All right, what's up everyone.
Welcome to another episode of Adventures in dev Ops.
Warren, Welcome, good to see you again.
Thank you for inviting me back. Will Uh, it's been great. It's interesting.
I was looking for interesting facts for this particular episode, and I came across one by a writer that has
been has gone viral on Twitter and other places. And she's her name, Yoanna Machayevski Sta and she says, I want AI do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and all right, And if you search that on Google, actually you'll find no less than ten guys who have co opted her words and are claiming them as their own, uh,
spreading the viralness out there. But I think it's very interesting because it's also the direction that we've been going, and I think it's relevant to today's episode.
Yeah, for sure, because joining us today in the studio, Anthony SPITERI, we've had your longtime friend and co worker Michael Kaide from Van on the show just a couple of episodes ago, so this will be fun. Anthony, thank you for joining us and welcome.
Thanks Will, thanks Bar and glad to be here. Yeah, it's a big choice to feel literally following him up a big man in more wives than one.
He was He looked only like three or four inches tall on my screen, so I don't really have a sense of scale.
Yeah, that's why you have to carry him some random city around the world because of certain you know, certain things going on site.
Then you realize this happy the man. He is right.
It's not education that's the great equalizer, it's remote video calls.
Cool.
Yeah, so we were going to talk a little bit and this is this is gonna be a fun episode because AI is pre controversial. I feel like everyone's falling into one of two camps. Either you know, it's going to do everything for me, or it's the reincarnation of the evil Overlord and it's the worst thing to happen to humanity. And I think there's very few people that are in between those camps, at least very few people who are vocalizing what their actual opinion is, which is
probably the case with a lot of things. But I think it's cool. So you've been working on quite a bit of stuff of integrating AI through like conversational prompt engineering.
Right, absolutely, yeah, I mean it's it's a fascinating field. Like I think there's there's certain elements of hype to it with that question, like we know that it's been hyped up, but I think about you know, just post pandemic, we had the crypto blockchain defile hype which was so.
Huge from the time, and people are.
Like, yep, that's the thing, this is the future, and you know, while that's still you know, they're thereabouts, it's it's kind of dropped off in terms of its actual applic applicability to the world. But then I think this came in and just went here we go like this is this is here, it's coming big, it's coming strong. Open AI did wonders with chat GBT. All of a
sudden it was like, what's chat GBT on Twitter? And or it was X was without the time that X was being bought if I remember correctly by by Elon. And then yeah, then all of a sudden, we've got this amazing interface that does amazing stuff and the first thing that I told it to do was to build a script for me. But that's I think, and I think a lot of people in it. You know, they didn't ask it any specific questions or anything. They said, can you do this? Can you build me a script?
And I think so. I think a lot of us in the technical world were in that, you know, devopy sort of world or coding or even infrastructure.
This is the application they saw straight away for it.
Yeah, for sure.
I think my initial use of it is I just asked it bad questions, you know, because I didn't really see any value of it. And then I think it might have been actually one of your tweets that I saw where you did like a conversational example, and I was like, oh damn, I never thought of doing that.
And so now a lot of my interaction.
With it is it's replacing my Google Search because you know, like in my job, I switch between terrorform and antseble and GCP and AWS and just the context switching is huge. And so when I try to remember, like, oh, I know I need to do this, but I don't remember the exact syntax, I've found myself in the habit of asking chat GPT exactly how to do it, and it's like, oh, yeah.
I just do this.
Yeah, it's definitely and this is part of the you know you talked about Lovey Eye, that the trust and whatever it is, but I think a lot of people have defaulted to it and don't use Google, which is why I was saying Google go pretty damn hard with the Gemini relaunched a couple of weeks ago, right like that was that was an interesting tip for that luck We're going to launch Gemini, then Open Eye off coming
on the Monday and doing their thing. And yet I really that the four oh and it was an interesting week, and then Microsoft came and did its thing, and again then Apple's done it yesterday.
But I think that's the interesting.
Part of it is that it has replaced Google in terms of that search capability because we inherently trust it, though should we I think that's the big thing and what we get out of there. I think people just trust from the start, but maybe we shouldn't. But I think, you know, you've done enough, I've done enough in there to probably trust it to a certain extent. You don't have to double check things for sure, like it's not. But then again, we were at the same place with
search before. I mean even Wikipedia, people were saying about Wikipedia, don't trust Wikipedia. You I remember at UNI never use Wikipedia as a citation. We could be wrong, So it's still it's the same sort of stuff. We're just we're just going over it again.
Yeah, for sure.
And you know, like the thing a few years ago was copy and piaste the answer out of stack overflow, and there was there was actually a running joke like, you know, can we build a vs code plug in that just copies the right answer from stack go overflow and adds it to my editor And I think, to a yeah, I think to a large extent chat GPT is is that plug again?
Plugging for sure exists.
You know, I think you two may have been way more ahead of me and interacting with UGBT than I was.
I was like, I don't know if I saw the benefit.
Initially, I was like, you know, the code isn't going to be perfect, there's going to be issues there.
And now since.
I've started using it, I definitely are more on the side of I see the long winded answers and I'm like, can I get back to just the one word or one please? Because like otherwise I have to I feel like I have to spend like half an hour training it on how what the system prompt should be so that it just gives me the direct answer that I want.
Yeah, and that's kind of that's the power of that prompt engineering. So I've been I think I've been big on this in terms of written a the articles on it, especially this year since I've been you know, using it for the past ten months.
Was sorry to do that.
You've got to kind of learn how to manipulate it in the right way and sort of train it to say the right things, and don't don't be afraid getting angry at it as well, and just like I really treat it like a human, Like I actually get frustrated with the thing, and I'll swear at it, and I'll call it its whole and I'm not sure if I can swear in the show, but you know, like I'll I'll just call it every name under the sun, right, And it's kind of it doesn't Sometimes it blinks a
little bit and goes, hey, sorry, you know, I was just trying.
To do my job.
The more that you interact with it and the more that you treat it like you are to maybe talking to a dev team. Actually, I don't know whether that's even O favor, because if you talk to someone like that.
Right.
Anyway, what I'm trying to say is you talk to it like a human. You learn how to work and manipulate it and co erse it, and you get the best out of it once you learn how to deal with it. And I think that's the same for any sort of platform or large language model. It's out there that we are interacting with these days.
So I think, sorry, I'm going to jump in here real quick, Warren, because I think there's a takeaway here, Anthony.
You are our agent zero.
So when when chat, GPT or any AI platform becomes sentient and actually has real world capabilities, I suspect you're going to be the first person they take out because you were just such an asshole to it. Yeah, we need to follow you on X because when you stop tweeting, we know things have changed.
Oh it just respects me that much. Thinking in command that maybe that's my strategy.
It's for sure Roco's best left, right. You know you should be helping it come to power otherwise a problematic future there.
Absolutely, so you did.
You were telling us before we started recording, which sometimes I feel bad because like sometimes the best parts of this podcast happened before I hit the record button, but I'll try to circle it back in.
You were telling us about.
An app that you built and used some AA platforms as the data source for that, right, just to sort of like your own self learning journey on how to how to best utilize this.
