Prompt Scale

>> Anouk: Welcome to fusion talk with anouk and steve.

>> Steve: Do you want to be able to look at the future?

>> Anouk: The future?

>> Steve: Do you want to be looking at the future, knowing what's coming to be prepared for?

>> Anouk: Show me the palm of your hand, cross it with silver, and I will tell you your future. Actually, funnily enough, um, I remember a few months ago or weeks ago, I can't remember how we did it, but I took a photograph of my hand, all right. And asked AI to read the poem. Was it. Did we do it together or was it with somebody else? I can't remember, but that was quite funny. So she was able to sort of. I, uh, didn't need to cross anybody's palm with silver, but she told me my future. So, yes, AI is the future.

>> Steve: AI is the future.

>> Anouk: Maybe. Because I also read a report this week that says that the AI bubble will burst big time soon and a number of companies are going to go bust. But the one that's really going to go bust is the one that's supporting the other six big guys, which is Nvidia.

>> Steve: Yeah.

>> Anouk: And. But what they were saying was, um, basically that it's costing about 250 billion a year to actually generate the AI data centres. But they've only got 50 billion a year coming in and you can't sustain that model.

>> Steve: Yeah, true.

>> Anouk: And so that was a finance guy. I mean, yeah. You know, people do this, sit there and if they've got the money, they'll keep it going anyway.

>> Steve: But it's something a lot of people already telling you, that the AI bubble will burst.

>> Anouk: I'm saying that about the dot com for years, so that's possible. The other thing is that strike me is, um, could you go back? Let's assume it bursts. What actually does that mean? I mean, we have, in less than, uh, 18 months, created a whole brand new way of working.

>> Steve: Yeah.

>> Anouk: Already in the future, in 18 months, we were just looking at, uh, AI Azure.

>> Steve: Yeah.

>> Anouk: I mean, and you were blown away because I know we've not a chance to play around and look at it. When you start digging in there, I mean, and you think how long it

>> Steve: took power, I will go nuts.

>> Anouk: I know you will, you will. You know how long it took to get workflows and stuff working and stuff. So this is not only that, but

>> Steve: if I think, um, I was five, coding with cloth for a power app.

>> Anouk: Yep.

>> Steve: And in one day I created a full power app that normally took me about. I think if I should build it like that, a week to build it.

>> Anouk: So Is your job dead? So is the future with Noah Nuck?

>> Steve: No, you still need to have the ability to have that human there to make changes and to do things really.

>> Anouk: So at the moment I can put a prompt in that says, hey, create me a report or a PowerPoint presence or an image on X and Y Z. And if you think about the images that were created three years ago and the images that are done now is huge, surely I can just say, hey, and create me an application that does this and this, please, and then ignore it. And it understands what I need. And though, uh, I don't need you anymore.

>> Steve: Yes, you probably are able to, but what if something is not reacting like it should be and AI is not helping the future?

>> Anouk: AI, we already know all those things that you changed and created before and on you and people like you. Not that there's anything dangerously bad about you, but other developers that are updating and working in theory, AI is learning.

>> Steve: Yeah, true. But to answer that question, can we go back to the previous way of working without AI? That's going to be very difficult.

>> Anouk: Mhm. I think a lot of people have been able to use computers in a different kind of way where they've never been able to do that before.

>> Steve: Yeah. I also think, um, the AI bubble will maybe burst, but what we have in this moment in time will not disappear anymore. It's so used to a lot of people in their daily jobs, in their daily lives, that people don't want to give up on it.

>> Anouk: Yeah, yeah. Ah, you're not wrong. You're not wrong. Well, we're trying to look at really the future of organisational AI, so within the business, within the company. Um, and kind of, I like the idea of the theme that when we put this together, what happens when AI remembers everything, so it remembers the errors that you made on those apps and

>> Steve: it will learn from it. It will not do the same thing, but it can get maybe some dangerous stuff as well.

>> Anouk: Um, okay, so we're saying the future is not positive.

>> Steve: No, we are not saying that. We are.

