Gold collapsing. Bitcoin UP.

[sickpig]

This is actually a really neat idea.

Those of us who have powerful PCs can run several Cash clients with their own datadir and ports for redundancy during the fork.

drawback: the seeder will list only client that use the default port.

 

[awemany]

Expected Value is a mathematical term E[X]=r((1-p)/p)
It gives a measure of the center of the distribution of a variable
It is not the same as the average.

Both betters were right about different things.
Both were wrong to make such a bet.

Mining is almost memoryless. Nonce are not reused so it is not perfectly memoryless, only so close to memoryless as to be memoryless for practical purposes.

I was thinking along similar lines today. Per SHA256 engine, serially going through nonces is very likely how it is implemented on ASICs today.

But between different SHA256 engines it would make a lot of sense to actually just select the extra nonce at random, to avoid communication overhead over dividing up the search space, wouldn't it?

Is that how it is done with the current ASICs? Is it also how it is done on the chip level between different SHA256 lanes?

 

[NewLiberty]

Nonce space is divvied up between them.
It is passed out in big batches so that all have enough to probably last until a block is found.
When they run out they request a new batch of nonces and get back to work so communication is minimized.

 

[Norway]

Mining is almost memoryless. Nonce are not reused so it is not perfectly memoryless, only so close to memoryless as to be memoryless for practical purposes.

@NewLiberty Would you say the following is a good metaphor for mining?

Actually, I think we are talking about 3 different statistical scenarios:

1) You have a dice with 10 sides. One of the sides is red. You throw the dice until you get the red side up. This is not mining, but what mining is approaching.

2) You have a bag with ten balls. One ball is red. You take out balls until you find the red ball. If this was mining, it would make sense to not put the balls back after you looked at them. But this is the scenario that is most far from mining.

3) You have a bag with 10^8 balls. 10^7 of the balls are red. You take out balls until you find a red ball. You could improve your odds slightly by not putting the balls back in the bag, but it doesn't really matter because the improvement is so tiny. I think this is how mining works today. (But I don't know, never looked at the algos, lol!)

 

[Norway]

I also wonder if this theory of mine is true:

If mining increases to insane (non-realistic) levels, the number of possible correct nonces becomes small, and mining can no longer be considered a memoryless process.

 

[majamalu]

Teaching the Spanish speaking world to love the clusterfork:

 

[Peter R]

People are fooling themselves into thinking that mining has memory without realizing it.

In mining, you are not guaranteed to find a solution even if you scan through the entire nonce range.

Firstly, the nonce range in the block header is only 32 bits, which is regularly exhausted and so the miner must adjust the block contents a bit to get a new Merkle root and start back with nonce = 0.

Secondly, even if the nonce space were 10,000 bits, you still wouldn't be guaranteed to find a solution.

The only exception to this that I can think of is if you argue that SHA256d has a weakness and that the more nonces a miner tries, the more he learns about the patterns for his particular puzzle, and the better he gets at picking "good nonces." But if this were true, Bitcoin would be in trouble for other reasons

The only answer to the question is t = 15 min. Craig Wright assumes that mining has memory without realizing that he's making this assumption, and is why his conclusion that SM is a fallacy is incorrect.

 

[Peter R]

Last night I wrote a simulator to demonstrate that selfish mining works in theory. Here is the simulator:

Here I'm showing how the simulator works by tracking the public and private chains:

Here I'm showing that the simulator confirms that the selfish miner earns more than his fair share of revenue for alpha = 40%. I also show that this matches Eyal and Sirer's analytic result:

Here I show that the simulator matches Eyal and Sirer's analytic results for different values of alpha.

Note that I'm assuming gamma = 0 (the selfish miner loses all block races).

The simulator is based on the description of the Selfish Mine algorithm given in the Eyal and Sirer paper. I tried to use the same variable names so that it is easy to cross check this algorithm with my simulator.

 

[AdrianX]

Mining is almost memoryless. Nonce are not reused so it is not perfectly memoryless, only so close to memoryless as to be memoryless for practical purposes.

@Peter R this is how it looks to me. You're also making an assumption. If it was perfectly memoryless then your wouldn't need to assum it is. we could not have difficulty adjustments.

