She Replaced Her Analysts With AI and Beat the Market

Most fund managers find their best stock ideas by accident. Armina "Arms" Rosenberg built an AI that reads 30,000 articles a week so her fund doesn't have to rely on luck, and a human still makes the final call on every trade.
Armina is co-founder and portfolio manager of Minotaur Capital, an AI-led global equities fund she started with Thomas Rice, the rare fund manager who can also code (he built their AI system, Taurient). Before Minotaur, Armina spent eight years in sell-side research at JP Morgan and ran the global equities portfolio for Atlassian co-founder Mike Cannon-Brookes' family office, Grok Ventures.

This one is a proper look inside a business that has put AI into every step of its process, with experienced investors shaping the system rather than handing it the keys.

About the guest:
Armina "Arms" Rosenberg, Co-founder and Portfolio Manager, Minotaur Capital 
minotaurcapital.com

About the show:
Experts in the Loop is a podcast about how real experts put AI to work. The actual builds, the trade-offs, the workflows, no hype.
Subscribe so you don't miss the next one.

In this episode:
- How Minotaur Capital scans 30,000+ articles a week across 174 sources to spot companies going through real structural change
- Why Minotaur keeps a human in the loop at every decision gate, and how the AI learns the way they think
- The research that used to take an analyst 5 days, now done in 2 minutes
- The three mistakes most people make with AI: treating it as an oracle, trusting one model, and skipping source documents
- Why you should treat AI like a junior analyst, not a portfolio manager
- The "SaaS apocalypse" debate: what software agents will use, and what they will replace
- Where Minotaur sees the real opportunity: AI infrastructure and the memory thesis
- How they run 20+ models and route cheap versus expensive to control cost
- How to actually start building, from Codex agents to Claude Skills (and why you should never buy a skill)

Chapters:
00:00 The AI fund that reads 30,000 articles a week
02:33 From public housing to global equities
04:35 JP Morgan, a family office, and the Mike Cannon-Brookes story
09:20 Why finance was an early home for AI
12:29 Inside Minotaur: the Axon stock they almost missed
15:03 Taurient: 30,000 articles and idea triage
18:36 A human in the loop at every decision gate
21:28 "AI won't fix a bad fund manager"
23:41 The mistakes people make with AI
25:42 The hallucination double standard
28:38 Build your AI a brain
30:34 The "SaaS apocalypse" and a 40% short
34:07 Will agents replace software, or use it?
36:20 The memory thesis: betting on AI infrastructure
39:39 Hyperscaler spend and the real cost of AI
41:20 How to adopt AI: ask better questions
43:18 Running 20+ models and managing cost
46:46 Where to start, and Claude Skills (don't buy them)
50:39 Prepping for talks, and opening for SoftBank
52:36 Making active funds great again

Website: eitl.show

If this was useful, hit subscribe and tell us in the comments which part of Armina's workflow you'd build first.

Keywords: AI in finance, AI hedge fund, AI investing, agentic AI, AI agents, large language models, AI workflow, human in the loop, fund management, build vs buy, SaaS and AI, AI infrastructure, Claude, Codex, Claude Skills, Minotaur Capital, Armina Rosenberg

#AI #ArtificialIntelligence #AIagents #Investing #FundManagement #FinTech #LLM #AIworkflow #ExpertsInTheLoop #BuildWithAI

Transcript

Chris: 0:00

From public housing in western Sydney to managing billions for one of Australia's most powerful tech billionaires. Amina Rose, aka arms, has never and it's paid off. From JP Morgan managing a global equities portfolio for Mike Cannon-Brookes family office, she is now the co-founder of Minotaur Capital, Australia's first AI led global equities fund. An AI is at the heart of every decision they make from the brain down. And today we're going to dive business the investment Let's go. Thank you so much for joining us I know you've been quite busy You've had a few, few presentations and episodes of shows. You were on equity markets. Yeah. Yes. How was that? Yeah, it was good. Yeah. Good.

Arms: 0:50

Very, you know, finance stock But I like those guys.

Chris: 0:54

Hopefully we can be a little bit more focused on some more fun technology type stuff in stage chat.

Mark: 0:58

Do you dive into the tech with Or is it more on the, you know,

Chris: 1:03

Things.

Arms: 1:03

A little bit? I think on most podcasts, it's Okay, cool. So yeah.

Chris: 1:08

And why is that? Tell us.

Arms: 1:10

So we run a global equities So we invest in stock markets Um, and that's literally very So all capsule sectors, all Um, and I think the reason we can go so broad is because we use AI and large language models to go through, you know, the entire universe. So there's sixty thousand listed Yeah. Um, it's actually there's still to billion dollar plus market. I didn't realize that there's ten thousand billion dollar companies out there that are listed. That doesn't even include all

Mark: 1:40

Need a higher measure.

Arms: 1:42

Yeah, exactly. Uh, and so, you know, most fund managers, like, they get ideas from really random sources, uh, and we use AI to help us with idea generation. We still have those random We can add alongside, but yes,

Chris: 1:55

It's interesting because I can't Oh yeah. It's super interesting what you

Mark: 2:00

That's my background as well, like for um, less using AI in capital market space, but doing like I'm a former equity researcher. So I kind of know more about that world and, you know, doing a different quantum factor kind of models. I wish I had AI ten years ago to guys can do now.

Chris: 2:16

Yeah. Could you imagine what you now if that was the case?

Mark: 2:19

It would literally be, um, not What's that thing? Um minority report I'd be doing like that with the meta glasses and stuff and just like moving things around.

Chris: 2:27

Moving financial, a bit of ETF It's like.

Mark: 2:30

Oh, delete, delete, you know,

Chris: 2:32

Reinvest, sell, sell, sell.

Mark: 2:33

But, um, before we get into may not know, you've got a background, um, in terms of your could you tell us a little bit

Arms: 2:45

Yeah. So I have a very different bros in finance. Uh, you know, I grew up in I grew up in Parramatta. Um, I grew up in housing So my mum came out here. She was a single mother, had Wow. And now, like having my own I know how hard that is. Like, I just never had a full would have been. And she, like pretty much single Single mom.

