The Chaser Report - CharlesGPT

Episode Date: March 24, 2025

Charles makes a $10,000+ request to Dom, which if successful will forever change the course of this podcast. Meanwhile, Dom shares how his material has been used to misinform AI. Plus we find out why this election is going to hit new levels of boring.Watch OPTICS on ABC iview here:https://iview.abc.net.au/show/opticsCheck out more Chaser headlines here:https://www.instagram.com/chaserwar/?hl=enGive us money:https://chaser.com.au/support/ Hosted on Acast. See acast.com/privacy for more information.

The Chaser Report is recorded on Gatigal Land. Striving for mediocrity in a world of excellence, this is The Chaser Report. Hello and welcome to The Chaser Report with Dom and Charles. Charles, you're back in Sydney. You survived Adelaide and all I can gather is that many Adelaideans don't listen to the podcast because, A, they came to your show in large numbers, I gather. Yes. But also you weren't ripped limb from limb for the comments you've made about Adelaide on the feed recently.

So welcome back. I'm pleasantly surprised to say. I think the thing is, if you live in Adelaide, you just expect a bit of, like, derision from people. Yeah, I think that's part of the trade. Like, sure, everything costs less, but on the other hand, you've given up on hope and you've given up on hope. You've basically decided to die. So, you know, it's just one of those, but it's a fair trade because your house, you can

afford a house. Charles will be returning to Adelaide in November or December for the war on 2025 and frankly, No, we've cancelled that to it. Oh, is that happening? No war. No war. No war. Well, the war is over? Yeah. Well, the problem is we're going to the UK, so. Oh, you've finally shafted the Montpelankonomics.

Abandoned. Abandoned, all the sort of... I might do the war on 2025. You should. It feels like a good business. Yeah. All right. Anyone wants to be on board, get in touch. Anyway, Tal's what are we talking about today now that you're back?

Okay, so I have an idea for why I need to buy a Mac studio with 512 gigabytes of RAM. Well, this is going to be nice. Charles, I, having looked a little bit at, because as people know, we like our tech, we like our Macs, I cannot possibly see a use case for such a thing, unless, hang on a second, are you about to do your own AI? Yes, yes. Charles GPT, after the break, we're going to know. Don't encourage him.

Don't listen any further. So, I've been reading up about deep seek, you know, the one that was done by all. Yeah, the Chinese one that A, apparently was much cheaper than. and chat chip Epte and smoked it in performance and B, has been banned by the Australian federal government on all government computers. Yes, and have you read some of the outputs of Deep Sea? No.

It is the most hilarious thing. It was done by over 100. It was basically this guy who'd actually been an entrepreneur and done a whole lot of other sort of tech ventures decided a few years ago, I think in 2023, that he was going to pivot to AI and do something there. Like everyone else. So he hired 200 Chinese researchers, AI researchers,

All of them were trained exclusively in Chinese-run universities. So there was no sort of like, we're going to get... No MIT or any of this stuff. Yeah, yeah. Smart asses to think they're better than everyone, right? And then they came up with this model which, yes, absolutely, you know, learns from the breakthroughs that, you know, GMIPT.

But then has a whole lot of innovations that make it much easier to reason, right? So that's the whole thing. And it only costs $5.5 million US to train. The crazy thing about it, because all the big tech companies from America have spent countless billions, I mean, the amount that Apple spent creating a crap Apple intelligence is far more than were spent on DeepSeek. So, Open AI spent $9 billion with a B dollars last year, and Deepseek spent $5.5 million. And none of the money that Open AI spent pretty much was spent on acquiring rights to intellectual

property. No, exactly. By the way, I discovered some of my comedy books are in that massive cache of pirated data. LibGen. Yeah, in LibGen. which means that at least some of the facts in that database are deliberately wrong. So I'm looking forward to those results filtering through. Okay, so this person who's writing for the London Review of Books

decided to ask it just a simple question about colour blindness. So basically, in a population where red-green colour blindness affects one in 200 males, what is the likelihood that an unaffected couple, you know, who don't have colour blindness, will have a child with colour blindness? That's the question. right listen to the output of deep seek okay well like it starts out and then it goes okay if she is a carrier the chance of having an affected child is half since each son has a 50% chance of getting the gene wait but actually the chance of having a son is a half and given that the child is

a son the chance that he has affected his half therefore the probability is a quarter wait no actually wait no wait if the mother is the carrier each gamut she produces has a 50% chance the father then contributes y or x and since the father is unaffected he must be x c y but wait actually no wait the mother's contribution is independent of the child's oh my god this is like a conversation with you it's it's a deep first and the funny thing is so it goes on for two thousand words like this but comes up with the right answer and and it's one of those things where it's like you're literally watching the train of thought of it yeah yeah show you're working and just how slow and ineffective.

