Futuristic #35 – DeepSeek, Deep Research

Futuristic 35

[00:00:00] Steve Sammartino: What are you doing here, Steve? Sorry, I’m wasting your life.

[00:00:10] Cameron Reilly: This week on The Futuristic, tries to tighten up a microphone.

[00:00:16] Steve Sammartino: Yeah, yeah, that’s exactly what he’s doing and he’s failing massively. What’s going on with this? Why is this doing this? I did it, man. So good. Cameron, I did it.

[00:00:29] Cameron Reilly: You did it.

[00:00:30] Steve Sammartino: Can you hear me now? Is that good? Is that working? Cause I’m bad with microphones. It’s one of my

[00:00:35] Cameron Reilly: Yes.

[00:00:36] Steve Sammartino: suboptimal performance arenas.

[00:00:39] Cameron Reilly: Welcome back to The Futuristic, ladies and gentlemen, boys and girls. Um, this is episode 35, we’re recording this on the 7th of February 2025. Apologies, I’ve had so many people in the last month, Steve, say What’s going on with Futuristic?

[00:01:00] Cameron Reilly: Uh, there’s all this stuff happening and you guys aren’t explaining it to me and I’m like, uh, you know.

[00:01:06] Steve Sammartino: Steve,

[00:01:07] Cameron Reilly: Steve, Steve got sucked up into an AI vortex. Oh, hold on. I’ve got something. Hold on. Hold on. I nearly forgot. Um, there once was a chap, Samatino, who predicted the future bravino. He learned how to trade and he never got played.

[00:01:26] Cameron Reilly: Now money just flows like the vino.

[00:01:30] Steve Sammartino: he’s so fucking good. I wish you did that. It’s

[00:01:34] Cameron Reilly: Yeah. Well, I did tell the AI to do that and we went through like five or six drafts until I was happy with it. So, uh, it’s basically

[00:01:45] Steve Sammartino: So that’s a really, that’s, that’s a really important thing. People, I renovated a house once and people say, did you renovate it?

[00:01:53] Cameron Reilly: Mm

[00:01:54] Steve Sammartino: And I said, yes, they said, so you did it, you did all the bits, I said, no, no, I said, I renovated it, alright, because without me, it wouldn’t have happened. Don’t worry about who had the hammer in their hand, worry about the fact that the house went from this, to that.

[00:02:07] Steve Sammartino: And we need that mindset with AI.

[00:02:09] Cameron Reilly: Yeah. Well, Steve, it has been an absolutely bonkers month since we last talked. I, somebody was asking me, I think yesterday, about Futuristic and, um, I was like, like, every time there’s, I think, oh, we should do a show about that. A new thing happens the next day and it’s like, Oh my God, now we have to do something about that.

[00:02:31] Cameron Reilly: It’s almost like this needs to be a daily show now. There is so much going on. It is really, you know, I think I said this in our last show about the NVIDIA launch late last year when my jaw was on the floor listening to all the stuff that Jensen Huang was saying that the stuff that they’re coming out with.

[00:02:53] Cameron Reilly: And it’s been like that. For the last, ever since we last spoke, almost on a daily basis, I’m like, holy shit, I can’t believe that just happened. So, we’ll get into the news, there’s a lot to catch up on, we can’t cover everything, we’ll cover some of the highlights, but before we do that, Samatino, who flows like the vino, what, uh, tell us a little bit about what’s been happening in the world of futurism for you in the last month or so, since we last talked.

[00:03:28] Steve Sammartino: More corporate panic people ringing us up, we need someone to explain how this AI works. That. But the one thing that was interesting, one client. That I’ve been working with for a while, who is really bad at doing stuff. Some of the tools are so easy to implement. Now they crossed a little chasm in just saying yes to stuff because it’s an easy implementation.

[00:03:50] Steve Sammartino: It doesn’t seem to be a great deal of infrastructure. A lot of plugins, pretty easy. They’re working on a chat bot, not, not a new idea, but one that answers phones and gives specific directions to customers when people can’t answer phones in warehousing and that kind of situation, which they miss something like 7, 000 calls a year.

[00:04:08] Steve Sammartino: And they’re It’s a big organization with lots of outlets and all that type of stuff. I mean, it’s, you know, a handful a day in each location, but it adds up. And we, we figure that in the average order that they get is something like 700 bucks.

[00:04:23] Cameron Reilly: Yeah.

[00:04:24] Steve Sammartino: you know, you, the upside there is, is pretty huge. You know, we’re talking.

[00:04:29] Steve Sammartino: About big revenue. So I used, uh, ChatGPT in the verbal sense to design a chatbot script and conversation flow on 10 different types of topics. But I was literally working with it verbally in the way that you would work with a staff member.

[00:04:53] Cameron Reilly: Mm hmm.

[00:04:53] Steve Sammartino: Imagine you’re working in a wholesaling organization that sells these types of products with these types of customers.

[00:05:00] Steve Sammartino: Here’s what we need to do. Together we need to design, uh, chatbot scripts. Um, that’ll include the persona of the chatbot, the way it speaks, its verbiage and turn of phrase. Uh, the type of answers it has, the things it can answer, the things it can’t answer. I need you to remember all of this chat because you’re going to go away after the chat and take the pieces of it and put it into 10 different scripts explaining what the brief was and our outcomes and then the scripts of the areas.

[00:05:28] Steve Sammartino: And I need you to go into the AI that I built for these guys, I built them at GPT, and get Uh, the different types of products with the classic examples and substitutions and when people will call you back and when to ask for a phone number or an email blew my mind because I was doing it verbally. I haven’t got access to Operator yet.

[00:05:50] Steve Sammartino: But I thought, why don’t I try and describe what I’m doing while I’m doing it verbally, and give it instructions which are part of the verbal. And then it went back and it sent through the report after it. Um, it started writing up. I said, okay, now do it. And it was like, it was like 60 seconds. It had all the scripts, these 10 different scripts.

[00:06:08] Steve Sammartino: We talked for about 10 minutes and that was about 60 seconds doing it. Blew my mind though.

[00:06:14] Cameron Reilly: how are you going to do the voice on the end of the phone? Are you

[00:06:17] Steve Sammartino: Yeah, we’re going to use something like HeyJen and develop a voice, develop an Aussie style voice. Uh, for it. And you know, it’s an introduction. They even come up with really good things that said, Oh, I think the guy should be called Trevor. G’day, this is Trevor. Uh, I’m the, I’m the AI from And it was just really good because tradespeople are the customers.

[00:06:34] Steve Sammartino: What I thought was, even though we don’t have agents right now, there are kind of quasi ways to invent agents that aren’t really agents. By the way you speak to it, you can create an agentic output through careful briefing.

[00:06:53] Cameron Reilly: Yeah. Well, that’s very cool, man. That’s very cool. And this is definitely going to be the year of agents. If any, if, if January taught us anything, this is going to be the year of agents, as we predicted in our last show of last year. Um, well, for me, literally, uh, about two hours ago, I finished phase three of the project that I’ve been working on for the last six months.

[00:07:22] Cameron Reilly: So, my QAV podcast, I’ve talked about this before, we’ve got a very, sort of, Somewhat complicated checklist process that we go through to create our buy list. A year ago it used to take me four hours to generate a buy list. I wrote some scripts, now it takes me, um, 15 to 20 minutes. But the scripts are running in the background.

[00:07:48] Cameron Reilly: What I’ve been working on is writing a script that will get out of Excel. Because it’s still done in Excel. And just do it all. in code, with one click of a button. Not so much for me, but for our members, our club members, QAV members, so they’ll just be able to click a button and it’ll generate a buy list for them using all of our back end tech.

[00:08:10] Cameron Reilly: So I’ve been working on this using mostly Claude 3. 5 Sonnet for the coding. And I finished Phase 3 today, so Phase 3 was, Phase 1 I finished a day or two ago, which was to take a single stock and be able to generate a score for it, a buy score.

[00:08:29] Steve Sammartino: Yeah. Right.

[00:08:30] Cameron Reilly: 3 was to have a list of 10 stocks. And to be able to do it for the 10 stocks and have it exactly match the scoring that I do when I do it manually in Excel.

