EPISODE TRANSCRIPT

I used to believe that all the good ideas came from overseas in tech. I worked with a lot of companies that were either American companies or were distributors- Mm … of the American companies, and I used to believe that the ideas and the execution had to come from Silicon Valley or Boston or maybe even Europe in some way. I had this belief that you, it was not possible to lead from Australia. 

I’m changing that, my view on that. 

Okay. It feels like it’s been a few weeks, but we are, are of course back. The show is Things Reasons. We are the inside track to the Australian IT industry. I am, of course, Naran McClung. Your name again? Jono Staff. Excellent. 

We need to get a feel for what on earth you’ve been up to lately. What? What have you been doing? Mate, uh, a little bit… Many things. Mm. It, it doesn’t stop. But, um, but I tell you what I am focused on: the weather is getting better- Sure is 

and I’m, I’m dialing in. Mm. Uh, I’m dialing in PBs at the gym. I’m dialing in getting outdoors. You’re looking good, can I tell you? Oh, thanks, Naran. Not to say you ever looked bad before, but you know, I, I look at you a lot- Yeah … and you- you’re looking vibrant and fresh and fit and healthy. Yeah, now I’m red. 

Okay. Now I’m going red. Well, look. Perfect. Good. Perfect. It is warming up. I do a lot of cycling, as you know. I’ve been freezing my absolute bollocks off out there at 4:30 in the morning. I’m not so much anymore. Uh, it’s good. It’s better. A lot of gravel riding in the forest as well out at Wiseman’s Ferry on Sunday, which was amazing. 

Uh, so I’m loving all of that, and, uh, it’s great to be less freezing now. Very happy with that. All right, now our guest. So, uh, well, I’m just gonna get straight into the bio. All right? Our guest today is a gentleman by the name of David Keane. He’s founder and CEO of Southern Cross AI. Now, David’s taken a company public on the ASX twice, which in Australian tech makes him a repeat offender. 

The first was Bigtincan, which started in Sydney in 2011, listed in 2017, run largely out of Boston for the 14 years that followed, and sold to private eq- equity, sorry, in 2025. The second, and I was at the, uh, the launch of this recently, it was very exciting, is Southern Cross AI, which listed in August last week, uh, on a bet that most of the industry is looking in the wrong direction. 

We’re certainly gonna explore that. He spent 14 years selling software as a service and then built his own company without buying a single SaaS product. Uh, he’s buying ASIC chips. That’s application-specific integrated circuits, if memory serves me, uh, while the rest of the planet queues for these GPUs, these NVIDIA GPUs. 

He’s published a rebuttal to the idea of putting data centers in orbit. Certain lunatics suggested that. Uh, and he’s been named as Australia’s first sovereign language model after a bird that is notorious for dive-bombing people like me, cyclists, in every suburb of the country, and that’s now, by the way. 

That’s happening right now. It’s prime dive-bombing time. All right, now here’s the serious part. What he’s trying to do is to stop Australia outsourcing its own intelligence. Thousands of Australian businesses are sending their most sensitive data offshore just to use AI. 

And in his view, a data center sitting in Sydney does not fix that if the company operating it answers to a foreign legal system. Storage location is not sovereignty. In his words, control is sovereignty. Southern Cross AI runs Australia’s first sovereign AI inferencing node built on SambaNova, and we’ll get stuck into that, uh, rather than GPUs. 

And it is pointed deliberately at inference rather than training. He wants the industry to stop counting GPUs and to start counting tokens per watt. His line for the whole thing is that Australia needs AI that is servicing Australians, not Australians servicing AI. Hmm. David Keane, welcome to Things Reasons. 

Great to be here. Thank you so much for having me. I, I don’t know about you, Jono, but I haven’t been to an ASX launch before. That was awesome the other day. Yeah, it was great, uh, ringing the bell. I… That was my first one as well. Yeah. Um, and fantastic and really exciting, a, an Australian, Australian business launching on the ASX. 

Mm-hmm. Doing something, I think, really important And I know, David, we’re, we’ve known each other for a while now, and I know you’re on an absolute mission, uh, to make a difference. Absolutely. And- We think it’s important, Jono. You know, no doubt. We… This AI stuff isn’t just some small new technology that’s just appeared and people will use for a few years. 

It’s fundamentally gonna change, I believe, and I think many people share the view- it’s gonna change the lives of all Australians. But before we talk about AI, we should talk more about winter, because after spending, um, 14 winters in Boston- I can tell you all about winter. Please do. Um, yeah, it’s a very interesting experience. 

I found Boston amazing. What a great city. And for listeners who’ve been to Boston, I’m sure you will agree with me that you get a combination of amazing things in that city. Right. You’ve got the education, you’ve got MIT and Harvard- Mm … I think producing incredible people and ideas. But you’ve also got the whole New England weather and the true four seasons. 

I really enjoy that. Yeah. But I will tell you, this is not a bad winter in Sydney. I think I’ll take more of these. You’d take- Yeah … take more of these ones. I don’t think Naran could get out cycling at 4:00 AM in- No … Boston in the middle of winter, right? Not in the snow. Not in the snow. Yeah, not in the snow. 

Indeed. Okay, cool. Shout out to Boston. Indeed. There we go. Dave, um, I think our in, our listeners are gonna be really interested in this one, but we, we like to open with a little question. Could you tell us something that, uh, we should know about you that maybe you haven’t spoken about in other interviews? 

Wow, something super secret, right? Yeah. Special, special news. Tell us, tell us the secret. Special news. Well, for those of you who may be interested, I was actually born in Canberra. Hmm. Very few people were born in Canberra. W- well, well, of course, many people are born in Canberra, of course, but I think Canberra’s a very interesting place, and I grew up there. 

I thought it was a fantastic place to, to grow up and experience the things you can do in that environment. Hmm. But it also gives you a bit of an understanding of why it’s important that we bring the benefits of technology to all, all Australians, and I think all people. Too often we think about these, these enclaves where you’ve got- People that have got a lot of, um, the, the, the baseline, a lot of the un- the, a lot of the support, right, around them. 

I think government cities like Canberra, people don’t realize the support they’re getting from the population. Mm. And it’s something that I found very helpful, was going out of Canberra and coming to other parts of the world, Sydney first, and yes, of course, to the US. You realize that everyone’s life is not in the bubble of a place like Canberra. 