Chris: 3:09

Yeah. True. That's amazing.

Mark: 3:10

Shout outs to.

Chris: 3:11

Moms. Yeah. Shout out to.

Arms: 3:12

Indonesian immigrants though. Immigrant story just generally. Yeah. Um, and she was like, so Like, I remember we had this kind of computer and I was like, games, buy me a computer game And she was like, oh, there's actually computer games on the computer already. Like you should try. It's called Lotus one, two, And I was like, awesome. And I did these tutorials for And for any younger people in two, three is the predecessor Yeah. So like, I was literally I was nine years old.

Chris: 3:46

That was the game that you were.

Arms: 3:47

That was the game.

Chris: 3:48

She was training you to be. She was training you for finance

Mark: 3:51

Were you like the class It's like, no, guys, we can't spend this Susie and blah, blah, blah. Like working out playground

Arms: 3:58

Yeah. Playground finance. Um, so yeah, so I, you know, and was a need, you know, like every to be rich. And, you know, I grew up with Street, so it kind of got, you that sort of scenario. But that was met with a true passion for just looking at companies and looking at stocks and investments. And I've always just been a technology as well. And so like the intersection of technology and finance is always where I've kind of loved to dabble. And that's how I've been able to bring it together at Minotaur as well.

Mark: 4:35

And the work that you, I mean, JP Morgan that. Yeah, yeah. So, so from there to, um, I can't remember when you got to grok and then the other um places before Minotaur, but that journey, did it have technology along the way as part of your, your work.

Chris: 4:51

Yeah, sure. Beyond, uh, Lotus and in Excel.

Mark: 4:56

It's always been there.

Speaker 5: 4:57

So I was in equities research.

Arms: 4:59

Like you, I was I started my career at JP Morgan did eight years um covering Australian small caps. So made buy and sell outside of the ASX two hundred. And that included like early So like Webjet and what if.

Mark: 5:14

Wow.

Arms: 5:15

Car sales and real estate. So that was where the technology Um, and you know, Australia has marketplace side of things. Um, and so that's kind of the, the stocks that I was looking at at JP Morgan and then I joined the family office world after JP Morgan. So I worked for a guy called He made his fortune through funds management and property development. And, uh, I had never, I never knew what a family office was before then. So, um, you know, a family firm of one super rich family. And they actually, they're so rich that they hire like experts to run their investments for them. So I worked speaking first, running his global equities portfolio. And it was so funny. He you know, he wasn't I'd say But I remember him coming to me the stock tip from, but he was this company called Shopify. I think it might be worth market point of view. And I think Shopify was like like it was. And, you know, no one knew what Uh, so we ended up buying it. I think the funny thing about that is I think we ended up selling it at like forty dollars. It's like gone on to obviously sort of e-commerce software

Chris: 6:32

Um, it's going to be down at the Yeah. AI hasn't done too well.

Arms: 6:36

So we'll get into that.

Chris: 6:38

Yeah. Yeah. Yeah, exactly.

Arms: 6:39

Um, so, so yeah, so did that for And then obviously, um, joined co-founded it last year.

Mark: 6:46

Some people may have heard of

Arms: 6:47

Yeah. You know that guy?

Chris: 6:48

Little company. Nice.

Arms: 6:50

Well, so the funny thing. So I started Minotaur Capital, my co-founder, Thomas Rice. And so I'll just kind of run well because it kind of And that's interesting. So, um, he started his career at a place called PM capital, which was one of the first hedge funds in Australia. And I was broking like Webjet. And what if to him when he was there and we found we were kind of aligned in investment philosophy and got along really well. So we stayed in touch. And it was actually Thomas who introduced me to Mike Cannon-Brookes.

Mark: 7:22

What the.

Arms: 7:22

Heck? Yeah. So he, um, he met Mike before anyone knew who Atlassian Yeah. Um, Mike went to a Macquarie to sit down next to Thomas, and oh, you're Mike Cannon-Brookes. I use Jira and Confluence in my So, you know, one weird thing about Thomas is he's like this kind of unicorn guy because he's an amazing software developer and an amazing fund manager And like, like those two skill sets don't usually come in the same person, particularly in Australia. And so Mike Cannon-Brookes was conference knew what Atlassian And so, you know, they got It led to Mike and Scott practicing their pitch on Atlassian to Thomas the next week. What the heck? So I think Thomas is probably to hear an Atlassian pitch. Jeez. And then Thomas, uh, I think like one of the only Australian the IPO of, of Atlassian. Um, and they stayed in contact. And then when Mike was looking office, he said to Thomas, do He goes, oh, you should, you Um, so, um, that is crazy. And then Mike was one of the biggest investors in Thomas's old fund when he was at perpetual. And so what that meant was I became Thomas's client effectively.

Chris: 8:44

Yeah.

Arms: 8:45

And so we.

Chris: 8:46

Were. Was he happy? Was he happy about that? Yeah. He's stolen my stuff. Enough to be a.

Arms: 8:50

Co-founder of mine.

Chris: 8:51

Yeah. That's right.

Mark: 8:52

On Atlassian, it's interesting smaller and they weren't small afford PwC as an auditor. But in a former life, I was an auditor on Atlassian in their old, old offices on Sussex Street. I went to some chili eating contests and I just remember having like, whatever the Texas Reaper was. And that was that was fun. I couldn't eat for the rest of It shows what curiosity does in connections in the community. You never know where they end Yeah. Jeez.

Chris: 9:20

The thing I love about the market, or the sort of area that you're playing in, is if you look back, the history of AI and machine learning and how it's evolved. I mean, there's a few industries where it did originate from military, all these other places. But finance, particularly stock stages of introduction of AI. If you take it back to machine learning, because you are algorithms, you were predictive markets, all that type of stuff was all come from the finance sector. So when we talk about AI, we It is like pretty much touched where you are now.