>> Anouk: Well, you seem to be, you seem to be saying, yeah, but, yeah, but, yeah, but we don't know. Um, that's right. You know why?

>> Steve: It's the future.

>> Anouk: The future. Yeah, yeah.

>> Steve: It's a little bit like the article that was being there recently about that GPT or One of the GPTs from ChatGPT did the hacking, uh, of the other GPT.

>> Anouk: So they, yeah, so what actually happened was they wanted to test to see how smart it could be. So they locked it into a sandbox.

>> Steve: Yeah.

>> Anouk: And then, um, said basically escape. So it had no access to Internet, it had no access to anything. And the first thing it managed to do was find a, um, uh, end day error.

>> Steve: Yeah. Is it end day n day error?

>> Anouk: Yeah, end day error. That gave it access to the Internet. Then it managed to connect back into chat GPT by breaking through their firewall to get the information it needed to be able to get itself to escape.

>> Steve: Yes.

>> Anouk: And they all went, oh, I'm sure.

>> Steve: Yeah, of course. And that's. Those things will happen in the future and are happening at this moment.

>> Anouk: They're not just going to make us

>> Steve: nervous, they will make us nervous. But I do think it's a very good learning school for people to learn from it. Yeah.

>> Anouk: But it moves so damn quickly.

>> Steve: It does.

>> Anouk: I like the other storey about where they created like a, uh, an environment and they put all the different AIs in there. So they put in the big ones, chat copilot, uh, Grok and everything else into different villages in this world to see what the villagers would do over a period of time. Um, copilot was the most productive. So it got things done. Chachi PT was all just bit like flower power. Hey, man, this is really cool. You know, what if we build some flowers? Anything else? Grok killed each other, though. They just argued and murdered each other. And there was something else on the set. Fire to all the houses. That's right. So. But uh, it also reminded me of that Waymo storey. We've mentioned this before in San Francisco where, you know, where they release these waymos and then they all started going at night to the same place or the same two or three places and grouping together like as if they wanted to do it for company.

>> Steve: Yes.

>> Anouk: Um, and you know, that was. And they had to effectively put a command in a. A guide rail that says guys at night time choose a completely random location and then go there rather than letting them decide themselves.

>> Steve: And I think that's exactly what we need to think of those guardrails for AI.

>> Anouk: But it's a little bit like putting a bunch of kids together in school. You know, you put. You say here's playground and away you go. They will do what kids do. They will group together. They will. They will almost certainly start fighting. The boys will go and pull the girls pigtails. Um, you know, some people will push people around, some people won't. And then you have to put rules in for the playground to say, hey guys, play well together. Don't do this, don't do that. Other than that, you can do what you need to do?

>> Steve: Yes, I think so. That's what we will need in this.

>> Anouk: So the future will not be about telling it what to do, it'll be about telling it what not to do?

>> Steve: I think so, yes.

>> Anouk: What you can't do.

>> Steve: I think so.

>> Anouk: And, uh, it kind of makes some sense.

>> Steve: I think it's more logical to tell something or tell AI what not to do than what he can do or what it can do, because we already know it can do so much. It can do coding, it can write documents, it can read, uh, the numbers faster.

>> Anouk: Yeah, but these are all tasks that's got. That's not. That's not an organisational AI, it's just tasks. So, I mean, we went to a conference on Thursday, so we did live tomorrow on Thursday. And the guy, uh, a report in the US or from somebody, had said that 18, on average, of everybody's jobs can be done by AI. Yeah. And that's because those tasks can be done. That means that there's 68%. No. Um, 72% of work it can't be done. 72, 80, 90. 82% can't do maths today anyway. But there's 82% that can't be by AI. That needs humans in there.

>> Steve: Yes.

>> Anouk: So if we give the organisation everything and we tell it what not to do, then we also need to give IT structure and, you know, like humans need. So my job is that I stamp the boxes when they come in and make sure they get delivered to the right place. All right. So if an AI is going to do that job, that's all it can do.

>> Steve: Yes.