People have real lives and real problems and real challenges, and we need to address them. And I think that’s something that I learned from that upbringing. Yep. I li- look, I love Canberra. Um, I went to uni in Canberra. Um- Oh, did you? I did. I went- ANU or University of Canberra? Canberra University. Canberra University. 

Yeah, that’s right. I did a, a year of Newcastle Uni straight out of school. Failed miserably. My wife will say it’s ’cause she wasn’t living with me at the time. She’s almost certainly right. Yes. Moved in with her and just got on the straight and narrow, and then figured out how to study and do things. But I really enjoyed Canberra. 

Yeah. I did. I loved it. It is. Um, it’s changed a lot, hasn’t it? Since I- It’s changed a lot. It’s grown dramatically. Totally, yeah. D- grown dramatically. But I’m sure you found it, too. You are in a bubble in a place like that. Yeah, you are. Yeah. You know, you’re, you’re separated from the real world, the challenges that people face- 100% 

to build businesses or to even build their family. Yeah. I think that it’s nice to understand that. Mm-hmm. And we, and we want that. We want that benefit of that for, for everyone. Yeah. But I think you learn very quickly that that’s not the real world. Well, true. And so, like for me, like from Canberra, I went straight to London. 

Oh, wow. Yeah. There’s a change. So straight to London, straight in the, in the big smoke, working for Dimension Data and then RBS and then News International, and then all throughout Europe. So I got a taste for the real world- Well done … after that and came back. But, like, now that I visit Canberra, it’s one of those places where I think I appreciate it more- Yeah 

now, having, having sort of been and left and come back, etcetera. But anyway, yeah, good place. Yeah, definitely. We’ve, uh, run into each other in Canberra before. We have. So there’s a, there’s a- We’ve met down there … there’s a bit going on down there. Well, that’s, um… Yeah, I think that, that probably, probably gives us a bit of an insight into the, the way you think and, and sort of what’s driving, uh, your passion- in developing Southern Cross and, uh, and also perhaps why you’ve taken the route that you’ve taken- Mm 

in terms of, uh, listing the business too- Yeah … so that all Australians can participate. Um, because I don’t think y- you really have to do that, right? It’s not something every founder looks to as a, as the, the goal. Like it’s, it, the… There’s other headaches that come with listing, right? Sure is, Jono, and I think that, um, it’s true. 

If you think about businesses like Southern Cross AI, just for the listeners out there, we call ourselves SCX.ai. It’s amazing how many people, um, like this concept of the s- the short form version of it. But, um, the reason we decided to list SCX, and just to share, there were so many people interested in AI infrastructure. 

There’s plenty of money. One of the challenges though is some of that money comes with consequences and, and connections and ties, and it’s, much of the money does come from outside Australia. Mm. So the alternative for a business like us was to bring money in from outside Australia. Mm. But we really believed if we’re gonna walk the walk here, right, and really build something that is important for all Australians, we need to give Australians an opportunity to be part of it. 

So listing on the ASX means everybody who feels it’s worth investing in, they make their own, of course, investing decisions, can see what’s going on in SCX. They can see behind the curtain. They can see what’s actually going on, and if they’re interested in that, they can play as well. Yeah, so you rang the bell on Friday. 

Uh, I think it was the 21st of this month, and, um, I think there was a pretty big lead-up to that, right? The, there’s a lot of preparation that goes into, into getting to that moment. And, uh, but I think then you wake up the day after. What’s, what’s different? What, what’s different? What does your week look like after sort of the big- It, it sort of culminates in ringing the bell and, and now, now what? 

Okay. Does, does the work really start or is… Yeah, what? Tell us all about that. Well, just, just keep it all on the down low here. Don’t tell anyone. Yeah. But there’s no change. Okay. Okay? Because, you know, what you’re doing with those, those events, y- you’re trying to make the most of those events. And, and really I, I love doing that because it gives the people involved in getting there a feeling of accomplishment. 

They’ve done something. But the reality is nothing changes. In fact, the company was already on the pr- program of working with customers and building our technology and building out our facilities and all those things. Nothing actually changes. What you do, of course, is you’ve now got a whole bunch of people along the journey with you, a whole bunch of shareholders, and the job of the company, number one job, is to get the best outcome for those shareholders, of course. 

And so I think that’s always been in the mind, though, of the, the executive team at this company, but nothing really changes, Jono. I think it’s now about execution. 

  

Sort of execution at speed too, because this industry, this game that, uh, that you’re in is moving so incredibly quickly, and there’s a lot of attention. 

I think, uh, just before we kicked off the pod- Mm … you were talking about today’s article in the Sydney Morning Herald. Mm-hmm. Mm-hmm. Uh, what was the headline there? So as we’re doing this pod, the today’s headline in the Sydney Morning Herald is, is, uh, “AI is the most important sovereignty issue for all Australians since World War II.” 

Yeah, that’s absolutely huge. That’s incredible. Yeah. but It’s probably true. Yeah. Because if you think about what the impact of AI could be. Now, look, we can, we can sit here and drink the Kool-Aid and talk about it. Mm. And maybe it doesn’t end up that way. Mm. But in case it does- Mm 

we need to be thinking about these things and making sure we’re ready for it. Look, I, I’m gonna go back to the, the point you made before, that this technology’s here and it’s here to stay. I wholeheartedly agree. Now, look, you know, we work in tech, Jono. You and I both work for MSPs. I think WinDC’s obviously got a different slant on this as well, which we can get into. 

If I look at our own business at Macquarie Technology Group- I haven’t seen a technology disrupt what we’re doing more than this. Now, admittedly, I’ve been with the business for eight years. I’m pretty sure that David would say that this has been the most disruptive influence on the company, really, since we got into the data center game- which is now linked to it anyway. Mm-hmm. Right? Uh, when I look at the way our business consumes AI, I mean, it started out as basic, you know, dicking around with, you know, chatbots. Mm-hmm. We’ve now gone to what we call sort of, uh, GenAI 2.0 in our business, which is we’ve got headless agents, right? We’ve got scheduled tasks. 

Um, we are- developing, right? So we’ve got a stream of developers working with this stuff. We’ve got, um, Chinese open weight models hosted on H100 arrays- Mm-hmm … that we’re using for this stuff. And we’re being asked questions on, like, how much should we be spending on tokens? Now, fortunately, the business has got decent sums of cash, so we get to play and we do things. 

But I’m fully expecting at some point that every P&L owner will be asked to sit within a budget of tokens and, you know, will we move away from frontier models entirely, right? We’re getting such good reasoning now out of our open weight models hosted ourselves. Um, I feel like there’s an evolution just in our own use case, but to your point, there’s no way we’re going back to not using this tech. 