 

[Peter R]

Craig Wright says these kind of simulators are wrong because of the way they calculate the probability of whether the SM or the HM finds the next block:

if Random[{0,1}] < alpha, SM finds next block else HM finds next block​

He says you need to factor in the "negative binomial" based on multiple independent identical processes taking place at any instant in time. But (at least as far as I can see) the only way the probability model above can be wrong is if mining indeed does have memory. Thus, Craig Wright is implicitly assuming that mining has memory without realizing he is doing that.

 

[Peter R]

Based on the suggestion of @Roger_Murdock, I've enhanced the Selfish Mine simulator so that the question of who finds the next block is settled by a true mining process. The commented out line (SM = Random[] < alpha) is replaced by the block of code below.

Not surprisingly, the results are identical to the previous simulator. This one just runs a lot slower, because it is actually mining blocks (albeit at very low difficulty).

 

[Peter R]

I've just learned that Craig Wright had recently offered a bounty for a convincing simulation and explanation of selfish mining. The bounty pays 15 BTC if you can prove that SM is a fallacy and 30 BTC if you can prove that it isn't.

 

[Christoph Bergmann]

Love this thread. Tomorrow we will fork to bigger blocks, finally, after all those years of debating and fighting, and the whole thread is about a bet about average blocksizes :=)

If I remember correct from school, the likelihood to get six eyes when you roll a dice is always 1/6, no matter how often you do it. But on the other side if you rolled it six times in a row and there was no success, the likelihood to have six eyes feels bigger than 1/6. And when you throw 20 dices ...

 

[AdrianX]

A peek into the YouTube of Bitcoin and the economy: It's reassuring to see what I think is happening confirmed on all these layers by different players:

Patrick Byrne bang on the money with what's happening: Go for a 45 min walk and listen to this it's essential for understanding why bitcoin base layer is important.

What Luke thinks he is doing: (building layer 2 because "reasons" undisclosed)

What a Cheef Market Strategist thinks the BIS and other Central Banks have as a plan B, Layer 2 enabled by smart contracts coincidence exactly what the BS/Core developers are working on:

rewind that Luke-Jr interview - its torture but it's clear he is building the tools for central bankers so they can stay in power on top of Bitcoin and he doesn't want to say why. Ask him why he is into bitcoin he'll tell you it's complicated, and leave it at that.
[doublepost=1501538579,1501537483][/doublepost]

Love this thread. Tomorrow we will fork to bigger blocks, finally, after all those years of debating and fighting, and the whole thread is about a bet about average blocksizes :=)

I was reminded yesterday of this prediction I made October 22 2014

I see a political standoff approaching, soon, where those secretly invested in sidechains seek to limit block size.

LN and even SegCoin can be defined in a way that they identify as a sidechain.

 

[Peter R]

Great article carefully explaining why the answer to the question from my bet with Craig Wright is t = 15 minutes, along with other interesting tidbits:

https://eklitzke.org/demystifying-the-selfish-mining-bet

 

[Bagatell]

I'm a sucker for omens. I just felt a ML 3.9 earthquake and minutes later see this -

 

[NewLiberty]

Craig Wright says these kind of simulators are wrong because of the way they calculate the probability of whether the SM or the HM finds the next block:

if Random[{0,1}] < alpha, SM finds next block else HM finds next block​

He says you need to factor in the "negative binomial" based on multiple independent identical processes taking place at any instant in time. But (at least as far as I can see) the only way the probability model above can be wrong is if mining indeed does have memory. Thus, Craig Wright is implicitly assuming that mining has memory without realizing he is doing that.

Norway is pretty much spot on with his view of the disagreement, I have only a few insights to offer in addition.

As I recall from the discussion, the talk with Craig was nothing to do with memorylessness, but he was discussing independence and the Expected Value computation. The memoryless discussion was happening about the same time with other folks in the discussion.

Craig was looking to show folks that the SM and HM probabilities are independent.
Thus P[HM|SM] = P[HM]
Expected value for HM is computed from N-1 and unchanged by SM finding a block or not as it is not released. HM doesn't change expectations. To HM Nothing has happened, nothing is recalculated.