[00:08:41] Cameron Reilly: I, I nailed that this morning. It went from phase 2 to phase 3 was like 24 hours. Phase 4 is to be able to do it for the full list of, um, 600 stocks. Um, and to be able to, uh, Generate all the other scripts that I’ve got going. So anyway, I’ve just started working on that. But it was so cool. Chrissy’s been laughing at me because every day for the last week, I’ve been going, I’ve just got to crush this one more bug.

[00:09:08] Cameron Reilly: I’ll be, I’ll be on my laptop at like 2 a. m. She was, what are you doing? I just got to crush this one more bug. And it’s always one more bug. And I said to her, she like, she left home this morning. She goes, I’ll come back tonight. You’ll still be trying to crush one more bug. And I’m like, no. And I, I, I nailed it, but it was an amazing feeling to have built this thing that has a lot of moving parts and to have coded that for somebody who’s not a coder, to have used AI to

[00:09:34] Steve Sammartino: You are now.

[00:09:36] Cameron Reilly: yeah, I am, it’s like building a house, right, I’m the guy that told it what to do, worked with it to debug it and get it right, it’s an amazing feeling, a feeling that Everybody will have, although it won’t be the same.

[00:09:50] Cameron Reilly: Like both Sam Altman and Zuckerberg I’ve seen in the last day or two giving talks where they’ve said, software development at the end of 2025 will look very different to software development at the start of 2025. And you know, today I, it’ll help me, but I still need to manage the process. I’m the project manager, right?

[00:10:10] Cameron Reilly: I need to manage the whole thing. It goes off script, it gets it wrong. I need to go, no, no, no, no, no, no. That’s you’re, you’re missing this bit. You need to do that. Where we get to at the end of the year will be, Hey, uh, build me an app that does X, Y, and Z and test it iteratively and don’t report back until it’s working flawlessly and producing this result.

[00:10:31] Cameron Reilly: Right?

[00:10:32] Steve Sammartino: Right. I mean, yeah. And, and it’s so interesting that you mentioned that software development would be so different. You know, less than 12 months from now, it’s already quite different to what it was six months ago. For me, that hearkens to some of the stuff we’ll talk about today with AI investments in that three, four, five years ago, everyone was saying STEM, code, you gotta learn to code.

[00:10:55] Steve Sammartino: And now, eh, do we? Of course, we’re going to need software developers, just fewer of them. Uh, I also wonder if this giant investment in chips, oh, it’s an arms race in chips. It’s like, well, is it? Because there are unexpected curve jumps. Right, and forks, and that is one of the things that is always surprising, and we just don’t seem to learn that there are forks that change things.

[00:11:24] Steve Sammartino: One of my favorite ones to refer back was in the 70s, the population bomb was the big thing. We will never be able to feed 5 billion people, or 6 billion, and all of that kind of stuff, and then now we’re in the opposite direction.

[00:11:37] Cameron Reilly: Peak, peak oil in the mid seventies. We’re going to run out of oil.

[00:11:40] Steve Sammartino: these things, and we, we managed to find ways to circumvent them, uh, so it’ll be really interesting with tech, with energy, with chips, with food, with various forms of abundance that we’ve spoken about, gee, we’re getting in deep, but, but I find that comment on software an interesting thing to look back on where we were with kids, develop code and so on, and they were like, okay, well, maybe things have changed.

[00:12:07] Cameron Reilly: Not maybe. Things have definitely changed. Um, so let’s talk about some of the, Oh, I’ve got an app I’m working on too, that I’ll talk to you about off air. Okay. Because it’s a little bit, uh, Secret Squirrel, but I’m building a thing called the FutureMaker app that, um, I, I, I plan to go out and raise financing for

[00:12:28] Steve Sammartino: Ooh, exciting.

[00:12:30] Cameron Reilly: Yeah, yeah.

[00:12:30] Steve Sammartino: Podcast off, cut it. Let’s go to the, let’s go and talk about it right now. Exciting.

[00:12:35] Cameron Reilly: It’s one of my other coding projects I’ve been working on for the last couple of weeks that, um, I want to share with you. So anyway, let’s talk about that later on. Uh, so look, big, lots of big announcements, um, but let’s just tackle, I, I think one of the biggest thing that’s happened in the last couple of weeks was the launch of DeepSeq R1. Now, that absolutely, uh, Blew everyone away, um, and took the market by surprise. Wiped a trillion dollars off of the U. S. market. Uh, I don’t think NVIDIA has recovered yet from that hit.

[00:13:15] Steve Sammartino: Let’s go to the tapes on that.

[00:13:17] Cameron Reilly: so for people that, Haven’t followed that for whatever reason. There is a Chinese company called High Growth Flyer Hedge Fund or something, I can’t remember the full name, but it’s basically a quant firm, a hedge fund that’s a quant firm in China.

[00:13:35] Cameron Reilly: It’s been around since 2016. Run by, run by a young, uh, super smart guy. And for people who aren’t investors, a quant firm is basically a firm that uses software and AI to analyze millions and millions of data points of stocks and sectors, and then use the software to not only decide what to buy, but usually to trade very, very quickly.

[00:14:02] Cameron Reilly: And they’re, you know, using. Various, um, tools to get in and out of the market quickly. And sometimes you don’t need to make a lot of money on a trade, but you do it fast enough, it all adds up. So this company came out with their own ChatGPT, uh, style clone, um, called DeepSeek, but they came out with, uh, R1 reasoning one, uh, about, I don’t know, a little bit over a week ago,

[00:14:32] Steve Sammartino: Week and a half, two weeks. Yeah.

[00:14:34] Cameron Reilly: And there were two big things about it. Number one, it blew away all the benchmarks. So it was as good as the then state of the art models coming out of Um, OpenAI and Anthropic and Google, as good as if not better in some metrics. Secondly, they made it free, the web slash app version of it, so the same level of, Intelligence that you have to pay Google or Anthropic, uh, or open AI to get their highest level model.

[00:15:12] Cameron Reilly: They made it completely free, unlimited free. Their API they, the right, they had to pay for the API, but the rates were about one 30 of open AI’s API rates for oh one, uh, or four oh or oh one mini. So very, very cheap. The other thing was they, the, the. Story that was being spread was that they developed it for about six million dollars, versus hundreds of millions of dollars that, uh, the state of the art models out of the U.

[00:15:48] Cameron Reilly: S. Now there’s some debate over that, there’s some lack of clarity over that,

[00:15:52] Steve Sammartino: No one really knows.

[00:15:54] Cameron Reilly: Yeah, they had said that they bought 10, 000 NVIDIA chips a year or two ago, and people are, you know, using that as the basis for it. There are some suggestions that they’ve actually got a lot more. But, so there’s some three things there that China has caught up.

[00:16:09] Cameron Reilly: A Chinese firm, has caught up to the state of the art models in the US, at least in some metrics. It’s not multi modal, it doesn’t do video, it doesn’t do photos very well, it doesn’t do voice, you know, there’s still some gaps between what the state of the art models in the US can do versus it. But in terms of the text based chat and the reasoning, it was very good and outstanding because it shows you it’s Thinking process as it’s doing its reasoning.

[00:16:39] Cameron Reilly: And it still, yeah, after a week of playing with it, it blows my mind watching its thought process. I think, I think I sent you a text on day one. It came

[00:16:48] Steve Sammartino: Day one. Yeah.

[00:16:49] Cameron Reilly: check this out, dude. What, what have, what have your, what are your thoughts on it’s, uh, thought process?

[00:16:56] Steve Sammartino: That for me was the, it wasn’t just astounding. It was fun. It actually made something that is just, here’s an answer. It actually. It kind of anthropomorphized it far more. And it kind of helped you see inside the neural network. What I’m really interested in is what are the layers of abstraction that it’s doing that in?

[00:17:23] Steve Sammartino: Is it actually thinking in English in pictures or in the words or in the code and then translating the code to the words? I actually, I don’t know the answer to that. And I just, I was so astounded by it, but I feel like in a way that opens it up to different kinds of references. And the ability to find a hallucination, which might, you might not be able to find, if it just gives you a single answer.

[00:17:52] Steve Sammartino: That was what I thought was significant, but I loved it. Oh my God, did I love it.

[00:18:00] Cameron Reilly: To give people an example, um, here’s a question I asked it a week ago. I asked it, I said, when was Bowie born? And here’s its thinking process. Now, this isn’t its answer. This is its internal thinking process, but it exposes that to you. Um, so it says, okay, so I need to figure out when David Bowie was born.