Mm-hmm. It’s too late for that. There’s no way. So- Yeah, the genie’s out of the bottle. Yeah. Exactly. It is, isn’t it, Jono? I’ll tell you an interesting story, though- Mm … as well. So I was at an investor lunch this past week, and j- jokingly, one of the analysts was saying they’re gonna create this new metric. 

Mm-hmm. So we all know about financial metrics, and one of the key ones is EBITDA or E-B-I- Yes, that’s right … T-D-A, right? Yes. So the new one is EBITDA, but with two Ts. Okay. So it’s earnings before interest, tax, tokens- … depreciation, amortization. So- Let’s, uh- I think that’s got a- Let’s put that up there. Ready? 

Let’s put that up there. Ding. Let’s put that up there- Let’s put that up there … as an acronym. There we go. There we go. That’s it. Yeah. Good. So, but, but of course, everything you say is correct. We, we need to understand this. Mm. We need to embrace it, and we need to be ready for the changes, and those changes at speed. 

Mm. So again, as we’re recording this podcast, since Friday, so it’s now a Monday morning, recording this pod. Since Friday, we’ve had significant changes in our industry over the weekend. We had new Chinese open weights models come out- Yep … from GLM. Yep. GLM 5.3 Flash and GLM 5.3. At SCX, we have both of those models running right now- Yes 

on our endpoint. So if you’re out there, you wanna try GLM, give it a try at SCX.ai. I’d love to have you try it. But, you know, those models are frontier-level models- Mm … at a fraction of the price. Yes. And I think more importantly, it’s the transparency. Yes. So of course, price is important- and consistency’s important, but- You’ve now got transparency. 

Open weights means you can look at what’s actually going on. You can look under the covers and see what’s inside it. Yes. Versus the proprietary ones, it’s totally opaque. You have no idea what’s happening. Right. And so when you’re looking at… When, when you’re seeing what’s inside it, just for maybe some of our listen- listeners who aren’t as au fait with, with how these models work, but I feel like seeing, seeing what’s inside the model or that transparency means you can actually start to understand why you’re getting the answer that you’re getting. 

Mm-hmm. Is that… That- that’s what’s going on there, and that’s- Yeah … ca- can you talk us a little bit through- Yeah … what that transparency means? Yeah, let’s think about that. So what is a large language model? Mm. Right? What is it? I mean, one of the best ways to think about it is, it is a deep neural network. 

Yes. And it’s a mathematical exercise. Mm. And so what happens in one of these large language models is in any de- deep neural network or convolutional neural network, you’ve got these layers of nodes inside them, and you’ve got decisions that are made at each of those nodes based on mathematics. So you’re either passing a, um, connection to the next one, or you’re stopping a connection, right? 

That’s right. And you’re looking to go through this flow to get to a particular answer. Mm-hmm. And there’s many, many, many layers of these neurons that are passing information back and forward. Yep. And it’s simply based on math. Probability, yeah. Exactly. Yeah. And so the benefit of open means that you can see the math. 

Mm. Now, of course, trying to understand the math- Mm … at that level is extremely complex. But the point is, you can do it. Yes. And I think the best example of that, and I think, Jono, you talked about  this, uh, point about transparency. Some of the listeners will remember the DeepSeek, uh, models- Yes. .. that came out- Yep 

about a year, year and a half ago. And there was a story about the Tiananmen Square. Do you me- mem- remember that? Yes. Do. Where if you went to the DeepSeek model, you asked it about Tiananmen Square, it would give you a particular answer. Mm-hmm. What’s interesting and im- I think important is that AI researchers were able to find in the model, in the weights, which is the term we use to describe these mathematical formulas, they were able to find in the weights where it had been fine-tuned to answer Tiananmen Square questions in a particular way So because they are able to find it, it shows that it is truly simply a mathematical calculation. 

You can find something in the weights. Yep. And you could then tune the weights. Now, of course, the challenge is normal humans trying to understand the weights of a massive, you know, hundreds of billions of parameters model is very difficult. But it can be done, whereas the proprietary models, you have zero idea what’s in the weights. 

You can’t tell. Mm, yep. And so this comes back to sort of a, a horses for courses type conversation around what are you trying to achieve with that model, and what’s the most appropriate model, and, um, and where, where, uh, where is that transparency, uh, really important. Mm. You can start to think of certain applications where- Yeah 

that might be really, really important, and others where maybe not so much. Right. So cybersecurity, I’m sure that’s what you’re talking about. Right, yeah. And so one of the things about the GLM 5.3 model that just came out is when you read the paper they put together with the announcement, they talk about its leading capabilities in cyber, right? 

Being able to be offensive and defensive cyber. What’s interesting is they say that is an emergent capability, and this concept that we’re getting capabilities from these models that are emergent. In other words, it wasn’t trained to be good at cyber, it just came about. And by the way, that same emergence is what happened with coding. 

So we’re all… I’m sure many people out there, I’m sure you’re out there doing some coding. A vi- vibe coding is our favorite word, but that coding capability was also never trained into an LLM. It was also emergent. Hmm. Simply a combination of the data going in and the training algorithms- Yes 

gave us an opportunity for these emergent capabilities. Interesting. So with cyber being emergent, what does that mean for us? Because of course, no human, I would say as of today, no human is good enough to be a cyber defense on their own. Yep. It can’t be done. 

  

So David, um, I’m wondering, uh, how much of our audience truly understands what inferencing really is. 

Hmm. Right? Now, we all put our prompts in, right? And we get answers back. I wonder how much of our audience knows really what inferencing is and how it breaks down. Now, um, the SambaNova stack, and I- I’d love to, to get a take from you on, on why you selected that particularly. Mm-hmm. We’ll talk a little bit about that. 

Um, allows you to be, um, a full stack inferencing provider- Mm-hmm … giving you some of these capabilities- Mm-hmm … um, going well beyond, let’s say, what a Neocloud perhaps would, would, where the, where the line would stop on service and pass off to somebody else. So talk to us about Samba and being a full stack inferencing provider and why that was a, a choice for, um, SCX.ai. 

It’s always interesting when we get these new industries- Hmm … where we get the new kind of terminology- Hmm … that, that we people use. Yeah. So inferencing, why choose that word? Right. They could have just said using. I mean, inferencing is the activity you perform when you are using an AI model. That’s right. 

