A LessWrong Crypto Autopsy | Hacker News Reader

A LessWrong Crypto Autopsy

lesserwrong.com

63 pointshenryaj57 comments

Oh boy. For supposed rationalists, there's a lot wrong with this sentiment. Biggest issues:

1. Confirmation bias. How many things were discussed on rationalist websites that people were sure were going to be yuge, petered out, and were forgotten? You can't claim victory for rationalists without factoring in the ratio of failed predictions.

2. You can make a strong argument that the fundamentals of bitcoin itself are irrational (power consumption, transaction throughput, etc). Is accurately forecasting the irrationality of others itself rational? Maybe. But given how unreliable people are, feels more like augury than rationality to claim you saw it coming, post facto. Especially when your actual chain of reasoning turned out to be incorrect, despite a correct conclusion.

3. I did mine myself 20 bitcoins in the early days. But then I cashed out when they were $20 a pop and bought and iPad, and was well pleased with myself at the time. Obviously not a great idea in hindsight, but the point is that in order to make a 10000x return requires a strong belief that Bitcoin will actually go that high. Because to reach 10000x you have to resist the strong temptation to cash out at 10x, 100x and 1000x. I would argue that cashing out at 100x IS more rational in most scenarios than holding out for the moon shot (even if that eventually happened.) It could have just as easily crashed back down to zero.

> How many things were discussed on rationalist websites that people were sure were going to be yuge, petered out, and were forgotten?

This is addressed in the article. The potential upside with Bitcoin was enormous, and so many people had predicted it amongst the rationalist community, that it made sense to make a tiny bet on it. The fact that other tiny bets might not have paid out doesn't really matter.

The problem with huge potential upside arguments is that they are equivalent to Pascal's wager, which pretty much no one today takes seriously. And, if you accept them, you have to also accept huge potential downside arguments, which (along with accepting some other dubious premises that are popular in the "rationalist" community) lead you to Roko's Basilisk.

> huge potential upside arguments... are equivalent to Pascal's wager

I see this a lot, and I continue to disagree with it.

Bitcoin was comparable to long-odds betting or penny stocks. $10,000/btc is a large number, but it's completely tractable. The risks of Bitcoin were a high chance of failure and an unbounded context where calculating odds was harder than, say, making long-odds sports books.

Pascal's wager (and in rationalist terms "Pascal's mugging") is about generating a number so large it can't be reasoned with. More specifically, about arbitrarily increasing value until something beats any other consideration. Heaven and hell are infinites, duplicate simulations of a mind are unbounded, human extinction 'kills' an arbitrary number of future humans. And so you start getting nonsense results where risks are literally incomparable.

If someone in 2009 told you 1 BTC was going to be worth $10,000, you'd have said that sounds very unlikely. If they told you it would be worth $10,000,000, you'd have noticed that this would make the value of BTC comparable to the gross output of the entire planet, and declared that no, this was impossible. It's not Pascal's wager if you can slap an upper bound on it and start doing meaningful odds calculations.

(Roko's Basilisk is such a case, but I should reiterate that it was written to display why such claims don't make sense.)

> Confirmation bias. How many things were discussed on rationalist websites that people were sure were going to be yuge, petered out, and were forgotten? You can't claim victory for rationalists without factoring in the ratio of failed predictions.

If there are indeed more than 10,000 of these, then I agree it's selection bias to only look at the successful one. Do you think there were? Personally, I'm having trouble coming up with even a single equivalent case, though I'm sure there are at least a few.

> If there are indeed more than 10,000 of these, then I agree it's selection bias to only look at the successful one. Do you think there were? Personally, I'm having trouble coming up with even a single equivalent case, though I'm sure there are at least a few.

Isn't this basically every startup? If it grows up to become Google or Microsoft, you 10,000X your money. And then 10,000 of them don't.

The average person never gets the opportunity to be an early investor in startups. You could buy 10 bitcoins when they were $1. You couldn't buy Google stock until after they were successful.