[00:18:19] Cameron Reilly: Let me start by recalling what I know about him. He was a famous British musician, right? I remember he had some iconic personas like Ziggy Stardust. Now his birth date, I’m not exactly sure, but I think he was born in the 1940s. Maybe around 1947? Wait, wasn’t there something about him passing away a few years ago?

[00:18:37] Cameron Reilly: I think he died in 2016, and I remember hearing he was 69 at the time. So if he was 69 in 2016, that would put his birth year around 1947. Let me check that math. 2016 minus 69 is 1947. Yeah, that seems right. Now, the exact date. I think his birthday is in January, maybe the 8th? I recall some references to January 8th.

[00:18:59] Cameron Reilly: So putting it all together, David Bowie was born on January 8th, 1947. Let me make sure there’s no conflicting information. I don’t think he had a different birth name that might confuse things. His real name was David Robert Jones, but he changed it to Bowie to avoid confusion with Davy Jones from the Monkees.

[00:19:13] Cameron Reilly: So yeah, January 8th, 1947 in London, England. That seems to be the consistent information I’ve heard before. I don’t remember any sources contradicting that, so I can confidently say his birth date is January 8th, 1947, which it is. goes on to tell me. Um, now, I would have assumed it would have gone, well, let me, yeah, let me look it up on Wikipedia.

[00:19:37] Cameron Reilly: Okay, it was January 8th, 1947. Here is January 8th, but it, it doesn’t. It’s got this convoluted,

[00:19:44] Steve Sammartino: does,

[00:19:45] Cameron Reilly: anthropomorphic thinking process,

[00:19:47] Steve Sammartino: which is cool. It’s cool. For me, it’s cool for a few reasons. The reason that thinking process can be really interesting is that it makes it different to search by default, because if it was search kind of related AI, it would go to the web, go to Wikipedia, read it, and send it back. It makes me think a couple of things.

[00:20:10] Steve Sammartino: First of all, If it has all of that stuff in its memory, it’s behaving the way you and I might if we’re having a coffee around the table and we just don’t have web access and we’d go through that’s really almost exactly the way we would go through that. And in fact, I can almost. That there, think of conversations I’ve had about, Oh, when was this event or who was that?

[00:20:31] Steve Sammartino: Well, it was this year. Well, I was in university first year, which was 1994 or whatever. That, that exact process,

[00:20:38] Cameron Reilly: Cobain’s death. I always go, I remember I was driving to my first wedding when I heard that Kurt Cobain died. So, you know, I can sort of try and remember when I got married the first time. I roughly know how old I was,

[00:20:51] Steve Sammartino: Yeah. Yes.

[00:20:52] Cameron Reilly: year Kurt Cobain died, you know.

[00:20:54] Steve Sammartino: But what it does for me, that makes me really excited. If this is the process now, it could be performative is the first one, which is, which is interesting. You wonder if they do that as a way to kind of make it look amazing and blow minds. Which if it is, oh my god, what a stealth trick.

[00:21:08] Steve Sammartino: And yes, it’s working. But the thing that I get super excited about is this idea, we don’t hear this term much now, edge computing. Where something isn’t really connected to the web, it’s kind of looking at a hard drive of information and having a mind sitting on top of that. So I could, on this laptop I’m looking at right now, if I was not connected, but I had an AI sitting on top of which had.

[00:21:33] Steve Sammartino: A level of compute that’s big enough to do an AI within what my stuff is and my information. Uh, and if they can do the whole world for six million, maybe I can do Steve Sammatino for The hardware on a laptop, it might be able to go through a thinking process without any external help and anthropomorphize that.

[00:21:52] Steve Sammartino: And Steve’s asked me this, I’m going to go through his files. He was talking about this customer here. I remember he did a presentation. If I look through his emails, he’s got that. And then bam, give me the answer. That could be really cool.

[00:22:05] Cameron Reilly: a couple of the other things that are important to note about it, number one is it’s open source. So they made the whole thing open source. I downloaded a version of it that I can run on my MacBook, but it’s not very useful. Because

[00:22:20] Steve Sammartino: How does that, cause I was going to do that experiment. What did you do?

[00:22:25] Cameron Reilly: I

[00:22:25] Steve Sammartino: Like, like in terms of, no, in terms of. What you got it to do and how it works and doesn’t work like that. Like, how did you test it?

[00:22:32] Cameron Reilly: Um, I used, uh, gee, which one of the models, um, wasn’t typing mind. It was like, there are, there are platforms that they’re like ChatGPT for all, uh, or no, GPT for all, um, or I think I used Olama. So. Olama’s like a framework that enables you to download models from Hugging Face and then you can run them locally.

[00:22:55] Cameron Reilly: It does it all for you. It’s, you know, pretty much plug and play. You say, you have Olama, you say, which one, here’s a, Olama says, here’s a bunch of models you can download. Which one do you want? I tried one, first of all, that was like 36 billion tokens. Installed it, but it crashed my Mac because my Mac doesn’t have enough RAM.

[00:23:15] Cameron Reilly: My Mac’s got like 16 gig of RAM and you need about. 30, 40 gig of RAM to run it. So I crashed my Mac. So then I downloaded the 8 billion parameter model and it runs okay, but it’s kind of useless. It, it, I asked it some basic questions that it couldn’t answer very well. So, I mean, At this stage, it might be useful for some things, but it certainly wasn’t useful for anything I was trying to do with.

[00:23:41] Cameron Reilly: The, the, one of the big things is they made it open source. Obviously OpenAI and Anthropic and Google aren’t making their top tier, uh, systems open source. Meta are with their Lama models. And Sam Altman has since come out and said, I think we’re on the wrong side of history. And I think we need to start making our models open source.

[00:24:02] Cameron Reilly: And this is a thing that. You know, I, I know I’ve said this on the show. I don’t want to toot my own, um, horn. If, you know, if no one else is going to toot it, you know, you got to toot yourself. That’s what I always say. If you

[00:24:15] Steve Sammartino: I’ve been big on toot. You got to toot yourself. I’ve always said tooting yourself is one of the main things in life, I’ve always

[00:24:22] Cameron Reilly: need to work on your flexibility, so you can toot yourself. Um,

[00:24:25] Steve Sammartino: I

[00:24:27] Cameron Reilly: You know, when people, uh, say, oh, there’s going to be a handful of billionaires that are going to control all the AI and no one will have it, and it’ll only be for the elite and the rich, I don’t think that’s how this is going to play out.

[00:24:41] Cameron Reilly: I think there are enough forces in the world that are going to try, and a lot of really, really smart, People out there who are going to try and make this stuff as available as possible for everybody and open source, not just with software, but when we get to robotics and nanotech, and I’ve got a nanotech story to get to later, uh, those will be open sourced as well.

[00:25:04] Cameron Reilly: Now, they may not be the very, very top tier models that end up. Available, but there will be really, really good second tier models and systems that I think might be, there might be a lag between when the top tier models are available and the second, like, I’m sure DeepSeek, or someone like DeepSeek, like Quen, the, the um, Alibaba version, which also came out with a benchmark killing model a day or two after DeepSeek, DeepSeek.

[00:25:35] Cameron Reilly: They will come out with voice. They will come out with video. They will come out with, you know, all of the bells and whistles that the state of the art models have. They might just have a bit of a lag. The, um, the other thing that we want to point out about this, though, is, apparently, DeepSeek was trained on ChatGPT and as I saw one guy on X wrote, OpenAI stole from the whole internet to make itself richer.

[00:26:05] Cameron Reilly: DeepSeek stole from them and gave it back to the masses for free. I think there’s a certain British folktale about this.

[00:26:13] Steve Sammartino: love that. That is so good. That’s old Twitter right there. That’s old Twitter, isn’t it? Like Twitter used to be.

[00:26:20] Cameron Reilly: Shout out to Suspended Robot for that. So yeah, so they basically trained it, it seems, on ChatGPT, which was a lot faster than having to train it on, you know, doing its own, um, RAG or its own reinforcement learning. It just used ChatGPT to train it. So these models, like, we know that. OpenAI used, uh, used 4. 0 to train 0.

[00:26:47] Cameron Reilly: 1 and then they used 0. 1 to train 0. 3. Um, there are, you know, the other firms can use that, their models to train their models. So we have models that are being built by other models and other companies models and open source models that are built on the, on the closed source models. So it’s, I mean, it’s, it’s crazy.