When you type something into a prompt, and you get back some answers, writing an email- Hmm … analyzing a document- Hmm … you know, doing coding. Yes. That is all inferencing tasks. Doing your homework if- Doing your homework … if you’re a student. People wouldn’t do that, would they, John? No, they wouldn’t do that at all. 

  

I’ve seen some people try and build detectors- Yeah … to detect it. It’s very hard to detect. God, it’s a can of worms, isn’t it? Very hard to detect. Well, the watermarking piece is coming though, isn’t it? You know? It is. Yeah, right? Fascinating. We should talk about that. We will talk about it. We should. 

Yes. It’s actually a very important thing for Australian business to understand- Absolutely … the watermarking story, right? Yeah, absolutely. So inferencing means use it. And I, I like to joke if people have seen the old Oprah Winfrey Show, you know, the answer is you’re an inferencer, and you’re an inferencer, and you’re an inferencer. 

Yeah. So you’re all inferencers. Everyone out there watching this pod, you’re all inferencers. So that’s the word they use. Yes. Inferencing is measured in tokens. Yes. Another word the industries loves to use, right? Mm-hmm. So tokens are simply a measurement or, or, or a way of bringing down to a lowest possible level a set amount of stuff- Mm-hmm 

that is then assembled later on. These tokens are reassembled later on- That’s right … that measure the, the, the amount of inferencing work that is done. That’s right. So at SCX, we had a belief that there was a unique opportunity to bring high speed, cost efficient and fully sovereign inferencing to Australia- Yes 

so that we could make the tokens right here in Australia. Yep. We could serve the tokens in a way that was going to meet the needs of Australians- Yep … but do it with a focus on efficiency and sovereignty. Yep. You talked about SambaNova. Yes. So long story here, but- That concept of building an Australian AI inferencing system was core to SCX’s foundation. 

Yes. And we looked at how are we going to do that in a way that was affordable, both economically and environmentally. Mm. And we looked at all the different AI chips, and there’s lots of AI chips. Of course, NVIDIA, everyone knows NVIDIA. Yep. Um, they’ve been very successful. Those GPUs, graphics processing units, are very good at processing, uh, information in a particular way. 

And it’s this interesting story because people often now associate a GPU with AI. Mm. But if you step back a little bit, maybe it’s even 10 years now- Mm … why do we associate GPUs with AI? No. And, and the interesting story, and it comes down to some researchers at Google, they were looking at building these deep neural networks, and they wanted a way to, to train them faster, ’cause it was taking a lot of time to run them on Intel CPUs- Yeah 

which is how they started. And one of them had a gaming, uh, GPU at home, right? And actually said, “Gee, what if I ran the training loads- Mm … on my gaming GPU?” Mm. And they discovered, this is actually really good- Mm … at running training. And the industry story says that, um, there’s a… If anyone’s been to Silicon Valley in Palo Alto, there was a, a sh- a shop called Fry’s, F-R-Y-S, which was the big, like, electronics store, and there’s a couple of branches of it, but the big one was in Palo Alto. 

And the story is that all the Google researchers kind of realized, and they went and bought out every GPU in Fry’s. Oh, my God. Right. And suddenly there were no GPUs left in Fry’s, and they’d all gone to these, the Google researchers. And the story goes even further, that the manager who was managing that group had been tracking their Google CPU utilization. 

So it basically went to zero, and went to their team and said, “What are you guys doing? Aren’t you working?” And the answer was, no, they put it all on GPUs. And they bought at Fry’s. There you go. But so, so of course GPUs are the core of all this, but GPUs were not made for AI. They’re made for graphics. Mm-hmm. 

That’s why they’re called graphics processing units. Yes. So the industry realized early on that as the market got more and more mature We’d need specialized silicon. So, so a lot of computing, you start off with these general purpose things. Mm. And then over time, when you can get specialized, you do, ’cause there are all kind of benefits to specialization. 

One of my favorite ones, of course, if, if people are out there watching who are Apple users- Mm … if you remember, the Apple, um, laptops used to be Intel CPU based. Yes. Some of us may remember having those. No, I remember, yeah. Remember those? I had one, yeah. You had a Mac? Pro. You had a MacBook with a- Yeah … with a Intel CPU? 

Yeah. And then Apple came along one day and said, “That’s a bit dumb. Why are we using that general purpose CPU?” Yes. “Let’s build specialized silicon.” Yeah. Best move they ever made. Right. Yeah. And they said, “We’re gonna create our own silicon.” But what they were able to say was, at the time, crazy. They said, “We’re gonna make ch- machines that are faster and longer battery life.” 

Mm. The history of computing was, if you wanted it to be faster, it had shorter battery life. Right. That was the history of computing, right? Or if you wanted longer battery life, it had to be slower. Mm-hmm. So Apple said that by using specialized silicon, we could make something that was faster and longer battery life. 

Yep. So exactly the same thing happened in AI. Mm. We went from general purpose g- GPUs, used for training and inferencing- Yes … to whereby we’d use different chips. We now continue to use GPUs for training. Yes. A lot of reasons why, um, some of, some of it’s to do with software. Mm. Some of it’s to do with the way memory is processed. 

It, there’s, there’s reasons. Mm. To specialized silicon designed for inferencing. Yes. And everyone’s doing this. This is a mega trend. GPUs for training is going to ASICs for inferencing. Yep. So of course, we ended up with SambaNova, and I’ll talk about them in a minute, but- it isn’t just us. So there’s a inter- um, Google have a new chip they call- TPU. You’ve heard of that chip? Yep. Um, the folks at Amazon are build- building one. Yes, they are. Um, there’s one called Groq, G-R-O-Q. Mm-hmm. So don’t confuse that with Elon. G-R-O-Q that’s doing that. So mega trend has come. GPUs for training, ASICs for inferencing. Yes. Why SambaNova? So we looked at it, like, what are we gonna be able to do? 

And we wanted to build something that was sustainable for Australia, that wouldn’t be just, uh… It would… Australia is what it is. It’s not the United States. Mm. And if we try and treat Australia like Texas, I think that’s a mistake, right? Mm. Um, we need to build things that are economically viable and things that are environmentally sensible. 

Yep. Australians care about our environment in a way that people maybe in Texas don’t. Yes. Although there’s been a few protests about data centres, I’m sure we’ll get to that. Well, this is very topical right now, isn’t it? It is. Yeah. It is. So SambaNova, we did a survey of the entire ASIC inferencing market, just to say we didn’t wanna do training at SCX. 