That's fascinating. And apparently the thing is that Open AI specifically tells its models that they are not allowed to show their thinking. But all models go through this sort of weird thinking chain. That's so interesting. But the reason why Open AI has hidden its reasoning is because it's so plagiarized that it will just reveal how illegal its methods are. Oh, that's so funny.

You know, I actually was part of it, well, I was listening to an AI expert talk last week kind of in my academic life. And she was saying that one of the biggest problems with AIs is that it doesn't show, they never show how confident they are in their results. Like if the output said, I'm 100% certain that this is true. And other times it said,

look, this is just on the balance of probability maybe 60%. Wouldn't it be much better? Wouldn't we have much more confidence in it? If actually, as in a conversation, someone says,

oh, look, I'm not sure about this. Yeah, wait, no, or wait. Or this is a Charles Fir theory or something. Just give a sense of how reliable it actually is, but they never show they're working. No, no, exactly. So why does this mean you need to spend $10,000 on a computer?

Well, 15, please. And the answer is because, okay, so DeepSeek also generated this little mini version of DeepSeek, which you can run on a laptop. Like, I could literally run it on the laptop that we're recording this podcast on. Yeah. But then their full R1 model, which is fucking, like, advanced. It's like literally the most advanced thing. Can be run on a Mac studio, admittedly, two Mac Studios.

that are hooked up together that you can get the full the most advanced AI engine to be running on your own setup right so there so and what i want to do is i want to upload every episode of the chaser report to it as it's learning data right so we're not plagiarizing anything we're only using our own training it in ourselves and my i don't know but my guess he didn't having read the reasoning that the deep seek does is it will produce perfect podcasts on any top of that we ask it to based on our sort of thing and we can just retire and start and just count our money like basically this podcast suddenly becomes a sort of passive income strength starting 15,000 in the red having bought this computer system well I think 30,000 oh because you need two of course

of course can you please just get one of your AI models to try and justify why that makes any sort of business sense I know I realized that 26 years into running a business with you charles The chaser was started in 1999. It's a bit late in the piece to say when you spend money, the concept should be to actually generate more money than you spend. And I know that we've never run any business according to those rules.

And frankly, it's amazing that they're still even technically solvent. But I just, I'm not entirely convinced that that is a great idea. I'm very glad that you think we're technically solvent. But you know what? You know what I've just realized? You know what you can do?

You can go to the Apple store. You can pay your $30,000. Yes. Get the computers and bring them home for a two-week trial. Yes. And you can run it for two weeks. Yes.

And if it can generate, if it has any promising results at all, you can even just ask it, the first question, it's a bit like deep thought in Hichiker's Guide. Why is this worth $30,000? How can we make money out of this setup? Yes. You've got two weeks before you can return it. That could be the first podcast. In fact, the first podcast could be about us theorizing about wanting to get to Mac Studios and hook them up to do.

That could be the first one that we put out that's actually AI generated. And the brilliant thing is we've always. Already, just through human-generated content, made the podcast incredibly boring in this episode. So no one will notice a quality drop when the things become completely AI generated. But also, I feel like if we could then hook it up with facts, like you could hook it up to some news feed, like, I don't know, Fox News or something. We've done a thousand episodes of this podcast, and you want to bring facts into it. Well, I feel like you sometimes look up Wikipedia. I do.

I do occasionally fact-checking. So we just upload Wikipedia as well. The whole of Wikipedia. Great. You can put a Wikipedia on a thumbnail drive. It's actually remarkably... Yeah, it's just small. Sure.

And then link it to key events. So every time, say, you get a news alert on your phone, it will tell the engine to put out another podcast without that event. Does this mean, hang on, hang on. I've just realised what's going on here, Chas. I've just realised why you're doing this. What? The Chaser Report.

Less news. Less often. This is a way to get out of having to come. cover the federal election, isn't it? This is the way to avoid having to cover Albanesey versus Dutton, the most uninspiring contest in living memory. Because, yeah, no, and also the depressing thing will be, it feels like Alba's going to win, isn't it?

