Cameron explains his MANIFESTO FOR AUSTRALIAN AGI, Steve explains why he supports the Australian government’s ban on social media for kids, and they discuss NVIDIA’s downloadable local LLM.

FUTURISTIC 33

[00:00:00 ] CR: Welcome back to Futuristic. It is episode 33. It’s the 29th of November 2024. Steve Sammartino, welcome back.

[00:00:16 ] SS: Reilly.

[00:00:19 ] CR: We have, uh, some big things to talk about, Steve. Big ideas that I want to get into today about the future of AGI in Australia. Social media bans, personal AIs from NVIDIA, but before we do that, why don’t you tell everybody how you have been futuristic since our last conversation.

[00:00:44 ] SS: One was with VUT and the brief that I got was, you gotta help convince university that we need to embrace AI more, which is kind of crazy and interesting. You know, they’re still using it a lot to stop people cheating or using the tool in some capacity. And I just had two things to say.

[00:01:11 ] SS: I said, not using it is insane. And I said, I don’t know the answer, but I know the question. And the question is, how do we help our curriculum rise above the AI? Because that’s what we do with every tool. You’ve got to rise above it so that you’re working on top of it. Now, of course, you need to know what good looks like.

[00:01:30 ] SS: So there’s that. So that was the one thing that was interesting. And I did a workshop with CBUS. Which is a big building company, Seabus Property, who owned the superannuation in the building. It went really well. Incredible Workshop was on MVP’s, Minimum Viable Products, and I want to say that for any company trying to innovate, Or a startup.

[00:01:50 ] SS: MVP’s are still the ticket. That is still the secret sauce. And it’s really, really hard to do. You know, I really had to thin up some of the ideas for these guys that just cannot think small enough to get momentum. And that was an insight that that’s been true for 20 years. And I can’t see it ever changing.

[00:02:09 ] SS: Is that if you do something small, you can find a path to success and iterate. And I was really driving

[00:02:16 ] SS: that home.

[00:02:17 ] CR: Think small.

[00:02:19 ] SS: Think small.

[00:02:19 ] CR: 2024. Well, I’ve done a lot of coding, uh, as usual, just coming up with ideas and coding stuff. Um, just trying to automate a bunch of things. Like, I’m still trying to automate the, um, investing checklist that I do each week. So it’ll be one click button and off it’ll go. I had some success with that this week.

[00:02:39 ] CR: Did a first phase of it that worked. First test worked. Now I’m trying to expand. And again, think small. So it’s like, I wrote all the scripts around one stock. That I had the data for. I said, right, generate a score for this stock. Got that working. Now I’m like, trying to give it a list of stocks and get it to do scores for a list, like 10.

[00:03:01 ] CR: And then I’ll give it 200 and see if it still works. But it’s just, yeah, starting off building it piece by piece. But, um, Yeah, I think that’s, that’s mostly it from a technology perspective. Just a lot of using the latest coding tools and it’s frustrating. You know, I’ve been using Claude 3. 5 Sonnet, which a week or two ago was freaking amazing.

[00:03:28 ] CR: The last couple of days just really sucked. Uh, and I see people talking about that on Reddit. It’s this constant frustration where these tools, uh, have weeks where they’re just like genius level, amazing one shot solutions for everything. And then all of a sudden they’ll just be dumb and, you know, like they’ve had half their brain removed.

[00:03:50 ] CR: They’ve been lobotomized.

[00:03:51 ] SS: Why do you think that is? What do you think

[00:03:53 ] SS: causes that?

[00:03:55 ] CR: I think they’re. I don’t know. I see lots of theories, but it seems to be some version of, they’ve got a certain amount of GPUs that they’re using to build their frontier models, maintain their existing models. They’re quantizing things. They’re trying to, um, Save money at different stages. They’re moving the, you know, we’ve got 50, 000 GPUs applied to this.

[00:04:23 ] CR: We’re going to take 10, 000 and move them over to this thing. And so this thing becomes dumber. I don’t know. They’re just messing around on the back end as they’re continually building new models, training new models, trying to save money, trying to, I think, reduce The algorithms as well, like, okay, if it’s costing them a dollar to give you an answer that they’re charging you a cent for, they’re trying to figure out ways to reduce the cost of that by optimizing things, and it doesn’t always work, there’s a lot of experimentation, like, a lot of people get really shitty on Reddit, Um, I’m paying you my 20 a month and this service is terrible.

[00:05:06 ] CR: And I’m like, dude, you have an artificial intelligence for 20 bucks a month. Like shut the fuck up.

[00:05:12 ] SS: Hunched a year with excitement, the greatest tool ever invented you have for less than the price of a

[00:05:19 ] SS: pizza?

[00:05:20 ] SS: Look, let’s complain here. We’ve got the most incredible AI, mostly free, or the super version for 20 bucks a month. You should so be complaining. Your lot in

[00:05:29 ] SS: life is

[00:05:30 ] SS: horrible.

[00:05:30 ] CR: Yeah. Shut the fuck up and stop complaining. Yes. Yeah. It’s going to be up and down. I just accept it. It’s going to be good days and bad days, but you know, the good makes up for the frustrating things. Um,

[00:05:42 ] SS: plausible ideas. The one that, uh, all of a sudden, let’s say, a few people get onto the new version, there’s a couple of tabs open, next thing you know it’s amazing and there’s thousands of tabs open and the machine slows down. I mean, that’s highly plausible. The idea that, you know, GPTs are generative pre trained transformers is a big one because one of the things I see on Reddit is often a new version has, let’s call them scary capacities, where they do things that are a little bit, Oh my God, wow.

[00:06:11 ] SS: And they throttle those back or change those to an extent, which knowing the way the algorithms work that they’re imperfect and it’s not an, if this, or then that protocol that could infect other parts of the system. Uh, the answers that it gives. So they were like really plausible and the cost one, the third one that you’ve mentioned, that’s another great one.

[00:06:31 ] SS: Gee, this is great. We want to launch and show everyone how amazing it is. But now that everyone knows how amazing it is, let’s just, we’ve launched the plane, let’s pull back there and save a bit of fuel in the, in cruise mode.

[00:06:45 ] CR: a couple of other things that, uh, I know you introduced me to an AI. There’s a video editing tool this

[00:06:52 ] SS: Yeah.

[00:0