I don't think the ratio of failed predictions is the correct basis for assessing the quality of the prediction, particularly not with something the predictor himself explicitly states is a low probability event. If I only give a handful of horse-racing tips but one of the long shots comes in leading to a high ROI, your prior for me just being lucky should still be high. Especially if none of the factors I've suggested which might make that horse more likely to succeed than predicted come to play, but it incidentally wins the race thanks to a massive pileup that takes out most of the other horses.

It's that second bit which really matters. Sure, if you had taken gwern's advice to mine a few BTC in 2009 and sold in late 2017 you would indeed have made the $10,000 per BTC mined hinted at. So it's easy to take that as confirmation of the quality of his reasoning.

But what gwern actually says is that BTC has a +ev on the assumption that it has up to a 0.1% chance of "eventually replacing a decent-sized fiat currency" and in order to do so, the value of an individual BTC would need to hit $10k. This replacement of "a decent-sized fiat currency" hasn't happened. The status of the prediction that BTC had a low but not trivially low chance of replacing "a decent sized" mainstream currency is indeterminate.

It's entirely plausible that when deciding whether or not mining was worth a bit of time and electricity gwern also weighed up the possibility that BTC would have great potential as a speculative asset due to an ideologically committed base who would want to hold it, public interest due to novelty and hype, and readily manipulable markets, but that's not in the post. This might even mean that he considered the situation that has come to pass as less likely than a scenario which hasn't. (Maybe he'll be along to tell us what he thought about that at the time!)

From the point of view of people who listened to gwern being richer than people who thought he was wrong, that might be an irrelevant detail. But ignoring the possibility the reasoning about the probability of x being systematically underestimated by markets might be entirely wrong because x occurred anyway is also a form of confirmation bias.

> in order to make a 10000x return requires a strong belief that Bitcoin will actually go that high. Because to reach 10000x you have to resist the strong temptation to cash out at 10x, 100x and 1000x.

Indeed. The article gives no rational explanation for why someone shouldn't have sold at $11,600, or why they shouldn't hold on and wait for $1,160,000. Saying "If you bought at the start of the bubble and sold at the peak you would have made a ton of money" is pretty meaningless, especially when the reasons for buying at the beginning weren't great ("If Bitcoin becomes the new global monetary system"/Bitcoin "eventually replacing a decent size fiat currency").

You could have a system for pulling money out (as a comment mentions), but that involves a different understanding of Bitcoin and also involves earning less money than the article mentions.

You can always buy an asset, bitcoin or whatever when you have money to invest and sell it when you need some to spend and not worry about the price fluctuations. It often works better than chopping about if the asset has good fundamentals. (with hindsight, bkr, amzn, appl shares, bitcoin, etherium etc)

Not that I made myself rich by this, but, when asked by noobs "How high can bitcoin go?" I told them it was roughly analogous to gold. So bitcoin and perhaps a few other cryptocurrencies could, in total, come to within an order of magnitude of the total market value of above-ground gold.

That means a $10,000 bitcoin is plausible, but definitely not guaranteed. Whereas a $1 million bitcoin is unlikely because something would have to happen to make it plausible. Being somewhat analogous to gold would not be enough.

The plausible values for bitcoin are supported by more than "animal spirits." So are the limits to plausible values.

>The total value of all gold ever mined would exceed US$7.5 trillion (wikipedia)

Approx number of bitcoins 16 million (21 max - 1 to go - 4 lost)

So to be equal btc would have to hit about $500k each

True if bitcoin is the only successful cryptocurrency. It seems likely there will be more than one.

Scott argues that Bitcoin is an example of why rationalists need to be more confident in their results — I’d argue it’s an example of where they should be much more modest. Hindsight bias is in play here.

The line of reasoning that “this investment has a 1/10000 chance of going huge, and a 9999/10000 chance of being at best a waste of time, so I should invest” is not a rational strategy by itself. If you applied this logic to every moon-shot investment you came across you’d barely break even between the few winners and thousands of losers. You need to know something more interesting about the investment than “this could go to the moon”. Otherwise you’re spending all your time buying and then losing money in penny stocks.

