Futuristic #36 – Fight me, Grok

Steve and Cam catch up on the latest FUTURISTIC news, including Elon’s new gaslighting AI Grok 3 writing porn, Microsoft quantum computing not totally made-up Majorana 1 chip, the fact that the world now has 1328 LLMs, and Zuck can now turn your thoughts into text. What could go wrong?

FULL TRANSCRIPT

FUT 36 Audio

[00:00:00] Cameron: Welcome to Futuristic. Steve Sammartino, uh, first of his name. Uh, it is Futuristic episode 36. We’re recording this on the 21st of February 2025. I think it’s been two weeks since we’ve spoken, and a lot has happened in two weeks, Steve.

[00:00:21] Cameron: How have you been, buddy?

[00:00:23] Steve: Good. Busy, traveling around. Did one bit of interesting work, Cameron. We can’t talk about a great deal of it now, online, but let me just say this. It’s right up your alley. I did some work with the National Security Apparatus on evolving risks involving technology.

[00:00:43] Cameron: Well, that’s

[00:00:44] Steve: a good conversation.

[00:00:45] Steve: And it was very mind opening in a whole lot of areas. Mind opening. And, uh, uh, I think the net outtake was, is that the biggest risks aren’t the behavior of the people. And this is not a secret. The biggest risks is the behavior of the corporations and the lack of regulation around them, which then facilitates all other risks, which has been clear for a really long time.

[00:01:16] Steve: And the lag effect geopolitically on that. Within this security realm is pretty apparent anyway. So hopefully that landed and they were, I think, grateful to get some new understanding. Let’s just leave it at that because they’re listening, Cameron. I think we know that, and it’s all there to be had. And I had to go through a significant, uh, vetting process to

[00:01:42] Cameron: do the work.

[00:01:44] Cameron: Hold on. You do a podcast with me and you still got approved? I couldn’t believe it. They dropped you. Drop that ball.

[00:01:51] Steve: Please don’t bring Cameron. No, I think that’s why they got me. I was in a Hercules. I was like, no, .

[00:01:59] Cameron: No, really? They’ve got you because now they’ve got you to spy on me. I know how this works. I can’t talk about these things.

[00:02:05] Cameron: Embedded. Embedded. Yeah. Well, uh, that sounds exciting, Steve. Um, I think there. Shuffling deck chairs on the Titanic. I knew you would say

[00:02:17] Steve: something like that. But by the way, do you remember once that you had on your bio on Twitter or somewhere, it said, twisting the nipple of industrial complexes since 1987?

[00:02:27] Steve: That said something like that, didn’t it?

[00:02:30] Cameron: I don’t know. Possibly. I can’t believe you don’t remember it. I remember it. I don’t remember. Yeah. Um, I, uh, I finished my big coding project, uh, I think since the last time we spoke, uh, last week. The, the QAV checklist, fully automated it. The thing, that took me months on and off to do.

[00:02:50] Cameron: Now it’s, what, what, a couple of years ago would take me four hours to do. Uh, now it takes literally one click of a button. Then I go, then, and it runs for four hours overnight. But I click it on a Friday night, I wake up on a Saturday morning and I’ve got an entire buy list with hundreds of companies analyzed and, and, uh, scored and in a spreadsheet.

[00:03:14] Cameron: It’s, um, it was a good feeling to finally finish that, get it code complete.

[00:03:21] Steve: Just on that topic, the coding thing, remember I did an AI portfolio? I tricked ChatGPT into creating a portfolio, it wouldn’t give me one, and then I asked it about Warren Buffett and his principles, and which sectors, uh, those principles follow, who’s doing well in those sectors, who’s likely to, and it gave me a portfolio.

[00:03:40] Steve: Well, it’s two years have passed, and the portfolio, I did an assessment, I’ll send it through. And I compared it to the S& P 500. It didn’t outperform, it underperformed the S& P 500. But all, but interestingly, all of the stocks that it picked within those sectors were the best performers in their sectors.

[00:04:01] Steve: Now the reason the portfolio underperformed the S& P 500 was because the tech sector did so well and it only gave me 40 percent of my stocks in the tech sector.

