Carbon Brief is now giving 4%.

(Yes, this means PredictIt is mispriced. I don't know if that's a big deal after fees, but I guess if you know people who are on PredictIt and allocate money effective-altruistically, please consider letting them know.)

— edited by steven0461

Based on [the data cited by Fruo](, no earthquake since 1985 technically counted as major. Loma Prieta and Northridge met the damage requirement, but at 6.9 and 6.7, they narrowly missed the magnitude requirement. Loma Prieta barely did enough damage, so let's say damage and magnitude were both coin flips and if we rerolled it we'd get a major earthquake 1/4 of the time. Northridge easily did enough damage but only came within 0.3 of the magnitude requ...
[Our World in Data]( has a good graph of US hurricane landfalls through 2017. A quick look at Wikipedia suggests a Cat 2 and a Cat 4 for 2018, and just a Cat 1 for 2019 (so far, but we're far into the year and it won't mislead us too much to use that as the full count). Assume the number of US landfalling hurricanes in a year draws from the past 20 years, so 2/20 chance of 6, 2/20 chance of 3, 3/20 chance of 2. By my count, 13% of landfalling hurricanes since 2000 have been Cat 4+, but let's use 20% to be safe in view of recen...
According to Metaculus, the [75th percentile is 60M infections]( That's under 1% of the world population. Presumably the fraction among Metaculus users would be lower. Only some of the infected will have a doctor's note. There have been 36 predictions, so I'll wildly guess 20 predictors now and 50 predictors eventually. I'd guess fewer users would go to the trouble of posting their doctor's note by the...

It would be interesting to see a question of the form "will one of the following ten improbable events, each of which is too improbable to get its own question, happen this year?"

* civil war resulting from the 2020 elections * other unlikely wars (within EU, within NATO, US vs Mexico or Canada, US vs Russia or India or China in the next year) * unlikely technological breakthroughs (cold fusion, faster than light travel) * surprising but widely accepted scientific results (ESP, relativity disproved, life on Mars) * unlikely presidents (Oprah, The Rock, any Libertarian or Green Party candidate) * humans landing anywhere other than Earth/Moon/Mars, or anywhere other than Earth in the next year * average 2020 temperature at least hal...
The event in question has happened only once so far, which suggests "normally" the probability isn't much above 1% per year. That it happened in 2017 suggests this probability has been rising. But for the community's 68% in 4 years to be reasonable, it would have to have risen a lot. If I look at various other hurricane-related statistics, I see some support for increasing numbers of intense hurricanes, but I don't see any support for the kind of order-of-magnitude increases needed here. Ocean temperatures haven't made a sudden huge jump either. The "at ...

98+% chance to be in the top 2 according to Berkeley Earth. 2019 is currently #2.

The question text says:

This question will close retroactively to the resolution time.

The resolution time is still a few days away. So how come it's currently closed?

If there's a human-level AI by 2075, it will probably have existed for decades, which means an intelligence explosion will probably have happened, which means a huge amount of optimization will have been poured into getting more intelligence out of less computing power. Current estimates here seem too high by a lot.

I would love to see questions about Drexler-style nanotechnology / molecular manufacturing / atomically precise manufacturing. I don't personally know enough to formulate such questions.

@olliebase22 It's hard for me to imagine an AI system capable of killing >10% of the population that isn't also capable of self-improving and self-replicating to the point of having a decisive strategic advantage over all human actors. (Not impossible, though. For example, maybe it doesn't take a superintelligence to hack nuclear weapons systems.)


I haven't been following the plot, but a 0.9% chance of Daenerys Targaryen dying sounds absurdly overconfident.

I think it has to be the all-things-considered view, or else Metaculus becomes one of those games that ethically require their players to play badly in such a way that they constantly have to think about exactly how badly they're ethically required to play. I'd be in favor of eliciting inside view predictions separately.

— edited by steven0461

@Jgalt "When will the Bulletin of the Atomic Scientists announce we've reached 1am?"

Metaculus currently assigns a 37% probability that under 2% of the world population will be infected, so I think that's still/again not consistent with the 93% estimate here.

Sea ice decline slowed down a lot in the last few weeks and it now looks virtually certain that this will resolve negatively.