Arms: 9:50

Yes. That's right. And I think like quant funds have used AI machine learning forever. Right. You talk about your Renaissance Um, but I think what's interesting is like fundamental equity. So doing research on companies been a pretty tech light, uh, last three decades. And I think what you're seeing now is AI and large language models are leveling the playing field between the fundamental guys and the quant guys, because now fundamental guys can like pass through lots of information very quickly. And that's something that quant So it's like, I think that and convergence between the two.

Mark: 10:29

That become more and more, uh, analyst, you're an equity that you have a bit of quant to That's fascinating because we see it with other industries as well evolving.

Chris: 10:41

It's like the entire, like you you know, back in, you know, like five years, maybe even to build any kind of thing, you an architect, an engineer to be kind of technologies. And nowadays that this is what It's leveled the playing field generalist and the, you know, So now we can do things that we've never been able to do before without the need of an engineer. Everyone can do it. It's, it's, it's so much fun.

Mark: 11:11

In that world. And I see it in the legal AI And for folks that haven't been paying attention, um, to, to that space, there's these big billion dollar companies having the stuff that they build and there's a whole heap of lawyers shout outs to the legal quant community. They call them legal quants. They're not quants as we know it, but, you know, very software driven. So hence the name. But they um, have been a bunch, a bunch of lawyers building solutions in word in other tools to do the things that these bigger companies, billion dollar companies do. So there's this big question around, well, should I pay for that? Um, when the capability is But the difference is those bigger companies have data, and they've got more experience in process and know how to manage all the risk and stuff like that. And I guess it gets to the approach that you guys have to and we'll get into, you know, how you're using AI in your space, apart from just being in investments and managing all of that. Um, you guys have built, uh, witnessed and we had a great year, um, where yourself and was that you guys were building to this day, talk about it as application of AI into a

Arms: 12:24

Thanks.

Mark: 12:25

Can you tell us a bit more guys are doing with AI?

Arms: 12:29

Yes, we can definitely. I mean, we use AI for every step of the investment process and actually all of our operation stuff too, which is a whole other conversation. But on the investment side of with idea generation. So, uh, well, actually, uh, taking a step back from that, you start with investment philosophy. So every fund manager has their our centers around. We think companies that are change, some sort of like long That's an interesting time to potential for that company to be higher than normally.

Mark: 13:03

Are we talking longs and shorts? Like as in buying because you think it's going to go up or buy, you know, um, shorting it and people can look up how that works. Is it for both sides?

Arms: 13:12

Both sides? Yes. So a strategy change you could say is good or bad for the company, or you could agree with the company's strategy change or not. But to illustrate with an stock that was in Thomas's old called Axon Enterprise. Okay. And axon, uh, there was an seventeen that Thomas happened axon makes all the police tasers And he was like, oh, that's Didn't realize Taser was a He'd seen The hangover. He knew what a Taser.

Speaker 6: 13:43

Was.

Chris: 13:45

Right?

Arms: 13:46

And the article is talking about free cash flows that they got monopoly business, but investing and the way that body cameras body cameras to police for free. Police would wear it all day and It would upload all the image called evidence dot com. And then axon would charge subscription fee for evidence dot com. So it's probably one of the business model kind of ever. Um, and existing investors in axon hated it because they were used to, you know, this free cash flow business, but it looked like they were wasting all this money on the body camera business. Um, you know, it looked like it was super expensive, if you guys know what, like ps r it looked like was trading on a p e of seventy times. But if you stripped out the losses of the body camera business, it was on twenty times. And actually, if you thought body cameras was going to be a thing, it was very cheap and like body cameras in the US in twenty seventeen, like it kind of went from like being a nice to have to we very much need this. Yeah. Right.

Chris: 14:49

Uh, so now they have to weather. Yeah. Yeah.

Arms: 14:52

I mean, long story short though, that stock twenty thirty X in the time that Thomas owned it in the fund. But it occurred to him that if day, he would have missed his

Mark: 15:02

That is crazy.

Arms: 15:03

So one of the first things we do with with um Taurean, which is the name of our proprietary um AI software. Um so Taurean goes through about articles a week. Wow. From one hundred and seventy Yeah. Uh, and looks for has this, uh, prompt that's, uh, sits over the top of large language models that's looking for companies that are undergoing some sort of structural change. And like in the old world, the way that would happen is like just through keywords like looking up, looking up the word strategy and change in the article, which is pretty blunt, right? Yeah. Whereas now, because large to, parse natural language.

Mark: 15:43

Um, it's more semantic now.

Arms: 15:44

Yeah, exactly. It's like you can, you can be very detailed in what you're looking for. So in the prompt is examples line, or maybe changes in

Chris: 15:54

And it can also understand context, which general search doesn't. So, you know, stringing it could be random set of words, but it understands that that means strategy or something like that as well.

Arms: 16:03

Exactly, exactly. It doesn't have to be a Yeah, exactly. They say like, oh, we've had a because we're using AI like a

Chris: 16:11

Exactly.

Arms: 16:12

As an example.

Mark: 16:12

And I guess you're adding in gone well or not. I remember seeing something and hopefully I'm not giving one, no one can really see. And we'll talk about the AFR screenshot of something there. Um, but I remember there being this like way, it was almost like a dating app in my head for, for stock selection where it's like, okay, we like this swipe right, we don't like this swipe left or something like that. Does that learning still feed

Arms: 16:40

Yeah. So. So what you're So when we do that search on the um, effectively a funnel of undergoing a strategy change. And the way that it works is it brings in the name of the company a paragraph about the strategy change. Um, and then we can swipe left So it's funny because that's how Okay. Now, um, we can actually get an bring me ten ideas in the funnel would vote on it based on And it tells me and then I can this, this and this, but you And it learns how I, you know, And so it's getting better and So that's like the next