[00:04:12] Cameron: Well, I don’t know about that, because So, the stocks it

[00:04:15] Steve: gave me, the stocks it gave me, like it gave me some banking ones, some retail, some infrastructure, and, and all, and the companies that it gave me beat their competitors in that sector.

[00:04:25] Steve: But the overall portfolio, because only 40 percent of it was weighted in tech, underperformed the S& P 500, because tech is such a big chunk of the S& P 500 now.

[00:04:35] Cameron: But the S& P 500, I mean, the results have been good, but not Mag 7 good. I’ve got a US portfolio that I run for QAV. It’s doing three times the S& P 500 over the last 18 months I’ve been running it.

[00:04:51] Cameron: That’s really good over 18 months. And we have no tech stocks, because tech stocks, uh, shipping, banking and finance, um, Land’s End, a retailer. Um, you know, the best that’s done like 300 percent is a company called Willis Lease Finance Company. They’re up 300 percent in the last year that we bought into. Um, but no tech stocks because the tech stocks are way too overpriced for a value investor like us.

[00:05:22] Cameron: But S& P’s up about 30%. Over that period of time, our portfolio is doing about 100 percent

[00:05:27] Steve: over that period of time. They’re both extraordinary. Of course, you know, long term average of the S& P is 11. Yeah, it’s been a bonkers couple of years, but we digress.

[00:05:38] Cameron: So, um, big news, uh, I guess this week, Steve, is Grok 3 launched.

[00:05:44] Cameron: For the people who don’t know, that’s X, formerly known as Twitter, X, AI’s, uh, AI, Elon Musk’s AI, let’s just put it that way. This is version 3 that they built. The data center for this, extremely quickly, and built the whole thing extremely quickly. And, you know, Elon, when it launched a few days ago, Elon was out there tweeting saying that it beat all of the state of the art models.

[00:06:15] Cameron: People then got into it and reviewed it, and the feedback that I’ve seen has been pretty good. Pretty good, I gotta

[00:06:23] Steve: say. Did exclude though, it did exclude some of the models in the assessment tests. Like OpenAI03 was omitted in certain areas, there were certain things that were omitted. And when you put those back in, I looked in Reddit, then it didn’t beat all of them, but it did really, really well on average.

[00:06:41] Cameron: Doing really well. And, uh, you know, the way that Elon is positioning it is that, A, it’s very, very smart, but B, it’s not as woke. Let’s say that it’s not as locked down. I hate that term. It’s not, uh, it doesn’t have the same sort of guardrails. That’s a better way to say it. It’s not as

[00:07:05] Steve: regulated. It’s a little bit less regulated.

[00:07:07] Steve: I think that’s a better way to frame it because the reality of the world is that large chunks of the world, certainly when it comes to thought, wishes, behavior, what we write, the internet is filled with porn and all sorts of stuff. So it’s a little bit more of a reality engine, but I did like what he said.

[00:07:24] Steve: The quote was, it’s configured for maximum truths. Which sounded like something out of a comic book in 1965. Configured for maximum truth seeking, Grok 3. 0, the AI that you can rely upon for the world as it is or you want it to be in a non woke nature.

[00:07:44] Cameron: According to the Grok website on X. AI. com, Uh, trained on our Colossus Supercluster with 10 times Colossus Supercluster?

[00:07:54] Cameron: Geez, that was a Colossus Supercluster, wasn’t it, G Rod? That doesn’t happen again. With 10 times the compute of previous state of the art models, Grok 3 displays significant improvements in reasoning, mathematics, coding, world knowledge, and instruction following. Tasks, Grok3’s reasoning capabilities refined through large scale reinforcement learning, allow it to think for seconds to minutes, correcting errors, exploring alternatives, and delivering accurate answers.

[00:08:25] Cameron: Has leading performance across both academic benchmarks and real world user preferences, et cetera, et cetera, et cetera. Now I put it to the test, uh, Steve. Now I, it, they also have a. You know, OpenAI launched their deep research a couple of weeks ago, then, uh, everyone’s got their own deep research model now.

[00:08:47] Cameron: You gotta go deep, Cameron. It’s all about how deep you can get. It’s all about deep. I’ve always said that. I’ve always

[00:08:52] Steve: said that.

[00:08:53] Cameron: Depth is key. They’re all calling them the same thing. Deep research, or I think X calls their