[00:27:08] Steve Sammartino: way back? Two things, and I’m pretty sure you’ll agree on this. Pretty much everyone who had inordinate wealth stole something from someone. If the most profitable business strategy of all time is steal your, your raw materials, whether that’s data, people, labor, Raw materials, digging stuff out of the ground, whether it’s frankincense and myrrh or oil and gas, it’s all the same stuff, right?

[00:27:30] Steve Sammartino: Or land. Yeah,

[00:27:31] Cameron Reilly: Land grants. Yeah.

[00:27:32] Steve Sammartino: yeah, I mean, it’s just steel stuff and you get rich, but, um, it seems as though technology has had this particular habit of things developed in public, then it gets Um, it gets entangled by private forces who build barriers around it and, and privatize what was essentially something that was public, you know, make small literations and then privatize it.

[00:27:59] Steve Sammartino: And that kind of happens again and again and again. Um, and I feel like we’ve been in a deep cycle of the web has become something that a few powerful companies have kind of managed to in, in circulate that. Um, Keep the power. So I’m kind of hopeful that this is part of a, an opening up where we don’t have five big tech AI players.

[00:28:23] Steve Sammartino: I’m super hopeful on it. I hear that it came from

[00:28:27] Cameron Reilly: of it. We’ve got a good example. You know, if you go back to the mid nineties, you have, and we may have talked about this before, but I lived through this. I was at Microsoft at the time, you know, before the mid nineties, if you wanted an encyclopedia, you had to go buy Encyclopedia Britannica, which would cost you a thousand bucks, or you’d go to the library and, you know, borrow something or look it up.

[00:28:53] Cameron Reilly: Um, then. You know, for people who don’t know the story, Microsoft, Bill Gates, went to Britannica when they were coming out with the first Windows PC that was going to have a CD ROM on it, and tried to convince Britannica to make the encyclopedia available as a CD ROM, and Britannica said, no, we don’t want to do that, it would cannibalize our book sales.

[00:29:20] Cameron Reilly: So Bill went, get fucked then, and Bill quit. Microsoft Incarta, which killed Britannica’s business within a couple of years. Britannica was fucked. I mean, they’re still around as a brand and they’ve got an online thing now, but it just absolutely destroyed their business model. But again, and Microsoft gave it away for free for, I

[00:29:46] Steve Sammartino: They did with, they bundled it, didn’t they? It was part of

[00:29:48] Cameron Reilly: Bundled it, yeah, but you had to buy a microsite, you had to buy a Windows computer, and that kind of stuff, and then, Then they tried, when the internet came along, they tried to make it available on the internet as a premium service, but then, Wikipedia, Wikipedia came along, and basically, is still today, completely free and available, so it took, Britannica, so you go from Britannica, For profit model to encounter sort of a hybrid profit model that was bundled.

[00:30:19] Cameron Reilly: It’s like a freemium kind of a deal to Wikipedia, which has been completely free to anyone and and is user managed and moderated as

[00:30:31] Steve Sammartino: Brilliant. It really is. It really is.

[00:30:33] Cameron Reilly: yeah.

[00:30:34] Steve Sammartino: is probably one of the most, not as valuable now as it once was because I’m now using GPTs where I once used Wikipedia.

[00:30:41] Cameron Reilly: Yeah, and it’s been trained on Wikipedia too, obviously. But, you know, we do see that progression, so there is a, there is a template there. But the other thing, before we move on from DeepSync, I want to talk about is almost immediately after it came out and crashed the US stock market, um, uh, it got DDoS’ed, we think.

[00:31:01] Cameron Reilly: I mean, they’re saying they’ve been DDoS’ed. There’s also, you know, probably got 100 million users, um,

[00:31:06] Steve Sammartino: it did. I couldn’t get in, but I found it interesting. I couldn’t get in on the, on the browser mode, but I could get in through the app mode on the, within the same five minutes

[00:31:17] Cameron Reilly: And the API was shut down. So for the last week, week and a half, it’s pretty much been unusable to a large extent. They stopped new registrations. But, um, I just checked their server status an hour or so ago and for the last couple of days they’ve been all green for the API and the web. So they’ve, and I did see a story this morning that they’re moving to a big Chinese computer cluster that’s going to help them with security.

[00:31:43] Cameron Reilly: But it seems like they’ve got a massive attack from, you would imagine, someone in the West that was trying to, uh, Absolutely crushed them. So it’s, uh, it looks like it’s a bit of cyber warfare between the US or other Western alliances and China on the AI front. The first time we’ve seen

[00:32:07] Steve Sammartino: Just on that,

[00:32:08] Cameron Reilly: war.

[00:32:09] Steve Sammartino: IO warfare, just on that, given your geopolitical proclivities, your Cold War experience, you know, as much as anyone about this stuff, where do you think that would, if you were a betting man, where do you think the most likely place that that would have come from is? Who, who, what, what kind of, uh, Israel

[00:32:33] Cameron Reilly: I mean, Israel is basically a US, um, proxy for lots of evil shit. They’ve got a very, very high, um, tech capability in Israel. So yeah, combination of the US and Israel working in

[00:32:48] Steve Sammartino: at a governmental level, not a private industry level.

[00:32:51] Cameron Reilly: I imagine it’s probably a combination of both, um, you know, I don’t know if Sam Altman or Demis Hassabis or any of those guys are

[00:33:03] Steve Sammartino: Well, they would, they would do

[00:33:04] Cameron Reilly: Dark Hat

[00:33:05] Steve Sammartino: yeah, dark hat kind of,

[00:33:06] Cameron Reilly: but,

[00:33:07] Steve Sammartino: and, and that would take the form of dark hat lobbying. Hey, we know that there’s a risk here for you geopolitically. If you let them grow too much, maybe, yeah, that kind of

[00:33:18] Cameron Reilly: But the other thing I want to point out is I’ve heard a lot of people, you know, Online, I have to keep pulling people up on this. They assume that DeepSeek is owned by the CCP. And as everyone says, well, China’s doing this and China’s doing that. And you’re giving China your information and et cetera, et cetera.

[00:33:34] Cameron Reilly: And I’m like, look, there’s a lot of businesses in China. There’s a lot of people in China. There’s a lot of businesses in China. I think it’s a fundamental mistake to assume that everything every Chinese company does is being orchestrated by Xi Jinping at some level.

[00:33:50] Steve Sammartino: not, none of it is. Well, here’s what I think happens. And a good friend of mine ran a hundred million dollar business there. And once you click over a certain amount, they have a CCP member sits in your office. They, they literally that becomes a job, which is fine and good. I think what people think is that it’s all, it’s like conspiracy theories or let’s call them conspiracies that emerge or emergent and become comfortable and people want to maintain them.

[00:34:15] Steve Sammartino: I don’t think they’re ever designed to ground up. I think if they get to a certain area where it suits certain parties, they try and maintain them. And my understanding of the CCP is they let things grow organically. And then once something gets important and big enough, they go, Yeah, about that. We’re still in charge here now.

[00:34:30] Steve Sammartino: Let us just insert ourselves and use you. I think that’s a more strategic approach in any case. And, and

[00:34:39] Cameron Reilly: and, and not that different from how things work in the West as well, you know, we, we, we call it, you know, there’s regulatory bodies that Trump has tried to dismantle, but we have regulatory bodies that will get involved and have a look at what you’re

[00:34:52] Steve Sammartino: are significant enough and

[00:34:53] Cameron Reilly: when they’re significant enough.

[00:34:55] Cameron Reilly: Yeah, yeah, yeah.

[00:34:55] Steve Sammartino: yeah, that’s right. It’s the same there. And I think, you know, if the government bans it on Gov devices, which is really reminiscent of what happened with TikTok. First, they banned it on devices for the government.

[00:35:07] Cameron Reilly: that, but the TikTok red note thing that happened in the last month or so,

[00:35:12] Steve Sammartino: Well, we’ll talk about that quickly, but I think that if they do get banned in China, it’s, it’s, it’s economically useful for big tech in America

[00:35:21] Cameron Reilly: in the U. S.

[00:35:22] Steve Sammartino: Yeah, ban DeepSeek in the US if that happens, and there’s, there’s a non zero probability of that. It’s

[00:35:28] Cameron Reilly: Yeah,

[00:35:29] Steve Sammartino: almost a 50 50. It’s, it’s higher than that. So it’s, it’s really, it’s really convenient for big tech operators from Western markets to point that out to the government and get a beautiful little policy that, that suits them.