Chris: 17:25

So that's something, I mean, intelligent, but I do something I've built a little application that the home for a home for, for myself. So it uses open claw. I use perplexity and a bunch of other tools that does a whole bunch of article scraping as well, Nowhere near the volume that you're talking about because I don't have that much money, but it pulls in a whole bunch of articles relevant to AI. So I can stay on top of what's happening, all the trends, like all the latest updates, and then it pulls into a triage, finds the things that things are trending the most, and then presents those articles in like, I think like twenty separate pages depending on whether it's business targeted, just general or some other interest that it associates to. And then I have a secondary thing that automates going, okay, based on sort of what's happening in social media and other elements, what would make the most important or the best social posts. And I have another tab that then that I could post with, with all And on top of each one of those, think this is a bit iffy. I can click on it, say, hey, why this is wrong. And then it feeds back into a sure that it's optimized on on

Mark: 18:28

When we all got Neuralink flow to go, okay, upload this to Keanu in the matrix. You just sit there and it's

Chris: 18:36

Yeah, exactly. one hundred and thirty five So the reason I brought that up is the show is experts in the loop. So a lot of the thing that you talked about then was, or it was all automated, you're using agents. So is there any sort of human in Because one of the examples is where you mentioned Thomas Price. It came across that article Like, is there a chance that the AI could similarly miss one of those articles? So therefore, is there someone what's happening in that as it's

Arms: 19:04

Yeah, that's an excellent And yes, if there is some sort that isn't picked up, then we'll you didn't pick up this change? And so interesting self-learning There is a human in the loop at So, you know, I just talked The next phase is we generate like a quick view of the company. It's called a snapshot. And it says like, what does the What are its revenue and What makes it interesting? What's the bull and the bear Can it double in three years or Because that's how we kind of So we're looking for something Yep. Um, and you know, that report sort of investment analysts, Probably takes like two minutes Um, but then like going from like that's a human that reads Thomas and I, and then we'll, do further investigation on. And so yes, you're right, there single decision gate. But like you, like I kind of mentioned with idea triage where simultaneously training Taurean to like know how we think and why we push those decisions through. And eventually we think that Like we think it'll get to the point where it starts to just generate like, here's, you know, a list of fifty stocks that we think should be in the portfolio. Mhm. Um, now I think we're a little Uh, but like, I think what's to a future where, you know, I'm cocktail in hand, and taurine is Yeah. And it's interesting with my investor base, I'd say it's split fifty fifty between investors who are like, please tell me you're not diluting the AI with your like bad human judgment. That's half of my investors. The other half of my investors is, please tell me you're not just letting AI run on the portfolio, and you make sure that you know what's in the portfolio. And I'm like, obviously the

Chris: 21:00

Like it's yeah, exactly.

Mark: 21:02

But, you know, this is all Yeah. Yeah. Into the loop. Um, but has AI actually changed Has it gone the other way? Have you had feedback from it might not have seen before, but Uh, not all seeing, all knowing, does things in two minutes Are there things that are affecting the investment philosophy going back up the chain?

Speaker 5: 21:28

Yeah, it's interesting.

Arms: 21:29

I, I don't think so. So like, I think the underlying investment philosophy hasn't changed over the last four decades. And, you know, I think one of that Thomas and I have is that investment philosophy into AI. And it's interesting because, you know, a lot of people ask us like, oh, is AI your competitive advantage? Like, if everyone had access to wouldn't everyone just do as And taurine embeds our investment philosophy, but that's very different to another fund manager's investment philosophy. And I like to say, like you add AI to a crappy fund manager and they're still going to be a crappy fund manager, just amplified.

Mark: 22:08

They do crappy things in two That's what it is, right?

Speaker 6: 22:12

Yeah. Their returns are probably

Arms: 22:15

Um, so I mean, the other discussed before that I get ever license out to Oregon? And I think the answer is like high degree of customization. So even, let's say another fund look at all of those articles. They might have a completely different prompt that they want think articles is a great place to start with, you know, annual

Mark: 22:41

Macro or whatever.

Speaker 6: 22:42

Yeah, exactly.

Mark: 22:43

Everyone's got their different Okay. It's how you use the, the And you guys have done like a guess, ahead of the curve. Because at the time when we and But at the time when you showed me this literally, I had not seen anyone. And just because I knew the whole like, sorry, working on cell side, so I wasn't, I dealt with a lot of fund managers, but I never crossed over to that side. But I knew about the workloads So just being so intimate with any group use AI as much as it So, so again, congratulations

Chris: 23:19

Particularly in the space that industries, it definitely is. It definitely is occurring, but in finance, it's pretty innovative what you guys are doing or investments in stock management.

Mark: 23:28

Even for a space that's been

Chris: 23:30

Has, like the.

Mark: 23:31

Pre GPT kind of AI. Now going on to the next thing, it's like, where do you kind of see others making mistakes when it comes to using AI, given that you're so well experienced in this?

Speaker 5: 23:41

Yes.

Arms: 23:41

I think that, um, uh, there are a lot of people who treat AI as an oracle, not like an investment analyst. So like kind of take every kind of thing that it says as, as the truth. I think also a lot of people model, like, you know, you language models because they're Um, you know, the famous example of like any Chinese model doesn't answer anything about Tiananmen Square. Yeah. Like even the US models like won the twenty twenty election And you're like, what the. It's crazy. Um, so I think it's like, it's of models and people, people And then the last bit is like, So, um, a lot of people just ask ChatGPT something and you know, when you ask a large language model a generic question, it goes out to the whole of the web effectively and like gathers information from the whole of the web. And I find that you get better it where to look. Yeah. Yeah, exactly. So like, you know, with parts of feed all the articles, but we transcripts and, um, you know, So for any ASX stock, it goes to SEC, etc. like any exchange. Um, so that's also a really But I think the macro point is skeptical around, around AI. Um, and you can't, you can't, that you get as the, the truth

Mark: 25:15

Yeah. Treating it as an fourth year kind of thing. So it has to be checked, but it So like, you know, it's still very valuable because I find that, you know, the polar opposites is just crazy where it's like, no, we don't use that because we can't trust it because they saw one mistake or on the other side, we trust it implicitly. And then something big hasn't happened to them yet, but it will because it's a stochastic probability model. So that's that's the middle is

Speaker 5: 25:42

Yes, one hundred percent.

