The Chaser Report - ChatGPT's CEO Sacked And Replaced By AI

Episode Date: November 21, 2023

Charles Firth, who has spent the last 18 months ingesting all the world's data on LLM's and Open AI, tells you the (maybe) real reason why the company that created ChatGPT have sacked their CEO. 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. Hello, Charles. Now, today you're in luck. Oh, yeah. Because today you can talk about your favourite subject, Open AI and Chat GPT, because it's actually in the news.

Yes, and the thing is, I just want to warn listeners that this does not involve me replacing Dom with AI We had a huge number of complaints about that episode. That's so sweet. I'm actually genuinely surprised, if I'm honest. It even made me long for you, Dom. Oh, my God. I know. Wow.

No, this is about the massive, massive news coming out of the US over last weekend, which was that Sam Altman, who's the CEO of OpenAI, was fired. Of course he was fired, Charles. Well, because Americans don't like success. No, because they've got chat GPT. What can he do that he's built his own destruction? Well, I think you'd find if you listen to the episode where I placed you in the AI. I will do that.

I will do that. It sounds like the most and possibly only affirming thing ever to happen to me. To you. I'm saving it up to a special occasion where I feel particularly miserable. Hey, let's get into the world of AI and the bizarre decision to fire the CEO of the world's most successful AI company after this. Okay, now, Dom, you know how I often come on this podcast and update you with information? Yeah. That then a few weeks or even months later turns out to be absolutely true.

That does happen. It's true. Particularly the submarine, I was really worried about how morbid and detailed you were, and I thought the chances of you being wrong were about as great as the designers of the submarine. But in fact, you were spot on. I was right. You were the Oracle. Same with the Wagner thing, Evergrand.

I mean, actually, that's true, Evergrand. I must say the Chinese property market, which, you know, like, if you're a devoted list to this podcast, you'd be across. There's a couple of, to be honest, it was the best investment advice I've ever got Charles is, don't go long on Chinese property stocks. You've got to sink all my life savings. But that's again in the headlines today, because the rest of the market is now collapsing. Of course it is.

It's like a Chinese skyscraper. It's just sort of, it's imploding. Dear. Anyway, point is, I'm going to give you a similar sort of analysis of the Sam Altman firing because I have to acknowledge that this is a bit of a conspiracy theory. No one actually at this point knows why Sam Altman was exactly sacked from OpenAI. So let's just recap for those who, unlike you, don't read every single article in the news about OpenAI. So this is the company that makes ChatGPT, the world's most prominent generative AI.

Product is the thing that students have been using to write essays for about a year now. Yes. That's changing the world. It's changing the bullshit marketing emails that you get in your inbox that you don't read. The great is no one writes them now. No one used to read them, but now no one writes them either. And the whole thing is that Open AI was originally a non-profit company. I mean, it still is a non-profit company, right?

And part of the reason for that is that the actual principles behind LLM, which are large language models, which is what Open IA is based on, is actually an academic theory that is, was written about 10, 15 years ago and is extremely well understood. It's actually a very simple concept at heart, which is simply like, it's almost like, imagine if you could make a massive spreadsheet

about everything that's ever happened and then run a pivot table about which word will next happen or which thing will next happen. I'm so impressed that you brought a pivot table. It's a lot of times that I've heard about. So basically, yeah, just to try and explain, they've indexed all the text ever, basically. And all the arts and all the music. Completely stealing, essentially, works from every crowd in history.

And based on that, they can predict what the next most likely word is in any sentence. For instance, you might write Ben Robert Smith and the next likely words are clearly war criminal, right? And the thing is that this whole theory is not exclusive to open AI. This is just an academic theory, which turns out to be really full. fucking amazing and quite sort of groundbreaking. But any company, and almost every company is, can set up their own LLM, large language model and do exactly the same thing that Open AI doing. So this company, Open AI, decided to set up as a non-profit, I think partly because they

acknowledged that it wasn't necessarily a very novel or commercial thing. They wanted to change the world with this amazing revolutionary technology for the better. and they acknowledge that actually commercialising it might sort of, you know, it could be used for lots of bad reasons and things like that. And that's not the vibe of what, you know, this whole academic theory was about. I'm surprised to hear that, Charles, because when you say large tech company ingests all the data ever, the next words in my brain that come out are ends in terrifying dystopia.

Yes, exactly. Yes, I think that's right. And that's been part of the reason why it was for the first four or five years, before Sam Altman came along, because Open AI was actually formed in 2015. It actually was right. Wasn't Elon Musk the founder? He said he was. He put a bit of money in the beginning.