You also have to hold onto Bitcoin much longer than seems rational. It’s rare to find that someone buys Bitcoin at $1 a coin, and doesn't sell at $1000/coin. A 1000x return is the definition of moon-shot. Scott claims that “most people” agreed Bitcoin had a more than 1/10,000 chance of achieving $10,000 a coin — but this is a totally non-obvious result to me. Maybe it’s 1/1000 chance of a price of $1,000? A 1/100,000,000 chance of $100,000? In 2011, how could you reliably make such a precise prediction? For a community centered around epistemic modesty, such a finely calibrated probability like that seems unreasonable.

And for the record — Bitcoin was expected to grow in price proportional to its value as a currency, tool, technology, etc. Nobody said “I’m going to buy Bitcoin because it’s going to enter into a giant speculative bubble and I want to be on the bottom”. Nobody seems to know why Bitcoin has exploded in price, so I’m hesitant to award points to rationalists. Seems more like a case of “correct by coincidence”. You made statistical inferences and acted on the result, and ended up being right, but not for any of the reasons that your statistical inference is based on.

> If you applied this logic to every moon-shot investment you came across you’d barely break even between the few winners and thousands of losers

You would become... a VC

Their method is to just select the moonshots on the basis of some common characteristics with previous moonshots

I was active on LessWrong in '09-10 (and not-as-active until '14). Bitcoin was mentioned rarely. Hal Finney and Wei Dai themselves barely talked about it there, if at all. IIRC they didn't mention it until a few years later. My awareness of Bitcoin came from various other tech/programmer forums. I mined 0.7 BTC in early 2011 and held it until a few months ago. Not life-changing, but certainly the biggest ROI on anything I've ever had by a few orders of magnitude. But my point is I don't remember becoming interested in Bitcoin because anyone on LW was talking about it. They weren't. Even though I regularly interacted with Finney and Dai, I didn't even know they had anything to do with cryptocurrency until last year.

Edit: To be clear, I'm adding context, not arguing with the article. Yvain/Scott was certainly there.

I mean, the article claims the first mention of it was in early 2011 on LW, so your timeline makes sense. Maybe it was on IRC or something else that you heard about it at all.

According to the link, the first mention of Bitcoin on Less Wrong was in early 2011.

Because this is Less Wrong, I'll say that they're making a logical fallacy. A prediction was made that Bitcoin would go well, and it did, but the fact that it went well doesn't mean that the prediction was rationally founded.

I have the impression most predictions on modern companies or technologies are completely random.

Some people tell you X will change the world.

Some people tell you X won't change anything.

We wait some time and look how X did and then say one of the two groups were totally right and saw it coming, while the rest was completely elusive to how the world works.

To me it seems like there will always be some people who get it right if the questions are simply answered by enough people...

I mean, just look at companies like Kodak or Nokia, who were filled with really smart people that simply failed to acknowledge the fact that their industry, which they knew deeper than anyone, were chaning.

How's the old scam go? Mail 500 people your promotional flyer claiming Stock A will jump and 500 people a version that says Stock A will plummet. Split the 500 people who got the correct prediction into two groups and repeat with Stock B. Then sell 250 people your third, 100% accurate prediction. You'll have 125 loyal followers to evangelize you.

Sounds awfully familiar.

There aren't that many predictions being made on LW though. If you expect this has only come true by chance, there should be a large amount of similar predictions that haven't come true in the LW archives. Are there?

As another comedic twist, the LW community is probably the biggest proponent of publishing long-dated predictions and calibrating their confidence based on their results and priors. If ANY group of people has a defensible history of trying to make verifiable/actionable predictions, it is LW.

If you think my predictions are bad in general and this is just cherrypicked, one can always review my 2,000+ expired predictions since ~2008: https://predictionbook.com/users/gwern (I also have a profitable track record of personal & prediction market betting going back to ~2003.)

