AI sustained innovation more than a disruption

Ashleay: Welcome to the “In Your Interest!” podcast. My name is Ashleay and this week I'm joined, as usual, by my chief strategist colleague, Sébastien McMahon. We'll be talking about the growth of artificial intelligence and its impact on the stock market with Maxime Houde, who is a director and portfolio manager in thematic investments. So, hi, Sébastien. Hi, Maxime.

Maxime: Hi, Ashleay.

Sébastien: Hello, Maxime. It's good to have you here. And it's not your first time on the podcast. You've been here before.

Ashleay: Yep.

Sébastien: Right? Ok, just making sure I didn’t mix that up.

Maxime: Yeah. Second time on the podcast.

Sébastien: And you know, we've been talking about AI everywhere. ChatGPT also is all over the place. We're talking about Gemini now, I think, from Google, which is maybe the next thing, maybe not. But, we thought it would be interesting, Ashleay, to bring Maxime in today to just say, all right, what brought us here? It’s not like AI just appeared out of nowhere. No, there’s story behind it. So, can you maybe just give us a bit of the history that led to ChatGPT?

Maxime: No, absolutely. Look, it's been a big journey, a long journey. Actually, it started in the mid-1950s. When we think about a large language model like Gemini, like ChatGPT, behind that is what we call an advanced neural network. And that was actually invented in the 1950s. Now, if we look a bit closer to us, in the year 2000 plus, the big step forward that we took was actually in 2012. And what's interesting is it was here in Canada, actually, it was in Toronto. So, each year, there was a competition that was called ImageNet. The goal of the competition is that anybody around the world could try their model there. And the goal was to… you have a bank of images, and you would use an artificial intelligence model to try to say what you were seeing on the image. And there's one thing that was made by two young grads and their PhD supervisor that won this contest by the biggest margin we ever saw. And what was really interesting is that they used a GPU, so a graphic processing unit instead of a normal computer unit, a CPU that we have in a computer.

Sébastien: So, GPU, that's kind of the graphics card in gamers’ PCs. At least until that moment, that was mostly the use of GPUs. Am I correct?

Maxime: Exactly. That's exactly what it is. So big kudos to the video gamers out there because, I would say, the evolution of AI is really coming from them, overall. But why it's super interesting is GPUs are really good at repeating the same process multiple times in comparison with the normal PC CPUs where it's good at doing multiple steps, one after the other. And so, when you think about AI, usually you do the exact same step multiple times, so that's why GPUs are perfect for that.

Sébastien: And behind those artificial intelligence models, there's a lot of mathematics. And, you know, multiple operations done very quickly in parallel. So that's where the GPUs come from, and as we're recording this, yesterday, we had the new Nvidia event, showing their new products. And you know, now, it's 280 billion transistors on one chip. I'm not an expert in that, but you can use them in parallel. So, you have like millions of billions of transistors that can work together. So, this is the acceleration that we're seeing?

Maxime: Absolutely. Two things to point out there. Just to put an image in people's minds, the new chip it sits in the palm of your hand. So, when we say 280 billion transistors, it's at the atomic level. So that's probably the piece of equipment that humans are able to make that is the closest to magic.

Ashleay: Yeah.

Maxime: Now, the other thing that is interesting is that they have a full-scale system where they combine a rack of those GPUs. And what they highlighted is that it was able to digest all the available information in the world in roughly one second.

Ashleay: Oh, my goodness!

Sébastien: So, in 2012, someone showed the way that if you use GPUs rather than CPUs, you could, you know, accelerate this thing and make something very useful out of that. So since then, Nvidia is becoming pretty big, GPUs are getting stronger and, you know, the computing power is getting stronger and stronger, but on the software side, there was also other milestones.

Maxime: Yeah. The other biggest milestone actually happened at Google, in 2017. And it's really the architecture behind how we think about the model that they changed that's called the transformer model. That was the title of their paper. And really what the transformer model did is that you were able to repeat the same process, but do that multiple times in parallel. So what it allowed to do is to shrink the computing time that you needed to do to use those models. And that was really the biggest milestone to bring us to the ChatGPT moment that we had. The only thing that we needed after that was more computing power, and that was really coming from Nvidia being able to upgrade their GPU each year. Just to give you a sense of how to think: the rule of thumb for model improvement is if you want to improve your model two times in terms of consistency, you need to train it on ten times more data. And right now, we're thinking about trillions of inputs of data. So, you need a lot of computing power. So, all of this coming together last year led us to the ChatGPT release.