[00:35:45] Steve Sammartino: It thwarts competition, which may be bad for society in terms of opening up technology. But it’s good for the shareholders of the dominators who already exist, you know.

[00:35:59] Cameron Reilly: Yeah. So keep an eye on DeepSeek and Quen and the Chinese space. I mean, they are still dealing with the issue that they have limited access to the top level platforms and chipsets coming out of NVIDIA and TMSC and things like that. But, and, and these are non trivial, like I’ve, I’ve, We’ve read quite a lot on how hard it’s going to be for China to replicate the manufacturing process that TMSC uses to build NVIDIA’s stuff, and I think a bunch of it comes out of Sweden or Switzerland or somewhere like that, a lot of the technology for, um, Laser printing using lithography, light lithography to build these chips is, all comes out of somewhere in like Scandinavia, but, um, You know, on the flip side, China’s got a lot of really, really smart people, a lot of money, and a lot of focus, and they know how to get shit done really, really quickly when they want to, so I wouldn’t count them out, um, just yet.

[00:37:09] Steve Sammartino: Well, if you’re one in a million in China, you’re really just one in a thousand,

[00:37:14] Cameron Reilly: Yeah, that’s right.

[00:37:14] Steve Sammartino: right? So you’re going to have more smart people. If you’ve got more people, it’s easier to get them.

[00:37:21] Cameron Reilly: And they’re very, very good at, uh, Um, not only breeding smart people, like training them to be smart, but identifying them and isolating them and focusing them on stuff. You know, they’ve, they’ve, they’ve got a, they’ve got a system that’s been designed to find the smartest people in the country and get them to be as productive as possible.

[00:37:48] Cameron Reilly: You know, they’ve been doing that. very, very effectively since, you know, since Dong Xiaoping really

[00:37:55] Steve Sammartino: 79, round about that error.

[00:37:57] Cameron Reilly: 70s. Yeah, his whole thing was, you know, because, you know, during Mao’s years there had been this backlash against intellectuals and in favor of working classes and the proletariat and, you know, Dong was no, no, no, no, no, we need to get all the smartest people that we can and, and.

[00:38:16] Cameron Reilly: Laser focused them on making us the most powerful country on the planet, and they’ve done a pretty, pretty good job of that in the last 45

[00:38:26] Steve Sammartino: good. Who’s that off? That’s a comedian. He’s a pretty, pretty Hey?

[00:38:30] Cameron Reilly: Larry David.

[00:38:31] Steve Sammartino: Larry David. That’s, that’s, it is too. It is too. When he was doing, um, he was the New York Yankees guy. Hey, Georgie, it’s a pretty, pretty good little thing you got going there. Hey, just quickly on that. China versus US intellectuals and technology.

[00:38:47] Steve Sammartino: I think that America and Western markets do have sufficient institutions to gather smart people, universities, and so on, and industry. I think the big miss or the big risk that you have in Western markets is how people are influenced in what they value. And I’m talking about Kardashian world. The fact that the type of things that we look at and we value in the media.

[00:39:14] Steve Sammartino: Celebrity gossip and all of this kind of stuff that tends to influence a world, all of that stuff in the last 20, 30 years has become really difficult for us to, for Western markets to empower and Encourage the types of behaviors that lead to excellence is almost like this kind of war against it a little bit.

[00:39:44] Steve Sammartino: We, uh, I think our media has too much of an influence and we don’t shape society. Like, you know, what have we got? Strangers getting married on islands. And I haven’t seen TikTok in China, but apparently the things that are most viral are people doing amazing science experiments and all these. Whereas it’s, what did the Kardashians do in this country or who’s on Love Island?

[00:40:04] Steve Sammartino: I think that has an impact on the Western markets.

[00:40:08] Cameron Reilly: I think the other aspect of it too is, you know, less so here, but certainly in the US, is order to get to a top tier educational institution, and I’m not even talking about tertiary level, I’m talking about primary and secondary, you have to have money. If you’re, if you’re born in the backwaters of bumfuck Virginia,

[00:40:29] Steve Sammartino: Like you and, like you and I were.

[00:40:31] Cameron Reilly: yeah, your ability to get a top tier education, uh, and then end up as a result of that, You know, being part of a startup or being part of a top corporate and getting a job, you know, that is limited in China.

[00:40:47] Cameron Reilly: They actually go out looking for the smartest kids in the country at the youngest ages and go, Oh, you’re smart. We’re going to put you in this fast track thing because we need all the smart people in here, right? It’s one of the advantages of a. Socialist with Chinese characteristics system versus a capitalist system is it’s engineered to make the country as smart and as productive as possible because everyone benefits from that rather than, I mean, I know they have, there are, there are charter schools in the US and there are grant systems and that kind of stuff that poor people can get into, but

[00:41:24] Steve Sammartino: you also even got it, you wouldn’t even know about it unless you had proactive parents or you were proactive, you just, you just wouldn’t know. And even in Australia, too, our school, our schooling system is very geared towards people with money on the right side of town, and even if you just look at public infrastructure on the east or the west or the north or the south side, every city has it’s good and it’s bad side.

[00:41:43] Steve Sammartino: If you’re on the wrong side of it, you just don’t have the exposure.

[00:41:47] Cameron Reilly: Anyway, moving right along, um, because I got a doctor’s appointment I got to get to, um, OpenAI launched Deep Research in the last, I don’t know, week. Um, this is building on, so, so immediately what happened when DeepSeq came out is OpenAI just started pulling everything that they had out of the cupboards and making it available.

[00:42:08] Steve Sammartino: I love that. That’s funny. That’s funny. Panic mode.

[00:42:12] Cameron Reilly: right? Yeah,

[00:42:13] Steve Sammartino: what competition does. Competition is good.

[00:42:16] Cameron Reilly: And it was funny because they just did their 12 days of OpenAI before Christmas that we talked about in our last episode where they were launching lots of stuff that they had. But they saved the best stuff for, oh shit, DeepSeek just stole all of our thunder, we need to come out with stuff.

[00:42:30] Cameron Reilly: So, they came out, they made O3, which they had talked about before Christmas, they made it available on a limited level. But O3 Mini and O3 Mini High for coding you can get if you’re a pro or a plus user and you get more access if you’re a pro user. I’m a plus user, so that’s the like 20 bucks a month, whatever.

[00:42:51] Cameron Reilly: I’ve been using O3 Mini for coding and stuff. It’s their deeper reasoning model and it’s pretty fucking impressive, I’ve got to say. It really is a step up from O1. Really, really good. Still a bit too verbose for me when I’m coding though. I still prefer Claude. If I ask O3 to write some code or to debug some code, it’ll give me a fucking dissertation.

[00:43:16] Cameron Reilly: And here are 20 different things you could look at and 20 different ways of handling it and I’m like, too hard. Yeah, yeah. Claude, 3. 5 Sonnet, will say, oh shit, um, I think I worked it out. Uh, let’s, let’s try this. You know, it’s much more. Concise. It’s much more user friendly, I find, for that kind of stuff at this stage.

[00:43:38] Steve Sammartino: Yeah.

[00:43:38] Cameron Reilly: But then, after O3, they launched ChatGPT Tasks. Well, you can give it a task to do and it will do it automatically. And it took me a few days to come up with something, but, uh, I figured out a way to use it. Have you used tasks yet, Steve?

[00:43:56] Steve Sammartino: I haven’t.

[00:43:57] Cameron Reilly: What I came up with was, I gave it a job. I said, one of the things that I do for fun, whenever I need some intellectual stimulation, is I play the old, um, Debono random word association.

[00:44:13] Cameron Reilly: Game, right? I used to do it with my clients when I was running a marketing consulting business or if I need business generation I just get random words and I stick them together and I go what is You know, um, Apple, Apple Scissors. Okay, what does Apple make me think of? What does Scissors make me think of?

[00:44:30] Cameron Reilly: And then I try and come up with some business ideas that come into my head out of that. So I said to it, um, every morning at 7am, I want you to do 10 random word association exercises, come up with a bunch of business ideas based on those random word associations, then I want you to write Each of those business ideas against the following metrics, how easy would they be to do, what kind of capital investment would be required, what’s the size of the market opportunity, score them on about half a dozen metrics, then take the one with the highest score and write a one page white paper on what it would look like and how you would pull it together for me and send that to me every morning at 7am. And it did it!