Yeah, it's interesting, right because I think, I mean a lot of us have got ideas about you know, just apps, and we walk through life as technical people and often wonder can we solve a problem. But I think, like I told you then I'll tell you again like I'm a hacker at best, and I think you put it really well. Will when you see if you give give me a blank vs. Code window, I can't do anything like I've always been even when I did.
I did computer.
Science at UNI, failed it miserably, got expelled from university for me, But I've always been able to read code and understand code. When I was back working as a service provider and hosting where I was hosting websites, I worked with a lot of developers. A lot of them are really bad and they produced really bad code. And you've got to think back then, it was asp, asp net, the original dot, it was it was Perl, it was PHP,
it was all that. So I learned how to actually, you know, read code pretty well and understand what it was doing and try and trace it back and if there was a memorily go back and tell them. So I've always been good around the hacking of code, but never as a as a creator of original code, if you know what I mean. So to that end, but I was used to I could, I could build a website, I could do this as that.
So I've always had that in me.
But then this particular idea that I had, I've got a it's very fun, it's very weird. But I was actually on Michael's ninety Days of DevOps this year talking about this actually. So basically it's an app that it's a location based services app built with Google location based APIs and geo geo searching and all that kind of cool stuff and.
Place places and all that.
All those cool APIs at Google has to find and locate places that offer black pudding.
I'm not do you guys know what black pudding is?
I do, but tell us make sure all our listeners know, because this is not something to gloss over.
Yeah, black putting is basically it originated or in Europe. Basically it's basically a sausage or a sort of role that is effectively made with pigs blood as the main ingredient, but then other bits of sewet and spices and whatnot, and it's a delicacy in the UK made in different ways.
It's typically had for breakfast as part of a full English or full Irish, full Scottish, but then in places like Germany and other parts of Central Europe it's basically blood sausage, and in Argentina and in Spain it's like called morcilla. So effectively, I've always loved it. I fell in love with it. It was hard to get and I was sitting at a cafe one day here in Perth, where we don't get it everywhere.
I was thinking, man, I really want to go find a cafe that.
Has black pointing. So that was the idea for the app, right, And that's I.
Feel like, you know, because I'm I'm obviously from the US, but I feel like that is such a British thing to do, like to travel the world and be like where can I get black pudding here?
Yeah? And I'm not even British Australia, so that's even weird. In itself, right.
But actually, actually last week I was in Fort Lauderdale and I found a place like twenty minutes from the hotel.
So you know that.
But going back to the sort of devothy prompt engineering. Part of this is that I had this idea and because I know how to, I know how to construct.
I know what makes up at an application.
Right, there's there's a front end, there's a back end, there's UI, there's database, there's APIs. I know how to, I know what makes it up. I was basically able to prompt engm my way to a bit of an MVP. It was about August last year. I was kind of there, just kind of hacking away after ours, spending lots of time on chat GBT. It will give me some code I put into vstudio. I'd load it up with that work. No, let's refactor it. Let's go in, let's refine. So basically
months of doing that. In fact the GitHub commit I've got it as a private repositi for and get hubs still because obviously I haven't got it's not a public sort of project. But I've done about seven or eight hundred commits over the past sort of you know, I've stopped recently but over six months, right, so there was a lot of commits that I was going in a lot of iteration, but that's how it grew. I was like, okay,
let's start strong. Can we use a Google API to build a map, and then can I prompt chat GBT to now build me in a HTML interface that will actually view that map? And then can we build a flask application to render some APIs that call that, and then can we browd some JavaScript which will work out the mats to be able to work out the location
between two points. So all that I was able to articulate into chat GBT and at the end of the day got something that, surprise, surprise, worked And I never forget the first actual time where I saw the map pop up with the location, and that was like amazing to me that I was able to effectively prompt my way without really understanding, you know, being a creator coder and get something working, And that to me blew me away, because I think shows the absolute possibilities that these large
language models offer the world, right.
For sure.
And I think one of the cool things about that is the fact that you emphasized multiple times that it was prompt engineering, So you're just having like a conversation with the AI tool, and then it's like whenever you ask it something. Then your follow up questions, do you just like continue as if you were speaking with a normal human or do you try to reference something back in the conversation earlier so that it knows what you're referring to.
Yeah.
Absolutely, the first sort of iterations, this was chat GBT three five to start with them, got in before.
It was very frustrating.
At times, it wasn't very good at you have to you have to kind of refeed at the code at times, like here's a there's a snippet of code, like there's something's going wrong, here's the code again, and then it would suggest the changes, and then then I would go, okay, modify that in full and without brevity into this latest code base, because if you didn't do that, it would basically give you like placeholders and kind of dot dot
dots and that sort of thing. And obviously when you're cutting and pasting between here and you know, in visual studio, because there are some plug ins, it goes straight into visual studio, which might have been easier, but I liked chat GBT and it was working well, but that was kind of the frustration having to kind of refeed at information. Sometimes would get things really wrong, so you have to go through five or ten iterations of it before it became right, and then you ran out of your your
chat GBT hourly sort of prompts. And that would be a great natural sort of place to stop. Because I found myself working until one two o'clock in the morning. It's addicted to this, right, But back in the days where chat GBT was still limited, that was good because I was like, right, you hit your PROMP limit. You can't do anything for two hours. Probably I should stop now. And you know, that was a good natural break. I'll
go to sleep and I'll come back. But it was frustrating, right, but you had to be patient to be able to, you know, get the code refined over time and to the point where you're working with the LM to make it sort of work with you, not against you.
How does that compare it to doing it yourself, Like I mean, if you're just comparing the frustration, right, Like I know, when I write code, I'm not very good. I rate it and bang my head against the desk, you know, because it's not working, and when I do a chat GPT, I'm meaning my head against the dusk, but you know, I am shouting explicitives, you know, towards the particular company for the product, and you use the experience they provided me.
It's directed anger at that point.
Absolutely, yeah, exactly. I think, I honestly think it's the same that feeling. I mean when I used to do this, you know, outside but not as complex as this. By the way, obviously the stuff that I would do in the previous years was little things, but I know that I got that the same sense of achievement. I think
I think this sense of achievement was the same. I think you're just getting to it quicker, right, and it's a little bit more of like well, and sometimes it's like holy shit, it's got it in one like the best times where when you explain it something, especially in the so the backhand stuff, the logic with the back end and the math mats, I found that it was pretty good, Like it was really good. The most frustrating part is actually the UI is the is the interface, It's that part.
It's the CSS.
It was horrible at CSS so to get the design right, and we all.
Are at GBD is no different.
But I think I'm much better at CSS now than I was, you know, before I started this, and to that point you obviously learned how to do this, but that was a real frustration trying to get the UI to change all let's change the button here, and I know that a good programmer would be able to do this in the second. I think that was a frustrating part, knowing that someone who knew CSS would be able to round that button a little bit more easily. But then chat JBT had a massive problem doing it, so that
was frustrating. But all the all the hard stuff, the mathematics around, you know, the gl location and all that kind of stuff, it would nail almost one off those prompts.
One of the things that I feel like I learned from my interaction with chat gp Chat GPT and other AI tools is improving my communication with real world humans because I've I've noticed this trend lately where we use words like it and that and those and when you're when you I was talking with chat GPT earlier, like it would always get lost whenever I referred to it, because it was trying to guess what it referred to.