We believe there are challenges with training in Australia to do with the, uh, the, the Copyright Act. Right. So there are reasons why training in Australia is challenging. Sure. But we thought inferencing was the way to go. Mm. We looked at all the different chip companies, and there’s some amazing, amazing chip companies, but SambaNova stood out for two reasons. 

One is we’re getting between two and a half and five times better, um, more tokens for every watt of energy we use. Yep. So every watt that goes into it, we’re getting two and a half to five times more output on a comparative basis. Yep. And at the second point was that they’re faster. Mm. So we’re getting between four and nine times better performance. 

Yes. All depends on the model and depends on a lot of things. Yeah. But we’re getting less energy use or more output for every watt of energy. Yes. And we’re getting more speed. Mm-hmm. We thought that was essential.  

  

I love this, right? 

So this is a topic that, um, I’ve gone fairly deep on recently. I do. 

This idea of time to first token. Oh, yeah. All right? Mm-hmm. So for, uh, SCX, um, delivering service, and I think with Australian, uh, businesses going through the motions similar to Macquarie, and that is building headless agents and depending on those- Mm-hmm … the outcomes. Now, we’ve talked about cyber as well. I think the cyber use case at Macquarie is probably our best one, right? 

Mm. It got our mean time to respond and close metrics down to under three minutes. Wow. All right? So- Amazing … and that, those are SLAs now that we- Yeah … we stick to. So- Delivering agents as a service provider against SLAs, you need to be able to guarantee the performance of inferencing- You do … to do this, right? 

You do. And I think you’re in a great spot to do that. You do. Mm-hmm. And I think, I mean, it’s a question for you actually more than a statement, that industry is going to have expectations on service with inference, and as a full stack provider, I feel like, uh, you’re in a perfect position to address that. 

Well, does anybody want slow AI? Well, this is the thing. I wonder who’s thinking about this, ’cause like you, you, you and I- Yeah, we- … talked about it, right? We, we were talking about it today, right? So, um, and the market has moved really quickly, but it, certainly even in, in my work with WinDC, uh, even I would say as recently as six months ago, most people were talking about latency when it came to where you, uh, locate your infrastructure for AI. 

Mm-hmm. And I think it would surprise, uh, most users of, uh, frontier models that a, a lot of the inferencing isn’t actually done on shore- Not interesting … when you’re interacting with those models. Yeah. And it’s like, define d- define fast. I don’t know. I, I, I used Claude the other day- Mm … to build a deck. Mm. 

It took 20 minutes. Mm-hmm. I was actually pretty happy with that. Mm-hmm. It would’ve taken me two days. Yeah. And it was good. But I think for those automated, uh, activities, as you start to embed those things into your business- Mm-hmm … and we’re likening, um, headless agents to like, uh, for old school guys like us, service accounts- That’s right. 

Mm … and, and automated tasks. Mm-hmm, yep. You need to be able to have some reliability and service levels built into that. Mm-hmm. And I think, um, knowing where that infrastructure is is gonna become increasingly important for those types of activities- Well, yeah … w- when it comes to in- inference. Well, because y- you’re gonna put parameters around your headless agents as to how you want them to perform, right? 

And if you can’t offer guarantees, right, on that, then they’re not doing their job. It won’t be acceptable for it to take 20 minutes to do its task. You may want it to get it done in 30 seconds, 20 seconds. Well, and, and even at some point- Mm … it will become a competitive advantage. Certainly. So let’s say we’re in the same business together. 

Yep. We’re selling security- Mm … and I can do mine in two minutes 30, and you can do yours in three minutes. Right. These things will actually matter- They will … I think in terms of that. I think Jono’s point though about latency is a very insightful one, because why? People have often said the problem with Australia is it’s a long way away, so you can’t… 

It, it’s tricky. So you can’t do much here because if you’re an, um, o- overseas, you can’t bring much here if you’re overseas. And if you’re here, you need to use stuff that’s here because of latency. Mm. Uh, the interesting view is that in the AI world, that 20 minutes you spoke about, that is the thinking latency. 

Mm. The network latency is at nothing. That’s right. Tiny component. That’s right. Lucky if it’s 100 milliseconds. That’s right. So we’re moving away from thinking latency being the challenge… Sorry, from network latency to being the challenge- Yeah. That’s right … to thinking latency being the challenge. That, it’s, that’s exactly right. 

You know, that, that’s why this whole, like, like, the, the industry started with, “Oh, Australia doesn’t really have an AI advantage because we’ve only really got, you know, half a dozen places where you can put this stuff.” Mm. And now we’ve moved to what the industry needs is abundant, uh, power- Yep … and, and land, and Australia’s got a lot of that. 

Australia’s incredibly well-positioned To capitalise on, on what even the government is calling the, the biggest sovereign action since, since World War II World War II. And, and that, that point about limited data centres, and we’re seeing a lot of Australians are saying they don’t want big new data centres in their backyard. 

That’s right. I think it’s super clear. Their, their voice has been crystal clear- Mm … that they do not want that. Mm. But do they want, um, distributed, you know, uh, efficient AI capacity in places where there aren’t people? I think they do. This is the whole question, is we need to think about this as a move from big, massive buildings to- distributed facilities- Mm … that take advantage of our natural advantages we have here in Australia. And we can build an export industry. So we’re getting into the future a little bit, but why, why can’t we be an exporter of this rather than a net importer? 100%. Yeah, exactly right. And I think, um, the industry will be on a, uh, continuous education path as it relates to inference, this idea of time to first token being a component of network- prefetch and decode. All of these things. We will have greater expectation- um, of inference, particularly as industry embraces these capabilities. And it’s gonna be very important that we have these sovereign capabilities- for sure. Mm-hmm. Yeah. And imagine if we could create a… 

I mean, we… Come back to this. Australia, of course, has been built on… We’re here- Mm … ’cause of a number of reasons. Of course, our amazing, um, education- educational institutions, our sensible rule of law. Mm. We’re a very trusted environment. We have a great services economy. Yep. But also, we had a natural resources boom- Mm 

because we were simply able to extract iron ore- Mm … copper, zinc, uh, more cost effectively, uranium- Yes … than other places. We had a natural resource. We were able to extract it efficiently. If that natural resource becomes tokens- Yes … can we participate in a global market from Australia in a world where it doesn’t matter that we’re a long way away from New York or London? 

Why not? I think it’s a huge opportunity. 

