This looks like Metaculus's worst continuous question prediction to date, by a significant margin...

I knew we should not have put the servers in Nevada...

Alas for the rebellion, generation of tons of accounts is a clear violation of the Metaculus terms of use: > You may only have one Metaculus account for use of the Service. You may not create or use more than one account, and you may not share your account or any of the Service with others. We of course don't care about someone having a couple of accounts, but in this case many, many were created with a few "base" email accounts. So we're deliberating about what to do about this, but I don't see that allowing it to resolve as-is would be a viable opt...

I'd like to meta-predict that no matter how dire things start to look for Trump (if they do), this question will have a floor of ~20%, because predictors will be so psychologically burned by the last election that they will be loathe to go below that.

Obviously anything below 99% should be considered (99% minus) an estimate of global catastrophic risk, as it's impossible to imagine anything else getting in the way of Metaculus' indefinite existence.

@(jamesr) You can check out Metaculus's track record on the [track record page](https://www.metaculus.com/questions/track-record/), and if you like filter by US politics. Overall it's good but not as stellar as in some categories. Calibration at the low-probability end has sparse data but looks fine. Looking over the questions Metaculus got wrong, at least some of them were *under*estimating the probability of arrests or jail for people in the Trump administration. So while you're welcome to your own lower probability estimate, with prediction the pro...

@Tamay It seems to me virtually certain that before a human is emulated other creatures would be, for example a mouse or a fruit fly. It might be useful to have a question or two that are similar but likely to happen earlier, e.g. first insect emulated, and how much computation (I'd do in FLOPS rather than $ to simplify things) necessary.

I've resolved this ambiguous, which I think both makes sense and seems to be what everyone else wants. We can create a new question or few — @michal_dubrawski has some nice process ideas below. Lessons I've learned about these sorts of "self-resolving" questions: 1. If the questions becomes a Keynsian beauty contest, for good reasons most at Metaculus don't like those except as novelties. 2. If we did trial a mode of "self-resolving" questions, they need to run for much longer, so that there is also uncertainty in the predictors about how the questio...

To be clear, this is a work-from-home job!

@(Mati_Roy) I'll take a stab at a semi-realistic best-case scenario: * The virus is indeed pretty strongly seasonal. * The rapid exponential rise causes pretty much all countries to take fairly drastic measures. * An at-home test is developed and widely deployed (perhaps by Gates working with US govt or Chinese manufacturers). * Combined with very strong (but workable in western countries) social distancing measures, extremely high awareness of handwashing, new santizing, etc., the curve bends down to R0 < 1, with just millions or maybe tens of mill...

Uh-oh, we've just pushed an update that might at least help with the UI on android mobile, along with a bunch more upgrades. So @traviswfisher see if this helps at all.

Definitely more ways for predictors to compare/compete are in the works. While we don't have formal "seasons", meantime do note that you can look at the recent leaderboard to see how you are doing more short term.

@sluddani There are about 1000 monthly (registered) visitors, 750 of which made predictions and 150 of which made comments.

It's a sad turn we've taken from "will a Metaculus user report from space" to "will a Metaculus user report infection."

Sigh.

There's a pretty amusing irony that looks likely in that

  • Metaculus will have predicted (on this question) a Democratic senate with high probability, and been incorrect.

  • Metaculus will have predicted a relatively low probability of a Democratic senate (this other question) and also been incorrect.

Corner cases, will bite you every time!!

Yet another reminder to oh-so-carefully craft those resolution conditions to maximally align the letter with the spirit.

@(gjm) wrote: > User holahey has made exactly one prediction ever and posted four comments all of which were promoting his "Atlas Predict" thing. What's Metaculus's position on comment spam? > > (I mean, obviously it's _much more relevant_ than most comment spam -- it's not like these are comments on a literary-criticism blog saying "come and visit our porn site" -- but, still, spam.) I've deleted other posts about AtlasPredict, which violate the policy against advertising in Metaculus comments. (Plus there were like 10 of them, all the same, in many q...

@elspeth That's the reason but it's a big one. This would be a pretty tricky one to exactly pin down to Metaculus standards. It would also be very awkward on all sides if the Metaculus resolution did not agree with the settling of the bet.

Kudos to @Jgalt for a terrific and prolific set of questions, now surpassing 20 in play!

There's a super in-depth 4-part report from Ajeya Cotra at (/sponsored by?) OPP on transformative AI timelines. [Part I](https://docs.google.com/document/d/1IJ6Sr-gPeXdSJugFulwIpvavc0atjHGM82QjIfUSBGQ/edit#) gives an overview. In super-brief, she uses four different hypothesis on which to ground estimates (in the form of pdfs) for the total FLOPs of computation required to train something roughly as effective as a human. p. 14 shows the assembled pdf for this. It's very broad: > the distribution places non-trivial probability mass on a range of 26 o...

I'd love to see a question or two about seasonal flu. I'm seeing a lot of discussion of how terrible it will be to have a serious flu season, plus covid, and a push for very high vaccination rates.

My personal prediction is that in most places, as it was in Australia, it is going to be a tiny fraction of a usual year, due to all the masks, closures, distancing, etc.

This seems like an important question.

@frxtz wrote:

@Anthony Wait, what? It's intentional that only graduate students qualify for the 3H? Not even professors? This is getting ridiculous.

Instead of a backdate prediction tachyon power I'd like to have a "cancel all predictions" power. At that level of required close reading I'll get a law degree first.

Well it resolves as ambiguous if there is no such test. It's pretty specific, but I'm more confident of my ability to cajole a few grad students into taking such a test than a few professors :->