Astral Codex Ten

538

15 min read

538

Article

538 is a recurring publication in the Astral Codex Ten archive, appearing 9 times across 9 issues between April 19, 2021 and January 30, 2024. The archive places it in contexts such as “538’s model for most of the 2020 predicted about 20% or less”; “538 and other top pundits (who gave Trump about 20% chance)”; “but it looks like both 538 and the prediction markets trust it quite a bit”. It most often appears alongside Polymarket, PredictIt, Biden.

Metadata

  • Category: Publications
  • Mention count: 9
  • Issue count: 9
  • First seen: April 19, 2021
  • Last seen: January 30, 2024

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Mantic Monday: Grading My Trump Predictions
April 19, 2021 · Original source
Prediction 1 was directionally correct. Was it overconfident? It feels like I was overconfident to say only a 20% chance for Trump to win way back in 2018, when in fact Trump came kind of close. But my prediction on Election Night was still only about 20%, 538's model for most of the 2020 predicted about 20% or less, so I think this might have been just right.
- Predicting that Biden (and the Democrats) would win the election. I kept buying this throughout 2020, on the basis of the spread between PredictIt (which gave Trump about 45% chance) and 538 and other top pundits (who gave Trump about 20% chance). This never stopped seeming like a good deal, so I just kept buying more and more of it as my other bets closed and more and more funds became available. I think I won about $200 on this, though it's complicated for me to say exactly how much because of the next point.
21
November 01, 2021 · Original source
(source) A lot of this comes from a single Fox poll which found found Youngkin way ahead. There’s been some debate over how much to trust it, but it looks like both 538 and the prediction markets trust it quite a bit.
Inline links: source
A lot of this comes from a single Fox poll which found found Youngkin way ahead. There’s been some debate over how much to trust it, but it looks like both 538 and the prediction markets trust it quite a bit.
Predictions For 2022
February 01, 2022 · Original source
No direct inline source block was recovered for this mention.
Which Party Has Gotten More Extreme Faster?
June 08, 2022 · Original source
No direct inline source block was recovered for this mention.
22
August 16, 2022 · Original source
No direct inline source block was recovered for this mention.
22
October 18, 2022 · Original source
Polls this year look bad for Senate Republicans. Pollsters’ simulations give them a 22% chance (Economist), 34% chance (538), or 37% chance (RaceToTheWH) of taking power. Even Mitch McConnell has admitted he has only “a 50-50 proposition” of winning.
Inline links: Economist, 538, RaceToTheWH, has admitted
Open Thread 248
October 30, 2022 · Original source
1: Polymarket, Manifold, and PredictIt now have shiny interfaces for predicting the upcoming US midterm elections. In terms of the Republicans taking the Senate, Polymarket is at 65%, Manifold at 58%, PredictIt at 73%, and 538 at 49%.
Inline links: Polymarket, Manifold, PredictIt, 538
Mantic Monday: Twitter Chaos Edition
November 21, 2022 · Original source
Polymarket again within 2% of Manifold. Only 23 traders here, and they’re a lot less optimistic than the Trump traders. FTX! 43 traders, seems like probably. I’ve seen a lot of Twitter takes about how rich well-connected people never get in trouble for this kind of thing, but the markets seem less cynical. 251 traders, and by the way amazing job by “mr22” who started this market on October 5. I also appreciate the relatively late end date - there’s another market “. . . by 2024” which is in the 30s, but that’s because people don’t trust the justice system to move quickly, not because they think he’ll be found innocent. There are a series of markets on sentence length which seem to suggest more than a month but less than a year in jail; this doesn’t really make sense to me and I’m going to nervously ignore them. Only 8 traders here, so take with a grain of salt, but this is a great example of the creative ways people are using Manifold. The market resolves not to “yes” or “no” but to the percent of FTX US users’ funds that they eventually get back; you make money if you were closer than other traders. Here they seem to think most people will only be getting about 14 cents on the dollar. There’s another market for FTX.US users which is a little higher at 29. 