But that's because it was like a philanthropy thing. It was a non-profit. I did not know that. There you go. One thing already. So the thing is, then Sam Altman comes along and he goes, look, you guys just are not serious about AI, you guys at Open AI. Because the whole thing is, what you've got to do is you've got to ingest the whole

of all information ever created. Oh. You don't have nearly enough money or computing power to actually unleash the power of the Skynet that can lead to the dystopian future that this actually implies. So they were going for sort of most of the text or at least enough to give you a kind of indicative thing.

Yeah, he just went all. Yes, and the thing is that that required unbelievable amounts of capital. Like we're talking, like I think they got $7 billion from Microsoft or something like that. I think over the course of the last few years, he's raised sort of 20 or 30 billion dollars. There's certainly another $8 billion sort of that he was about to raise. Wow.

The valuation of that money and that thing. So what he did is he immediately, as soon as he got, you know, appointed CEO, he then set up a separate company called the for-profit amount of Open AI. Of course. And he said the reason why we need to do that is because we need to actually ingest all of the world's information. That's billions and billions of dollars worth of computing power. So he raised all this money. The latest valuation that he raised it out was $86 billion.

Which is over 100 billion Australian, right? Very, very amazing sort of thing. And that's, that's, let's use our preferred metric. That's a quarter of a submarine, isn't it? And that is why, open an eye, suddenly, as soon as he arrived, suddenly became incredibly exciting and interesting because suddenly they were actually unleashing the power of this theory. Like, it's almost like...

Anyway, I think I've underscored that enough, right? It's a big, bold idea that turned out to be true. Which, by the way, is why AI is not reliable, because even when you ingest all the text in human history, all of that means is that the next sentence, like, it's like a shit comedian. Like, I'm not even going to say the things, but if you think of the bad stereotypes about, for instance,

who in society is bad at driving, who in society is ugly, or whatever. It imports all our prejudices. All our deepest fears, all our most mediocre jokes. Yes, all of our bad shorthand for everything is in Open AI. And that's why you can't actually use it, particularly for assignments, because it's just wrong. I'm sure Rodney Rood could get a routine out of it. Oh, shit, yeah.

Isn't that what he trades on, his fulfillment of expectations? Well, actually, I wonder if it is the case that because it's happened so often, if you write Santa Claus You, it just goes, can't, where's my fucking mic? That's the finish the sentence. Because that's the most likely next text. Yes, that's right. I don't know. They could just last movie read back start of the podcast.

I do like the fact that somebody came up with the theory that, well, if you could know absolutely everything, then you'd probably be able to put together a fairly media consent. That's right. But that is literally, instead of going, because for years, I was all like, okay, we've got to understand the semiotics of what's going on. And the revolution was to go, I don't need that. Let's just sort of, let's just add a few more GPUs to our

Kauks. Yeah, and just just basically guess. Do a statistical analysis. It'll be kind of right most of the time. That's what they've achieved. Yeah. They've achieved kind of right most of the time. Yeah. Which is kind of like this podcast. Not unlike you, Charles. Yes, that's right. They've made a digital fit. But it is interesting because that's why I don't want to bang on about this, but you can't use it for research because it doesn't know what you asked it. It doesn't understand the question. It just, it's basically like a smarter version of Google that's ingested everything else on the web. Although Google references, by definition, references its works, because it's linking

you to actual information. I mean, that's the other example that's so fascinating in this space is that the inventors of Google went, whichever article has the most links to it is the most authoritative and so on and so forth. So these insights can make your company massively valuable. So I can, why don't we had an insight like that, that child? That, again, Google doesn't know that it's the, what your query is. It doesn't know, it doesn't actually understand the semi-

It's the meaning of what you've written in. It just goes, okay, what is the best match with the most links to that chain of text? Yeah, to that idea. Yeah. Okay. Yeah, maybe. None of these things are smart, it's my point. But it's funny that, yeah, no, yeah.

Dumb Google, dumb open AI. But they found it, but this is the whole genius of it is. They found a way to approximate that stuff. Oh, yeah, yeah. Yes, it's a very convincing white man, is what it sounds so. Yes. It's somebody who doesn't know anything. Great confidence and no knowledge.

Yes, exactly. It's just drunk. your uncle who goes, well, I read it on the Facebook. Yeah, and stereotypes absolutely everything around them. Yeah. Yeah. But it's funny we should mention Google because that's the next

part of this yarn, which is, okay, so Skipford, Sam's been very good, he released chat GPT 3.5, and that sort of seemed to be working well. Yeah. So then he decides, okay, chat GPT 4. This is where I've got to hit it out of the park, right?

