00:31:50 · Nate Silver, Nate Silver, Joe Weisenthal, Tracy Alloway
[00:31:50] S1: Okay. [00:31:50] S2: The models keep missing low on Florida because you have a steady flow of red voters into the state, and we're using benchmarks based on the last election, but Florida's a moving target for that. [00:32:01] S3: It's interesting, if you look at polymarkets, because I keep pulling them up as we mentioned them. And everyone has shown a little bit of a bump for the Democrat in the last several weeks. So Mississippi, South Carolina, quite a big one. Annie Andrews at 28%, which is way higher than she was just a few weeks ago when it was at 13%. No bump at all in Florida. Angie Nixon is at 10%. And that's actually lower than where she was during much of the summer. [00:32:28] S2: Yeah, and this follows polling, right? The New York Times had some polling that was very strong for Democrats, but not in Florida. They had Ashley Moody ahead by six or seven or something. [00:32:34] S0: To what degree are prediction markets just like reflecting polling now? [00:32:37] S3: Yeah, talk to us about that. [00:32:39] S2: I think they're reflecting polling plus sentiment in kind of a, it's somewhere between smart and midwit. [00:32:51] S1: Right?
00:32:51 · Nate Silver, Nate Silver, Joe Weisenthal
[00:32:51] S1: Right? [00:32:52] S2: So again, people in prediction markets do have to quantify ultimately their price, right? And I think that exercise is helpful. When they go play poker, they can be like, oh, I think that guy might be bluffing. I'm saying, I say, well, I'm getting two to one from the pot, right? Last decision I have to make, do I think is a greater than a 33% chance or not? Like, I think the prediction markets have a slightly Republican-leaning clientele, right? Like Democrats are a little preachy about them sometimes. Republicans have more money, right? [00:33:21] S3: Well, we saw the famous, like, huge whiff in the L.A. mayor race where, like, Spencer Pratt, like, who in their right mind thought a Republican was going to win a mayorship at a major U.S. [00:33:30] S2: City? Yeah, but, like, they have, like, these, like, the recognition that polling error is correlated, like we talked about at the top of the show. That's an important recognition to have. The markets were actually overconfident relative to my model in 2016 on Hillary Clinton, right? So everyone knows that, but, like, I think The belief that, oh, polls always underestimate Republicans is not the most sophisticated version of that necessarily. [00:33:55] S3: Is there anyone systematically making money? I mean, I'm going to put you on the spot here or the Silver Bulletin on the spot. But do you feel that someone could make money? betting on prediction markets anytime there's a deviation between silver bulletin model for a given race and the prediction market?
00:33:55 · Joe Weisenthal, Nate Silver
[00:33:55] S3: Is there anyone systematically making money? I mean, I'm going to put you on the spot here or the Silver Bulletin on the spot. But do you feel that someone could make money? betting on prediction markets anytime there's a deviation between silver bulletin model for a given race and the prediction market? [00:34:14] S2: Well, look, I would say when our model came out on August 11th, we moved the markets by several points, right? And so like, if you have markets that can look at our model and are influenced by our model, also look at other models, right? And on top of that, incorporate some degree of vibes. I mean, I helpful. They can also just get to things faster, right? If there's a scandal, the New York times reports, on Thursday morning, then it will take several days to show up in the polling, right? And so like, yeah, I mean, I went through recently and looked at every race where you have a gap between Flipper, my model, and Polymarket. And like, I agree with Polymarket on half of those, you know, although ironically, some of the ones I agreed with, I'm like, oh yeah, Flipper's way too high in South Carolina. And then there's this good poll from South Carolina that brings, you know, so maybe you're kind of arguing it to yourself a little bit, but like, yeah. So the test would be like, you have to create like some type of laboratory environment where prediction markets, aren't allowed to look at polls. And my expectation is that prediction markets would be extremely terrible.