Arms: 25:43

Sorry. That's the other mistake that they they ask AI something or a something, and they get a well, we'll never use it again And that's like, you can't do That's burying your head in the Right. And the funny thing about that I always find is like, you know, one question I get asked a lot is like, how do you deal with hallucinations? And, you know, happy to answer that, but you never get asked, like, how do you deal with hallucinations of your third or fourth year investment analyst who is like this Patagonia vest wearing finance bro, who probably doesn't have a lot of context to certain things either, right?

Speaker 6: 26:18

Like, that's so true. It makes more mistakes. Like I've never been asked. Yeah. I've never been.

Arms: 26:22

Asked like, oh, do you check the work of your fourth year investment analyst?

Mark: 26:25

Like we never use them again. We put them in the back room.

Chris: 26:28

Yeah.

Mark: 26:29

We don't we don't do that. Right.

Arms: 26:31

Yeah, exactly.

Chris: 26:31

It's the same argument you see with like autonomous vehicles, right? With the rise of EVs and crash and everyone's like, oh, It's like, yeah, but people are well as like.

Mark: 26:42

Have you seen Chris? Right.

Chris: 26:43

No reverse into a pole the other day was, um, but I mean, even from the simplest point of view though, like calling out the fact that models do treat things differently. You have, you know, Codex and that we've had quite recently, like Claude code isn't for isn't up to date with the latest You have to prompt it to action the latest thing.

Mark: 27:08

Um, yeah, you asked it about Tell me about dispatch and it you're talking about.

Chris: 27:13

And it's like.

Mark: 27:13

Look online for the feature that

Chris: 27:17

It didn't know a thing about it.

Mark: 27:18

Oh, okay.

Chris: 27:19

Yeah, there's that thing.

Mark: 27:19

But we love Claude, by the way.

Chris: 27:21

Yeah, we do love it's the main

Mark: 27:23

But still.

Chris: 27:24

Yeah. But so when you're and then to your point though, it's like, okay, we're understanding where the gaps are, how do we overcome that? And then you start. Okay, now we have to make sure I Maybe I'll do my initial research in something like Gemini or, um, or Codex and then bring that into Claude for it to do the actions that it needs to do. And that's a really important Whenever we're using any kinds of AI, the, the risk there also though, is, um, and it's touching on the point you meant before you mentioned before, which is around data and history and knowledge, you know, the difference between your junior and someone like yourself was experienced. Um, AI only knows what it knows or what it has access to and it's not good at at the moment for people building things that's self-learning and self-teaching. And if it is self-learning and self-learning off of its own or it's self-learning off of bad Um, so unless you have someone feeding back information or has prompt and learn and teach it end up with these hallucinations validated all people use it that type of stuff. Um, so you did touch on the data What, how did, how did you come and your data into, I guess what

Arms: 28:38

Great question. And I have actually, I say this fund managers, because a lot of experience lives in their heads. Yeah. Like even something as simple as an investment philosophy, people don't sit down and articulate, this is what I'm looking for in stock.

Mark: 28:53

I walk into a room where we were We were selling our research and we're like, well, what's your investment philosophy? And the guys like you pointed up here, he was being cocky, but he was like, yeah, this is this is no, no.

Chris: 29:04

People don't invest in that. Like that's the other thing. Like.

Arms: 29:07

Yeah, yeah. And I think like with AI, a lot of language models, like it's really important to articulate everything in words, in documents so that they do understand and know what you're looking for. And, and so yeah, we are Everything we do is in markdown every single stock to hear it. Like it's and and like, even to and we'll say, okay, here are Here's what we learned from that Um, yeah, we've like everything. We're basically trying to build

Chris: 29:36

A brain.

Mark: 29:37

Disciplined in doing that.

Chris: 29:38

Yeah. I'm going through the out a brain. It's like it's so much fun to do have you put mind maps on top of do you mean by mind maps? So it's like a, I guess a brain And it shows where all the discussions and etc. it's like perspective all the information It's more of a side thing of it's um, really. Yeah. And you can like zoom in on connect and see how it's between certain elements. And you can even then start to connection is not right, etc., Um, but yeah.

Mark: 30:14

With the mind map stuff, you can connect it to software that does you can just ask the AI to build

Chris: 30:22

Like. exactly.

Mark: 30:23

And speaking because there is we build or buy? And it's especially there like, space, um, what are your views Oh.

Chris: 30:34

Here we go.

Mark: 30:35

What is it for folks that maybe

Arms: 30:37

Yeah. Um, so obviously all the software, so I should say, um, and I'll tell you about the Minotaur journey with, with SaaS. So, uh, at the end of December beginning of December last year, oh, a lot of software names have We think there might be some We were long things like publicly, heavily about it. And literally in the space of completely reversed our view and flighty, but just because the say, to be short software. So we went short, which is like when you like sell a, sell a sector or think that the sector's, you know, about to fall. Um, and so yeah, we went short software at the beginning of January, and that's because I came back from Christmas, New Year's. I, like most people, was like, kids and, and drinking and Thomas is like, I have coded

Chris: 31:35

Okay.

Arms: 31:37

I have like five new agency like a PA agent who like goes moves that is.

Speaker 7: 31:45

Yeah.

Arms: 31:45

Um, and, you know, he was step change in the ability for code and develop software. So we saw this step change from Um, and you know, we were seeing Deirdre Bosa who's this famous up Monday dot com. And so we went short. Honestly, the whole sector, the anything we imagined. So we made forty percent on our think four weeks or something. Um, and I think now so, and like, you know, we're going to think about. Software was one of the best business models in the world, right? Like like recurring revenues, eighty, ninety percent gross margins. People were basically banking And that's why they traded on They did. So they I think the whole sector Now, we're at this point where I think we can afford to be more discerning. And you're seeing that with our reporting season is playing out right now. So you're seeing stocks like Atlassian had a really good Um, I think software is getting question's gone from like, will will agents use and what So that's where we're at now.

