@(D0TheMath) On the one hand, this is extremely impressive. On the other hand - come on, there's no way, right? I think this would qualify as an *[impossibly](http://nautil.us/blog/impossibly-hungry-judges)* impressive forecast.
Looking more closely at the paper, what it's saying seems to be that empirically, US political instability is cyclic with peaks every ~50 years, going back. That's pretty reasonable; maybe it's true for generational reasons. But it gives a rough estimate of "superimposed cycles with periods of 50 +- 10 years", which means that i...
Ay
@krtnu!
@jgalt!
Oh
Oh no, oh no
Oh yeah
Diridiri, dirididi @alexrjl
Go!
Sí, sabes que ya llevo un rato esta pregunta mirándolo
Tengo que predecir hoy (@Jgalt!)
Vi que [Wikipedia](https://en.wikipedia.org/wiki/List_of_most-viewed_YouTube_videos) ya estaba llamándome
Muéstrame el camino que la línea de tendencia va
Google Sheets es el imán y yo soy el metal
Me voy acercando y voy armando el plan
Solo con [graficandolo](https://i.imgur.com/jxnp1sB.png) se acelera el pulso (Oh yeah)
Ya, ya las vistas están aumentando [menos de lo normal](ht...
I think it's difficult to come up with a truly justified Bayesian prediction for this question, as a lot depends on how exactly you structure the priors and conditionals. Let me start by telling a story.
Suppose that wild game meat becomes a popular delicacy in the US, so that every several years, an epidemic of some kind emerges in the region. One day, an avian influenza virus emerges less than 10 kilometers from a biological research facility. This facility has a [BSL-3](https://en.wikipedia.org/wiki/Biosafety_level#Biosafety_level_3) laboratory, whic...
There are several questions like "Will X happen before 2021?" where X is some vaguely Poisson-like event, i.e. on any particular day it may or may not happen with some probability - an earthquake, for example. The right thing to do is roughly to set the probability to p, then update to 364/365\*p on January 2nd, then 363/365\*p on January 3rd, etc. decreasing to 0 (this is only approximately optimal, but never mind that).
However, this is an enormous hassle and I don't think anyone actually does this. You often see such questions where the median predic...
I think the most reasonable way to break this down is that compute used = nominal dollars expended * TFLOPS-days per dollar. My basic argument is going to be that expenditures have increased rapidly but are likely to plateau.
AlexNet, which seems to have kicked off the current era of deep learning on GPUs, was trained on two [Nvidia GTX 580s](https://en.wikipedia.org/wiki/GeForce_500_series#GeForce_500_(5xx)_series) for 6 days. For simplicity let's amortize hardware costs over 1 year. The nominal cost then would have been roughly 2\*$500\*6/365 = $16. T...
(warning: ridiculously long post)
Reading the comments, I feel like everybody thinks this is much more likely than I do. I'm especially confused that I haven't seen any top predictor, or actually anyone at all, argue for "no"; with such one-sidedness I'm concerned I'm just making a fool of myself by not abandoning my position.
There's a substantial probability I look like an idiot in, like, 24 hours, but here goes. This is my position:
- I believe the rioting should be greatly concentrated in the first few days.
- My not-so-thorough [Wikipedia sca...
That's a good point (I would guess the number of Metaculus users working from home right now is very high), although I think a substantial amount of the probability mass is on "someone was infected earlier this year, but hasn't gotten a serology test yet". Once serology tests are widely available and a majority of the population is tested, it will become more clear.
GPT-3 is now available to the public. I asked it the questions in the "default-factual-answering" preset, which I checked was able to accurately answer simple questions like "How many legs does a horse have?" etc.
Of 20 prompts, it correctly answered three, and additionally had partially correct answers to another two. Its responses are below.
```
Q: The following sentence doesn't quite make sense. But if you replace a single word with the word apple, the sentence will make sense. Here's the sentence: "James took a bite of the cake, it was soft, so Jam...
The FiveThirtyEight snake chart currently shows the modal outcome as 186 votes for Trump, 352 votes for Biden. The FiveThirtyEight model outputs place a 0.9875% probability on this value. Presumably this is an upper bound on FiveThirtyEight's prediction of this question resolving true, because an electoral vote count could correspond to multiple "maps".
The current EV count with the highest single probability is 413 for Biden, with probability 2.5975%.