[00:45:20] Steve Sammartino: That’s so good.

[00:45:21] Cameron Reilly: it didn’t do a great job of it, um, and I, I tweaked it as I went along, and I found that it kept using the same words over and over and over again, and then it would go, then it would use a random word with a job role, it’d say, um, Apple Lecturer, Apple Coder. Apple, you know, something, you know, Lawyer, it would come up with the same like job.

[00:45:46] Cameron Reilly: And I was going, no, no, don’t come up with job titles, just do random words.

[00:45:49] Steve Sammartino: Yeah.

[00:45:50] Cameron Reilly: And it, it, it was kind of lame. So I stopped it after about a week of doing it, but it was a way of getting it to do a thinking exercise or a task for me and just present me with something every day. So that was Apple Tasks, sorry, OpenAI, ChatGPT Tasks.

[00:46:08] Cameron Reilly: Then they came out with. Operator. Do you want to explain Operator to everyone?

[00:46:13] Steve Sammartino: Yeah. So Operator is their first agent that’s come to the market, which we’ve spoken about before, where you’ll go in and you’ll give it a task to go and do something for you, and it will take control of your desktop. and go and perform the pieces needed in the same way that you might do a task yourself.

[00:46:34] Steve Sammartino: Let’s say you’re going on a trip to Sydney and you need to book a hotel and a flight. It can do that. It, it does it, and I haven’t used it yet. I’ve just watched videos because you can’t get in Australia without, or you can do it with a VPN. I was going to do it, but I just watched some videos anyway to get a taste.

[00:46:51] Steve Sammartino: It opens a browser, so it’ll do it inside. It doesn’t do it in your browser, it does it inside their client. They have a few partner Uh, websites that they can, uh, work with, you know, like, recipes, develop a shopping list on Instacart to make me a meal that’s high in protein and is vegan, for example.

[00:47:12] Steve Sammartino: That’ll go, here’s, here’s some recipes I found, here’s the ingredients, uh, you might be able to say, I’ve got this, and it’ll tell you what to do and order it and get it delivered to your house. And it does, it’s not fully self directed, but at various points, it allows you to do logins if needed to go to external websites and different things.

[00:47:31] Steve Sammartino: Um, but it’s basically a thin version of an agent, I would say. I wouldn’t say it’s, uh, like a, a full agent model. It’s kind of, it’s pretty small scale from what I’ve seen. And I just have to tell users, I haven’t used it in the first person. I’ve watched people doing various experiments with it, but this comes down again to where we are with recursion.

[00:47:59] Steve Sammartino: Okay. In a month, where’s it going to be? It’s going to be twice as good. In two months, it’s going to be twice as good as that was. So it’s going to be 400 percent better. Yeah. By mid year, we’re going to have agents doing, or I reckon, almost anything. Is that your understanding of where it’s at?

[00:48:12] Cameron Reilly: yeah, yeah, and, and, to be fair to OpenAI and their launch, they said, look, this is very, very simplistic. And I don’t think they were ready to launch it, right?

[00:48:23] Steve Sammartino: they brought it forward.

[00:48:24] Cameron Reilly: Brought it forward because of the DeepSeek stuff. Um, but it does simple things on your behalf. It can go out and look stuff up, it can analyze stuff, it’ll check back with you if it’s got questions.

[00:48:37] Cameron Reilly: But it’s the beginning of being able to give it, uh, you know, a set of complicated things that it needs to do, and it’ll go off and do it by itself, and then come back to you when it has something to report back.

[00:48:49] Steve Sammartino: Yeah. Or needs your input. And you could say, don’t ask me unless it’s this, or ask me if you need logins or, uh, or you need me to spend money, or you can spend up to a hundred dollars each, each time with a maximum of 500 this week on these projects. You can, you can give it instructions. Uh, I mean, I, I think it’s, it’s the start.

[00:49:13] Steve Sammartino: Again, it’s, I think these have been available for a pretty long time. I’ve been using AgentGPT, which you can, you can get to buy things on your behalf as well. And BabyAGI and a few of those other ones. Uh, I imagine in a month it’s going to be twice as good and in two months, the recursion now, it’s monthly, sometimes weekly.

[00:49:35] Cameron Reilly: Well, daily, because a few days after they came out with Operator, they came out with Deep Research.

[00:49:41] Steve Sammartino: Yes.

[00:49:43] Cameron Reilly: Deep Research,

[00:49:44] Steve Sammartino: thing. Different thing, a little different, not really out doing tasks and production for

[00:49:52] Cameron Reilly: Well, it is, it is, yeah, it’s research. So, you

[00:49:57] Steve Sammartino: a research oriented agent, whereas that is, or the other one was, let’s call it, it’s not this, but let’s call it a consumer focused agent.

[00:50:06] Cameron Reilly: Yeah, a little bit more open ended. The research one, and again, I haven’t been able to play with it because it’s not available yet to everyone at our tier. It’s at the higher tier, which I should pay for because I probably spend 200 bucks easily on bloody the API. Um, You know, I was, it’s funny, like I, I’ve been using Claude to code with and I’m easily, it’s, with the API, I’m spending five, six bucks a day on it.

[00:50:34] Cameron Reilly: And I’m like, part of my brain’s going, oh shit, that’s really adding up. And then the other part of my brain’s going, you’ve got an expert programmer for five, 5 a day. And I’m spending like hours, six hours a day, eight, 10 hours a day, coding with an expert programmer for five bucks. And I’m like, really?

[00:50:54] Cameron Reilly: Like,

[00:50:54] Steve Sammartino: This is how we forget.

[00:50:56] Cameron Reilly: costing me five bucks a day.

[00:50:58] Steve Sammartino: It’s like when I watch whinging about how terrible modern life is. And yeah, there’s challenges, and not without challenges. Uh, now it’s can you afford a house? You know, uh, 200 years ago, am I, am I going to die of disease X in London because the water’s polluted? Yeah, cholera or whatever it is.

[00:51:17] Steve Sammartino: And here we are saying, should I really spend five bucks a day on someone with a PhD in all forms of code and language? And geez, it’s a bit much that five bucks a day.

[00:51:28] Cameron Reilly: Hmm. so deep research, you can, you can give it a job, um, and it will go out. And research, whatever that topic is, it’ll scour the web, it will then compile a PhD level report for you, that’s fully sourced, um, got all of the references. And, you know, it’s well written, well structured, and it will go and take half an hour or several hours to do the work, then it’ll give you the report and come back to it.

[00:52:09] Cameron Reilly: And I’ve seen, uh, some of the stuff that people have produced out of this. Somebody asked it to look at all of the executive orders that Trump has signed since the inauguration and do a, do a deep dive on the implications of each of them

[00:52:26] Steve Sammartino: Which ones aren’t? Yeah, I saw that. Here’s a question for you. With Deep Research’s ability to do PhD level research with sourcing on any topic of your choosing, does this make McKinsey More profitable or put them out of business. Over to you Cameron Reilly.

[00:52:45] Cameron Reilly: in the short term, more profitable, uh, in the long term, out of business.

[00:52:50] Steve Sammartino: Wow. You heard it here first on the Futuristic

[00:52:54] Cameron Reilly: Well, look, I think it’s the, I think it’s the same with, you know, any cognitive based profession, whether it’s legal or accounting, certainly software development, um, consulting, marketing consulting, business consulting.

[00:53:11] Cameron Reilly: Initially, People will still want a human to sue if something goes wrong, or a human to talk to and have a coffee with. And that human, the principals of the firm, or the partners of the firm will increasingly get rid of lower level staff and replace them with AIs. I mean, there’s still going to be a hurdle.

[00:53:35] Cameron Reilly: where we want to make sure there’s no hallucinations. You know, even people using deep research have said that there are hallucinations in deep research. So we’re going to have to get through that level. And I’m very, very confident that the AI firms will figure out how to deal with hallucinations. And one of the ways they seem to be dealing with a lot of the deep reasoning stuff at the moment is they have bifurcation of the systems.

[00:54:02] Cameron Reilly: Um, where you’ve got, you know, one system will do some research and it’ll feed it to the other system and that system will clarify that and go, hold on, that doesn’t make any sense and they’ll go backwards and forwards. I know that’s sort of how DeepSeek seems to have built its reasoning system as well.