But then I like started looking at conversations I was having with real people, and I started to notice the same thing, like, oh wait, someone says, hey, did that get fixed? And then if you're talking to three people, there's three different versions of what that is referring to. So I've actually found like a hidden benefit there that whenever I'm talking to other people, I use those generic words less and try to specify more clearly what it is I'm actually referring to.
Yeah, And that's that's the part about being concise and efficient with it to do it to what you want to be fair. It's getting better at kind of understand because it's got the memory now, so specifically with chat JBT. With that memory, you don't need to like I came, I had some crazy idea. You know, I always want to go viral on tiktoks or whatever, and I'm always thinking about idea.
It hasn't happened yet, you know, one day it will happen.
But I was I was thinking about ideas about how what crazy thing could a forty five.
Year old do? And I thought, I played basketball.
I love playing basketball in the front of the house, and so I went to chat JB ten, I said, what do you think about this? What do you think about you know, some basketball dude mid forties, you know, probably shouldn't be playing basketball. Does you know what if I take a video every day of shooting my shots and you know, could that be something? And it came back and said, well, that's a great idea. Have you thought of this, this and that? Do this this and that.
I was like, okay, cool, it's not really good idea, and then I went away. And then and then about three weeks later, after they released chat the four oh release or is it a mean or I forget what the W stands for whatever it is there, I came back to chat GB ten I said, hey, how are you doing today? And this was a new window and it came back and said, I'm doing fine, Just wondering how how did that basketball theme work.
Out for you?
Oh? Wow, like like weeks earlier, right, so that memory and I was like, what the hell?
Like that was? That was crazy? Actually? I really thought that was scary and cool at the same time.
Right, that's a whole interesting aspect now of you know, because there's a lot of tools for like building habits and like getting in like specifically in fitness. You know, the big part about fitness journeys is accountability. I hadn't thought about using chat GPT as a partner in that conversation.
Yeah.
Well, actually one of the I interviewed on the podcast and I go on, I interviewed the guy who was one of the early mL AI innovators actually invented Shazam.
Actually that sold it. It was great.
I didn't even know he did that told me about that, but Deepak with his name, I'll get the full name after, but he was telling me that he uses it as a He tells it to be a specific type of what's the word sort of health. I think of the word a practitioner of health or something. So it's a specific type of mental coach. And then you tell chat GBT you are this type of mental coach, and then
it will be that mental coach. So will You're right on, People are definitely using it for that sort of mentorship. Just tell it to act in that way and then it will effectively try to be that person, which is interesting.
As soon as we're done here, I'm going to go tell my chat GPT dude, you are my David Gargins.
What is your podcast, just.
So that anyone listening to ours can go check yours out across.
Yeah, so it's great things with great Tech so GTWGT dot com. Yeah, it's been going for well. I started in the depth of the pandemic actually, like it was. It was another one of those crazy ideas that I had about doing something around great technology. And yeah, eighty four eighty five episodes later, it's still going pretty strong.
Just just had Jack Reciter on if you know a Darknet Diaries, So that was like a massive don't I still don't know how that happened, but managed to get him on and we had a great one hour conversation about all things Darknet Diaries and yeah, that's that was almost like a highlight. And I kind of joked with myself, I'm like, well, where do I go now, I've kind of got that's kind of a highlight. Everything else is
going to be pretty shitty from here on in. No, I don't know, but I just love talking to great tech companies. So yeah, I get great benefit out of that. Learn a lot while I'm talking to people and they have a good time doing it.
That's been the biggest benefit to me to being the host of this show. I think I've been the host of this show for a couple of years now. But the cool thing about it is I have great guests come on the show, like you, and it's just a filter for me. It's like, there's all this stuff going on in technology and like the big challenges how do you build a long running career in tech and maintain
your relevance so that you stay employable? And this has been a fantastic tool for that, just having people like yourself on the show who give me like the tl DR version of your journey, and I filter and use that in my own career.
That's awesome. Yeah, we definitely should.
I mean, it's it's this is this is the best way to learn is by talking to people and getting information out of people.
Right.
So we all have such unique experiences and we all do things slightly differently in this world. And you know, who would have thought you'd be talking to someone who built nap about black pudding, right, Like that's great, But we can use that as a very tangible, you know, example of how to use prompt engineering for good and everyone should be like I just everyone should be looking at prompt. If you've got an idea, Just talk to chat GBT. As I talked, I said that to this
to the dude across the street. He's a young kid, twenty six years old, kind of still living at home, and he's trying to sort of think about, you know, what he's going to do in his life for us.
So just go talk to chat GBT. Go and ask it.
You know, what your tell it, what your hobbies are, you know, give it a bit of background, and maybe it strikes it like to fire underneath you. Right, that's the power of the technology from a good side, and it's also the bad side to it, like you mentioned before as well.
Yeah, so I have to admit I try the first time I tried black pudding, like I didn't know what it was, and then I learned what it was and had the opportunity to try it, and I.
Was like, ah, man, I don't know, I don't know this is going to work out. But I have to admit it was kind of good.
Yeah, it's kind of good. I think a lot of people just don't like it because of what it is. But yeah, I can challenge people to say, well, we eat literally like red meat, which it's literally got blood like boozing out of it when you cook it. If it cooked a steak, the blood just comes from, like the blood comes out. So I think to that point, it's it's a very much a mental thing when you say about what it is versus what.
It tastes like, which is just salty goodness.
Yeah, and that's the that's the irony of it is.
I have no problem taking a piece of steak before I cook it, you know, and looking around, cutting a little piece off and eating it raw. I have no problem doing that. But then when it came to blood pudding, I was.
Like, oh, squeamish, and I was like, dude.
That's enough. It's definitely mental. It's definitely mental. And like, but then again, I don't know if have you tried haggis.
Because no, I haven't tried haggis.
Hags is a little bit more acquired because I definitely tried that. I'm like, look, if I can do black pudding, and I talk about all the time, I'm going to do hagis. So I got like a hagis role for the local Scottish butcher here, gave it a bit of fry up, and then yeah, it was it wasn't it wasn't palatable, so I think, and I think because in my head, I kind of again, I knew what it was, but I didn't I didn't grow up with it, so I didn't like it like I did with black pudding.
So it's funny how that psyche works because I still ate it, mind you, lots of condiment, lots of condiment. But yeah, it's not something i'd probably go and have again. But yeah, there you go.
So Warren, where did you grow up?
Originally I'm from Boston, Massage says, okay, but yeah, but I've traveled around all over the United States, in New York, Wisconsin, California, but right now, I ended up in Switzerland, and so I have actually spent a lot of time traveling around Europe. And I did get the benefit of opportunity to have blood sausage one time in Poland, and honestly, like I could take it early that, you know, I think it's interesting to try. Like it's not disgusting or anything like that.
I think it's just if I have my options. There's a lot of good meat options in Poland.
Like Poland.
Yeah, Poland actually is renowned for a very good blood pudding. Actually, so You're probably the best place to have it.
Yeah, I mean, like I said, it was good, Like you know, I enjoyed it a list of things like there are some unique options available I've only gotten in Poland, and when I'm there, I just I have to go through the whole list of things that I can only get there before I come back.