> A prediction was made that Bitcoin would go well, and it did, but the fact that it went well doesn't mean that the prediction was rationally founded.

This just looks like a naked assertion to me.

It's of course possible that Bitcoin had less than a 1/10000 chance of hitting $10,000, and we're just living in an extremely unlikely world. But given that it _did_ happen, it's probably fair to bet that it was possible at the time to give it better than 1/10000 odds.

LessWrong is all about the Bayesianism, which means being comfortable with probabilities, even though it means you can never be "sure".

I strongly believe that the prediction was rationally founded. I think casting a lot of doubt on this is a sign of regret/motivated reasoning.

Yeah. The prediction of $10,000 was predicated on:

> I estimate that bitcoin has somewhere between 0 and 0.1% chance of eventually replacing a decent size fiat currency, which would put the value of a bitcoin at anywhere upwards of $10,000 a bitcoin.

Likewise the other prediction mentioned:

> If Bitcoin becomes the new global monetary system

Let's see the results of the people who are holding on with that expectation (and some still are). That $116,000 isn't there until after you move money out of Bitcoin.

But maybe you put money in knowing that it's a speculative bubble, and are willing to pull that $116,000 out now? Sure, but bubbles are irrational. If you had people advocating a schedule for pulling your money out of Bitcoin (the first comment mentions one), than that makes sense, though the amount you would be holding would be a lot less than $116,000. Without that schedule you're left with trying to time a bubble, which is the issue with every bubble. The article doesn't give any indication of why people should sell now, and they shouldn't have sold at $11,600, or why they shouldn't try to hold on until $1,160,000.

> 3: Arguments-from-extreme-upside sometimes do work

I have seen a bunch of less wrong articles on pascal's wager and variants so I assume they are familiar with them:

- https://en.wikipedia.org/wiki/Pascal%27s_Wager

- http://lesswrong.com/lw/kd/pascals_mugging_tiny_probabilitie...

- http://lesswrong.com/lw/h8k/pascals_muggle_infinitesimal_pri...

With that fallacy in mind, is bitcoin still a good call? It could've been any other cryptocurrency winning and they all satisfy the same tiny probability with astronomical returns. It doesn't make sense to dump all your resources onto every new coin that pops up. Just like how you wouldn't adopt every religion to guarantee that you will never burn in any hells.

> It doesn't make sense to dump all your resources onto every new coin that pops up.

No, but if the odds work out, then it definitely makes sense to dump a _small_ amount of money into the ones you think have a shot.

This is exactly how startup investing works: http://www.paulgraham.com/swan.html

Rationalists are familiar with Pascal's Wager, arguably to a fault. (There's a lot of discussion regarding how it ties into strong AI. The darker theories look something like Stross's A Colder War.)

It's certainly true that long-odds betting isn't an inherent winner, and crypto better is long-odds betting. But BTC was the first blockchain currency, and is credited with being the first decentralized answer to the double-spend question. People who understood the tech and the importance of that idea at the beginning might well have made a prediction that this was different than your average investment bubble, and given that BTC and ETH are clearly the 'safest' cryptocurrencies they'd have a solid point.

It was still a long odds chance. It could have dried up regardless of technical merit, or crashed when the selfish miner problem was unveiled, or been supplanted by a more-usable version of the same insight. But it's quite reasonable to argue that 2011-2013 BTC was a much more justified bet than early-stage Ripple or Litecoin or other cryptos.

I find it really funny that even though the author of the original article from 2015 turned out to be fundamentally right about price movement, a lot of people here are nitpicking the reasoning to death.

One factor that has contributed to crypto's explosive growth that does not get mentioned as much as I think it should is how controversial it is. I can't remember if it was PG or Mark Cuban or some other VC-related person, but I remember a quote that says (roughly)-

If a new company is based on an idea I like, I want to buy. If it is based on an idea that makes me bored, I don't. But if it is based on an idea that I hate and can't stop thinking about how much I hate it, I want to buy.

