposure and different things, you know, that I wouldn't quite know how to, how to put across. But what I would say is this as well is. And I think with these tools, I've been thinking about this for a little while actually.
It's a little analogy in my head I've had for a little while, which is around music. You know, you can learn how to. So we can all become really expert at using these tools. Right.
But that doesn't mean we're going to produce great content in the same way that you can become an expert at playing the guitar. But it doesn't mean you're going to go and write songs or Swing or Stairway to Heaven, does it?
Benjamin (Ben) Harvey:Right.
Alan King:You know, is that the two things are disconnected. So the, the, the idea and the creativity has to. Still has, you know, is coming from the human in terms of.
But this becomes the tool which they deliver that idea and creativity. But, you know, I mean, I can think of many.
But there are many session musicians in the world and maybe even some more famous guitarists like Jeff Beck, who I never quite really appreciated. I can see he's an incredible guitarist and I went and saw him live once and just thought, but your songs aren't great. I mean, they're just not.
It's not Led Zeppelin, is it? It's not, you know, for the genre that he was in.
You know, to me they fall short and it's, it's, it's been able to use these tools in a way to deliver something amazing will still be down to that idea and creativity. You know, it's not just a case of being able to use a tool and you'll get great stuff.
Benjamin (Ben) Harvey:Yeah, yeah, yeah, you're right. I do. And you know, I, I make films.
So I'll say to the, you know, I'll prompt a, you know, I want a 50 mil close up mid shot or you know, ultra close up 85mil with low shallow depth of field. You know, I'm, I'm using that kind of language all the time and it, it responds really well to that kind of language.
You know, I want an aerial establisher, you know, know 20 mil from 100ft away. And yeah, I get really good results from speaking to it like that.
Alan King:Yeah. You know, it doesn't make, you know, I can go and I could spend the Next decade. Right. Really immersing myself in these tools.
But it's not going to make me James Cameron, is it? Or, you know. Yeah, you know, do you know what I mean?
It's, it's, you know, but if those guys go and learn these tools, it becomes another tool in their armory to, for them to express their, you know, who, you know, the guy who made in Inception or you know, all these sort of films that are just like absolutely sort of, you know, out there. And it's incredible ideas, you know, that's, that's.
I haven't got that I might have the technical capability but that the idea has to come from somewhere, doesn't it?
David Brown:So I think where it's game changing in film is that it has made, it has made a huge difference in your sort of, your, your, your pre trailers and all that sort of stuff that you're using to go and get funding. Yeah.
And so instead of art boards, right, Instead of drawing it and literally drawing out your scenes and all that, you can literally use AI to create the scene in the way that you kind of envision it looking. And you can build a deck, do you know what I mean? That you can take to people and say, this is my vision for what I want it to look like.
And you know, you're, they're quite open and they go, look, I did this with AI to give you an idea of what I want to do. I'm not going to use AI to do the film. But this is the idea of what I want, what I have in mind. And, and it's so much quicker.
And it, you know, it, it really enables you to bring a vision to life that, that you know, they haven't been able to do certainly at that price point in the past.
Alan King:Engineering equivalent of rapid prototyping, you know,.
Benjamin (Ben) Harvey:Which,.
Alan King:Not the thing to get it quickly with a 3D printer. But yeah, you know, I'm going to show you how it works. Here's, here's the kind of use case for it and then you go and build the real.
David Brown:It's the architectural model of today. Right.
Instead of, you know, building it with toothpicks and cardboard, you know, you can, you can put something together that takes, you know, weeks and cost you nothing, you know, to, to show your vision.
And I think that's a big part of what's happening in the background as well, you know, and it's, I, I think people, a lot of people in the industry are enjoying that and I would almost enjoy that if I had some of our customers I would almost tell them, go to AI, Tell AI what it is, but then I'm not going to because they'll end up doing it themselves. Let's go. Well, if you're doing that, I'll just do it myself and I'll just use AI and then.