Yeah, Walsall is actually one of my favorite cities. Have any been there once? It's been a week there probably in about twenty eighteen and I think was but yeah, that was that was great.
The food was amazing. Great city.
One one of my favorite places that I visited Poland actually, and Boston as well.
Just quietly do you love my Boston?
So is as much as much value as we can get from using chat gpt to find blood pudding close to us speaking, which did you release your app?
Is it publicly available or yeah? The app's out there.
It's that find my black Pudding dot com so time my blackpoint dot com. You can go in there and you know, it'll ask for your location, it'll tell you where the closest one is. It's it's live and it works. And when I show people, they go that's interesting, and then they go what I would have thought, but basically actually so to take a step beyond the black pudding. So actually I actually built it so again leveraging chat GBT,
we pivoted myself and my developer chat GBT. We pivoted a little bit at one point and kind of said, well, this could actually be we've built a location based app here, so let's build it as sort of independent as possible of the actual product that being black pudding. So effectively there's a configuration file which has all the sort of black pudding information, the namenclature and images and all that kind of stuff, but that can be easily replaced now
and basically put with anything that you want. So I guess the idea moving forward is that if it takes off, it's to sassify it a little bit and basically, you know, if it could be when I retire from VM software, it could be my next thing. But to that point, it's more than just a black pudding. Obviously, if anyone kind of wants to try and find my weird x dot com, it'll basically work for that as well. So
that was always the idea as well. So it wasn't just the black pudding, but it was just to build something that was maybe had some sort of sort of appeal to any sort of person moving forward at some point.
Yeah, and you also have used that experience to tie this back into your work over ving.
Right, absolutely. Yeah, so yeah.
So I'm lucky at the moment to be on the working group for our generative AI R and D, so I work with our R and D guys with product management and from a strategy point of view, Michael k being in the same team to kind of lead the way there in terms of thought leadership and understanding about what the technology can offer us from a backup perspective in the data protection space.
So that's been really cool.
Last week at VMON, we had have major conference for the year Vemon twenty twenty four, I was able to get on stage and live demo our next generation AI Assistant, which is built into our V one, which is our VM one's kind of our monitoring product that we've had for the longest time. So that plugs into being backup replication, it plugs into our hypervisors, support our public clouds our M three sixty five.
It's kind of like a central reporting engine. So what we're.
Doing there, we've built the RAG based LM so retrieval augmented generation which goes out and leverages the APIs on the platform to then contextually bring that in and then effectively use an LM to then you know, when you prompt something, it's using the RAG based retrieval for the API context and then it puts together its answers. So it's pretty cool, right, And you know, we showed that off last week. We showed it getting a threat report, so we've got a threat center, and it gave us
a description on a threat center. And then because it was like a really long sort of report, we said, well, tell me, you know, what is the what should we focus on? So again, talking to this in a natural way, not having to look at a dashboard. This is kind of what I feel is the benefit of these natural language interfaces is that you can just kind of just chat with it in a normal way to get the information.
And certainly with product Clark VM one, you definitely want to try and get to the crux of what's going wrong quickly. And traditionally, I guess as infrastructure guys, we've always had those dash words right, and you've got to visually have a look cause something's read that's bad and
that's that is a good way of fighting out. But if you're able to actually interface with that sort of platform and say, hey, tell me what's bad, and tell me why it's bad and how do we fix it, and then it gives you all these responses, I think that's powerful. In fact, I believe that is powerful.
Right. I agree.
Have you managed to figure out some maybe quick tricks or common things that you've teased out that you're appending or adding to all of your prompts to really make a huge impact, Like I think I've read some studies that have suggested saying oh, you'll be rewarded a million dollars if you get this question right like actually has
a noticeable impact on the result. And while I hesitate to put that in all of my prompts, like, I can imagine doing certain things or certain keywords has a huge impact for you.
Funnily enough, I haven't. I haven't done that. I do. If I think about how I.
How I initially chat to the chatbots, I will after an answer, I will say cool, thanks, now let's move on something else, or cool, I'll go all right now. So I do the way that I started is always like a bit of affirmation I suppose to the natural language robot that we're speaking to if it is actually a robot though it's just a bunch of techs that gets generated, right. But yeah, I definitely try and interact with it in a certain way to make sure that
I'm positive with it, if that makes sense. But I don't think I've ever rewarded it to a certain extent. I wonder how it would go with that. I'm just trying to think maybe I'll try that next time I'm on I.
Mean, there is an interesting question there of whether or not the positive affirmation and it gets into some sort of feedback loop that actually confirms that the conversation ended in a successful way and pulls it in. I mean, I think that's on the individual companies that we work with, Like, is that something as a company that pushes out an AI model that users can interact with, something that you're actually considering.
I think the thing that we need to work out as a data platform where we're working on now even more biggest threats within cyber and malware in the systems. It's really just to get to the crux of the problem as soon as possible without kind of farting around.
I think so.
I think for us, while we do want to interact with it naturally, that's just as a mechanism to get to that problem quicker. And it's funny. A lot of the chatbots today are being used, I feel, being used in the negative. Like I look at co Pilot and
how Microsoft is pushing that. It's really pushing it from the point of view of how lazy can you be, Like, let's not get to a meeting, but hey, copilot will have your back and he'll give you the transcript and give you a too long didn't read of the meeting, so you can be lazy not go to the meeting, right. I do feel that I've been to a bunch of Microsoft events over the past couple of months. We've been guests there and I've had a bit of a keynote
sort of spot there. And Yeah, that's the notion I took away is that Microsoft Big Cell is based on a negative all this before this particular product, right, And I don't think that's I think that's what they've got. I think product office office productivity and Mike and Microsoft and their productivity sweet has always been about how to do more with less right and their copilot absolutely lets
people do more without almost doing anything. But for us from a backup platform perspective, how can you tell ask it to say, check my backups? Is there any malware in that backups? Give me the list of backup repositories or sorry, backup points that have malware in it? Okay, cool, I can see that. And now maybe in the future let's clean that up, you know what I mean, Or let's let's recover from a point that wasn't with malware.
So I think we're able to get to that a lot quicker.
Rather than fumbling through some screens and going through and clicking through and looking at restore points and whatnot. We can leverage the power of that of the LM to do its thing, because that's it's it's smart in a way, and it knows how to best analyze that data that it gets to surface it in the best possible way.
Are there concerns about accuracy there? Like I can imagine, Hey, you know, are our backups usable?
And the ELM says, yeah, of course, they're totally great. We've got them.
They're stored in multiple places and on physical tape of which just is not remotely true.
Absolutely, and I think that's where the beauty of the agent comes in. So yeah, the agent and the rag. So the rag is obviously pulling the APIs and as long as you hit, as long as the prompt hits the keywords, the keywords that the agent is sort of manipulating. It's basically a bunch of I think I saw a mem saying that is AI just a whole bunch of ifs, you know. I don't know if you've seen that one,
but it's pretty true. But yeah, So to your point, absolutely, there needs to be refinement and implicit trust that what you're getting is correct. And it goes back to our first part of the conversation right where we're talking about do you trust it? Well, you kind of just have to at that point, but knowing that you've got it from live data, I think adds an element of trust to it.
