@Roko Do you really think internet points are worth more than having an accurate model of the world that's publicly shareable?

I spent like 40 minutes trying to figure this out more. I feel like I do not have enough insight to form a clear model away from the 5%, but here are some things I discovered so far, in case other Metaculus users can "tag-team" this and we can do some [factored cognition](https://ought.org/research/factored-cognition). 1. There have [indeed been complaints about 3 Gorges flooding ever since its inception](https://www.google.com/search?rlz=1C5CHFA_enUS880US880&tbs=cdr%3A1%2Ccd_max%3AAugust+2019&sxsrf=ALeKk035TScS1Ls58w3V7xZCpSG_GWOmxw%3A1595826564319&ei=...

Metaculus currently thinks that there's a 3% chance of human extinction this century. Metaculus currently thinks that there's only a [50% chance of positive transition of radically smarter than human artificial intelligence](https://www.metaculus.com/questions/4118/will-there-be-a-positive-transition-to-a-world-with-radically-smarter-than-human-artificial-intelligence/). Metaculus also thinks that there's an [80% chance of smarter than human artificial intelligence by 2090](https://www.metaculus.com/questions/3479/when-will-the-first-artificial-general...

@notany

If you think that is too optimistic, I'm willing to bet against you at those odds, and bet $1 against your $900,000,000.

— edited by Linch

@(Sylvain) I looked into this more and I'm more confused. 1. This is one of the estimations I've seen, using extrapolated figures from market data: https://www.facebook.com/robert.wiblin/posts/884838271895 This puts the economic cost at ~100 trillion. 2. Another assumption is to ballpark it at the economic harm of the Spanish flu. https://www.reuters.com/article/us-reutersmagazine-davos-flu-economy/flu-conomics-the-next-pandemic-could-trigger-global-recession-idUSBRE90K0F820130121 https://files.stlouisfed.org/files/htdocs/publications/review/08/03/...

Conditional upon an unaligned superintelligence that's both capable and willing to kill >10% of humanity, I see little reason why it won't also kill >95%.

The main exception I could think of is if the superintelligence was aligned to a genocidal dictator, or otherwise used in a bloody war for decisive strategic advantage.

I did not give the probability of the latter a lot of thought; it's possible if I think/read about this for a few more days my probability for this question overall will go down to 80% or up to 98%.

— edited by Linch

Have the option to display more granular/coarse calibration plots.

Right now (at 5% granularity) some of the bars on my calibration plots have only one prediction.

Today I learned that I am not a general intelligence. :/

A long time ago (in pandemic time) I suggested that Metaculus should have more visible UI for closed and resolved questions, so people won't accidentally link expired Metaculus predictions and wrongly think that the questions are representative of current Metaculus opinion, especially for people from other communities (eg EA) who are not themselves forecasters. Some other people (including I think @Tamay?) agreed but were unsure what changes can make this visible. I now have a proposal: For closed/resolved questions: automatically change the title/he...

@D0TheMath

In the off chance that CadeMetz is reading this, I want to say that I've personally attended multiple SSC meetups and I am very much not white. Nor do I recall times where I was the only ethnic minority.

I have also never been recruited into white supremacy.

@JonathanRay 99% seems really damn confident here, given the uncertainties involved.

I'm curious whether Metaculus predictors would consider the following to be fundamental changes to the human condition: - Legally enforced limit of 15 hour workweeks for >50% of adults, lifestyle of >90% of people materially better than median 2020 American. - UBI being the default for >95% of humans. - Median newborn has IQ equivalent to ~200 in 2020 units - Worldwide abolition of sexual reproduction. - Abolition of death - Abolition of extended nonconsensual suffering (via genetic engineering, drugs, or very good psychotherapy) - drugs or genetic eng...

@Jgalt

That title was not the most fortunate.

Might be too much of a hassle to implement technically, but I'd like an ability to go below 1% or above 99% for some binary questions. I understand that this will be too punishing for newcomers, but maybe you can make it so you need to be above a certain Metaculus level to do this? I think this is important because good thinking around tail risks is important for the Metaculus agenda of foreseeing the future of science and technology and long-term risks. For example I'd like to predict below 1% on these questions, but I'd have to think a bunch about how...

Metaculus median has consistently been at April 2 since March 22nd; which is a much more respectable stability than on similar Metaculus/covid questions in the past.

Still too soon to tell if this is the right prediction (I place moderate credence in April 1st, as well as moderate-low credence for a long tail depending on testing capacity limits, but tentatively April 2nd +-1 day looks reasonable, so good work, team!

— edited by Linch

Thank you for having this very important question on Metaculus. Black Bean Crunchwrap Supremes have become one of my go-to foods this pandemic, so I'd be really sad if the price increases substantially.

— edited by Linch

[Base rate of suicide about 8.0 per 100,000 yearly for USA women of GM's age](https://www.nimh.nih.gov/health/statistics/suicide.shtml) [Women in prison are ~9x more likely to commit suicide](https://www.penalreform.org/blog/suicide-in-prison-a-new-study-on-risk/) [It was hard to find base rates for child sex offenders, but anywhere from ~10x to ~200x seems plausible](https://pubmed.ncbi.nlm.nih.gov/16040578/) Even at 200x, we're still "only" looking at 1600/100,000 per year, or ~0.8% in a 6 month period. So almost all risk continues to be from the ...

@Jgalt

Yeah, or "base rate of suicide for arrested child sex offenders with extremely close connections to Jeffrey Epstein." That's like 67% if you use Laplace right?

@johnwave

You and @Skyt3ch have huge gains of trade to be made by betting against each other.

avturchin from LessWrong estimates that AlphaStar used 2.8 x 10^4 petaflops.