not a fan of the resolution criteria here since it's pretty clear some people are voting just because other people have already voted for 1%. this has now become a Keynesian beauty contest rather than an actual question about coronavirus.

reminds me of a question about Elizabeth Warren being the front-runner that resolved because prediction markets briefly spiked above 60%.

It's interesting how Metaculus is divided into "plausible extension of current trends" and "Dude, the singularity is just around the corner". Metaculus predics >70% chance of [AGI]( by 2075, and superintelligence a few [months]( later, but "will there be a million people on Mars" is onl...

Perhaps we collectively overestimated this one?

One think to keep in mind is that the P(coronavirus | metaculus user ) < p(coronavirus | general population), since Metaculus users probably have above-average knowledge+ability+desire to avoid Covid-19.

@(ThirdEyeOpen) Estimating from some of the numbers [here]( and [here]( I think it would be around 1700. World population at the time was around [600M](, 1/10th of what it is today. By 1640, the 13 colonies alone had 50k. So maybe 150 years to reach a similar proportion. Of course America is 40,000X closer and much more ...

Anyone want to comment why this is so low?

It seems like: 1) Trump wants to run 2) He got more votes this election than any other Republican candidate ever

Are people betting he will be too ill/in prison to run, or are they betting that the GOP establishment will figure out some way to sideline him?

Another anecdote about how adversarial a human tester can be: In the mid 2000's (AOL instant messenger days) I received an unsolicited IM. Bots were just common enough at the time that I wasn't sure whether or not I was talking to a human. At the time chatbot technology was fairly primitive such that the following dialogue universally tripped up chatbots: > Me: Hello, how are you? > > Chatbot: Hello > > Me: What is the sentence I just sent? > > Chatbot: My name is Bob When my unknown IM sender successfully passed this test "You just sent 'hello...

"Bennett, Lapid reach agreement on forming coalition to oust Netanyahu"…

— edited by nagolinc…

"Starting today Teslas in the US can automatically stop for traffic lights and stop signs. 🚦 🛑

Just say where you want to go. Your Tesla will drive you through city streets, across highways interchanges, go around slow cars, and exit all on its own."

— edited by nagolinc

Updating downward on the likelihood that Trump will lose by a landslide. If he has to flip 1-2 states I can see him contesting, but not if its 5-6.

@Jgalt Time for "How many more Starship prototypes will be destroyed before one lands?"

— edited by nagolinc


I think a nice follow on question would be "when will an individual Starship prototype fly to at least 10km for a 2nd time?"


Maybe "what will the democracy index of Russia be five years after Putin ceases to be president"?


a large machine learning model [yes]

that was trained on both images and text [yes] , and other data [???]

using massive computational resources [almost certainly, although not explicity stated, based off of the parameters compared to GPT-3 this is virtually certain to be >10^4petaflop days]

I think unless someone explicitly comes out and says "no, this isn't the secret project there is another", this should resolve positive.


I don't think there are nearly enough presidential elections for the statement "in this domain 20% for them really means 14%" to be meaningful.

@Matthew_Barnett Given the time frame of this question (2200AD), I think it would be helpful to define what "United States" and "China" mean. If, e.g. the US and EU form a political union, is that still the "United States"? What if Japan conquers China and the territory that is currently China then exceeds the US in all 3 metrics? What if the CCP is overthrown and China is divided into several countries (Eastern China, Southern China, Tibet and Outer Mongolia)?

@michaelchen I felt physical embarrassment reading that paper. Littered with phrases like "One can only speculate" and claims of "linear specimens" that can be seen from space. Rather than focusing on one or two interesting cases, it's a grab-bag of pareidolia and wild speculation complete with claims of a NASA conspiracy theory.

I'm actually a strong believer in the possibility of life on Mars (p>0.5), but this paper isn't it.

We have a pop!…

I believe this means the question can no longer resolve <3

SN8 still scheduled for Oct 11?

— edited by nagolinc

How about a feature to incentivize creating questions with a high information content (i.e. 50/50 probability)? This could be implemented by giving points to questions based off of their entropy $sum_states{(log(p)*p)}$ one week after questions are opened. One advantage is this would encourage people to ask questions where the answer is likely to be surprising, especially if the resolution time was required to be in a certain range (1 month, 1 year, 5 years). One disadvantage is this would be easy to game: "Will stock for company XYZ close higher 1 we...
@(Tamay) I don't mind deferring to future experts, the thing that bothers me about this question is that (as an expert on machine learning in the present) I don't know how this question would resolve for an existing system (AlphaStar). The words "most of the system" and "significant portion" are too poorly defined to determine what another expert would say. In order for me to make a meaningful estimate, I would prefer a stronger dichotomy "essential component" "nonessential component" or a qualitative metric "percent of compute on SGD during training"...


@Matthew_Barnett Can you edit the question to clarify whether Korean unification resolves positive or ambiguous?