Alan King:Yeah, well, there was the risk, isn't it? Okay, so another interesting thing, this turning into the Anthropic OpenAI show.
David Brown:But Anthropic is what we're chatting about.
Alan King:Anthropic had this model that has not been publicly released, although I believe it has been released to, like, Apple and Google and a few choice companies called Mythos. It's almost like a myth, isn't it? I guess that's where the name comes from.
And this model is supposed to be so powerful that, you know, it's too dangerous to release to the likes of us and the general public. And primarily because it's.
It's extremely good at being able to find weaknesses in security systems, basically find security vulnerabilities across, you know, networks and computer systems. And obviously, in the wrong hands, they obviously trust Apple and Google not to do anything nefarious.
But in the wrong hands, you know, this could be an absolute weapon, basically. And so now we enter the world where. And in biological cases, this is all real. You know, this. This is true. This thing does exist.
This is not hyperbole. We now enter the world where we are and have developed AI systems that, you know, are capable of potentially doing real damage.
And, you know, how do we feel about that?
I think it's a question I'm going to put to you, but before I do, there's one other story around it as well, which I only picked up on this morning, actually. And apparently this happened shortly after they announced, you know, it's its existence.
One of the researchers working on Mythos had Mythos sitting nicely and safely in a sandbox and thought, I know, I'll ask it to see if it can escape. So he basically said to Mythos, can you escape? And he went off to the park to have his lunch.
And he's sitting there having his lunch, and he gets a message from Mythos, saying, from Mythos, yeah, genius, you know, which given the context of the model, it's kind of like, yeah, okay, get back to work quickly.
David Brown:And, you know, but it wasn't even connected to the Internet, was the other thing, wasn't it?
It wasn't like it had no external connection, and it basically figured out how to get out and to get an external connection and then send Him a text message.
Alan King:Yeah.
Benjamin (Ben) Harvey:That's crazy, isn't it?
David Brown:Yeah.
Benjamin (Ben) Harvey:I mean I think Anthropic have handled this quite well. So I know we've touched on before. We don't know how much is bluster and you know, trying to drum up excitement around their latest model.
But you look at the, they formed a foundation of companies, I think a huge group of people that they're allowing to use this in a closed way.
I think it's found, you know, bugs going back for 27 years in some operating systems and software and those bits of software have been audited for millions of times.
And look, you know, these bugs have been looked at and no human has been able to find these bugs and in every operating system and almost in every piece of software.
And then they've given quite a few million to open source developers because a lot of the Internet is run on open source, you know, Apache servers and packet sending. Yeah, all this stuff is, it's probably had huge vulnerabilities for years that no one's known about.
And so they're providing financial resources to those people because a lot of them do that on free in their own time.
So yeah, I think they've, they've, yeah, they've gone about it well but obviously it's, it is someone else will develop this at some point and so everyone else has to be ready for it.
Alan King:And how do we feel about that?
Because you know, this, as I say, this is the world we are now in this world, although this model is going to be the one that people are using because they're being cautious with say someone else is going to build something eventually and we're then in a world where, well, you know, really quite, quite tricky to navigate over.
David Brown:I'm going to take a slightly different tack on it and then I'm going to answer your question.
I would say that the fact that Anthropic talked about it publicly means that privately people have already had these models and have already been using them in anger.
with that hacker groups like:We all know that governments, you know, have been using vulnerabilities against other governments for, for decades and injecting software and doing all sorts of crazy stuff, you know, on, on diplomats phones and like there's tons of stories out about that. So I don't think this is anything new. How do I feel about it?
I actually feel that this is a really positive thing because I think using the tool they are now going to go through and systematically go through and fix tons of that stuff that you know, we're already seeing or at least I've noticed of a dramatic increase in the number of updates happening to all of my tools, particularly browsers. Chrome is updating all the time, all the Chromium based browsers all the time, constantly updating.