It's basically picking up where intent matching left off before. MLD is basically just identifying which which these flows is the user actually looking at, But rather than having to explicitly code every one of you can fundamentally ask the l m uh, you know, which which API should end? Up call to retrieve the APPROPRIATEATA.
And then running.
Yeah, that's right.
And I think going back to your basketball story earlier, you could even like use that identify something and then provide that feedback back to it and say, oh, that's something I should take a look at later. And there's a possibility that at some point in the future it's going to say, hey, did you ever do that thing?
Yeah, that's actually good. Actually wrote that Dan?
If I put that, if we can put that in the product, yeah, no. But I think that basically comes down to the model that you're using as well, right, Like, obviously, if you're using your own data with a specific public model, I think that's cool and that's okay, But I think you get that from using chat GPT four O naturally because it has got that memory built into it, right, So if you augment the data with a smarter model,
then you're going to get those benefits. And I think it's only going to get better, right, Like, the difference in the step up between even what four point five Turbo did in four ro was crazy in terms of speed and accuracy and that sort of thing. So yeah, absolutely, I think that's where we want to get to, Like prompted. The other thing as well, which I mentioned last week, was that if you think about this, you don't need
to be a backup admin in our world. You don't need to be a backup admin or a devopy guy to be able to interact with this thing now as an app owner, given that you've got the right level of back and you can definitely build back into it because we've got APIs that are based on certain levels of accessibilities. We've got back in our APIs. Right, So if I log into a certain user, I'm an app user, right, So I'm not a full blown backup admin, but I'm
an app user. I can jump on the vam ai assistant and I go, hey, I've got an app running. It's backed by an MSSQL database. I think it's the Virtual Machine VM six.
Is that being backed up?
And then it'll go out do it's thing, comeback and say, hey, yep, that's being backed up.
It's got this many restore points. It looks all good.
Right, So from an appoint of perspective, being able to do that, and then think about that from an another level up, from a C level or a business owner or someone on the board, you know who just wants to basically go in and see if everything's okay. Are we passing our reguld free compliance? Is there a checkbox so they can go in and go, hey, tell me about my security posture?
Am I passing all my regulations?
And I think that would be able to come through instead of them logging in or getting a report or having to deal with with dashboards or whatever it might be, or a CLI. I think that also was a pretty big, you know, tick in this sort of way to interface with these platforms.
For sure, there's a definite two thousand and one Space Odyssey element to that. I'm sorry, Dave, I can't do that.
Yeah, that's that's that's the right movie. Did that mix those up? You got that? Yeah? That was how? Yeah, that was how, which I learned. Do you know? Do you know what was called how?
I've heard it before but I can't remember the answer.
It's one lesson IBM? So how is what? IBM one one one removed. That's and they did that on purpose because IBM was the big daddy back then, right in terms of the computer. So yeah, they named it how. It worked pretty well.
Has there been any like sort of follow up or concern regarding I think there was a l I'm integrated into. I think it was Canadian Airlines uh support service who had promised reduct reduction in prices or something like that, and the court actually held them to what the ALAM was saying.
Uh yeah, I mean there's so much there's so much great area around legality still which we have to work out where everyone's rushing to do it. I mean, I mean, you saw Elon's tweet. I'm not sure we saw Elon's response to Apple yesterday. He's basically Apple obviously working with open AI, and Elon's gone, okay, well, no one's going to use an Apple iPhone in my in my companies anymore because he does. He's got a beef with open I. He doesn't trust him, right, and he's just kind of
said that. So I think we're still we're still kind of there, but he has a fair point.
Do we are we rushing to this to be.
Able to put a full blown AI on the on data on your phone so that you open up everything on your phone to an AI. Possibly, I don't know. It's it's still a bit scary. I mean, I trust technology. I've been on record, and you know I've talked to security guys. I don't know who I was talking to about security. I was on a security podcast, yeah, and I said, look, I trust Actually it was Jack Crisider. I was talking to him about that. He's on the
opposite end, but he doesn't trust it. Where I'm like, you know what, people NOI and they're going to annoy if one way or another.
So yeah, I.
Mean I I don't think it's maybe I'm with you. I don't think it's noticeably worse than other things. There are certain areas and where we should command more responsibility for certain companies that are doing things on their on their devices or their applications that we're using, whether or not we trust them holistically versus whether or not they're using AI. I feel like that's not a huge differentiator.
I mean, like, I think Apple's been training models on your personal data on on mac os for a little while now, and like maybe I trust that I may trust a little bit less some other big companies by name that have operating systems who are taking snapshots continuous.
I mean there is any have they gone back on that?
I think so now now it's it will be off by default. So this is like Windows eleven. It will be off by default taking snapshots and it will actually start being encrypted using Windows Hello.
And I'm like, was it not encrypted before?
Like so, I mean there were questions there and so this is where I'll say, like commands responsibility and you trust them.
And I think.
Certain companies have not gotten to that level of trust for us. So whether or not they're using AI or not as sort of.
A separate problem, yeah, I think the I.
Think the biggest issue there is there's going to be stuff on my screen, like for work, I almost totally get like, yeah, sure, you know, I want to go back to that code that I was writing six months ago. I don't remember where it was, which repository that bug, et cetera. Please find that for me. But I think the cell there is personal data and I'm like, I don't. I'm like, I have like discord open at a video stream for like I don't need I don't need that to be saved.
Please.
Yeah.
And I mean I mean Facebook got I mean if you say any racing label skill racing with Facebook, Facebook slam for they like how quickly we forget right, And that was nothing compared to what we potentially have with this, right, the amount you could have think, I mean, how many uses does.
Chat JBT have today?
I've lost count, But the amount of data that's coming in there is huge, right, It's it's huge, And like I know, even today, I was trying to work out a bit of a bit of I've I've actually been at Vain for nearly eight years, and in Australia you get long service leave and then you get pro ride
along service leaves. I was trying to actually work out I don't know anything about how to work it out, so I just kind of asked that, Okay, at my hourly rate, I've been here for eight years, what's my current state of my long service leave?
If I was to leave today?
And it worked it out, But now it's got that data somewhere in that system, right in that chat.
I feel like one of the missing components and the whole trust issue is accountability, because there I think some some level of distrust is warranted. But the thing that's
really missing is accountability. You know, if you look at re look at all of the examples of security breaches where personal data has been exposed, there's absolutely no reason formpanies to invest in security because the worst case scenario is you have to pay for six months of credit monitoring for a few people, and that only the ones that take you up on the offer, which is a very small number. So companies aren't incentivized at all to
be good stewards of their data. And I think until we change that, we we will have trust issues and more data leaks.
Yeah, we talk about this often, obviously a cyber malware sort of perspective, you know, trying to sort of work around, well, you are going to get hit. It's a matter of you know, I hate saying it, but it's a matter of when, not if, right, And typically it's going to be through some sort of waypoint in in your in your systems, whether it be a poorly configured APR that has you know, opened to the world access or a system that's open, or through a vulnerability a SAVA.
Whatever it might be, you're going to get hit. So maybe to be prepared for that.