>> Anouk: I was looking at AI as a service today with somebody. Uh, we're looking at putting some stuff together and we was talking about agents. And the agent needs to have guide rails in there so that you don't use the agent for the wrong thing. Because they're going to learn. Yeah. So if you suddenly start talking, uh, to an HR agent and about, hey, how can I update the antivirus on my laptop? They're more than capable of doing that, but you don't want them to learn that. So you need to be able to sit there and go, hey, you know, you got two things here. You either don't answer the question or, or you go over to the IT agent and get an answer and then point that person to the answer. And so you have to set them up that way. Just the same way as if I'm putting a team together to manage a network I need somebody that will do architecture, somebody that will do circuits and lines and somebody that do configuration.

>> Steve: Yes.

>> Anouk: Then potentially we have an organisational AI

>> Steve: that is helping your organisation to grow and do the things that need to be done for it. But is it going to replace humans, then?

>> Anouk: No, I don't think so, no. But. But if. If I've got. If 18% of it can be replaced by AI, you need to make sure that it's only doing the IT functions, so still need to be organised and a human will have to set that up and control that and manage it and make sure that it gets maintained

>> Steve: and supported also, because most of the time the human already know what the job is and what needs to be done and have better control of it. But will that disappear at one point in time when AI is growing?

>> Anouk: Well, that's the point about this, isn't it? I mean, we've got a few discussion items that we had on our list, like AI generated prompts, reusable thinking. Um, you know, are you going to allow it to. Well, it will actually send a prompt to another agent to say, give me that answer.

>> Steve: It does, but, um. My experience with AI generated prompt is.

>> Anouk: No, it doesn't matter. This. What you're about to say is completely irrelevant because we're talking about the future of. So you're doing it again. You're saying it won't happen. Because my experience is it needs to do this or it doesn't do it well enough in the future. It'll learn and it will do it well enough. Now tell me what you're going to say, or was I right?

>> Steve: No, you were not right.

>> Anouk: Damn, damn, damn, damn. I'm sorry, go on.

>> Steve: My experience with AI generated prompt is that the way you get that prompt is also something AI needs to learn from itself. But you do a lot of.

>> Anouk: Go on.

>> Steve: You do a lot of prompting, you have a great result, and then you ask AI what was a better prompt to start off with and how did I get Foster my results? Then you let AI generate the prompt and you start working with it.

>> Anouk: So why would it not learn? I've done this now for somebody else, I'll just use the same prompt. Because it gave them the right result.

>> Steve: Because in your organisation, it only gets the information from where you have access to.

>> Anouk: So again, in the future we will have fixed that. This is the future, we're saying. So are you saying that the future will never have AI generated prompts, just spuriously doing them themselves?

>> Steve: I'm not sure about it. When you ask Something to an agent. You ask your HR agent for updating the antivir virus of your computer, then you asked it to do something, then the HR agent will tell that the IT agent to do. But it's not a new generated prompt them.

>> Anouk: No, but the AI agent will have to write the prompt. The HR agent will have to write the prompt to send it across to it. Two things need to happen. First of all, the HR agent needs to understand this prompt is not for me. Um, and uh, secondly, it then needs to decide how it wants to pass that on and it will do. So for example, um, if you're going to do agents properly, will every agent generate images? The answer is no. All right, so if you want to have corporate branded content, you have to have a specific agent to do that. All right, so then, and if you're in HR and you say I need a logo for a presentation, it says oh, I don't do that, but I know that something else does. So I will ask that person to create a logo. Uh, and then the guide rails and the guidelines and the skills in that agent creates them based upon the corporate branding. And then, and then it needs to do something with it. So it needs to go and place it then in a uh, shared location for them or something.

>> Steve: Yes.

>> Anouk: Otherwise you end up with too many skills in one and then they, you end up getting them. They're too creative.

>> Steve: Yes.

>> Anouk: So you have to structure this.

>> Steve: But then again if you think about that one agent sending it to another one, is it going to create a fully new prompt or just saying, look, I received this prompt, fix this for me.