[00:54:19] Cameron Reilly: So, first of all, we’re going to have to deal with hallucinations. We’re going to have to be sure that there’s like a 99. 999 percent chance that there won’t be any hallucinations in, uh, the output. Once everyone’s confident of that, we’ll start getting rid of low level people, and you might still have a mid level person that’ll be reviewing the work of the AI system, and eventually, once they go, yeah, it’s fantastic, you get rid of the mid level person as well, because you don’t need them, then you’ll just have the, the, the principals of the firms that are, And the sales people that are dealing with the customers, the relationship type people, until the customers get to the point where they go, well, what the fuck am I paying you for?

[00:55:00] Cameron Reilly: You’re just using an AI system to do all the work anyway. I don’t really need you. And then they will go the way of the dinosaurs as well. And the businesses will just use their AI systems.

[00:55:11] Steve Sammartino: Yeah. And it does two things. The first thing is it points out that secondary desk research now is something that’s going to be increasingly difficult to make a living out of. Primary research is always still important because there’s always these two kind of cohorts of research.

[00:55:26] Steve Sammartino: Is it something that someone’s done the experiments, learnt on, tested? Empirical research and then there’s desk research, aggregated research, finding what’s happening in industry stuff, where is, which is where most of the money is made. That layer on top are all the consulting firms and not, they don’t do primary research, they just aggregate information.

[00:55:45] Steve Sammartino: So that, Kybosh

[00:55:46] Cameron Reilly: even the primary, even the primary research now is going to be completely AI’d.

[00:55:52] Steve Sammartino: Well,

[00:55:53] Cameron Reilly: I need you to,

[00:55:55] Steve Sammartino: But, just quickly on that, just quickly on the, uh, the deep research, when it did Humanity’s last exam, which is the testing method, without going into great detail, it’s an incredibly complex exam in all sorts of levels of

[00:56:09] Cameron Reilly: built specifically for

[00:56:11] Steve Sammartino: for, for AI, yeah, it came in at 26%, which means within two months it’ll be perfect.

[00:56:17] Steve Sammartino: Because it’s gonna get twice as good, it’s gonna be 50 percent

[00:56:19] Cameron Reilly: Well I think

[00:56:20] Steve Sammartino: And then it’s going to be 95 to 100%.

[00:56:23] Cameron Reilly: the state of the art models before deep research were doing like 3 or 4 percent

[00:56:28] Steve Sammartino: I’ve got the numbers right here. I actually looked it up. I did my homework, uh, Cameron. So before Deep Research, DeepSeek got 9. 4%. ChatGPT 03 got 10. 5. Chat, uh, GPT 03 high. Mini got 13 and Deep Research got 26. 6. But based on that

[00:56:50] Cameron Reilly: doubled.

[00:56:51] Steve Sammartino: 183 percent improvement on previous scores. Well, you’re pretty much almost going to get there in the next iteration.

[00:56:58] Cameron Reilly: Yeah.

[00:56:59] Steve Sammartino: Certainly within two, it’s going to be 95 percent confidence level. And if that’s not AGI, I don’t know what is.

[00:57:05] Cameron Reilly: Yeah. Anyway, so that’s what’s been going on, and of course, the other thing that we have to talk about is that, uh, day two, I think, of Trump’s presidency, he and Sam Altman, uh, And Larry Ellison and the guy from SoftBank co announced the Stargate project. Um, the Stargate

[00:57:31] Steve Sammartino: Hmm.

[00:57:33] Cameron Reilly: is a 500 billion dollar project.

[00:57:36] Cameron Reilly: Now, a lot of people seem to think it’s the US government that’s funding it. It’s not, and Donald Trump really just had nothing to do with it apart from They were, they were all sucking Donald Trump’s Penis on stage. We couldn’t have done this without you, President Trump. You’re so great, President Trump.

[00:57:54] Cameron Reilly: You made this happen.

[00:57:56] Steve Sammartino: You’re the

[00:57:56] Cameron Reilly: I’m not sure, not sure about that. But, um, basically it’s going to build out, it’s a joint venture that’s going to build out data centers and electricity generation for AI, for open AI. It’s going to be built in Texas. There is going to be a hundred billion in the first trench and then, um, up to 500 billion over the next five years to build it out.

[00:58:22] Cameron Reilly: Basically, uh, because open AI just needs get out of fucking dodge size. data centers to run the billions and billions of Nvidia chips that they’re going to be putting into these things. I mean, it’s crazy. And on top of that, there’s been analysis that I’ve seen that, um, the US and China investments combined in AI will exceed 1 trillion by 2030 in the next five years.

[00:58:55] Steve Sammartino: Could it be, Cameron, for our Australian listeners, two of the big newspaper corporations here in the 90s, both built giant manufacturing plants at the peak of the old media world, one out at Tullamarine and one in Port Melbourne? Both of which were decommissioned. Yeah, they were never ran at full scale and were decommissioned long after.

[00:59:17] Steve Sammartino: It sounds like a crazy thing to say, but with what’s happened with DeepSeek, who knows that there might be some sort of a curve jump in terms of the way these models work. It could be a white elephant investment. This stuff can happen. I wouldn’t be

[00:59:32] Cameron Reilly: Well,

[00:59:33] Steve Sammartino: if, if this race for chips and giant data centers Become some sort of old world model.

[00:59:40] Steve Sammartino: Maybe AI goes, Hey, I’ve been thinking on all these chips that you’ve given me. You know, there’s a better way to do the things that I need so that I can help. You’ve got all these chips here, but did you know I can tap into the DNA in forests and use the information in their neural networks to, uh, code my information on little DNA strands?

[01:00:00] Steve Sammartino: Like, I know that sounds like crazy shit, but we know we can already store a lot of things inside DNA. So. Who knows? There may be some biological means with which data could be omnipresent within the physical Arbery infrastructure. I don’t know.

[01:00:16] Cameron Reilly: well we know how this plays out, um, the data centers will be converted to human storage tanks, like in the Matrix, and they’ll have humans in vats, and it’ll be using our brains to run the infrastructure on.

[01:00:30] Steve Sammartino: But Cameron, aren’t you and I already in VATS just now? We know this. We feel this. You feel that things are not, I mean, what is real?

[01:00:40] Cameron Reilly: Did you take the red pill or the blue pill, Steve, when you

[01:00:43] Steve Sammartino: If electrical signals served to your brain creates reality, then this is real.

[01:00:49] Cameron Reilly: you wake up as if nothing ever happened. If you take the blue pill, you see just how deep the rabbit hole goes. Um, yeah, look, data centers will be data centers.

[01:01:01] Cameron Reilly: We’ll still need data centers. What I do think will happen is you’ll buy, it’s a bit like PCs, right? You’ll buy a state of the art, you know, bunch of chips, and then a year later, you’ll literally Trash them because they’ll be completely outmoded and useless with the next generation of chips that are a thousand times more powerful And then you’ll trash those or hopefully you’ll use nanotech to deconstruct them and get the raw elements and rebuild them.

[01:01:29] Cameron Reilly: Speaking of which,

[01:01:30] Steve Sammartino: I do think just on that, that their AI is going to invent a way to replace the current infrastructure systems,

[01:01:36] Cameron Reilly: Totally. I mean we’ve already got some of, um, AlphaFold, uh, you know, uh, Googles, DeepMind, uh, scientific tools, designing new elements and new materials, new crystal lattices that have never been seen before, so, um, you know, these things are gonna continue to happen at a crazy pace. Um, I’m just texting my wife to remind her that I need to leave soon.

[01:02:07] Cameron Reilly: Um,

[01:02:08] Steve Sammartino: bit and we can tidy this puppy up. Talker

[01:02:11] Cameron Reilly: yeah. So, um, I want to get out of AI for a minute and talk about a couple of other things. Um, China set a new cold fusion record in the last couple of weeks. Um, the experimental advanced superconducting Tokamak EAST for short, E A S T, dubbed China’s artificial sun, maintained a steady state, high confinement plasma operation for a remarkable 1, 066 seconds, setting a new world record and marking a breakthrough in the quest for fusion power generation.

