My 'never predict on a question where the main thing you're forecasting is how a third party will evaluate a thing rather than the thing itself' rule may be long and poorly-written, but every once in a while it saves me 300 MIPs.

— edited by j.m.

In my view, this comment section overweights: - single polls in isolation - news stories - conjecture about how certain events will affect certain voters in certain ways - prediction markets - articles from pundits that are non-empirical This comment section underweights: - the predictions of sensibly-designed empirical models based on polling 538 is at Trump 10%. The Economist is at 4%. If you differ from those numbers by more than a few points, I think you should probably ask yourself what your personal model of the election knows that these models...
##Some rough impeachment math ###State Assembly The NY state assembly has to pass articles of impeachment. **You need 76 out of 150 members** to do this. 40 Democrats and ~43 Republicans have said they're on board so far. This represents 83 achieved out of 76 required, although the Democratic leadership specifically would have to give the green light on this. ###State Senate and trial The jurors for impeachment in NY are made up of 62 state senators and seven court of appeal judges. Two thirds is required to convict, so **you need 46 yes votes**. ~38...

I really recommend that Metaculus, whenever possible, discourage these resolving-positive-on-a-negative formulations. It's confusing and makes prediction results less accurate, in part because you have to account for people who've misinterpreted the question. Also, as others have noted, it's not even clear that's what's happening in this question because the resolution language seems to conflict with the headline.

@mishasamin Everyone conscientious enough to report for this question was conscientious enough to avoid covid. Everyone not conscientious enough to avoid covid was not conscientious enough to report for this question.

Although it's not the question upon which resolution rests, the community forecast for this AI question dropped from 2025 to "uh, maybe in a few months" in a matter of hours on Thursday (down from 2032 a few days earlier).

Maybe the canonical example of AI progress literally surprising the community!

@kalos Yeah, it's a shame this question tried to use experimental resolution, because the actual topic turned out to be (i) fairly interesting, (ii) fairly uncertain, and (iii) without a lot of good existing sources that clearly resolve it one way or another, all of which make a good community prediction more valuable that it would otherwise be.

— edited by j.m.

I'm predicting below the community average for the first time in about seven months (when Trump was (in my view, inexplicably) around 55%). We're now past Labor Day; polls are more predictive, and polling models like 538/Economist have Trump between 16% and 25% as of this morning. Unless I think I know something those models don't, I should be much closer to their views. I've been quite conservative on this question but I think the weight of the evidence is now strong enough to trust the evidence much more than my priors (my priors nudge me closer to...

I wonder how Metaculus would evaluate the probability of one million Antarctic residents by 2075.

Another upset for the community tonight, as Warnock and Ossoff appear to have won in GA, giving Dems control of the Senate.

— edited by j.m.

I've seen enough.

Biden will pass Trump in PA tonight or tomorrow morning and the networks will call it soon after.

Getting horrible flashbacks to the “what month will Covid-19 peak?” question of 2020, when we were sure that every wave was the last one.


  1. Incumbency helps, but it doesn't swamp other considerations.
  2. There is little evidence that a 'shy Trump' effect exists/will tip the election.
  3. Polls in 2016 were reasonably accurate, in line with historical expectations.

The prediction history for this question might be the most depressing graph of the whole pandemic.

March: it'll peak in April.

April: it'll peak in May.

May: it'll peak in June.

June: it'll peak in July.

July: it'll peak in August.


December: it'll peak in December.

— edited by j.m.

Applying a (fairly conservative) 8% growth rate to current numbers yields 10k deaths per day by mid-April. In my view, this is mostly a question about testing capacity and case confirmation than it is about lethality, which is clearly well into >100k/month territory.

@ugandamaximum he's +110 yes on this question, trying to juice the council of three

— edited by j.m.

@metani This is an important question. I favor including the J&J vaccine for three reasons: outright vaccine efficacy is less important to mass vaccination than I had assumed when I wrote the question; I didn't intend for the vaccines in this question to be effective against variants, just against the original strain (and it appears to be >70% against the original strain--it's variants that bring the efficacy numbers down); and finally this is a US question so using US efficacy numbers makes sense to me.

@nextbigfuture No call by a news organization is "legally binding," that's not what they're for.

Given the [resolution issues we’ve seen in other questions that require users to self-report](, I think the first resolution criterion (five users out of top 100 must report having ridden) seems overwhelmingly likely to turn the resolution process into a get-out-the-vote campaign for the top 100 users (who could be a totally different group of people when this is relevant in 2022-2035), much the way the questio...