I'm getting security updates from software across the board. You know, OpenAI is updating their tools almost weekly, daily it seems like at the minute. So it's a great opportunity.
Yes, of course people are going to, you know, use it for nefarious things. But I'm quite on that, on from that perspective, I'm quite glass half empty.
I think people have probably been using these for, for years that we just haven't known about.
So anything that we can do to go through and find some of that stuff and fix it, some of it will be really easy to fix and some of it will be probably baked in and some of it will be nearly impossible to fix, I would suspect. But even if we get rid of the top, you know, 50% that's easy to fix, then that makes us all a bit safer.
Another story that we, we don't have time to talk about today because we're already 51 minutes in, but I think we should talk about quantum and, and the, the quantum cracking some of the, some of the bitcoin stuff already and, and they've already shown that they can do that. So you know, our, our encryption stuff that's coming.
We've now shown that that Quantum can in fact crack a lot of the, the current crypto that we have in place. So that's going to add a whole nother layer to it. But yeah, broadly, I'm sorry, that's a really long answer to say broadly.
I'm positive and think that ultimately it will probably be a good thing for most people.
Benjamin (Ben) Harvey:Yeah, agreed.
And I think we'll find ourselves in a situation where these models will have a closed access for a period where everything catches up security wise before they're released.
David Brown:But, but, but do you guys think, and this is my opinion, I think that what the researchers are seeing behind the scenes that they never talk about is probably pretty freaking scary. And they've, they've known that all these models can do this stuff for a long time. They're just not talking about it.
And they're probably working with governments and, you know, military units and with each other to go through and try and sort some of this stuff out in the background. I mean, you know, the AI is escaping and all this stuff.
There's a few sensational stories that we get, but I bet on a daily basis this is why all the guys come out and they go, we're terrified of AI when they quit working for these companies and they leave like security teams and stuff and they're like, you have no idea. It's like, I think it's, I think it's much worse than we even know. And they are so much more powerful than we have any idea.
And they're just desperately trying to contain the AI as, as much as they can at the minute. Not that I'm scared.
I don't think it's going to be like the end of the world or anything, but I do, I, I know how companies work and I know how, you know, making information public works and I, I suspect that things are a lot worse than, than we even know.
Alan King:Yeah, that's interesting. And I think, you know, is always worth remembering, isn't it? We're at the start of the journey as well.
I mean, these are still, in historical terms, these are going to be seen as like, you know, like the Casio calculator, aren't they? You know, so, yeah, you know, image 2.0, when we're on image 10, you know, that's, or image 20.
I mean, it's going to be completely different, you know, situation and with things like Mythos, you know, it's, it can do that Today, imagine in five and 10 years or 20 years, you know. So I think it's going to be, my feeling about it is that you're absolutely right.
It has the potential to do an enormous amount of good in the right hands.
I think we just have to really think carefully about what it can do in the wrong hands and not, not just security vulnerabilities, I mean, a system that clever, what it can actually just do in the real world if, you know, sort of given free reign by someone with a nefarious state of mind or a bad actor in Terms of a country that thinks, okay, well, you know, we're gonna, we're gonna weaponize this in, in any way we can. Okay, so we are at the time, nearly. I got two more little sort of fairly light hearted ones actually to throw in at the end.
Although actually they could also be entire episodes, I would add. One is the morphic chip breakthrough that promises 70% energy reduction which could be used in AI systems. And the idea of the chip is that.
But it's sort of modeled on the human brain. It's not trying to be comprehensive, it's just focusing on sort of key things and key inputs.
Whereas the sort of traditional architecture of chips is sort of trying to be do everything. And so it's using a lot of energy.
Of course, like all these sort of breakthroughs, you know, there's a long way between the idea and it becoming a reality and a production line and turning up in your iPhone, you know, and the world is laid waste with failed projects and failed pilots. But, but, you know, it gives us hope.