>> Anouk: Yes, it will fix it. You tell it to. If you have to hand off to a different agent, make sure it's got all the information. So it might ask me a question. Yeah, in the future it will ask me a question. Hey, I'm going to go and arrange this for you. And you want it to go into a uh, PowerPoint whereabouts on the PowerPoint do you want it? Do you want it on the whole page or just in the top left hand corner and then it will create the prompt and send it. Yes, absolutely. It's the future. You can do anything.

>> Steve: Everything can happen in the future.

>> Anouk: See how problem as technologists is we need to understand how it's going to do it. And that's restrictive. That's why you keep saying my problem with uh, or my issue with. Because our experience has been such that, you know, we can't actually. Unless we work it out.

>> Steve: Yes.

>> Anouk: So unless we work it out, we're not happy with it, probably.

>> Steve: We try to understand everything in detail.

>> Anouk: It's also the risks associated with it too. So we also go, do we really, really 100% trust AI?

>> Steve: Yes.

>> Anouk: No. No, because we hear those storeys, you know, do I trust AI to behave? Do I, do I trust hackers to only do ethical things with AI?

>> Steve: No.

>> Anouk: Exactly. So, um, the other thing I read over this weekend, I did a lot of coming up this weekend, but was that um, third party skills, so things you would put into your agent and we were looking at AI Azure. There's a way of saying, hey, where are my third party skills? And you can connect into them and, and load them when they're owned by somebody else. Right. Uh, it's a perfect medium and it's not tracked and it's difficult to, to actually look for malicious code. All right, so in your AI, it's going off to a server to do something. Okay. Um, and then they just change what the server does and there's no way of tracking that agreement. And even my security guy, when I was asking him over the weekend as well, I said, hey, is there any way of dealing with this? He went, no, the only thing you can do is put really strong governance in on third party at the moment. And the security guy that was writing this said it's the same thing with the um, uh, app catalogue. It took his years to get the app catalogue managed, so it was safe. And now we're starting all over again.

>> Steve: But they rebuilt on that experience they had with the app catalogue. They have a starting point and they are building on um, that even it's now for a different propose. But that's something in the future AI will do for you as well. We use those kind of things.

>> Anouk: But you're sitting there. Yes, I agree. I was about to do what you did. Yeah, you're sitting there assuming that that will happen. I was doing what you did because we both know it's got to be done. So yeah, I, I, I kind of, uh, I get it, I do get it. Um, but I, I, I think, I think you're right. We will get over these. There's no doubt about that. Um, and we don't even know how today. True, we get over it. And I think that agents creating agents is definitely a way forward that will happen. Um, but again, we've talked about governance in the previous episode of this. Um, we will have governance in there about which agents can create agents and what their limitations are.

>> Steve: We also talked about prompting, so people need to learn how to prompt correctly. Is it a skill they need to learn or is it going to get lost in the future? And everything you say AI can turn that into the great prompt that you need to have for your feedback.

>> Anouk: Today, yes. Tomorrow, no. I think, I think. I mean, I noticed ChatGPT just released a new voice module. I haven't tested it yet, but apparently voice is becoming more human, real. Um, and, um, something to play with

>> Steve: in the car home.

>> Anouk: You could do that. Um, uh, but I think that, uh, I think eventually you'll just be able to sit there and go, look, I need this and this to be done. And then it will add the bells and whistles and ask you a few questions, then take it from there. I think that's what will probably happen.

>> Steve: A future to look forward to, to learn and to understand what.

>> Anouk: That's a really interesting question. A future to look forward to. We'll go with.

>> Steve: Yes, I think it is.

>> Anouk: Yeah. So the closing thought, I think, is quite nice again on these here. So maybe prompt libraries and prompts are temporarily and, uh, maybe organisations are teaching AO how they think.

>> Steve: That's very one to close down with.

>> Anouk: All right, we hope you enjoyed these, uh, five short podcasts. Let us know you know how to get orders if need be. Nobody, uh, ever gives us feedback, so if you want to be the first, go for it. Sam.