I mean, the polls have been very positive, but again, it was the Morgan poll. He's a long way ahead in that today, and then a new gov is behind. So no one knows anything. But it does seem as though, yes, he's on a bit of a role in the polls. But I feel like, have we explained Cook's Law before on this podcast? No, I don't think we have. Cook's Law of Australian elections. I feel like we should get Ritchie Cook on to explain

Cook's law of Australian elections. We should. That's a great idea. Because Ritchie actually used to edit the chasing newspaper. He does Planet America quite often, and that requires a three to four hour time commitment. So I'm sure we can spare 15 minutes for us. Well, and also we can offer to AI generate his responses using our new rig. But Cook's Law of Australian elections, which I think is the most ironclad law of any involving Australian elections,

is that the most boring outcome will always be the outcome of the Australian election. And I think that you have to go back to 1993 to go... Keating. That didn't happen. But actually thinking about it, 1993 was when John Houston ran Fight Back. Oh, yeah. So that would have been very radical. So that would have been far more interesting than another term of Keating.

So actually, I think it just holds. I think you'd have to go back to maybe 73. So was Albanese winning last time around a more boring outcome than more of Scott Morrison? Yes, because Scott Morrison was weird and, no, but also I think Boring has two components to it, which is like predictable and foreseeable as well as just inherently dull, right. And I think it very much adhered to the thing of like, yeah, like there was no headlines out of, oh my God, Alvinesey one. So this time around, and it's true that actually, if you think back probably, certainly since Hawk,

Rud and Albo when they won, won by making them. amazingly, unthreateningly dull. Like, they're kind of the anti-Lathom. Yes, that's right. Yes. And so my thinking is that's why it sort of struck me. Obviously, Anthony Albanyese knows about Cook's law. Of course he would have read The Chaser back in the day.

Sure. And he knows that actually what I've got to do is absolutely jacked shit over the next three years in order to win re-election. And that's what he's been doing for the last three years. It's suddenly everything just became crystal clear. I know what's happened. If you're going to apply Cook's law It was when Peter Dutton came up with policy

Things went wrong for him Yeah, definitely, yeah, yeah, yeah, yeah, yeah, the nuclear policy Yeah, the referendum Way too exciting All that stuff Yeah, it was actually an idea of doing something different Whether you liked it or whether you liked that year or not It was a substantive major policy shift

You can't do that in Australia We don't like it But I have a theory about why This has been an eclectic episode I, well, that's what happens When you get AI to generate your episode for you I have a theory about Peter Dutton's whole, you know, brain fart around the referendum thing, which is, I think that he thinks that now it's getting close to the election,

he has to start getting serious and be more interesting. And that's a huge error. Like, it's actually, like, you know, he's been doing all the sort of focus group tested stuff, right, like, for the last three years. And now it's getting close. He's going, oh, maybe I need to just freeball it and add my own touch. Well, and all the soft-focused interviews about how lovely he is when you get to know him. The Australian public has no interest in getting to know anybody.

No, no, no, exactly. That's a huge mistake. Yes, yes. I mean, Anthony Albanyi has been Prime Minister for three years, and I think the average Australian would barely recognise him in the street. Well, did you listen to the John Del Menico episode? No, I haven't yet. Friday about Albo's 60 Minutes interview, which was quite funny,

which is, so the whole idea about those 60 Minutes reports is you're supposed to go behind the scenes and see what their personal life is like. Albo Let me just take one guess Yeah The dog No Oh really

Albo took Carl Stefanovic along On a photo shoot Like a press stop Right In Tasmania on a salmon farm Like literally he doesn't have A private life

He just literally That was like That was like the most That was it It was just It was like a press ball Come and see the real Albo At a press stop

Yeah Go interesting That's how he's spin That's his hobby It's press stops So are you telling me Not having heard the episode yet That when 60 Minutes

really wants to just cut to the quick and basically fill it a politician on live television they call in Stefanovic. Isn't he the guy who did the ukulele? Actually, you know what? He did. Carl got rid of Scott Morrison with the ukulele. Last time round. That's the reason why John

said it was so successful for Alba was simply because it didn't have a ukulele. It was basically identical to the Morrison thing. Maybe we should get the AI to generate episodes with John Delmenico in them because he actually does some research. All right, Charles, I'm not convinced that you should spend $30,000 on this,

but I am convinced that you should do a two-week trial and see how far you can go. If anyone wants to help bankroll this, and also has the technical skill to install DeepSeek on to linked Macs, please make contact, podcast at chaser.com.com. Especially if you've got the money part of the ad declaration. I'm looking forward to our inbox remaining empty, Charles. We are part of the Iconiclass Network. Go to you tomorrow.