[01:02:56] Cameron Reilly: The duration of 1, 000 seconds is considered a key step in fusion research. The breakthrough achieved by the Institute of Plasma Physics under the Chinese Academy of Sciences greatly improved the original world record of 403 seconds, which was almost impossible. Also set by EAST in 2023. The ultimate goal of an artificial sun is to create nuclear fusion like the sun, providing humanity with an endless clean energy source and enabling space exploration beyond the solar system.

[01:03:30] Cameron Reilly: So we’ve talked about, I think our very first episode, we talked about breakthroughs in cold fusion, and that was only, what, a year or two ago. We’re now, they’re just. Again, it’s sort of doubling the recursion of what they’re able to do. They, it seems like we’ve, we’ve cracked the basic science of cold fusion and now it’s just figuring out how to scale it up.

[01:03:54] Steve Sammartino: That that’s, it’s extraordinary if that happens. Cause it, it, it changes a whole lot of dynamics. Would that be if the Chinese have developed it? Is, is that knowledge shared and, uh, could that be replicated in Western markets as well? Or is this something that becomes a cold war of we’ve, we have the technology?

[01:04:20] Cameron Reilly: No, I think a lot of the basic science with all this stuff is either published or is easy to reverse engineer. I mean, and that’s the thing with AI that we didn’t talk about with DeepSeek. You know, One of the things that you can see happening in the AI space, and, um, I’m not an AI engineer or a gynecologist, I just pretend to be those things, uh, for profit, but, you ever heard that?

[01:04:49] Cameron Reilly: I’m not a gynecologist, I just play one on TV, um, the, I always say I’m not a historian, I just pretend to be one on my podcasts, um, even though I’m OpenAI doesn’t publicly disclose how they build O3 or how they build deep research or, you know, what the underlying tools are that they’re using. For everyone else, All the other AI engineers that are at Anthropic or a deep seek or a Google, it’s not that hard to reverse engineer how they’ve done it.

[01:05:27] Cameron Reilly: You know, the basic principles and the thinking are easily, are easy to determine. Once they see it done, they go, oh,

[01:05:35] Steve Sammartino: you see the product in action, it gives you clues to what went into it to bake that cake.

[01:05:40] Cameron Reilly: exactly, so you know what’s going on here? I think a lot is, you know, the other AI engineers are gonna go, oh, okay, they figured out how to make that work. It must be a way to make that work. Let’s figure out how to make that work. And, and in fact, I saw this. So, what’s happened with deep research this week?

[01:05:56] Cameron Reilly: So OpenAI came out with deep research, the engineers at Hugging Face, one of the platforms that hosts all the models, said, hey, let’s see if we can build our own deep research in 24 hours. So they did a 24 hour code a thon to try and figure out how OpenAI works. built deep research and I haven’t gone to play with what they built, but they felt the last I saw, which was a day or two ago, they felt like they were pretty close to replicating it.

[01:06:27] Cameron Reilly: And I was like, okay, let’s see if we can reverse engineer the thinking behind how they did this and build our own. Like that’s where it’s at now is like, Oh, okay, that can be done. Let’s figure out how to do that. And then 20 people do it. 20 organizations have it. And then there’s another thousand that’ll come on after those, right?

[01:06:42] Cameron Reilly: So it’s, it’s kind of this crazy level of. So, recursive thinking. If it can be done by everyone else, it’s like the ol three minute mile thing. Once someone proves it can be done, everyone figures out how to do it,

[01:06:54] Steve Sammartino: well no one’s done the 3 minute mile yet, let’s go with the 4, but having said that Cameron, the day after Billy Bloggs did it, and who knows his name, apparently like 30 other dudes did it within like 3 months or something, because it was, it was like oh, it’s possible.

[01:07:07] Cameron Reilly: yeah, it’s possible, right? Let’s go do it! The last thing I want to talk about. So, I talk about nanotech from time to time, and, and the, the, the father of, of technology. I mean, there’s a couple of fathers, but the guy who wrote the book, um, in the eighties that sort of defined our current thinking of it, K.

[01:07:23] Cameron Reilly: Eric Drexler, um, I, I watched a talk recently that he gave late in 2024 on the current state of the art nanotech. And, and in a way it was a little bit disheartening. He said, look, what gets called nanotech out there today isn’t really real nanotech. Like what, what is. Called nanotech these days is like that thing we talked about last time, the iron filings being able to use with magnetics or it’s, you know, it’s very, very simple sort of DNA structures that are able to do certain things.

[01:08:00] Cameron Reilly: It’s not. His vision of nanotech, which is nano sized robots that you can program and that can communicate with each other and that can build things and dismantle things, etc. And in his talk, he basically said, it was only a 15 minute talk, but he said, the problem is, is we don’t actually have anyone building the vision yet for for what these things should look like.

[01:08:27] Cameron Reilly: The first step is the vision. And the example that he used was when, when, when NASA stole the best German Nazi rocket scientists after World War II, like Wernher von Braun, and said, build us a space program. There was a, an article that he had that Drexler had, um, From Time Magazine from the early fifties, I think.

[01:08:55] Cameron Reilly: That was written by Wernher von Braun, the great Nazi rocket scientist. That

[01:09:01] Steve Sammartino: listen to you for the rest of the day just say great names like Nasa and then Wernher von Braun? Can we just do that for three hours or can you send me a recording tonight of just saying a whole lot of really good tech names and people through history and I’ll just listen to it as a, as a, as a sonnet to put me to sleep tonight.

[01:09:17] Steve Sammartino: It’s beautiful Cameron.

[01:09:19] Cameron Reilly: I don’t have time. Um, it was like visualizations of what a rocket and a space station would look like. So he, and, and, and, and, he was, Drexel was going through it, and pretty much, it looks like a rocket. And it pretty much looks like a space station. Like they, they thought about what these things would look like, and they visualized them.

[01:09:42] Cameron Reilly: And then other people could come along and go, okay, well, how would you build that? And then they built it. He said, with nanotech, we don’t, we’re not even at the visualization stage yet. We need a generation of really smart folks to visualize what these things would look like, and then we can go and start building them.

[01:10:01] Cameron Reilly: Now I was sort of shocked and surprised by that. I thought we were. Well past the visualisation stage for Nanotech, yeah. But he was also very confident that we will do it, and I know that Kurzweil is very confident that this will get done this decade. But, the bottom line is, according to Drexler’s talk, We haven’t even really got started on that level of nanotech yet.

[01:10:25] Cameron Reilly: Now, of course, I think with the way AI is going, we should make a lot of very, very rapid progress on how to build these things once we know what it is we want to build. And hopefully there are smart kids out there thinking about this right now. I’m not, you know, but all of the smart kids are getting sucked into AI and robotics right now.

[01:10:46] Cameron Reilly: Actually, my son Taylor is, um, he just got his U. S. visa, his, um, um, not a green card, but his working visa to go over there. So he can go and live and work there, um, for the three or four years. And he’s got a friend of his who works at, um, uh, What’s the big robotics company over there? Um,

[01:11:09] Steve Sammartino: uh, figure, figure robots.

[01:11:12] Cameron Reilly: who’s an engineer at figure robots, he said he’s gonna give Taylor a tour of the,

[01:11:18] Steve Sammartino: be cool.

[01:11:18] Cameron Reilly: of the robot, um, factory or the lab or whatever.

[01:11:22] Cameron Reilly: Yeah. So anyway, um, hopefully some of those smart generations of kids are gonna start to work

[01:11:29] Steve Sammartino: I think that

[01:11:30] Cameron Reilly: really need nanotech.

[01:11:32] Steve Sammartino: given that agentic AI is going to open up a lot of time, and I think what it does is it opens up imaginations of people without the Uh, education or skill base to execute against their imagination and vision and new industries will emerge because people who’ve got great imaginations but might not have the technical chops can now start to explore certain areas.

[01:11:58] Steve Sammartino: And you know, we could have a Cambrian explosion because of AI, where a lot of these other areas like nano top, uh, nanotech start to get opened up a little bit because people can explore it more quickly, both, both verbally, mathematically and visually design design wise as well.

[01:12:15] Cameron Reilly: I absolutely believe that that will happen soon. Alright buddy. Listen, I gotta go. Um. But glad that we found the time to catch up and chat. It’s an exciting time and I enjoy talking to you about it.

[01:12:30] Steve Sammartino: Thank you, Mr. Riley.

[01:12:31] Cameron Reilly: let’s promise that we’ll do another one next week and then maybe within six weeks, we’ll actually do it.