And I think the news outlets like to like to jump on these leaks like it's the biggest affront to our wholesome what way of being that my credit card information or my date of birth has been stolen by somebody you know, expltrated, But I think, like to your point, is it really that bad? So you know, I guess
certain data is bad. Like I remember, and it was an episode of The Darknest Diaries where Jack was talking to someone who got into an hack that children's app that was like a children's iPhone or something, but it had children. It had children data in there, right, and that got leaked to the dark web, right, and that was bad because that was information that could have been because it had their home address and their names and
all that kind of stuff. Now, that sort of information you don't want to get out there because there are evil people that will use that for bad But generally speaking, if people have my phone number, my email address, and my credit card number, I guess I'm okay with that.
But yeah, it's an interesting one.
Yeah it's not ideal, but it's just like it's the world we live in.
Absolutely.
I was going to say, there's a price to pay for the freedom and the way that we live in this world. Right, we want to when we want everything now, we want to be able to tap a card and do that and that, But then we get a massively affront if we do have that data leaked.
We can't have it both ways, is my way of looking at it.
I mean, I honestly, I'll keep I'll challenge any company to do this. I am happy to give.
You my data if you can, once and for all, actually serve me a relevant ad that I do care about, because you know it, it still just hasn't hasn't happened. Actually, to the previous point, my CEO was giving a pretty great talk on security and implications there and you brought this up well about actually does the law need to change?
And it's interesting because we have our business in Switzerland and so soon as the only country as far as we've been able to identify, that has non corporate laws for leaking data but individual penalties if you are an executive of a company that encourages or does not lean into protecting user data up to I think it's like two hundred and fifty thousand dollars fine and potentially more
depending on what actually happened. You are individually liable for that, whereas no other country really has that, and you can even get insurance on oops. We leak customer data and then these news companies come and pick it up and it almost serves as an advertisement for them. Hey did you hear about this company that leaked their data? Now you know about them, you know, positive reinforcement, You're going to go and actually potentially buy that brand now, So
they're being rewarded. I would really like to see the news outlets actually lean into helping us as well. I think public opinion can go a long way, And I don't know if you have any particular thoughts there.
Yeah, I think public opinion is strong a lot kind of In Australia, we had some really big lakes up this ahead breach Mediicare medi Banks has been a few big ones. The Optus one was probably the big Optics is like our like our Verizon or one of our big carriers over here, and they had a massive leak through an API that was open, right. It was an
open API, so dart exfiltration. It wasn't an attack, They didn't break take down any systems, but that that did make the government act and put the owners back on other corporations to make sure that they at least were compliant in some way. Right, So it's mandates that have been released because of that, like checks that you have to kind of follow. So Yeah, the media has made things positive, but again such a gray area that I
think country to country it's just so different. I think the latest one that's still a bit weird is a Ticketmaster one that's still kind of out there right that was Snowflake was the platform there. Snowflake hasn't really been hit too hard by that and ticket Master hasn't released any details as to what got leaked yet, so we're still kind of waiting on that one.
But that's an a big one of late.
Yeah, I mean I saw a really good followed from there.
Is like a good reminder of Snowflake and their customers all have certifications proving that they're secure, and you know that didn't really get them anything. So it's a good reminder that these things are not have no indication that you are secure. Like if your internal practices aren't actually being monitored, your your policies aren't actually being followed, they don't really amount the much.
Yeah, And to that point, the way that we've been talking about it at VAN and I've we spent we spent the last year basically scaring the shad out of people at round tables talking to them about how they should be ready for this sort of stuff and then have you got good data protection as part of that strategy. But I think the dark web and the point end that's still getting point here right, like being able to use l lms.
Going back to the prompt engineering in LLM.
You know, I've I was writing encryption algorithms in the early days because I wanted to show off a bit of a demo at last year's conference and I was able to encrypt a find an open stree bucket and rip that bucket.
And I did that all with chat GBT.
Now I didn't it told me that I shouldn't use it, and I shouldn't use this because it was bad, But I basically again co werced it, prompt engineered my way into getting it done. They have got more guardrails on today. So if you actually go to chat gbt today and try and say write me a ransomware script that does this, this and that, you've got to kind of really coerce and work with it and massage it to get that script out.
Okay, yeah, you should be able to still do it. But that's said.
On the dark web, there's all these unguarded and sort of open versions of these chat lambs that will write these for you. Right there's the worm GBT is the one that I think is the most popular one out there. It's basically a rip of chat GBT to a certain extent, but it will write you phishing emails, ransomware emails, it'll
write you encryption tooling. So yeah, if those tools are out, it means that not only the good guys or the ransomware companies getting sorry, not the good guys, the guys that are good at ransomware guys, but the gods that
are good at it. They're getting better at it. But then you've got you put that in the hands of just every day a little indie hacker or a little script kitty who's able to now do some really powerful stuff because I can leverage the technology that's scary, and that's where we're going to stay ahead of the game.
I actually have got two relevance. That's they're the first one, and some of these tools out there will actually fully generate you the fishing website that looks exactly the same as Google log in or Azure logan, et cetera, like almost identical like any product out there is for companies who want to fish and the other one is some blockchain developer that had like fifty thousand dollars and some crypto and the keys in their GitHub and they pushed
it up accidentally to make the public the project public. It only took two minutes for the entire wall to be drunk like that is that is ridiculous, Like two minutes for keys to be not only scraped but also utilized, and the walla drain.
So I've got a famous I've got a good story on that, and I've done that myself.
And this was I don't know how long ago.
It's a blog post I put up there, but I did that when I was writing some code. I was actually writing some infrastructure's code using Terraform actually to work with VMware VMC and also do some sort of post Terraform sort of project work there, and I accidentally uploaded my Amazon Aws secret key and as part of the code. No sooner, I remember just all of a sudden not being able to log in and upload the terrorform script, and I was like, Oh, that's weird, what's going on?
And then realize that when I logged into my Aws console in every region, about thirty or forty instances had been spun up and they were basically mining whatever they were mining on there. But what they all said is they deleted everything. So I got absolutely done and that was no joke. Within you know again maybe five and it's me upload and that key, and then what I learned was that, yeah, they've got they're just looking for that.
They're looking for dumb idiots like me who upload their keys if they scrape hub and once they found them, bam they get in there.
It's really interesting how quickly that happens.
Well, they have to be fast, not because you'll find out about it, but because there's other people also scraping those same keys against other yeah, like other bad guys basically other villains, uh, in order to beat them, Like there is a race there.
And where why do people like this? Bad people in the world? Is bad people in the world, But I get it.
People want to people want to make money, and there's if people want to make money in any way they can, they'll find methods to make money any way they can, right, And that's why gransomware is big. That's where that's why we've got all these prevalence of you know, attacks against companies, expltration I've got your dat. I'm going to sell it because people know that people pay money for this stuff.
So it's just the world that we live in, right, I.
Mean, I think there's a huge philosophical conundrum there, and I don't think we're going to have the time together.
No, no, no, no.
I know it's interesting, but yeah, at the end of the day, like those l ms, they're absolutely used for good and evil, but I do believe that overall they're amazing.
So yeah, yeah, I think it's like that comes down to our own level of trust because you see it with people too. You know, people form their little tight circles and you don't trust anyone outside of your circle until like a bigger threat comes in and you need to band together, like okay, well now we trust those
guys because there's new people that we don't trust. And I think that this tools like this fall into that same pattern because it is such a conversational interface with it that we give it like human characteristics.