And what is that? We'll tell you after this. The Chaser Report, news you can't trust. Number four. Okay. So, bit panicking, though. He's already ingested all the world's information. What do you do? Because the only theory is, well, you just need more information.

So what you do, and this is the theory, and admittedly, this is not confirmed. But one theory is that Sam went along and he talked to the people at D2. Right. I have no idea what D2 is. So D2 is the Dragon Army, which is the Chinese, it's a cyber army based in China that scrapes the web for all information across the web. It apparently has, and this is true, it apparently has 10 times more data and better scraping than Google. Well, this is what I'm thinking, Charles, is I'm thinking, well, hang on, if you want more than the publicly available web, Because it's true that what's out there that's publicly indexable is a tiny fraction of what is available.

And for instance, all you need to prove that is I'll log into any library database with full text, for instance, or email. If you imagine if it was able to get into everyone's Gmail, how much more accurate it would become. Or say, pre-tell their TikTok account. Yes, into all their social media. Let's say it could access all of our text messages, for instance. Using their Huawei phones. Yeah, all of the span, the people from this podcast have texted Charles on his personal one-mail phone number.

Yeah. All of that would be in the database. It would be that much more accurate. So there are, there's certainly indications that it does seem like they've used the D2 data set for chat GPT4. Wow. Researchers have already sort of started going, hang on. This knows a lot about X subject, which is not actually formally part of the data sets that we thought it was using and stuff about it.

So you're saying that sacking the guy could have been because he basically committed a massive ethical breach, potentially. Well, yes, I think the thing is that the implication is that the board then went, hey, wait a minute, did you cut a deal with this Chinese cyber hacking firm that is basically the most powerful cyber hacking firm on Earth? Hang on a sec, Charles. I'm going to index my long history of corporate behavior. Yes, the answer is he probably did, because that's what people do. And he said to the board, no, right.

And then Microsoft found out, and Microsoft, of course, you know, in bed with Open AI. But they are also massively in bed with the American military. Like, they are, it seems to be part of the American military industrial complex. And that's where it all unraveled because it was sort of like, okay, well, you know, he's lied to the board. I think the board, which is the board, the board is actually the non-profit side of the company, had to sack him because he'd lie to the board. The funny thing is that Microsoft has now turned around, having had OpenAI SAC Sam Altman, who's clearly the talented person in that sort of space, and offered him a job.

Offered him a job. Well, because he knows the people are D2, presumably. And imagine how much better Bing would be if they had a data set of information. I'd imagine how desperate they would have any information in Bing. I've just had, I've just made us a trillion dollars, Charles. Oh, yeah. What if we had a search engine that, um, that, um, Okay, this is going to change the world.

You put your search terms in. The search engine goes and runs that same query on Bing and excludes every single thing that Bing gives you. Yeah. So it's an un-Bing. Yes. So everything, and that would make it more accurate than any search engine that's ever existed. If you imagine that every Bing result is as wrong as it's humanly possible,

it'd be brilliant. It'd be much better than chat, GPT. I love it. What would it be called? Un-bing. Un-bing. I don't know. Anyway, that's really interesting.

So, well, what shouldn't see this space? Because I know that the vast majority of Open AI employee have written to the board saying they should all resign. They have no confidence in them. And they want the guy back. Yes. So what's going on? I think that they used the Chet GPT to write those emails.

That's right that email, right? And it was sort of very much influenced by the Chinese Communist Party. Well, doesn't it just make things easier? I know. I kind of feel like... I kind of feel like... To be fair, Charles, it's not a... as though Western democracy is covering itself

with glory at the moment. It's pretty much finished in America. Democracy's almost done. I, for one, welcome our dictatorial overlords. Would you rather the world was run by Donald Trump's Magorite fascists or the Chinese Communist Party? I think only one team has the runs on

the board. They're both going to be oppressive. At least one lot is competent. Well, we would have been ever grand. Well, except for the property market, but at least they make the trains run on time, Charles. Because at least they build trains. I'm like a Republican. And I'll tell you what, last time I went to Beijing, I mean, Tiananmen Square was very, very orderly.

It's a very good bit of central planning, Tiananmen Square, yeah, absolutely. Our gear is from, we are part of the Iconic-class network. Get you tomorrow. Coming soon to the People's Republic of China.