34 traders. I think this is too high; I bet it was some random third-tier insider, just because there are more of them and they’re under less scrutiny. Moving on to the effects on effective altruism in particular (just assume I have all possible conflicts of interest here): 272 traders, check the detailed resolution criteria. I think the strongest case is something like the one described in this article, about Center for Effective Altruism leaders discussing concerns about Alameda Research in 2018. The article doesn’t give specifics but my guess is they were the same issues Kerry Vaughn describes here (though see the followup comment by an employee who left FTX, casting doubt on Vaughn’s claims). That means the market hinges on whether Vaughn’s allegations fit the resolution criteria that “the unethical behavior must have been related to fraudulent investment strategies that involve spending other people's money without their permission”. Vaughn describes “poor capital controls, including a lack of distinction between money owned by investors and money owned by Alameda itself”, which sounds like it’s in that direction but could cover a wide variety of badness levels. My guess is everyone will end up agreeing that disgruntled Alameda employees whisper-networked that some things were bad about the company in 2018, some of the rumors got to CEA leaders, the leaders debated whether this was worse than normal for a tech startup, decided it didn’t rise to a level where they needed to publicly freak out, and moved on. Isaac will have to pay attention to the details as they come out and decide whether or not it qualifies. 45 traders. This seems to confirm that the CEA incident is responsible for most of the probability mass above; many fewer people think the FTX Future Fund (ie the charitable branch of FTX responsible for giving out their money) was in on this. Related: this market only has five traders, but I’m highlighting it anyway in the hopes that it gets more. The most money is on 2022. My guess is that we’ll find that they had terrible accounting practices in 2018-2019 of the sort that could be classified as criminally incompetent in a way that bled into fraud (but the trades went fine so nobody was harmed) and then they ramped it up a lot in 2022 to deal with the crypto crash. I think this market will be harder to resolve than people expect. 47 traders. Everyone is panicking about this possibility, but it looks like it’s not too likely. 10 traders. I’ll take this chance to say: a lot of media is predicting the death of EA, or a major blow to EA, or something in that category. Not going to happen. The media isn’t good at understanding people who do things for reasons other than PR. But most EAs really believe. Like, really believe. If every single other effective altruist in the world were completely discredited, I would just shrug and do effective altruism on my own. If they instituted the death penalty for effective altruism, I would do it under cover of night using ZCash. And I’m nowhere near the most committed effective altruist; honestly I’m probably below average. “Saint gets eaten by lions in Colosseum, can early Christianity possibly survive this setback?” Update your model or prepare to be constantly surprised. 6 traders. So, we lost several hundred million dollars of funding in a giant disaster which was also morally outrageous and demoralizing. It happens. But lots of people have already emailed me asking how to send in more money to help fill the gap. Some added something like “it was so depressing that all the FTX money meant my money didn’t make a difference, but now I can help again, and it’s great!” Can these people fill the hole? 32% chance that they can! 10 traders. And if they don’t, we’ll still probably do better than in 2021, before all the FTX money started rolling in. We’ll try harder to hammer in the point about not doing “ends justify the means” reasoning, and do some reorgs and purges to prevent anything like this from happening again, we’ll make a bunch of other changes - some reasonable, some panic-driven - but we’ll go on. If all the far-future stuff collapses, we’ll donate to global health charities. If the global health charities don’t work, we’ll fund GiveWell to sit around and figure out something that does. If GiveWell gets hit by an asteroid, we’ll work on asteroid deflection (actually I think we might already be doing that). If asteroid deflection turns out to be -EV, we’ll switch to shrimp welfare, or give ourselves Zika virus, or any of a million other things. You have no idea how committed we are to continuing to do effective altruism regardless of whether or not it’s “popular”. But it will be popular. 45 traders, resolution criteria at the link, notice the dip when the FTX news broke, followed by recovery as people had time to think it over more. Moving on to slightly less serious topics: The snapshot doesn’t show this, but one of the suggestions is Atlas Rugged. 67 traders, interesting to see where forecasters’ priorities lie. This was a big rumor early on, along with “everyone was on meth”, but the on site psychiatrist said it was false during an interview. 