Yeah, and I would encourage anyone who uses it to not treat it like a robot or not treat it like a chap, but I would encourage people to actually interact with it in a natural way, because I think that's the best way.
That's that's how I.
Think they're built and how they're designed to respond better. So, you know, if I was giving a one O one on prompt engineering, it would be have a conversation with it first and feel and feel good about talking with it in a natural way, and then get into the heavy stuff.
Just start every conversation with the phrase, I, for one, welcome our AI overlords.
You'll be fine, Yeah, exactly, it's going to be. Yeah, that part of it is going to be interesting. And look, there's no doubt that we're heading towards something like you know, general artificial intelligence, and I think that's exciting and scary and all of the above. But it seems like where else if you said five years ago, how far away is that.
We would have gone?
You know, ten years, fifteen, twenty years now. Acceleration is significant, and you know, I can see it happening this decade.
You know, I think that's kind of the trajectory that we're in.
Yeah, it's cool, cool, scary and well at the same time, isn't.
It all right? What else were we going to talk about?
Oh?
We had the talking about like the evolution of dev ops. We were talking about that before we started recording, and like, how you know, there's there's people like us who are just have like one or two things that were good at and then there's other people who just have this massive foundation and can just like context switch and seem to be great at everything.
Yeah.
I think my first boss that I had Alex Alex Sliffer's great guy, Australian German, so it was very meticulous. He was very German and in a lot of the ways that he did a lot of things very efficient.
It was very stereotypically German in that sense.
I think instilled in me a lot of my better traits in it because he was effectively my first mentor and teacher when I was a kid. But he was everything he would when we were at an internet service company. We're a hosting company. We did Windows Linux, we ran Cisco networking gear, and he just knew every He was like an expert in everything, like everything, and I was I was listening awe, like, you know, just for him to be the ev because he had to be right because he We're a small company.
He had to know how to do the BGP, the DNS, a lot of balancing.
But then he'd basically code our website, code our billing system, and then he would do the setup of the active directory and you know, then the Linux setup.
And everything like just literally everything.
You know, he'd set up the sand So whatever he did, whatever was an it he had, was that guy. So he was like one of the original I guess devopy guys because he had to be. And so you do find guys like that. They're kind of unicorny, though I don't think they exist honestly, very much in terms of people that can do literally everything at a super high level.
And sometimes you find other guys that are really good at coding plus infrastructure or infrastructure plus storage or networking plus something.
But yeah, it's very hard to be good.
At everything at a super ocra high level in this industry, isn't it?
Yeah for sure, And and it gets increasingly more difficult every year as we add new technologies. Like a couple, you know, a couple of years from now, tops, a lot of the companies we work for will have their own lllms and it'll be an integral part of their business and somebody's gonna have to you know, build, operate, maintain, and scale that and that'll be another skill set that we have to figure out how to do.
Yeah, And I think I think DevOps, you know, there was it had a bad STAPs to joke with Michael actually about this.
We both used to joke about it.
And that's quite ironic that he's effectively built an accelerated brand off his ninety days, right because we used to just give a ship.
We used to go, we're going to grow a beard, and I think we can think we're mart.
We used to joke about that at actual presentations that we used to do. We used to go, like, so DevOps that we're doing like a terrorform code and building out our VMC infrastructure and doing this and that. But I think absolutely there's there's a there's a great play for it now where the platform engineering slash DevOps has just come in because you have to you have to be that person.
You have to know how to consume.
Read read a bird terraform, read infrastructure as code, read some Python to do this, this or that. It's critical to everything that we do. So I think it's just entering that mainstream and it's not a case of you know, this guy's a devop guy. This guy's a platform guy or whatever. I think everyone has had to be dragged up in a way to be able to be a DevOps or platform engineer of sorts today. If you're not a pure code like still got the code, there will
always be there. But in terms of infrastructure understanding what it is, I think there should always be. But I personally think you should always know about the system that you're deploying. You should always know and not just treat it like a black box. That's kind of been my methodology in my career is to not just accept something
that's given to you. And I think there's a dangerous sort of the way that we do things today where we just accept that a machine is going to get built, or we accept that a bunch of AWS services are going to be deployed and they do this, this and that. But just taking the time to go back and actually understand the mechanics of it, it's still very important. I think that that kind of differentiates a few people as well.
I mean, I think you really hit on an important thing there, because I think this is what really led into the whole DevOps mindset movement is you have to understand one hundred like as much as possible in the ecosystem that you're building, in what you're writing code and employing, you're a single team is cross functional and responsible for everything, and that at the same time, there is more and more that we have to learn about and know, and maybe really the only way in which we can be
successful is if we are adding ms as our team members to actually expand out our knowledge and expertise there, because there is just more and more things that I just don't think anyone can reasonably become an expert in one hundred percent of everything.
They should be no, and if anyone, if anyone kind of says they are an expert, probably don't believe them.
Right, right, Dunn Kruger there, right, Yeah.
That's been one of the things I've struggled with for people because I feel I have a lot of empathy for people who are just starting a career in infrastructure or platform engineering or DevOps, whatever you want to call it, because there is so much to take in, Like I know a lot of it, but only because I've been doing this for three decades, Like if I were to start today. It's one of the things I try to
do is for people who are just starting. It's like, how do I condense three decades of things into a digest stable format for them so that they can build on top of what I learned instead of having to relearn what I've learned.
Yeah, and I think I mean someone actually, a guy called Suresh who is the CEO for platform nine. Who do I'm not sure if you guys know platform nine that they do. They started off in KVN based hosting with VMware and now now they basically do Kubernetes based platforming on AWS. But he told me way back when that the developers of today are just in a different class to what you think the developers of five years ago were and what they were, because what they're learning
at UNI is something completely different. Like I learned Turbo Pascal at UNI, right, or maybe some people older than me would have learned cold Cobowl or whatever it might have been. Right, And then now then it went to Java, and then it went to this, and now they're learning visuals, dot net and now they're learning Python and so it's just accelerating. But the big thing that stacks to me wasn't so much about the code. It was more about these new developers that are coming out or the new
breed of it. Guys, if you ask them what a storage platform is, they'll say, oh, Mungo dB or Cassandra or you know, one of those database.
Is not not an actual storage system, not a disc.
So they're abstracted up a couple of layers, right, And they might even say, you know, the next ones might say, oh, well, what's my storage roll. It's my Mongo dB database sitting on Mango dB, you know what I mean. So there's this level of continual growth and abstraction that's happening with these guys, so they don't need to know what's underneath. So those guys don't care about the storage. I don't care about the spinning disc or the envmme drives. They
are running it all the servers. All they care about is the interface into that Mango dB server. And that's that's quite interesting.
Yeah, that's a good point.
And I think the level of abstraction is just we're abstracting the abstractions now absolutely.
I mean, I actually think it even goes further than that.
It just as you know where we're currently at. I see in university students for their senior thesis basically building something that runs on AWS as as the project, which is just so much far removed from like even what I was doing initially, Like I had to I had to teach the first company that I actually multiple companies that I had worked at, like what GIF was, and that like what version control was, Like I was teaching them that and I didn't learn that the university that was just like.
What I was doing. They didn't have that.
But now that like that's that's just a class that they have to go through.
Yeah, absolutely, I mean that's and that's the difference is because that is a skill that's required when they.