Sébastien: Okay. So, there was a lot of research being done on large language models. But now that we have this accelerator on the technological side, this is why it seems like it's growing exponentially is because lots of resources are being spent now on this topic, now that the way has been paved. So that's kind of how I would say it.

Ashleay: And it kind of feels like there's always a new trend dominating the tech world. Just a while ago, everyone was head over heels about the metaverse, and now AI seems to be stealing the spotlight. So, what's happening with the metaverse? Is it's still a game changer? And how is AI different?

Maxime: It's an interesting question. I think metaverse at the time is a great example of what a bubble could look like, because we started to estimate a lot of opportunity in the metaverse even before it was built. And that's a bit the opposite when we think about large language models and AI overall because we're already using it. So, I think that's the biggest difference. Now, will we hear about metaverse in the future again? Absolutely. And I think, yesterday, like Sébastien talked about, there was this big conference from Nvidia. And one of the things that they talked about the most was what we call digital twins, which is a good parallel to metaverse. If you think about, we need to build a building, the way it's going to work in the future, you can expect that they're going to build the building first in a digital twin environment – so a metaverse – to make sure that everything works. And after that, they will build it in real life. So that's a bit, now that we have the computing power that we needed, that's the way we're using the metaverse. So, those two things are converging at the moment.

Sébastien: And for R&D, you're mentioning a building, but you also mentioned once when we discussed, you know, new materials. So, if you want to build new science, basically you can try out in a simulation environment if, in the real world, things are going to work instead of just experimenting with actually real materials. So, there's just an accelerator of research and development.

Maxime: Absolutely. Like the two areas where it gets really interesting is actually, like you pointed out, discovering chemicals and also healthcare. Because you can expect or think about a future where we're going to recreate your body in a digital twin, try medicine on it, and if it works, we're going to do it in real life. So that's one of the areas that is extremely interesting and could have a lot of potential in the future, especially when you think about the research. Development for new drugs is one of the less efficient R&D spend that we have, like, worldwide. So, that's a great opportunity.

Ashleay: And with all the hype on AI and the strong rally we've had with the stock market, do you see the market as being in a bubble?

Maxime: It feels more like 1995 than 1999. Valuations are nowhere near where we were in the 2000s. As a matter of fact, if you look at Nvidia, Nvidia returned more than 240% last year. It's up close to 80% this year at the moment that we record this podcast.

Sébastien: So in the first… not even the first quarter of 2024, we had 80 some percent.

Maxime: Roughly 80%. So, what's interesting is when you look at the valuation for Nvidia, the price earning, it’s average with historical; it's in line with their historical average.

Sébastien: So earnings expectations are growing so strongly that the price is reflecting that.

Maxime: Exactly.

Ashleay: And it's been quite a year for the stock market, especially for the Magnificent Seven, you know, including Apple, Amazon, Google, Meta and others. Can you tell us a little bit more about them?

Maxime: Yeah, absolutely. So, great year for them last year. Lots of them are impacted by AI and will be AI beneficiaries. But most of them are also… they have their own idiosyncratic stories. Now the question is: “Should we still call them Magnificent Seven this year?” And I think the answer is no. We've seen a big divergence in performance this year. Actually, the worst stock in the S&P 500 at the time of this recording is Tesla.

Sébastien: This year? 2024?

Maxime: 2024, yeah. So, there's a big divergence. At the end of the day, Tesla is a car producer. Yes, they have opportunity in robotics with AI, but they need to sell cars. And there's a lot of competition happening right now. So, I think we should probably call them more Fantastic Four than Magnificent Seven. Because the one that has been working is obviously Nvidia. Meta, Microsoft and Amazon have done well. It’s been a bit more difficult for Google and Apple, but there’s some opportunity out there for them. So, I do think the Magnificent Seven are no more and we could call them the Fantastic Four, but we'll see as the year evolves, maybe we're just going to have a Magnificent One or Fantastic One.

Sébastien: Yeah, but the leadership is changing pretty quickly here. So that's the message.

Maxime: Absolutely.

Ashleay: All right. Well, that's all for today. Thank you, Maxime, for dismantling this whole phenomenon that is artificial intelligence and showing us how innovative it is. And I think this subject needs to be discussed a little bit longer. What do you think, Sébastien? I think we should invite Maxime back next week.

Sébastien: Of course!

Ashleay: Yeah? All right. So, Maxime, thank you, but now you're on the hook! And thank you also to our listeners. Please don't hesitate to reach out if you have any questions and we will see you all next week. Loved this podcast? Want to know more about economic news? Follow our “In Your Interest!” podcast, available on all platforms, visit the economic news page on ia.ca or follow us on social media.