Mark: 33:05

It's interesting on the Atlassian thing because as much as people go, because one of the first things like you and I, we were like, oh, we can build our own CRM, we can build our own workflow tools and Kanban boards. And like, I've been still using But the fact that, um, the have seen so closely, it turns And yeah, because they're things down and say things and turn out that you actually need. They, they work better when So it's like, oh, they do need And this is the amazing kind of So the comeback of folks like, and I'm not saying that I, because I'm no investment advice from, from my side, not licensed, but, um, relooking at things in that different way, like Atlassian actually could be valuable because software tools actually need that sort of thing. And the learnings that they have It's, it's fascinating, but it's Um, you kind of have thoughts on where this, this kind of ends up think about software.

Arms: 34:07

Yeah. So I think like the, I replacing software was around um, you know, having using AI people anymore, so you're not And that whole seat based model, Yeah, that's the first one. And then I think the second one that's good enough, then you're competitors in your spaces. So those gross margins of like talking about, maybe they go to I think that second question is But that first question, I think Atlassian result recently is the users expanded by ten percent. So that was like a direct basic compression fears. Um, and I think it's because like, yeah, they, so they've been really smart in, they have this, uh, collection of products in confluence. Jira. Yep. Um, loom and Rovo, which is like And the users who use that do credits, uh, and they build So yeah, to your point, I think tools, potentially agents do But then there's like software human typing in instructions. That's probably the ones that

Mark: 35:31

Less valuable.

Arms: 35:31

Yeah.

Chris: 35:32

It's the, there will always be a space for, like I said, like it's going to be an ebb and flow in terms of what is going to be available for SaaS products, because there is even if you're using AI to build your own agents to build your own tools and systems, there's still a cost associated with infrastructure that goes with that. Tokens, servers, GPUs, depending if you're running things locally or not, and those costs can be quite substantial. And so the balance of those costs versus going with a SaaS product that already absorbs all that cost and you're paying twenty bucks a month is becomes a no brainer in a lot of those instances. So SaaS is never going to go It's going to be eaten up, like you said, by a lot of the things like the agents and the automation and all those flows are going to be absorbed for the businesses that just can't, you know, create that stuff and innovate as quickly as AI is doing. Yeah, but it's going to stay there because of just because of pure costs associated with everything.

Arms: 36:20

Yeah. I mean, that's an interesting point because I think in the listed world, um, we're more invested in the infrastructure side than the software side. So, and I know like people might think, oh, like Nvidia, isn't that kind of overdone now, but Nvidia's on thirty times its growing earnings still by fifty percent. And it's still doesn't really have a credible competitor in that sort of edge computing case. Um, so I think that's fine. But it's interesting you're side, those profit pools broaden memory and compute and, um, So we're super enthused by Like the biggest position in our portfolio is SK Hynix and Micron. Collectively, they're eleven

Speaker 7: 37:03

Yeah.

Chris: 37:04

Memory being.

Arms: 37:05

Memory being like high

Chris: 37:06

As a hydrate.

Speaker 7: 37:07

Yeah, yeah.

Arms: 37:07

HBM and you know, here's some interesting stats for you because I think these are really cool. Um, Samsung Electronics and SK one and number two most world next year.

Chris: 37:18

Not surprised.

Arms: 37:19

And then the top three micron, three memory companies in the of high bandwidth memory. Um, they if if the high correct, they will outearn the

Speaker 7: 37:35

You're kidding.

Arms: 37:36

No. I like blows my mind a bit.

Speaker 7: 37:39

And those are huge companies.

Chris: 37:40

Chips have gone up so much. Like so I building all my own systems or I build my own computers. I've got a bunch of laptops and I have all those running my own hosted servers with them for and different things like to keep stuff local. Um, and one of the laptops, I RAM for this laptop because it's And so I went shopping. Yeah. What was the sixteen gigabytes to upgrade this thing. And I was like eighty bucks. We used to spend one hundred and So I'm like, alright, one hundred and fifty dollars, whatever. But then in my head, I was like, gigabytes, one hundred and fifty computer set up.

Mark: 38:12

What does that look like?

Chris: 38:13

Which is like got sixty four know, forty, forty, ninety RTX. It's a nice little AI Um, I was like, I built this Like the ram, I think I spent three hundred and eighty bucks on it at the time for sixty four gigs. I was like, how much is it now? And I looked online. Yeah, twelve hundred dollars. Yeah. So in three years, it's gone up four times the price that I'd originally.

Mark: 38:36

And now it looks like gold.

Chris: 38:37

And it's like it's. Yeah. And it's, it's crazy like So I'm not surprised that helix and all these other because the, the one they're not the costs of like quadrupled. Yeah. And three, the demand is up

Mark: 38:53

Well, speaking of Samsung, this TV here, this is like the largest single screen Samsung TV in the southern hemisphere as far as I was aware, like two years ago. Yeah. Somehow they've got it in this We're in this CNBC headquarters

Chris: 39:06

At home, the FCC, the the Square is now. So second times the size just

Mark: 39:13

So. So second, but still cool. Samsung executives have actually to check it out. And that's so cool. Yeah. So there's close ties to those So, so great to see that the Um, but you know, listed AI infrastructure kind of companies like Nvidia. Do you think it's, um, do you to the investors in the room investor crowd that's watching Can you say anything about, um, overvalued or thoughts on that?

Arms: 39:39

Yeah. I mean, obviously the like there's no tomorrow. I think they're talking seven hundred billion in Catholics in total.

Mark: 39:47

Uh, I, Bob have.

Arms: 39:47

That this year.

Chris: 39:48

Yes, please.

Arms: 39:49

I think if you listen to them see a risk of overinvestment acting like the biggest risk is given how much you've seen this And I don't know any chart that you see of like token usage, like everything is skyrocketing, particularly in this world of agents and, you know, like obviously, um, Minotaur capital is like one, a data point of one. But for us, we have already usage this year than we did for MM. Um, and you know, I think recently and, uh, one of our get spending too much on, on, you, are you gonna look.