00:35:10 · Joe Weisenthal, Nate Silver
[00:35:10] S3: Well, it seems like actually the test would be, you would want to have prediction markets that aren't allowed to look at models, right? Because of course they're going to look at- Fair enough. Right? Because ideally you wouldn't want the day the silver bulletin numbers drop to influence the prediction markets. Ideally, you're looking at polls and creating a model and prediction market traders are looking at polls and creating a price- And the question is, which is the better way to derive a probability? [00:35:38] S2: Yeah, look, and like building a simulation with all these regional correlations is tricky, right? We are like, as of when we're recording this, we are like literally like 0.995 correlated with polymarket on individual states. We have the overall democratic win probability a bit higher. You know, I don't know if that's because it's attracting a different type of trader, right? Like that, like you get maybe more of the you know, the whales, right? Not like the French whale from 2024, but like whales, like degenerate rich guys with a lot of money to burn, right? [00:36:11] S3: The dream. [00:36:12] S2: Whereas they're maybe less likely to invest in individual races. But there's really not that much of a gap, right? [00:36:20] S3: You've been building models for a really long time. And of course, you first sort of became a household name in 2008. So that's 18 years ago. [00:36:28] S2: That's crazy. [00:36:30] S3: So that means there are people who are going to be voting in this election that were, like, born during the Obama administration.
00:36:30 · Joe Weisenthal, Nate Silver, Nate Silver
[00:36:30] S3: So that means there are people who are going to be voting in this election that were, like, born during the Obama administration. [00:36:35] S2: Oh, my gosh. Oh, my God. [00:36:37] S3: Some of these things freak me out a little bit. Can you talk objectively about, like, is it unambiguous in your mind that AI has made it easier for you to build models and to do your job? [00:36:49] S2: Yeah, for sure. Look, I get to hire two world-class programmers, coders, for programming. whatever I pay for the max plans, right? A hundred or 200 bucks a month. So I have these complex projects that you definitely couldn't one shot or even come close to it. But I know this really well and I've quoted these things myself. And I know when you feed GPT or Claude a 10,000 word program and say, carefully audit this, it's quite sharp about it. Now I would still say on like two thirds of prompts, I have to provide quite a bit of guidance, right? It's not like, oh, cool, man. Next step. [00:37:31] S1: Right. [00:37:32] S2: But like, it's great. It's a great teammate, you know, to take something in a more obscure example. So we have an NFL model. The NFL has all these complex playoff tiebreakers, right? They're purely algorithmic. And last year when I built our model, it took me two and a half days to code them up. And like some of the code was buggy. We skipped a couple of tiebreakers. It was very slow. [00:37:54] S1: Right.
00:37:54 · Nate Silver, Nate Silver, Joe Weisenthal
[00:37:54] S1: Right. [00:37:55] S2: You know, now you have GPT or cloud do it and they do it in 15 minutes while you're at the gym, right? And it's code is 10 times faster. Now, again, there are dangers, like they can introduce bugs and kind of hide them from you. We built a soccer model too. And like in one version of soccer model, it had the ratings like looping back around. So like the 2026 rating would inform like the 1900 rating and it got, it would do things that were, it was quite weird. But you know, the combination of having like a lot of initiative and entrepreneurial spirit and expertise. Cause there's a certain type of person I meet. You guys are probably in that category, right? Where it's like, it's like people are saying me, I feel like I'm the most productive I've ever been. [00:38:36] S1: Right. [00:38:36] S2: I feel like I'm getting two days of work done in, in a 10 hour day sometimes. And I feel like that some of the time, but, but again, I'm probably in the 99.9th percentile of like, these models are helpful with the exact, I'm not really a coder, right? I understand model architecture. I'm an empiricist. I understand probabilities, but like, my coding was always the weak point. So it's like, this exact thing they're really good at and it's really valuable to me is the exact thing that they do very well. [00:39:04] S3: Your FIFA World Cup model did have Spain versus Argentina in the finals.