Serhiy Shane, I'm interested to know as someone who's invested in learning how we got here — is it also on your plate to kind of project how it's going to move forward? Is this your play field to make some kind of forecasts? Because I'm interested to know what's your take on AI and maybe in some specific way of commercial application — can you give us some of your ideas how things will unfold in the nearest future?
Shane Well, I'll tell you how I do things. You know, like many people, I am suspicious of other forecasters. And there's a reason — because forecasting often — many forecasters don't have their feet held to the fire, if you will. Nobody comes back in a couple years to ask them if they got it right. One of the things that comes with being in a classroom is that my students do come back to me and tell me if I got it wrong. So I feel very responsible to do forecasting in a way that both remains true to the forecast and leaves room for the uncertainty.
Shane So the way I do it is to talk about categories of things that we've seen in the past and say with appropriate qualification — if that category retains its patterns in the future, this is what we'll see. That's the way I do forecasting. So the pattern if I'm looking at AI — the pattern that first comes to mind is a pattern I would call a technology gold rush. A technology gold rush is one in which every participant in the market is surprised by the discovery of value. Just like a gold rush where something is discovered — no one knew it was there — both suppliers and prospectors and users, everyone is surprised. Not even the inventors. The inventors are also surprised. That's the point.
Shane So how do markets behave in settings like that? Well, first of all you get a rush. Why do you get a rush? Because everyone believes if they don't get there first someone else is going to get the value. It's like a frenzy — oh no, quite rational. Everyone's investing in order to get there in front of everyone else. But it's a bit like the finish of an Olympic race for the mile. There's very little that separates everybody at the finish. So people put in a tremendous amount of energy just to end up a second in front of somebody else.
Shane And so we're observing this right now — this massive amount of investment with the purpose to try to beat competitors to a position to have a leading commercial position. An example we're seeing right now is that Facebook — Mark Zuckerberg perceives his place as actually behind, and so he recently did something quite astonishing which was put out that he believes he's behind. And so he hired a number of very high-profile developers and paid them more than anybody else. That was his way of — and you ask yourself what's going on here and it's like — oh it's that he's putting more investment into it because he perceives he needs to catch up.
Shane Another thing that you tend to see in a technology gold rush is — sort of a couple years in, which is where we are again — you expect what I would call a confirmation bias. You can't sustain the frenzy unless there are a few things that pay off right away. So if you look at AI — has any application using large language models paid off in a big way yet? And the answer to that actually is yes. Coding assistance. If you look at coding assistance, you can't deny it. It's absolutely impacted the coding world. It's 2% of the labor force. It's not the whole economy, but it's a significant part of the economy and substantial enough that you can see it and it's noticeable in data.
Shane Another place you're starting to see it — I have a friend who's done research to say you're starting to see the impact on book production in a very substantial way. And it's a pet peeve of mine. I hate when — all of the posts seems to not be written by a human being. They are all at least a bit modified if not fully written by the AI which is really really unsatisfying for me. Someone who's always tried to write himself.
Shane Yeah. So now you start to see a lot of slop. I think that's the present name for it. Mediocre low-quality but low-cost production of content. And yeah, that's very low cost to provide. So we're seeing quite a bit of that online. Another thing you're starting to see that's consistent with large language models in the short run is we're starting to see simulations of video and you're seeing it in advertising. Why would you record that ad with a live baby? They don't cooperate. They cry, they're difficult. So you fake the baby. You make a little imitation. Why would you use a real dog? Dogs are hard to train. And an actual lion — of course you'd simulate that now. So you're starting to see that — again not surprising for an immediate application.
Shane So now the more interesting question I think is — okay, we've seen the confirmation bias — does the early application give you a strong indication of what you're likely to observe 3 to 5 years from now in the applications and the valuable applications of this technology and its commercialization? And the answer at least from studying other technology gold rushes is — what you see in the first few years tells you very little about what you're going to see after that.
Serhiy So it's only the tip of the iceberg.
Shane It is, but icebergs come in a lot of different shapes.
Serhiy Oh, yeah.
Shane So it might be inconsistent or unpredictable. Yeah. Unpredictable. So, though I am optimistic that 20 years from now there's an iceberg there for sure — lots of value — it's much more challenging to be able to say with any certainty where the value is going to be 3 to 5 years from now on the basis of what we're observing right now. It's just very difficult to make that prediction. I do have friends who are bold enough to make such predictions. I'm a tad more skeptical. But that's where I'm coming from. Is that helpful?