At this rate you'll get to 90% by the time it resolves false

@GameTheory "99%"

This is why we need real money prediction markets


Man, I wish you could bet money on this site lmao



Apparently he's pro-homeopathy as well. Doesn't seem like a very reliable source to me.

Seems like the you could have a system that passed these metrics but wouldn't actually be "AGI".

Russia has stated that its war aims do not include taking Kiev or Kharkiv. Updating down a lot on this info.



1% seems vastly overconfident.

> "Russian victory: 70% - Either has more land or territory under its control than when the war started, gets international recognition for what it already controls, i.e. Crimea, changes the status of the Donbas, or gets a no-NATO pledge or some other guarantee of Ukrainian neutrality." Honestly 70% seems low for this, but I think it's wrong to call this a victory for Russia. In fact I would even go to 90%. Imagine if Ukraine overruns Donetsk and Luhansk, joins NATO, pushes the Russians out of all the captured territory elsewhere, Russia takes 90,000 ...
@(cameron_parker) > With half a month left, it seems very possible for Russia to take the city if it wants to. Urban warfare is extremely hard for the attacker, and is only even attempted when the attacker has a 3:1 numerical advantage or better. Recent sieges like Aleppo or Mosul took 6-9 months after the city was surrounded, with the attacker outnumbering the defender by 3:1 or even 10:1 for Mosul. Kiev allegedly has 80,000-100,000 defenders replete with various anti-tank weapons and with more supplies and gear arriving every day. Kiev may be in ...

Updating to 67% for Trump, several sources have said that riots are usually good for the right wing candidate. Not taking a moral stance here, just a prediction.

Also this:

"having less than ten SAT exams as part of the training data"

How many SAT exams are in GPT-3's training data? You have no idea.

So what would happen if a system like GPT-4/5/6 that used a truly enormous training set won this? Resolves ambiguous?

This question is nonsense.



What about for the next 12 months?

17% seems very high to me. This crisis has more time to run into 2023 and even adding it all up I think 17% is too high

In the long-run, Metaculus (or some other prediction platform?) should remove the manual review process for new questions and assume that there will be millions of different questions, assume questions have a range of quality and a range of relevance, and find ways to deal with that systematically rather than manually. I'm not sure exactly how that would work, but I am pretty sure that it needs to be done. One idea is that questions should be split into observables, akin to oracles in the crypto world, and aggregates, akin to ETFs in the finance worl...


The interesting region between now and late 2021 has been compressed to about 1cm on my 30cm screen. I can hardly see what I am doing.

A UI improvement would be to widen all these bars to take up the whole screen.


> we assume the MS is not an internet-accessible resource What if resources that are extremely similar to MS are internet-accessible? How similar are they allowed to be before it doesn't count? > but no party is allowed to consult additional humans, What if MS was trained on a large database of humans answering questions similar to the ones used in the test. Does that count as "consulting" them? What if MS was trained on a video/text/audio record of everything every human has ever said or done (or something that approximates that). Does that count a...

Russia has stated that its war aims do not include taking Kiev or Kharkiv. Updating down a lot on this info.



90% are you serious? You want to make a bet at or near those odds?