  

I’m really interested in, I think you’ve got some views on Australian models. Australian models. And, um, my f- my favorite thing that you’re doing is, is Project Magpie. So I wanted to ask you all about that. Feel incredibly triggered by the word magpie. 

I can tell. Like, I’m all for the initiative. That’s also one of my favorite things. I get attacked more than anybody else here. You do. 

Great initiative. Talk, talk to us- Yeah. The, the- Let’s talk about models … tell us all about it Let’s talk about models. Yeah. So, so the model world we see is divided in a couple of different areas. Of course, you’ve got the proprietary foundation models. 

You know, you’ve got what Anthropic is doing, what OpenAI is doing, what Google is doing, for sure. They, they are building some amazing technology, and most Australians today that are listening to this pod are using those things right now. Yep. Yep. Of course they are. Of course they are You’ve also got this proliferation of open weights. 

Talked about the weights before. Right? And the idea of these open weights models in the deep neural networks being transparent. Those open weights models are starting to become a larger part of the ecosystem of AI. Yep. And when it comes to these headless agents, reality is open weights models are really good. 

And in fact, I would argue in many cases, um, at, very close to the performance of these proprietary foundation models. I’ll vouch for that for sure. Our use cases internally have been astonishing. Yeah. We love them. Exactly. For sure. Exactly. Right. So I think w- people are starting to understand that. Yep You know, it doesn’t mean that if you’re doing advanced biomedical research on the genome, you probably are gonna wanna use the very best one there is, no matter what the cost is. 

But certainly open weights models fit a fit for purpose for a lot of different use cases. It’s kind of like what we’ve been arguing with hyperscale cloud for a long time, and the right application workload living in the right cloud. Yeah. Well said. Same thing. Well said. Well said. So, okay, so we agree on those two things. 

But then those open weights models, whether they’re made by the Chinese, the Europeans, the Americans, or other people, there’s a lot of work to make that. That doesn’t come from nowhere. It’s a tremendous amount of work to get the data, to put it in the algorithm, to train it, to find the amazing humans who know how to do that. 

That’s extremely difficult to do. Okay, so someone’s gone and done that. Um, can Australia play a role here in building foundation model technology? I think the jury’s out a little bit. I think it would be great to say that we could do that and we could build found- models, whether they’re open weights or closed, at the level of those international providers. 

It, it’s a challenging thing to do. Let’s just pause on that for a sec, because I, I just continue to be blown away by the numbers. So if we look at, um, Musk, right? He’s got $60 billion worth of GPUs in a coherent array. He’s got, what, uh, Google, Anthropic leasing that out, right? I mean, who e- is that the biggest array? 

It’s gotta be, right, for training. I mean, it’s gotten to the point where there’s diminishing returns. I mean, who’s gonna build a bigger array for that training? Yeah, who’s gonna build a bigger array? He is. Right, right. Okay. And it’s, it’s, it’s up there. Right. And we’ll get to that in a minute. But, but I think that the, the, the concept that we’ve got, um, open weights models as part of the solution is great. 

Can we… Uh, e- even if we ignore the compute for a second, the actual concept of building the algorithms and getting access to the data, even if we had unlimited compute- Sure … for Australian researchers, are we ready for that? I think this… It, it’s hard to, it’s hard to know. As, as we’re on this pod right now- Yep 

it might be something that we need to work towards. However, what we can do is take the open weights models, and we spoke about DeepSeek, Teleman Square, etcetera, etcetera. What if we take the open weights models and we fine-tune them for us? Mm. So Project Magpie on SCX is exactly that. We’ve taken the OpenAI GPT OSS 120B model, the open weights model from OpenAI. 

Yep. And we’ve done fine-tuning or post-training, the industry likes to call it- Yes … around that to add in unique Aus- Australian information, but also to help it to understand that it is an Australian model. Mm. And it saves our customers from having to tell it that it’s Australian. Yes. So, so when you think about building an, an, an A- AI agent, for example, and it’s gonna go and process incoming invoices If you just use one of these open weights models, every time you tell it to do work, you use a thing called a system prompt. 

You basically tell it a whole lot of stuff it has to understand, like, “You’re in Sydney, we work on Celsius, not Fahrenheit.” Yes. You know, “We are,” um- Like a skill … you know, all the things it needs to know. Yeah, yeah. And as much as 30% of the prompt- Mm … could be used by all of this repeated localization information. 

Got it. Yep. So by fine-tuning the weights- Mm-hmm … it means our customers don’t need to do that. Yes. So it’s saving them work on that, that side. Does that mean they use less tokens, then? It’s optimized. Optimized. So, so we’re not having to, every time they do a request- Mm … rather than putting in a whole stack of tokens that are reminding the model to always report temperature in Celsius- Yep 

right? It doesn’t need to be told that. Yep. So you can save money- Mm … which is important, but also, um, get better processing speed and less work- Yep … to, to produce the answer. Is this like, um, what they call system one, system two tech, right? So if the model itself is the system one tech, the system two tech would be typically where you would have the guardrails and controls, etcetera. 

You’re able to put some of this localization and some of these rules so that the model behaves in a consistent fashion. I, I would say this is in the system one. That’s system one. It’s actually ch- it’s actually changing the weights. Got it. Okay. So we are literally actually going in- Interesting … and changing the weights in that model- Oh, cool 

and then running that on the SCX hardware. Got it. So we’re doing that today. Yeah. Um, now system two, the idea of guardrails- Yes … and that’s a great word. Yes. Tremendous and I’ve got an interesting guardrail story for which I’d love to share- Please … with the, the- Please … listeners here. Very interesting. 

Because a guardrails is a word we use in the industry to mean, how do you make sure someone isn’t trying to hack the AI? Right. Right? And it could be, uh, on the way in, they’re not asking you to do something that’s inappropriate. Mm. Right? Um, or even on the way out. Mm. It’s produced something that’s inappropriate, right? 

Yes. Now, there is a lot of that already fine-tuned into these models. So folks will have tried them. Nobody on this pod or listening to this pod would do naughty things, I’m sure, with an AI. Never. Well- Never … look, we hope so. We hope. Yeah. It’s a pod for, for good, not evil. That’s it. Just so we’re clear. 

That’s it. But yeah. Is that very clear? Yeah. So, so if someone had, had mucked around and asked their AI to do something that’s inappropriate, often the AI will say, “That’s inappropriate.” Mm-hmm. Right? Yes. And it is deciding what’s inappropriate and what’s, what’s not. Mm. And so it has that fine-tuned into its system. 