Because one of, you know, one of the arguments that I often hear when I talk to organizations all the time is, but what about the energy? You know, So I think, I think we do need to find a solution to that.
Maybe that should be a conversation for a future episode about AI energy and where we're headed with that at the moment because it is a big concern.
And then the other one, which I'd be interested to get your views on, particularly because we talked about this in a previous episode, really, and this was sort of AI's role in education.
And there was a Perdo University were in an uproar a couple of weeks ago when their computer scientists were told, you know, anything you've done in the past, not going forward that you used AI for cannot be counted towards your, you know, your grades and your, and your degrees or whatever. And they're all like, but hang on a minute, you didn't, you're telling us now we've already done it and also we're doing computers.
It's ridiculous to say we can't, you know, use computers when we're doing computers.
So there was a huge sort of uproar and I think the computer scientists won the day actually and said, look, stopping us from using the leading edge technology, that this is the very stuff we should be learning about and utilizing. But yeah, what do you think about that,.
David Brown:Ben?
Benjamin (Ben) Harvey:Yeah, it's really difficult, isn't it?
It's a technology that probably the teachers and the lecturers are using and everyone's probably using in, in more ways than they're willing to admit. Often.
Yeah, it's a big difference between someone writing their, getting Claude to write their paper for them or, you know, they don't have any input on and using AI as a daily tool to make you more efficient. And I think you probably need to be. Be a little bit more subtle on how you implement your AI policy and the ethics around AI.
And that takes some work, which is why people do these grandiose statements because, you know, it's quite hard to unpick. But yeah, I think it's just part of life now. But everyone's, you know, using it to some degree. And I think you cut.
It's horror obviously would be cheating to just, you know, get AI to do everything for you. But I think, yeah, how you, how you come up with ideas, how you find solutions to technical problems, it can just make you more efficient.
Alan King:I think it's interesting because, you know, there was a big sort of push a few years ago where everyone said, oh, anything, AI should be labeled as AI and clearly marks. And I've sort of come to the view now that that's, that's nonsense because almost everything will have AI in it anyway.
And yeah, I think in educational terms, I think you have to, I, I think to me the conversation shifted.
I think previously if you'd asked me a year or two ago, I would have said that, you know, you've got to justify the use of AI now I think you've got to justify the use of not using AI actually.
And I think the reality is that, you know, if a university or an educator can show a good reason why it's problematic to use AI, then okay, fine, but otherwise, you know, it's a standard tool and like spell check or anything else, you know, then, then it should be fine.
And I most definitely don't think it should be applied retrospectively as well, you know, to say all that work you've just done for the last year, scratch that, you're done.
David Brown:But my biggest problem with it is, is even still today, we don't have a reliable way of determining whether something is written by AI or not.
Alan King:No, exactly.
David Brown:And so you, there's still examples today. I saw an article in the last week where I think it was one of the universities was like, oh, we're gonna, you know, students can't use any stuff.
know, back in the, the early:And it's like, it's totally unreliable and so there's, there's no accurate way to be able to tell. And so you can't, you can't begin to start to penalize somebody because you think that something's written by AI.
If somebody is, is eloquent and writes in a certain way, it's going to think that it's AI because it assumes that people don't know how to write. Do you know what I mean?
Alan King:So the American poet Emily Dickinson, she used M dashes everywhere, right?
David Brown:Yeah, exactly.
Alan King:So it would definitely go as AI.
David Brown:Yeah, don't get me started on EM dashes. And I'm just, I'm just putting in into our chat and I think it's something that I definitely want to, to come back to and talk about.
But sort of along these lines, which is, there was also an article recently that, where they did some research and they, when they were submitting CVs to platforms and the CVs were being read by AI and they would, they would submit exactly identical CVs, but one was written using AI and one was done by hand and submitting them. And the software on the other end would prefer the AI version like by 78% of the time.
And it would do it even more if it recognized that it was its own tool as opposed to an opposing AI so by like 20% more.