Chris: 40:33

At the profits we've had on the Yeah.

Arms: 40:35

And I was like, so yes, we have, we've forex our spend, but it's still way less than hiring an investment analyst.

Chris: 40:42

Yeah.

Arms: 40:42

The benefit that's the relative. Right. That's the that's how we do the Like it's, you know, an us, you know, two hundred grand. And, you know, our AI compute spend is still well, well, well, well, well.

Mark: 40:56

There's a lot of room for analyst, uh, type, type, digital So that's, that is fascinating. Um.

Chris: 41:05

But like, so with, with what you're seeing in the air space for. people either in your space or who are moving into this, you What kind of advice do you have for those people jumping into that?

Arms: 41:20

Yes, I think we spoke about it your head in the sand. So there's like still quite a few, like AI naysayers who are like, I just don't want to use it. And, you know, I've done, you conducted fundamental equities the last three decades. So why should I change it? And the thing is, like the So, you know, you still AI still initiation report for us, which look into a company and But now, like, remember when we analysts and doing an initiation of months and now it's built in It's crazy. And there are different, I someone who's trying to get into get a large language model to You do have to experiment very blunt question, like give it tends to come up with, you rating like a buy or sell. And then it finds all the evidence to support that in that order. Whereas, you know, your better way of asking that question isn't like, give me an investment view. It's saying, okay, what would need to happen for Tesla to fall fifty percent? Like what are the pathways to

Chris: 42:36

It gives itself bias in that

Arms: 42:38

And then you say, okay, well, so that's the falling fifty percent. What would need to happen for Tesla to double like that and give me the pathway pathways for that? And then tell me what you think probabilistically out of those two scenarios, you know, would happen has the potential to happen more like, and that's you've got to ask more nuanced questions.

Mark: 42:55

More your work or your actual process rather than Oracle AI know, it's like, here is my process. Do the work for me. So it's assistance.

Arms: 43:04

Yes, yes, definitely treat it a portfolio manager.

Chris: 43:08

And obviously, being in the space and doing a lot of investment in space, what are you seeing as the biggest trends around AI? Um, whether it's adoption, up, etc., or scaling?

Arms: 43:18

Yes. I think that, um, I think I think on the enterprise side, there's too much like individual AI happening. So people in enterprises are using different bits and pieces of tools and not integrating that. There was, um, A16z, Andreessen

Chris: 43:36

Um, love those guys.

Arms: 43:37

Shout out, they, uh, do this really interesting blog on AI on this exact problem and an institution, you really need So I think that at the moment is, is something to, to watch in your organizations. Obviously, AI adoption is going Like, as I said, like there's Uh, we were in, uh, Shenzhen a couple of weeks ago and like just the, the Chinese guys are just so enthused about everything happening in the AI world. And and technology still, like of a hub for technology. And like, that's still in this Um, and it's interesting, like the Chinese models are just so much cheaper than the, the Western models. And we, we're large language I think that's another sort of piece of advice for, for people is like, don't be beholden to one model. We use twenty different models in the back end and we switched it out. We don't like, you know, that earlier, we used to generate And then Claude got better. So we did it in Claude for a Like deep seek was really, Um, and you know, when we pass thirty five thousand articles, to do like the first pass of list and we put that list expensive model, like just to

Mark: 45:04

That's a good workflow and And I think like people when played around with it and realize that one of the biggest things and the trade offs there. And it's like, well, you don't or God, opus for everything and balance it out with the, you search retrieval of, you know, So we might use haiku or something simple like that, but we need more thinking for this one. So I think that's a really good

Chris: 45:34

To also understanding the costs instance, with the stuff that I Perplexity has a high reading cost, whereas OpenAI has a really low reading cost. So I will pass articles that then do the conversion or the there I'm saving almost like a long a set of articles.

Mark: 45:58

And then you go deeper.

Chris: 45:59

And then you go deeper, interrogate, I will pop out to a different model.

Mark: 46:02

I was just gonna say, you go Hang on a second. I could just build the search model myself or something like that. If it works out in terms of the metrics and stuff, because, you know, isn't it like Claude Code does? Like they use Claude code, in their marketing, but I could Claude code to build new Yes. So it's this snake eating Kind of, but in a good way. Yeah. I don't know. A better analogy and stuff are warranted there, but do you have any other kind of views on, um, for folks that are building in the community or like from, you know, that lens of entrepreneurship, people thinking, oh, it's too late, it's really hard to get into this or I don't know what to actually build and stuff like that. I think we're, I believe we're but what are you kind of have as watching out there?

Arms: 46:46

Yeah, I think just So I, I did an off site for, you know, a big firm a couple of days ago and I said, I'll just engage kind of the AI fluency in the room. I was like, oh, put your hand up if you use ChatGPT or another lens. And obviously everyone had their if you use more than one. And most people still have their But then I said, put your hands If you've experimented with an open claw setup and have a, you know, AI system platform kind of connected to your emails and your documents, etc., and there was two people who put their hand up. And that was that to me, blew my I thought everyone had tried an it's still pretty nascent.

Chris: 47:25

So it can be scary for people because I think you've also got the media around risk and data leakage. And so I can understand a little surprising that they wouldn't That's.

Arms: 47:35

Yeah. And I think just on definitely don't recommend

Mark: 47:42

Open call for Macquarie Bank.

Chris: 47:43

Yeah, yeah. Please check my bank account.

Mark: 47:46

So it's gone.

Arms: 47:47

And yeah, definitely don't information in, but like, but there's been a bit of a Um, but I like have found Um, and the time that I it Yeah. Again, I start with a, a small, um, seafood shop owner yesterday. Um, and he was just like, I was watching him answer, read and answer his emails and he was just doing so manually and it was like.

Chris: 48:14

Like stressing out, get out of He doesn't.

Speaker 8: 48:17

Do it.