Looking back, the "memetic contagion" of people debating the merits of bitcoin has only helped it. The fact that a lot of people hated it (and continue to hate it) and can't stop talking about it has only added fuel to the fire.

Friday is groundhog day. Phil might be right about his weather forecast, but it won't make me give any more credence to groundhog based weather prediction.

Having recently received a reasonable fraction of a bitcoin, I am interested in what the optimal ways to cash out on it from a theoretical perspective are. Say I have some guess as to what chance it will have X value in Y years and have some quantified percentages of risk/security. How much do I sell now, how much should I hold on to "in case", etc? I feel like there is some mathematical advice here that I don't know. I feel like the best strategy should be something like "sell 50% every year" where 50% will be dependent on what I expect future performance to be.

Edit: and ironically, it's the inconvenience of selling it that has kept me from doing so yet! The opposite of the author's problem.

Can mods please use "cryptocurrency" or "cryptocoin" ? How on Earth am I supposed to know which links are about cryptography now ?

I was looking forward to a sweet post about cryptography failure, with possible advice relevant to my day-to-day work... but no, yet another Bitcoin analysis I really don't care about.

Sigh.

> [...] one bitcoin purchased today (for 90 cents, last time I checked) [...] My financial advice is to not buy more than ten [...]

That's a really strange proposition. "It might be worth a lot more than now but don't spend more than 10 bucks on it"?

I imagine, back then buying bitcoin wasn't easy (it's still not) so going through all that hassle for 10 bucks is odd.

I would have expected a recommendation of maybe $100 or even $1000 if one was adventurous.

I think it's consistent with the rest of the post. The claim is that long-term bitcoin is either going to be worth a shitton, in which case having 10 is enough, or 0, in which case having more than 10 is pointless.

Buying more makes sense if you're planning to sell somewhere between, like if it gets to $10k.

It also has to do with risk and risk tolerance. I assume $10 was chosen as an amount that wouldn't hurt too much to lose for most people whereas some people might miss $100. The higher the upside is, the higher the chance you'll lose the money can be and still be worth investing. In this case the argument is that although you'll almost certainly lose all the money you invest into bitcoin, it's still worth it because the upside is so high if bitcoin does end up succeeding.

Is this about cryptography?

No, sadly.

Apparently "crypto" means exclusively cryptocurrencies now, like "cyber" means exclusively cybersecurity. Thanks, marketroids.

No, it's about crypto.

www.cryptoisnotcryptocurrency.com

See the world for my retort.

I was logically onboard back then, but the notion of Proof of Work mining seemed ridiculous and wasteful, so I didn't bother. I didn't even consider its purchase, which in retrospect would've been a good idea?

I'm happy to hear the field is moving to Proof of Stake. I would've been on board from the beginning most likely.

I had a bit of Bitcoin back then-- maybe 10 BTC. Some I mined back when you could CPU mine. I set up a process to mine in a pool whenever my 2011-2012 era MacBook was plugged in and had Internet, keeping it alive even when the lid was closed with another hack.

I spent it on various things between then and the big Bitcoin pumps: coffee (there was a spot that would take BTC in SoCal), open source contributions, etc., and cashed a little out just to see Coinbase work.

I never thought it would go anywhere near this high for the rational reason that Satoshi's Bitcoin is really not a very good currency. It's a significant innovation in distributed systems and the Bitcoin code base is a good proof of concept, but I expected it to stay semi-niche or to eventually crash if/when major technical limits were found.

I saw three killer problems with it:

(1) Proof of work inherently promotes centralization due to industrial economies of scale. I expected some miner to eventually have 51% and behave badly, or a government that felt threatened by it to invest in >=51% hash power (or nationalize a few miners) and wreck it. The latter would be cheaper both politically and monetarily than a full frontal assault, especially if done semi-secretly.

(2) It doesn't scale and won't scale. It's far too heavy. Some of the issues are merely Bitcoin's choice of parameters and some are deeper problems with the implementation.