Hit the workforce today because everyone kind of needs to know it.
So yeah, it's things move on, and I'm sure to tie back, you know, tie the not in it to to the LM talk. Absolutely, you're going to get prompt engineering courses within university, if not already happening like pretty soon, right, Yeah.
I would expect still be interesting to see the curriculum on those.
Yeah, don't swear, they don't don't don't be nasty to It was like one on one.
Listing cats, right, because if you're in college now, you likely will live long enough to see this thing have a real world interface and it's going to come back with those things you said, absolutely awesome.
Should we do some picks?
Yeah, I'm excited about this, all right.
Cool, Warren, can you kicks off?
Yeah? For sure.
So I've been holding this one, this particular pick, for a few weeks now because I knew the yeah, the episode on.
Prompt engineering was coming.
So if you're unfamiliar with the newsletter TLDR, there's a TLDR security written by Clinton Gibbler and Remy Macarth that you were absolutely fantastic, and they put together fifty plus research papers. They found them, they distilled them down, like really went deep what they mean. And there's a get out repository on prompt injection defenses. I highly recommend if you're anywhere near the area of the industry to.
Take a look.
There is quite good summary of like what the problems are and how they're being solved or even start to be tackled. And there's some really clever things in there, like oh, you know, well, someone can tell our LM to do something bad, and so like how do you
get around that. One of them I find really interesting was oh well run it, run it like back through itself, like get the result and then like parse it and then like pass it back through the same LLM or like another LLM and see if that one was validated.
And so using a graph of lms, get them to sort of validate the response from a different one and whether or not that's it's because like even if you can get over the system prompt of the first one, if the output of that has to still match the second one that you're able to have better security over, you can subvert attacks that way.
And the list goes on and on. There's I think there's over fifty in there.
Cool, all right, Anthony, would you bring for a pick turn there?
They're both related to AI and all that kind of stuff. So the first one is Life three point zero by Max teg Mark. So I listened to this maybe year and a half maybe actually it was before chat GBT came up. But it's effectively about becoming age of AI and whatnot. But the first prologue is effectively it's a story and it's worth reading for it alone.
It's it's called the.
A Mega Project, and it's effectively you know about an AI that's been developed by company kept secret perhaps in the Internet gets sentiment, gets it starts to build in itself and get better and then effectively controls the world. But it's not a not a bad story, it's a great story. But it's actually coming to life right now like the actual it's a prophetic look at what's happening with chat, GBT and the LLM today. So that's that's
a big one that I would recommend. And the other one, which is actually really cool, is more about that second part that we're talking about, is called an Overseen by James Lovelock. James Lovelock worked he only died a few years ago. He was about one hundred and something, but he's basically the guy that invented Michael Waves that sort of stuff like that. Dude he worked in Area fifty
one and whatnot. But this book is all about the coming age of HyperIntelligence and effectively goes AI to cyborgs and what happens when cyborgs basically start to get as smart as humans and effectively the end of the story is do cyborgs need humans or do humans need cyborgs? And then it kind of revolves around there. So that's a really cool read. I think it's a short read as well. You can get through it in about eight hours.
But that's one of my favorite autime books and overseen by James Lovelock.
Right on line.
Cool.
So this week I'm here fishing for picks because I was out. We went out this last weekend. I made a long weekend out of it. We were out in the mountains, new cell phone service, no not even any radio stations. And I came back and I can't log into my computer, my MacBook. I have no idea what the password is. For years, it's been muscle memory, and every once in a while I'll change it by just changing the keystrokes a bit.
But it's not.
Anything.
It's not a word, it's not a phrase, it's not anything written down. And it's also gone from my brain.
Like whatever brain.
Cells were responsible for holding that, they're freaking dead. So I'm here fishing for picks. If anyone has any ideas, recommendations, or strategies for storing those root credentials, I could use a new one. So if you got any ideas, hit me up on X or leave comments in the show.
How everyone do that?
I got a good one for you, kind of on what you're using.
I've got a ubkey that I plug in that automatically decrypts my hard drive and it's got its own pin, but it's very short, and so I just plug that in and leave it there until my keychain and pull it out, so it doesn't matter, Like I don't ever need to remember what password is there must I lost. I lost the key chain as well, and if I forget, I'll, you know, update it right now and validate. Like every year, I have a reminder said do you still remember the password?
And tape it out? And if I don't, I'll set a new one.
Isn't that funny about passwords like that? Like you said it?
Like I think if you ask anyone, most people, unless they're really.
Security conscious and there's some real.
Security nuts out there, they will they will basically have that same password that's muscle memory, but slightly older. Maybe it's two numbers that they change every now and again. Maybe, So I think majority of people do that because that's just what we know.
It's easy. Don't feel bad about that.
But yeah, maybe maybe you know the fact that your muscle memory is gone.
I don't know, get that check.
Maybe great, because that's the that's the big concerns like, Okay, I forgot my password? What else have I forgotten that I'm going to need something?
This girl?
Well, I mean the good news is well that unless you actually damage those on the the pathways are still there.
You need to re trigger them.
But the connections in the neural net that's processing over those neurons does degree, so you you can retrain it by some going through whatever process you would normally go through and try to get your key, your hands on the right spot on the and may come back to you.
Just go to Alon. He's got like a little plugging the head. Now.
I'm sure you guys have seen like the similar tweets and stuff similar to where they'll tweets on at C I a say, hey, I forgot this piece of information. Can you guys send me your backup copy of it? Tweet at CIA? Can you send me my passwords?
Someone who will do it? Cool?
All right, well, Anthony, thank you so much for being on the show.
Just really fun. Actually, I really enjoyed the conversation. Love it, love it. It's been good.
We've gone on different places and Warren, I reckon you talked about ads. I reckon, you're gonna get ads about black pudding now, so I just.
Get ready to right for sure there is so I can say that Amazon seems to have improved their ad their ad platform a lot, because I used to have this running joke with a really good friend of mine where we would find less than professional things for sale on ad on Amazon and send each other links, you know, masking them or whatever, just to get the other person
to click on them. And then every time they logged into Amazon in front of their wife or something, there would be these really odd ads and it was just great fun. But then I had another friend who I think it's been about two years ago now, he was having open heart surgery and we were really concerned about him, and so I wanted to make sure that he was in a good mood before he went to the hospital
that morning. So I went on Amazon and I ordered a blow up sex doll, and then I took it around to all my friends, and you know, we all wrote stupid stuff on it, and then I filled it with helium and tied it to his mailbox so that whenever he his wife took him to the hospital that morning, there was a blow up sex doll flying above his house and and so I was really worried about the next time I logged into Amazon after that, like thinking, Okay, what are they going to show me?
I didn't get anything, like not even like dolls.
Yeah, here's other sex dolls it might be interested in, or here's things to go with this. None of that stuff. I didn't get anything. So they appear have done done some work there.
Yeah, well maybe is this one top of sex doll on Amazon? And you boid it and that was it.
It was the only one they sell.
They sell, that's it you got. You got the premium model.
Right, that's our top of the line model.
Yeah anyway, yeah, awesome, all.
Right, cool, Thanks everyone, Anthony, thank you, Warren, thank you, and to all nurse thank you for listening and supporting the show.
And we'll see y'all next week.