Whilst attempting to explain this concept, "the bet" emerged, which I believe Craig was then going to use to talk about the difference between dependent and independent probabilities. This difference is a part of one of his six criticisms of the SM paper.

When I read the bet, I think that I read similarly to how Craig did, that the middle sentences were red herrings as they are not factors in the P[HM] expected value calculation.

However... abstracted from the discussion context, "the bet" image can be read as if "Assume..." refers to all sentences except the question at the end, and that the EV should be calculated from the SM event.

This is why I see both sides as right about different things.

Craig was using that channel to attempt to teach some things that would be helpful for understanding the papers he has written and expects to publish. Perhaps the opportunity to teach a broader audience about what distinguishes dependent and independent probability calculations is now lost because of it being overshadowed by "the bet", but perhaps not. The answer will depend on what the two Doctors can do with each other, and whether it is more important to be right, or to teach.

I am not at all confident that there will be any helpful resolution between those two.

This is why I see both sides as wrong to make this bet.

It was also wrong of Peter to publish a link to the draft of the paper without asking first. I'm not so sympathetic to Craig on that matter simply due to him very recently doing something similar to another. Academic papers that get pre-released can violate publication agreements, they can be shared for peer-review and comments, but publishing them before the Journal can cause the Journal to yank the publishing, so it could have been a serious breach with damaging consequences.

I suspect that these actions will make any sort of reconciliation, including collecting on the bet to be unlikely. I very much do not want to be involved in this discussion, because there is no win for me. I do see some benefit to the broad population of Bitcoiners from a reconciliation. The "learning opportunity" that could have been there for many might not be lost.

I've seen some folks pick up useful knowledge on memoryless processes which is good, but that is pretty basic stuff. I'd enjoy seeing some folks learn more about probability calculus also. Imagining that Craig doesn't understand memorylessness wrt Bitcoin mining is a pretty absurd notion.

Craig likely sees this as a personal betrayal. I am grateful today, that I am not Peter, (I've always been grateful that Craig is Craig, and that I don't have to be him). Pride and falling are not independent probabilities, and there is not any shortage of pride on either side of this debacle.

 

[Peter R]

As I recall from the discussion, the talk with Craig was nothing to do with memorylessness, but he was discussing independence and the Expected Value computation. The memoryless discussion was happening about the same time with other folks in the discussion.

Craig was looking to show folks that the SM and HM probabilities are independent.

Thus P[HM|SM] = P[HM]​

Expected value for HM is computed from N-1 and unchanged by SM finding a block or not as it is not released. HM doesn't change expectations. To HM Nothing has happened, nothing is recalculated.

Re-read what you just wrote. You (and Craig) are still falling for the gambler's fallacy. You just don't see it yet.

(And for the record, the paper was publicly available on SSRN [he has now removed it] and Craig also knew I was critiquing it on this forum).

 

[solex]

Really insightful article on Zerohedge.

http://www.zerohedge.com/news/2017-07-31/scaling-debate-hard-fork-highlight-several-key-differences-between-bitcoin-and-gold

Edit: and now another article
http://www.zerohedge.com/news/2017-07-31/secs-ruling-icos-are-securities-proves-bitcoin-has-staying-power

Maybe they should rebrand as "BitcoinHedge"

 

[Peter R]

"Craig was looking to show folks that the SM and HM probabilities are independent.

Thus P[HM|SM] = P[HM]

Expected value for HM is computed from N-1 and unchanged by SM finding a block or not as it is not released. HM doesn't change expectations. To HM Nothing has happened, nothing is recalculated." -- @NewLiberty

For readers wondering, the reason this is wrong is because something has changed between N-1 and the SM finding the next block. The clock on the wall has changed! And with a memoryless process, the expected value is reset continually as the clock's second hand sweeps. (Since the process has no memory of the past, the expected value must continually reset). Because the SM finds the next block (solution for height N) at t = 0, it logically follows that the HM must still be looking for a solution a this height, and so the expected time for the HM to find it is 15 minutes later (in other words at t = 15 min).

The reason they call it the Gambler's Fallacy is because it is counterintuitive and it is easy to get wrong -- even for people who should know better.