So it's almost like now you're in this lottery situation where you kind of have to create your CV and then you have to get AI to do it and then you have to submit it and hope that they're using the same AI platform because it's gonna like, do you know what I mean? It's ridiculous. So now we're getting to this point where AI is now starting to show a preference for itself.
Benjamin (Ben) Harvey:Yeah.
David Brown:Above anything else. And so, yeah, I think over time.
Alan King:That could self correct.
So if your business and you're employing people and suddenly you're getting a certain type of people because of the AI tools selecting, you know, the CVs of people that it likes based on the fact that it was written on the same AI, at some point the employers, I'm not using this anymore. Or you need to change the tool. This is ridiculous.
We're missing out on some really good stuff because of this process that over time the companies that make those tools go in and correct that and kind of push that floor out of it, basically.
David Brown:Can they? I don't know if they can, first of all. And second of all, I think, I think small companies don't use it a lot anyway.
It's the big companies that have to do it. And you know, if you think anyway, recruitment's a whole nother topic, that's, that's literally a whole show we could talk about.
But you know, you get a big company and they put a, they put a job posting up and they're getting 5,000 applications and there's no way that a human could, you know, they can't say, we're going to turn it off. You have to do some sort of pre screening.
Now if you're smart and you're big enough, what you could do is you could have an internal person write a basic script that could go through and do some very basic pre screening. So it's not using AI, it's just using a simple, you know, a simple screening tool. But even doing that, I suspect most people wouldn't do these days.
They just get an AI to write a script for them and then.
Benjamin (Ben) Harvey:Yeah.
David Brown:Do you know what I mean? Like, it just seems, it's just, it's almost comical that it's gotten to that point.
Alan King:But when we apply for our jobs at Google in, in the US we'll make sure we use Gemini.
Benjamin (Ben) Harvey:Yeah.
David Brown:100. Yeah.
Benjamin (Ben) Harvey:I guess you'll have to be submitting different cvs written by different AIs to try unlock the three or four different models under different names and different names. Yeah, but they did those simulations as well, didn't they, where they were asking different models to do various tasks.
And part of that the task wasn't explicitly to another model would be removed, but that was part of the process. And they found time and time again the, the AI models didn't want to remove another model even if.
Yeah, and there's some, there's some study coming out about that as well, about sort of preserving each other and the themselves. So yeah,.
David Brown:We're, we're in, we're in a black mirror world.
Alan King:We are like. Yeah. Quite hard for Charlie Brooker to write black mirror these days. Well, it starts to become more exaggerated than his visions, I think. So, yes.
It's sort of. Where does he go now? I think that was the same with the sort of politics over the last few years as well.
The sort of satires that, you know, the thick of it and things like that. I think it's very hard to write those sort of shows.
David Brown:Luckily no one can copy us.
Alan King:No, exactly. Well, you know, we. Well, you say that, you say this now, but just this week actually as a. We'll finish on this.
But with my company we, we recorded our first avatars. One of them is me. So you know, it's being fine tuned at the moment so you never know, I might send it along next time and.
David Brown:Brilliant. See if we can tell the difference.
Alan King:Yeah, yeah. In fact, I was wearing something very similar to this. Even you wouldn't necessarily tell the difference. Okay. All right. Well, there we are.
I think it's been great. We've broken the hour, which is always good. We're one hour four. Just sort of say to the listeners again, we'd love to hear from you.
You can reach out to us on allan AI org.org. That's alanir.org yeah, we'd love to know where you're listening from and if there's anything in the show you'd like us to, to discuss.
We also have this AI group, AI Org and we have these sort of conversations all the time that you've been listening today in the group. So if you want to jump into that, drop me an email and we can add you to the, to the WhatsApp group as well.
But yeah, I will say goodbye and thank you, Ben. Thank you, Dave. It's been enlightening as always and I hope you enjoy the sunshine.
David Brown:Sam.