***TL;DR this question is going to be total chaos, don't predict***
***EDIT: If you do choose to predict, please message me on [the Metaculus Discord](https://discord.com/invite/D69rP5k). This question requires very careful coordination, or we're all going to unmake and it'll be a huge waste of time.***
The question states that puppet accounts are prohibited. However, as far as I can see, it is probably not possible for the admins to definitively identify and remove all puppet accounts.
Here is a strategy that one could hypothetically use to gain a la...
Joe Biden has 4% of statements rated Pants on Fire; the last one was February 2020, and before that in February 2015. So the base rate there is pretty much zero.
Donald Trump has 15% of statements rated Pants on Fire. Currently this year:
```
Jan: 1 of 8
Feb: 1 of 11
Mar: 5 of 18
Apr: 5 of 12
May: 6 of 15
```
Both the number of statements rated, and the proportion of "Pants on Fire"-rated claims, have risen recently. I would guess that this is due to the increased activity and number of off-the-cuff statements made during the coronavirus pandemic, etc....
FiveThirtyEight's model outputs currently put the national vote share for all 3rd-party candidates at 0.67-1.92%. From PabloStafforini's spreadsheet, the voting population in states where Kanye is on the ballot should be 46*138.8/323.1=19.6 million voters. This leads to an estimate of 131-376 thousand third-party voters in those states.
I'm guessing Kanye gets ~10-30% of the third-party vote share. This results in roughly 13-112K votes.
@(mdickens) I think this might be a reasonable approximation:
Consider two American-style put options with a strike price of $100 and $101 respectively, and suppose they cost C1 and C2 respectively. Consider these two cases:
1. With probability 1-p, the stock stays above $101. You lose all your money. In this case, the options are equivalent.
2. With probability p, the stock goes below $100, and at some point you exercise (let's assume this is guaranteed, although it's possible you might not exercise at $90, thinking the stock will go to $50, and inste...
Wikipedia has a list of [deadliest earthquakes by year](https://en.wikipedia.org/wiki/Lists_of_21st-century_earthquakes#Deadliest_earthquakes_by_year).
I will construct a prediction on Elicit as follows. In the last 20 years, in 10 of the years, there has been an earthquake with at least 1,115 fatalities. I will assume that this event occurs independently each year; hence, in a span of 10 years, the probability that there is no earthquake with at least 1,115 fatalities is 0.5^10 = 0.001. Similarly for other levels. I'll also leave 10% extra at the top. ...
The market puts about a 2% probability on this happening. AAPL would have to go down to a price of about $230 for this to resolve positive. Below $230, AAPL puts expiring in January cost an additional ~10 cents per $5 decrease in the strike price, i.e. the estimated probability of AAPL decreasing below $230 is approximately 0.10/5.00 = 2%.
On the question regarding the [month with the highest number of cases](https://pandemic.metaculus.com/questions/3781/which-month-of-2020-will-see-the-biggest-global-increase-of-covid-19-cases/), in every month of 2020, the community prediction basically said that cases would start going down. And it never did, so the prediction just kept moving right every month.
Now on this question it seems again that the community expects a precipitous drop in cases with ~30M predicted for [Q4 2020](https://pandemic.metaculus.com/questions/3767/how-many-new-cases-of-...
A video of the talk discussed by Jgalt below is available [here](https://www.youtube.com/watch?v=e5mTW25EzLs), and an article he wrote is available in, appropriately, [Quanta Magazine](https://www.quantamagazine.org/john-preskill-explains-quantum-supremacy-20191002/).
Based on my cursory glance of the video and article, the talk is framed as "Google has now demonstrated quantum supremacy with the Sycamore quantum processor (assuming their results are not fraudulent). What's next now that this has been accomplished?"
E.g. in the talk, he says:
> We've ...
[This site](https://www.govtrack.us/covid-19) shows 19 members of Congress as having tested positive so far. Their ages are as follows: `38 44 45 47 49 50 56 57 59 60 62 63 63 65 67 67 67 72 72`.
Using [case fatality rates by age](https://ourworldindata.org/mortality-risk-covid#case-fatality-rate-of-covid-19-by-age), we would expect roughly 0.398 deaths. So, if the age profile doesn't change, roughly 48 more people would have to be infected for the expected number of deaths to be 1 (including the 3-4 active cases right now).
Note that this doesn't cons...