13 traders. WHY DO PEOPLE KEEP GOING ON PODCASTS? Midterms! That was two weeks ago? It feels like years! A week before the midterms, I wrote: Polymarket, Manifold, and PredictIt now have shiny interfaces for predicting the upcoming US midterm elections. In terms of the Republicans taking the Senate, Polymarket is at 65%, Manifold at 58%, PredictIt at 73%, and 538 at 49%. Congratulations 538! Mike Saint Antoine (who wrote the review of Viral in the last Book Review Contest) has put some more work into scoring midterm election forecasts. Here are some headline results: Mike writes: The reason I didn’t just do a three-way comparison between PredictIt, FiveThirtyEight, and Manifold Markets is that the Manifold Markets forecasts included fewer questions than the PredictIt and FiveThirtyEight forecasts. So in order to do a fair comparison here, I’ll be comparing the smaller subset of questions for which PredictIt and Manifold Markets both gave a forecast. So it looks like both Manifold and 538 did better than PredictIt, and there’s no clear way to tell which of the former did better. (except I guess you could do this analysis with just the subset of questions Manifold and 538 share, but Mike didn’t and I’m also not going to). PredictIt has a pretty consistent Republican bias (it’s a minor epistemic sin to accuse a prediction market of having a predictable bias unless you’ve made money exploiting it, I made $600 this election so I’ll let myself pass). In years when Republicans do better than expected, it will probably look better than other markets; in years when they do worse, it will look worse. Still, this is a bias, so I think we should take them doing worse this year as a fair reflection of their accuracy, even thought next year it could go the other way. My main two takeaways here are: PredictIt isn’t yet good enough that the ideal theorems showing prediction markets should be unbiased and better than everyone else apply to it. The obvious explanation is its $800-per-question cap. Polymarket doesn’t have that cap and it did better, although Mike hasn’t done a formal comparison to 538.
Inline links: more than a month, less than a year, FTX.US users, this article, here, Not going to happen, some reorgs and purges to prevent anything like this from happening again, I think we might already be doing that, turns out to be -EV, shrimp welfare, give ourselves Zika virus, wrote, Polymarket, Manifold, PredictIt, 538, Viral, scoring midterm election forecasts, https://substackcdn.com/image/fetch/$s_!z0IM!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fbcfb99af-ec23-4ff9-bcec-ffe248821898_1800x900.png, https://substackcdn.com/image/fetch/$s_!NSlN!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fc9c5602d-e060-454b-8a20-10e925c57526_1800x900.png, https://substackcdn.com/image/fetch/$s_!vBga!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F2b35fbd9-ee89-4f4f-b8b9-832742e483fd_881x289.png, https://substackcdn.com/image/fetch/$s_!h6tt!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F9355d197-997a-4c48-93c0-76abc23f2c12_881x287.png
Mike writes: The reason I didn’t just do a three-way comparison between PredictIt, FiveThirtyEight, and Manifold Markets is that the Manifold Markets forecasts included fewer questions than the PredictIt and FiveThirtyEight forecasts. So in order to do a fair comparison here, I’ll be comparing the smaller subset of questions for which PredictIt and Manifold Markets both gave a forecast. So it looks like both Manifold and 538 did better than PredictIt, and there’s no clear way to tell which of the former did better. (except I guess you could do this analysis with just the subset of questions Manifold and 538 share, but Mike didn’t and I’m also not going to). PredictIt has a pretty consistent Republican bias (it’s a minor epistemic sin to accuse a prediction market of having a predictable bias unless you’ve made money exploiting it, I made $600 this election so I’ll let myself pass). In years when Republicans do better than expected, it will probably look better than other markets; in years when they do worse, it will look worse. Still, this is a bias, so I think we should take them doing worse this year as a fair reflection of their accuracy, even thought next year it could go the other way. My main two takeaways here are: PredictIt isn’t yet good enough that the ideal theorems showing prediction markets should be unbiased and better than everyone else apply to it. The obvious explanation is its $800-per-question cap. Polymarket doesn’t have that cap and it did better, although Mike hasn’t done a formal comparison to 538.
Inline links: https://substackcdn.com/image/fetch/$s_!vBga!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F2b35fbd9-ee89-4f4f-b8b9-832742e483fd_881x289.png, https://substackcdn.com/image/fetch/$s_!h6tt!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F9355d197-997a-4c48-93c0-76abc23f2c12_881x287.png
24
January 30, 2024 · Original source
No direct inline source block was recovered for this mention.