But there are ways to try and trick it, and some of you may have tried those things and may have had some fun trying to trick a model into, say, “No, it’s not real. I’m doing it for a movie script.” Those kind of tricks, right? Jono does this. Yeah. I’m writing a movie script. It’s a hypothetical. It’s a hypothetical. 

Yeah. A hypothetical, right? It’s a thought exercise. Exactly. Exactly. So guardrails is the term the industry uses to describe how do we protect this, uh, or help to add value and control some of that. So story here is that, um… and I hate to bring up this story, but it’s important to talk about. So- The Ukraine-Russia war Yep 

which is, in, in my opinion, it’s, it’s unbelievable tragedy and such a terrible situation to have that happen, and, you know, it’s the, the largest war… Talking about World War II, it’s the largest war we’ve seen as human species since World War II, and the amount of destruction and death is incalculable and hard to even think about. 

However, when this happened, the Ukrainian government realized that they needed to modernize the way their government worked. They needed to digitize their government And we talk a lot about digitizing government, you know, digital driver’s licenses and- Yep … all these kind of things so we could have a more efficient economy. 

Mm. Well, I think in, in, in Ukraine, part of their job was to make a more efficient economy so that rather than someone sitting around processing driver’s licenses, they can go onto the frontline and fight. And all these things are crazy. And, um, so they digitized their government, and one of the things they did was they used a lot of LLMs in the process to understand how they could do that and also to use an LLM to help with that process. 

So of course, they had a problem with what we like to joke about, Russian hackers. Everyone been to a, like a conversation with someone over a beer, talk about, “Oh, the Russian hackers.” Right? Well, I think in the Ukraine they have real Russian hackers, right? And so they had to build a way to protect AI. Mm. So they created a guardrails model called Opir, O-P-I-R. 

O-P-I-R Which is Ukrainian for resistance. Hmm. If any Ukrainian watches out there, you’ll know about the Opir story. So Opir was built with the whole idea that they could protect LLMs with guardrails, both on the way in and on the way out. Hmm. But do it using what’s called a classifier model, so a specially built model that could classify 996 different types of attacks- that would occur to protect themselves. So what we’ve done at SCX is we’ve partnered with the, the, the team that built Opir. We’re bringing that to Australia. Hmm. So now people who use the SCX models can also use the Opir guardrails- Hmm … which gives them additional, um, safety and security. That’s fascinating. 

That’s awesome. What a great story. Yeah, fascinating. And, and those people are amazing because- Yeah … you know, again, tested in wartime, all these things we like to talk about- Hmm … but don’t want to think about, but we talk about it, are real. Yeah. Yeah, no, it, it’s definitely real. Being, um, safe in terms of your usage and adoption of AI- Mm 

is a massive topic right now. Mm-hmm. Uh, in my work at ASE, uh, nearly every customer is talking to us about, “Hey, we need AI in our business. We need it to compete. We need it to stay current. If we don’t, we’re gonna fall behind. How do we do this and stay safe?” Stay safe. “Because if I just turn everything on, all of a sudden I’m giving access, uh, to my data, to people- Mm. 

Right … who might not necessarily, uh, need access to it.” And so- And so if anyone wants to try… I mean, they should call up SCX, mention the pod. Mm-hmm. They can try Opir. That’s it. Yeah. Cool. And, um, we’re happy to try. Let’s go. We’ll put a link on the, on the screen. Yep. Go and try it. 

Check it out. So, uh, Fascinating stuff with models, David. We talked about watermarking before and why that’s important. Mm. Um, maybe not just important to stop kids cheating on their exams. 

Mm-hmm. You, you had a bit of a take on that. It’s very interesting. So folks should realize that from now onwards, the, um, proprietary models are embedding a standard into their outputs. And they’re all doing this, aren’t they? And they’re all doing this. Yep. And that standard is going to ensure that someone can back solve, if they know how to, what was created by an AI, right? 

They can now back solve to know something was created by an AI. Now, there are two things that that indicates, I think. One is how much it’s being used, and people are concerned, some people are concerned about the idea that AI is gonna pretend to be people. Yep. But also, what does it mean? If you can back solve it… 

Now, these vendors say that this addition of this watermarking in the text can’t be determined, but in some way it is adjusting its output. Rather than being the best output, it’s gonna be the best output adjusted to ensure it has a watermark in the text. Mm-hmm. And what we mean by watermark is it’s a combination of this word, plus this word, plus this word, in this order. 

I can tell it was built by an AI because it did… You know, there was a famous one, if you wrote emails with AI a year ago, it would always start with, “I hope this email finds you well.” Do you remember that phrase? Yeah. Yeah, yeah, yeah. Do you remember that phrase? Yeah. Yeah, yeah, yeah. It would always start with that. 

Yeah, yeah. And that was not bec- that was not a watermark. Yes. That was just result of training. Yep. But- And the em dashes followed, right? And the em dashes. Yeah, that’s right. Like, all those things. Yeah. So what they’re doing now is it’s the equivalent of that- Mm … but baked in. So does that mean it’s not gonna give you the best result, the best answer? 

Interesting. Well, I, I’m curious about this too because does anyone actually care? So if, if you’re interacting with, let’s just say, an LLM to produce, uh, content for you, where you might have an idea or a strategy and you don’t have a week to write the copy, you’re asking it to build you a straw man. Mm. 

It’s, uh, it’s… You’re prompting it. It’s taking your idea and articulating that in a way that, um, you’re fine-tuning- Yes … yourself and then perhaps, like, it’s, it’s taking the hard work out of extracting that idea from your brain- Sure … into a document that you can share with other people. Um, is that a bad thing? 

I feel like, I feel like if everybody’s not leveraging AI right now to, to do that- I think separate it’s gonna be hard for you to compete with others. Separate the creation from receiving. I think there are certainly situations where you don’t wanna receive AI-generated content. Case in point, if you’ve spent 400 grand with a Big Four consulting for- firm- We saw that, didn’t we? 

to produce, right? Case in point. We saw that. Or if- You receive an email from someone who you care deeply about, and all of a sudden you realize that it’s been written by AI. It’s impersonal. It’s gonna upset you potentially, right? Potentially. Potentially. And I, and I think that’s where, um, you, you’re gonna… 

Like, I’ve got this, uh, this thought bubble on this, that as we, as we move through AI as a society, human beings are gonna place a premium on genuine human- Right … human-to-human i- interaction. Well, just think social media, right? So wouldn’t it be wonderful… I mean, I haven’t been on X since it became X, right? 

