You can also listen to this podcast at iono.fm.
Download the free LiSTN audio app on Google Play, Apple, or here.
Simon Brown: I'm talking to Viv Govender from Land Suisse. Viv, thank you for your time. Of course you run the AI Fund at Rand-Swiss. We talked about this pretty late last year, so I put it on my list to talk about again. And last week got really weird. I asked OpenAI to launch Soro. This is a closed beta and converts text to video. Also, Google Gemini released his 1.5, also released a closed beta, but my understanding is up to 1 million tokens are running. The previous high was around 200,000 tokens. It feels like we're pretty far along the AI path. [it is] In a way, we're just scratching the surface.
Viv Governor: Almost certainly. For example, you just need to understand what these things mean. This was the token. What is a token? It's something you put into things. For example, let's say you want to put in a bunch of stuff to analyze. So I put the PDF into something like GPT. The number of tokens limits the amount of PDF that can be included.A whopping 1 million tokens – that’s what you get [Google’s] Gemini goes to more common places and studies them. There are up to 10 million tokens.
There are a few things to understand about this. First, as the number of tokens increases, the computation proceeds linearly and grows exponentially. The token will be exponentially more difficult than the previous one. So this is obviously an architectural change that will allow us to do things through a kind of reference under a more expert system, just like the human brain is also modular to some extent. Something similar is happening here.
According to them, this is very interesting. Because the biggest he puts in 10 million tokens, it's essentially like a significant amount of videos, or almost an entire library. It's much more than the entire Harry Potter series. A use case that immediately came to mind for me was loading South African tax law into the system and suddenly having a genius-level tax lawyer at the helm. And that's what you can do with 1 million tokens.
Simon Brown: I didn't realize it, but while I've been using OpenAI and building my own ChatGPT based on columns I've written over the past 10 or 15 years, I've run into capacity issues. They obviously haven't run 1 million tokens. They are running far fewer numbers.
The main points are: I expect Nvidia's results to be released during US market hours on Wednesday. For those who think Nvidia has gone crazy, NVIDIA has gone crazy. But there are basics. grace hopper 200 [superchip] It just came out. I'd like to say that Microsoft and even the open source meta are sitting on a gold mine, but it seems like there's more to it than that.
Viv Governor: yes. What people tend to forget is that obviously we… This was a long time ago, but when you look back and see what IBM was predicting when the first computers came out, they were all over the world. And he was talking about five computers. – you know what i mean? Two in America, one in Europe, one in Asia, and one in the Soviet Union or something. It happened that there were a lot of computers that were needed all over the world. And the reason computers have basically become such a huge market is because they've become so cheap and so powerful.
So if you have to spend this much money on each computer, you'll naturally have fewer computers. But the fact that you now have a computer in your pocket, probably even on your list at this point, that is much more powerful than anything they could have imagined when they started this endeavor. means you are now spending much more money. A greater portion of the world's GDP will be available on computers than would otherwise be available.
And the same thing may happen with Nvidia. The products Nvidia launches this year will probably be effectively about 10 times more powerful or 10 times cheaper for the same amount of computing power, whichever way you look at it. you know what i mean?
That means we're going to experience something like Moore's Law, and in a few years we could be running these systems from our phones, which is something we can actually do right now. There are people who have worked with mules. Mules are meta. There are different levels like token levels. The small token level should run like an upper-end desktop, not the absolute top-end. So you're talking about a home computer running it, right?
Simon Brown: That's all. So his Nvidia rebooted there can basically run locally if it has an RTX, which is a top-of-the-line graphics card. What this means is that everyone focuses on the great Sevens, but they are mediocre. So Arm Numbers doesn't make Grace Hoppers, but they do make chips, and the demand for them is just through the roof.
Viv Governor: Well, definitely. And what I think is really going to get noticed, and it's really going up as much as anything else, is the SML of… Because these are the ones who actually have all the monopoly rights in this field. Basically he says Nvidia has that desire, but this could be elsewhere. He can make anyone make anything. He can copy and invent many things. No one in the world can actually build huge machines that cost hundreds of millions of dollars each. It's not just a matter of copying a design and working from there. In manufacturing, it is not enough to have a basic blueprint. It requires a variety of techniques, all…, physical… and you always have to have little hacks along the way. And no one in the world has this.
In fact, AMSL's only competitor is China. That's because the Chinese are disconnected from the global ecosystem, so they develop locally. And right now in China, frankly, there is a lot of activity happening in this regard. They are still a few years behind the top teams in the world, but I think they will easily catch up in 5 to 10 years.
What China forgets, as I think Elon Musk once mentioned, is that no place in the world has as many smart, hard-working people as China. The leaders may be bad, the system may be bad, but no country in the world has as many smart and hard-working people as China. This will soon bear fruit.
Simon Brown: One hundred percent. I agree with that. You are bearish on China, but you cannot argue against 1.4 billion people. Many of them are very smart and many of them work very hard.
Thank you very much for your valuable time, Viv Govender from Land Swiss.
Listen to the full MoneywebNOW podcast here every weekday morning.