Chris: 48:18

So then I guess then the question follow on question is the how. Like for someone who, who isn't start getting into it?

Arms: 48:25

I think it's like as simple as like opening codex or, um, and saying, hey, how do I build an agent to do this workflow for me? Yeah. And then trying it to do that. And then skills is like the So I don't know if your audience is familiar with what a skill is, but basically, you know, Claude came out with these skills. I think it was like July last Mhm. Um, and skills, the way that I like this is how I explain it to like, you know how you watch The matrix before and neo like

Speaker 8: 49:00

Yes.

Arms: 49:01

What skills.

Mark: 49:01

Exactly?

Chris: 49:02

I know kung fu.

Arms: 49:03

Yeah, exactly. Like, I know financial modeling We have built our own. We've got, like, I can't even And every time we do something repetitive, we just make it into a skill. I think start learning how to, Um, biggest kind of piece of definitely don't buy skills. Like don't go to the skills marketplace and, and buy a skill because you just don't know what's embedded into that, like what bad actors are embedded into it. So I highly recommend not doing Just get, get up, get okay with, with how to write your own skills.

Mark: 49:39

It'd be really interesting if injected kind of skill. Um, you know, those things like sounds like it's saying but if you take every first going down, it's saying, um, something like that. Like it'd be really interesting. You know, you do have to be so careful with all of that kind of stuff. But I think, you know, you can get inspiration from those places and then just go and try to figure out and build your own and stuff, but most importantly, experiment. So I think that's a key kind of You do a lot of, um, we're doing a lot of talking here, but you get invited to speak at like AFR summits. Um, equity made shout outs to And we were just at an event uh, there was, uh, some other from relevance AI to other the space as well as yourself. Uh, um, my question is like, how all of these talks that you do Anything that you can learn some

Arms: 50:39

Yeah. And this goes back to your but I like get AI to help me, So, uh, every time I do something like that, like the Galileo Ventures one, like they've obviously sent me kind of a list of questions or thematics they want to cover beforehand. And I now have like literally a whole repository of every podcast, everything that I've done.

Chris: 51:03

Wow.

Arms: 51:03

The brain is the brain back to And so I say, okay, here are the I get their wants. And this is the sort of audience Can you nuance some of the last three years to, you know, want to hear? And obviously it's still all me. It's still all my IP stuff. That's, that's literally how I prepare for, for stuff like that. And I think if you are a builder in the AI world and you want to get your product out there like it is worth doing these events and putting yourself out there to or being on podcasts like yourselves. Um, I have, I famously don't say Uh, and look, you gotta watch Don't, don't burn yourself out. Uh, I also have, like, a four So it's like, oh, my God, I have Um, but it's also like, I Like I could talk, as you've Yes. Um, but I think it's also you Like, you know, you told me I'll go and do some research on can potentially incorporate that Minnesota as well. Um, it's always great to talk to I actually don't go to enough I like shout out to, um, you've

Mark: 52:15

Yeah. Australian blockchain and So shout outs to the new folks Um, because it was called We always did AI related things, but the name never really told that story. So and we used to run the Data So yeah, we'll have to get you And you've got, uh, you're going

Chris: 52:32

I am.

Mark: 52:33

Yeah, yeah. When is.

Chris: 52:34

That. Everyone's going to Japan I know the place to be.

Arms: 52:36

Uh, going in a couple of weeks. And so CLSA, um, which is like a boutique investment bank they've asked me to present on how, uh, how an investment firm can use AI to build a disruptive investment strategy. And it's really funny because I'm effectively going to be presenting to, you know, what people would say on my competitors. But like I, we are so open and Like we, I like to say that and quant and fundamental like this Like this is the fight against So like, instead of, you know, people were just investing in indices. I think AI has the ability to make active funds management great again, because you know, you can.

Mark: 53:20

Let's get that on.

Chris: 53:20

A hat. Yeah, not.

Mark: 53:21

A red.

Chris: 53:22

one. Make funds.

Arms: 53:23

Great. Yeah. Make active funds

Speaker 8: 53:28

Yeah.

Mark: 53:29

You're the opening act for you As you were talking before in of the preparation. Um, you've got an interesting Who's that.

Arms: 53:40

Oh yeah. SoftBank is going to be So you know Masters Masters

Mark: 53:47

Yeah that is amazing. Like you after he's done it's Um cool. Uh, we'll talk about the four our fun and stuff. Because, you know, obviously, if you guys have seen we work, you know, there was some famous investments that they've done there. But, um, you know, speaking to folks like that, how long are you actually like there for, for.

Arms: 54:06

And we're there for a week. And you know, what the kind of broader point around like presenting. So we've done a lot of this presenting and this is going to sound really maybe a little arrogant, but we've spoken to so many different people around the world. And I think what's interesting is we're still quite we seem to still be quite far ahead of everybody. Like, um, there was a guy who is part of like one of the biggest Japanese asset managers who I think we'll see when we're over there. And he runs like something like, I think it's two hundred and fifty billion dollars US dollars. And he's asset management firm. He's written a book on how generative AI is going to impact asset management. And he said that we're the most advanced people he's seen around the world.

Mark: 54:48

So he's got good vision.

Chris: 54:49

Like, yeah, literally Australia

Arms: 54:52

There's no tech that comes out

Mark: 54:54

Absolutely, sir.

Arms: 54:55

And like, obviously like the billions of the world and doing But I think we're pretty There's a hedge week, which is like a, you know, a hedge fund publication globally. They did an article about like way in terms of AI. And so we feature heavily there.

Chris: 55:12

Amazing.

Arms: 55:13

And then actually, I should say Minotaur, like in terms of nineteen percent per annum. You know, the market's done So I do think that AI can help, um, with the alpha generation of performance.

Chris: 55:27

I mean this is exactly why we're here to showcase the Australian businesses who are innovating and working on incredible technology. But, um, it's been amazing to Thank you so much for joining Amazing content, amazing Hopefully people will have a lot Um, yeah, that's it from us. Thank you very much for watching