(3) Hyper-deflationary behavior is not a desirable characteristic for a currency. (Hyper-inflationary isn't either, but Bitcoin will never have that problem.) A currency needs to have at least a somewhat stable price and the ability to grow in quantitative base if needed to act as a good medium of exchange. Bitcoin is far more deflationary than gold. Not only does it have a strict known global limit, but it's subject to breakage (loss of coins) to a much greater degree than gold.

So I didn't "hodl" because I thought it would probably top out at around $100/coin. I underestimated the irrationality of the rest of the market.

Moral of the story? Markets are not rational and are therefore pretty tough to predict using purely rational models that don't delve into emotionally driven human behavior, herd dynamics, and game theory.

Second moral of the story: I agree with point #3 here that sometimes extreme upside arguments do work. I should have held because I didn't need to cash out and there was the remote chance that... this... would happen.

The latter works with startup investing too. I've talked to some angels and several said they made their biggest exits on companies they'd mentally written off. In one case when it came time to cash out he forgot where he put his certificate info or how many shares he had.

> or a government that felt threatened by it to invest in >=51% hash power (or nationalize a few miners) and wreck it

Well, AFAIU they will wreck their fork, fine by me.

> Hyper-deflationary behavior is not a desirable characteristic for a currency.

It is a desirable characteristic for a store of value though. And eventually, every hyper-deflation stops and it might become a currency.

Bitcoin is a poor store of value too. It requires a lot of power simply to maintain it. It's a dissipative structure. Gold on the other hand can be buried in the ground and can just sit there for free. There are costs associated with moving and guarding or hiding it, but they're less than the cost of maintaining a vast network infrastructure and lot of energy-consuming hash computers.

To me the whole "store of value" thing is goal post moving. It was originally intended to be a currency but it's not very good for that, so people have decided to punt on that goal and redefine its objectives.

I also forgot to add that Bitcoin in its current incarnation is not quantum hard and will die when practical QC arrives. It could be rev'd to support post-quantum crypto, but the way it's dealt with scaling issues so far does not inspire confidence that it will get ahead of the curve.

> To me the whole "store of value" thing is goal post moving.

I agree and the bet is currently on LN to get that currency aspect back. But it troubles me and many others.

On the mitigating side, many people think that it it's not a currency, even with low transaction fees, because it's so volatile. Therefore, those people have never bought it in the first place and cannot exert downward price pressure right now either.

> eventually, every hyper-deflation stops and it might become a currency.

That's not what tends to happen though. If you look at collectibles (which I maintain are the closest analogue for Bitcoin), they get rarer and the price climbs, and then people lose interest but the price doesn't drop; rather transactions become rarer and rarer, but those transactions that do happen still happen at the "old" price. And eventually no-one cares about the thing, but the value is still nominally high, and the market never gets more active.

I don't think collectibles are the closest analogue for Bitcoin because Bitcoin is divisible and fungible.

Therefore with collectibles a small set of people (the owners) may determine the price and it's always a binary choice. Hence the illiquidity.

Just because you disagree with the rest of the market doesn't make the rest of the market irrational. How do you know you're not just wrong?

Page not loading for me.

Cached: https://webcache.googleusercontent.com/search?q=cache:FWjBUR...

You can always place low risk, high reward bets on the stock market using options.

Missing from this post: no matter how smart you think you are, the future is still unpredictable, because it involves other people and people are mostly gooey and not perfectly rational game theory state machines.

Wow. This is a very Wrong post. As much as I like crypto and bitcoin (and I'm an early investor), this post gets it all wrong.

Bitcoin is kinda like a ponzi scheme. It can only benefit a few that bought in early. If most of LW readers bought in, the price to get them all in will be much higher thus making their investment less rewarding.

Remember, in crypto we are not building houses but convincing the other guy to buy in. We are yet to have reliable instant and cheap transactions. We are yet to have stable non-volatile crypto. So from a usability point of view crypto is still not usable.

Reply on news.ycombinator.com