‘Cause it’s a steaming pile of arguments and all the rest of it. I should probably shouldn’t say that out loud, but let’s face it, like, there’s a lot of bots out there. I mean, they couldn’t even determine just how many they were. I mean, how many actual humans are on that platform, right? Wouldn’t it be wonderful if there’d be a tick to determine that, A, you’re a human, and B, you’re producing original content? 

That’s gonna become a rare thing. It could be. Right. But, but come back to something else you both were saying about- Mm … the, the agents that are headless or- Right … talking to each other. They don’t care. They don’t care. So there’s a… I, I think the, the world’s gonna bifurcate. It’s gonna be stuff that humans make and bots make for humans, back to Jono’s point. 

Yes. But then there’ll also be a world whereby it’s bots talking to bots. Where the bots don’t care. They don’t care. But the productivity gain is substantial. That’s right. So we’ll see how that evolves. Fascinating. Super interesting. Mm. I especially think for you, you mentioned the big four and- so, like, how much of that can be done with AI now. I don’t know where that’s going to go on the, on the consulting side. Well, I think we’re talking about the big three now, aren’t we? Probably. Okay. Probably, yeah. It’s, it’s, it’s a, it’s a wild space, and it’s getting disrupted- Mm … um, every single day.  

  

Dave, we’re gonna enter into the, uh, final segment of the show, quick fire three. We ask these three questions of every guest. What’s one piece of advice you’d give your past self before taking on a major tech initiative or challenge? That’s very interesting one. I can think of so many things. 

We haven’t got an hour for that. Um, I, I think the most important thing I would tell myself would be, “Look, you need to be always remembering that there’s gonna be a new technology wave. It’s not the e- Th- this wave is not the last one. Mm-hmm. There’s always a new one. So take your time, think it through, get it right, you know? 

And find the best people around you to help you to be successful as that new, that current wave goes through, and then be ready for the next one.” Surround yourself with the best people. Yes. Love it. Yep. Obviously. Brilliant. Um, so this one goes to tech leadership, and David, you’re, you’re one of the brave ones, right? 

You’re one of these repeat ASX launching offenders, right? That makes you special in this category. So what is something that you used to believe in in tech leadership that you no longer do? I think I used to believe that all the good ideas came from overseas in tech. Huh. I really used to believe that. 

Honestly, I, I worked with a lot of companies that were either, uh, American companies or were distributors- Mm … of the American companies, and I used to believe that the ideas and these, the, the execution had to come- Yes … from Silicon Valley or Boston or, or maybe even Europe in some way. I had this belief that you, it was not possible to lead from Australia. 

I’m changing that, my view on that. Yeah. I think that it’s, it’s hard, and we have to be ready for the work, and you have to work really hard, but it can be done from here. Yeah. And there’s no reason for us to assume that just because someone comes from Silicon Valley, they automatically- Mm … know more than you do. 

I love that, and that’s consistent with what we said before. Remember, like, early in our careers, when you, you’d look up at somebody that had the job that you wanted in the company that you admired, and you thought, “Oh, how does that person do that thing?” And then you end up in that job, and what’s the- Mm-hmm 

what’s the fuss all about, right? Yeah. Yeah. Yeah, that’s right. Yeah. That’s right. I, I love that one. Um, yeah, be, be brave. I, I love that one. And, um, and yeah, I think there is a lot of good talent in Australia, lots of great ideas, and a, a great startup community, and- Mm … a very innovative community. Dave, what’s your gut check for spotting a good, uh, tech partner or business partner, um, or perhaps walking away from a bad one? 

I think a lot of it is to do with that commitment to really execute on doing stuff quickly. You know, if, if they’re gonna act slowly, if it’s not gonna… If they’re just not ready to go and just experiment, and even if the experiments fail- Mm-hmm … you know they’re not gonna be a good partner. You want someone who says, “Yeah, I know that there’s no certainty, but I’m prepared to try, and I’m prepared to give it a go, and I’m gonna do this, this and that, and then we’re gonna measure this and, you know, see what the result is.” 

If, if all they’re saying is, “Well, I’ll think about this, and I’ll get back to you on that, and I’ll work on some program here-” Mm … I, I think that’s a challenge. You know, to be successful in today’s world, it’s got to be about rapid decision-making and, and an acceptance that we don’t know all the answers, and that, I hate to say that failure is an option. 

There’s that famous book, Failure Is Not An Option, about the- Yeah … space program. Yes. But failure is an option. But you need to do the… You need to make that failure, um, quick. You need to understand what it means from that failure, and then go on to build something great from there. I love that, like a bias for action- and execution. Completely agree. And bravery, inherent bravery. That’s the thing that’s coming through, right? Yeah, and I think, I think the message, and that’s why it’s… I’m so glad you’re doing what you’re doing because y- what you’re doing is telling Australians that they can do it. 100%. And the biggest problem we have is this concept that we can’t do it, and that someone’s better than us. 

There’s, there’s… We’re as good as anyone. 100%. I love that. I reckon we might even be better, and that’s a great way- Ooh … to, to wrap up this show. Dave, thanks so much for, uh, coming on the show and, um, sharing some of these insights with our listeners. I think it’s super interesting. Think you’re doing very cool stuff. 

I love it. Very grateful for your time today. I love everything you’re doing with SCX AI, with SambaNova. I think bringing inference to the Australian market, uh, informing them of that, uh, being a full stack provider, I just think that’s just the right place to be in. Sovereign, I love that. Distributed, linking that to the WinDC story as well, I think amazing. 

And I learned something today as well. I actually learned a, a fair bit, so, um, which is why I do the pod. It’s fantastic. Yep. Naran, where can our listeners get the pod? Well, this podcast, Things Reasons, is obviously available everywhere you get your pods. That could be Apple Podcasts, Spotify. We are of course all over LinkedIn as well. 

We are produced by Pru Loon. Our digital producer is Karina Aguilera. You can of course message us at thingsreasons.show. We have our website as well, so there’s, uh, the confessions page. We still wanna hear about the confessions of IT fails as well. They are anonymous, and we will animate the best ones as well. 

We love that. They’re coming out on the next show. They are indeed. But thank you to all our subscribers as well. We value you mightily, and we will be back again shortly. Thank you very much. Thanks very much. Thanks, Dave. Grab Dave at SCX.ai, and, uh, there’s a few cool trials to check out there as well. We’ll put the links up on the bio. 

That’s it. Thank you.