@(nhuvelle) It's old news, really, that Tether is propping up the "value" of Bitcoin; it was obvious already in 2017, although that did not stop it from causing a massive BTC bubble. Tether (USDT) is supposedly a "stablecoin" backed 1:1 by USD for use on exchanges. However, Tether failed to produce an audit proving they actually possessed the claimed USD reserves, indeed their auditor ended up quitting on them because the whole thing was too shady. In reality it's an elaborate scheme procured by the exchange Bitfinex (which *curiously* consists of the ...
Edit: I see now why this confused me. This is the z-score relative to a baseline distribution, i.e. a measure of the significance, not of the size of the effect. I think in that case a binary question asking if the z-score will be higher than some fixed value (5, say) would have been better. There really isn't much practical difference between \(5\sigma\) and \(50\sigma\) in terms of significance; the only difference is the actual numerical number of deaths (and if that is what we're after, then standard deviations from a different distribution is a poor...

@SimonM That seems like a bad take, imo. The measure theoretic foundations of probability theory completely break if you don't allow 0 and 1 to be probabilities.

— edited by sbares

My thoughts after skimming through the paper: 1. I seriously doubt their approach will scale. 2. Even optimistically, it can only speed up repeated evaluation of the same neural net. This is already a pretty cheap operation; the real killer app for specialized hardware is *training* the net. 3. Their impressive performance stats are not the result of any actual demonstration, but merely a calculation of the "potential" performance of a hypothetical chip. 4. Trying to read the supplemental materials gives a 504, so I can't even check if the calculatio...

@Anthony After conversion the average score is currently , which is below the scale in this question. The community prediction is way off.

It will resolve ambiguously [...] if the Prize is awarded for a proof that the Riemann hypothesis is undecidable in ZFC set theory.

I don't necessarily disagree with this, but it is worth noting that RH is a equivalent to a sentence (for example: "for all integers greater than , the sum of the positive divisors of is less than "), so if it is unprovable then it is true.

@gjm I think the spirit of the question was clearly interpretation 1, but it has resolved based on technically satisfying interpretation 2. I (luckily) haven't made a prediction, but I think the resolution is wrong. It should be ambiguous at best.

@(bitshift) A pretty clear indicator that this (and other commenters' random walk models) is not a great model is that it fails to predict the several times bitcoin has fallen >50% within a year during the very time period it's modelled on. The historical price of bitcoin is pretty clearly not an unbiased random walk. Specifically, it is not memoryless, as there are periods of weeks to months where the price is trending (far more so than should happen by chance) in one direction or the other. Call it bear/bull markets if you will, or better yet bubbles ...

@irving Is that really the case for the Collatz conjecture? For all we know, couldn't there be some such that "collatz(n) halts" is independent of ZFC (say)? In that case, collatz(n) would not halt, and the Collatz conjecture would be false. Or do we in fact know that the Collatz conjecture is a -sentence, and not "merely" ?

— edited by sbares

It feels wrong, but I'm going with 1% for the reasons explained in the other question. Even though chess is almost certainly a draw, that also makes it almost (completely?) impossible to solve. A forced win would be much easier to solve, essentially because you only have to check the winning moves for one of the players.


Honestly, I think anything other than 1% is unreasonable. There isn't a plan for how this could realistically happen within the next decade. Never ever has anything on a fraction of this scale been achieved on such timescales without a plan. There are numerous technological problems as well that all need to be miraculously solved.

Had it said 2050 or even 2040, then it would be a different matter. But as it stands, I would rate this 0.1% if I could. Miracles don't happen, and it would take multiple for this to happen.

Will Coinbase default on an obligation to hand over their users’ coins on request by the end of 2022?

I would be much more interested in whether they default on their obligations to hand over their users' fiat. The same holds for all the similar questions, of course, though I'm not sure which of the exchanges actually have fiat trading pairs.


Margin trading for USDt/USD pair is now available on Bitfinex!

USDt/USD can be traded with a maximum leverage of 10x, an initial margin of 10% and a maintenance margin of 5%.

Bitfinex is now effectively willing to take 1 USD as collateral for 10 USDT...

@Sylvain I can only speak for myself, but I find that the very idea of a Dyson sphere is completely confined to the realm of science fiction.

— edited by sbares


On the whole, not very different from how it looks now.

@Sylvain It doesn't really work, though. You can arbitrage the question by simply putting three narrow distributions centered at 1, 2 and 3. This wins ~50 points regardless of the outcome. Removing all the weight from two of the options doesn't improve the potential score much. This is because most of the result space is impossible, so the mere fact that you can rule it out "for free" guarantees most of the possible points.

@devetec Quite interesting indeed. I think it is quite likely that this is a case of overfitting, i.e. "I Want My Hat Back" is likely to be part of the GPT-3 training set. For the record, I have not made a prediction on this question.

>If no serious attempts are made to pass the Bear Eats A Rabbit (BEAR) test, this question resolves ambiguously. I think this condition makes the question very unlikely to resolve negative, and very much less interesting; I'd probably put 99% even though my actual credence that an AI will be capable of passing the BEAR test by 2025 is more like 20%. Consider the likely scenario in which we simply don't have the necessary technology (i.e. at least an AI with considerable semantic text comprehension skills compared to any that exists presently, along with...

@AngraMainyu @Jgalt

I stand corrected with regard to the Chinese space station (I knew about the plans, but somehow forgot). As for all the others, the only one that seems remotely plausible is the Bigelow station. I doubt that it will serve as a permanent space presence. Nevertheless, I have updated up slightly based on the Chinese plans.