Can we know for sure that the price drop is accompanied by a change in the model such as quantization?
On twitter, some people say that some models perform better at night when there is a less demand which allows them to serve a non-quantized model.
Since the models are only available through API and there is no test to check which version of the model is served, it's hard to know what we're buying...
You can have growing deficits without actually "selling the nation" when you account for growth and inflation.
Say foreigners own 50% of US assets. When someone creates in the US a company ex nihilo and sells 10% of it to foreign investors, even though foreigners bought more US assets, this new wealth creation causes the % of foreign-owned US assets to diminish.
When foreign investors own US bonds, as inflation increases and the money supply increases, they end up owning less in real terms.
This doesn't account for investments that are limited in quantity such as real estate.
Also just because those are opposing forces, doesn't mean that foreign ownership wouldn't increase. As a matter of fact, when looking at the foreign ownership of US stocks, it has been steadily increasing.
Does this support the Selfish Gene theory of Richard Dawkins? They look like the smallest self-replicating molecules that he mentions in the beginning of the book.
Absolutely, reading Selfish Gene in high school set me on the path to this type of exploration. Genes, in the pure abstract sense, are the unit by which we interrogate understand evolutionary change. There's a large grey area about the boundaries of genes, but after a certain point, genes assemble into operational units larger than themselves, a genome. Obelisks are some of the simplest, most rudimentary genomes described thus far.
It seems more likely that this is just the consequence of training over up-to-date websites (which would include plenty of discussions about LLMs, and particularly ChatGPT). Claude is probably conflating its own identity.
Yeah, there’s a weird thing with how they “interpret” system prompts for identity.
If you suppress tokens “OpenAI”, “Microsoft”, and “Google”, GPT-4 will tell you that it is Project December, a language modeling project to duplicate the deceased based on text history.
Also, training over ChatGPT outputs is terribly expensive and Anthropic already has a working competing paradigm for fine-tuning (Claude worked before ChatGPT was released).
To be honest there were countries that defaulted on their sovereign bonds in their home fully controlled currency. Russia did this in the late 90s and took down LTCM, which was collecting premiums for insuring something that they thought is too illogical to ever occur. Well, it did occur.
The reason why this is illogical is that there is very little distinction between "the dollar" and US sovereign debt. It applies to all countries in a similar situation, for example "the Yen" and Japanese Government Bonds. It doesn't quite apply to odd cases like, IDK, Spanish sovereign debt in EUR and "the Euro" since the Spanish government does not control all forms of the issuance of the currency of the bond.
What they were doing is identifying pairs of securities whose values had diverged and they believed would eventually converge. They would short the more expensive one and buy the cheap one. When they converged they would sell the no longer cheap one, use the money to close out the no longer expensive one, and collect a profit.
However usually the reason why one was more expensive is that it had a more liquid market. So people could safely invest in it with money that they might need back quickly. This shouldn't matter if you planned to buy and hold though..at least in theory.
But in the wake of the Russian default, liquidity became more valued. So people sought to get rid of illiquid securities and buy liquid ones. This meant that LTCM had shorted things that were rising in value, and bought things that were falling in value. So they had a loss. And as the shorts got higher, they wound up having to sell assets at a loss to cover their shorts. And now the temporary losses became very real ones, and drove them bankrupt.
However, infamously, their investments made money in the end. They just weren't able to last long enough to benefit from it.
> I don't believe that LTCM was insuring anything.
While most of their strategy was convergence arbitrage, if I recall from the book, they thought of selling short equity options as a form of selling insurance.
People buy these options to insure against some event and they expected more buyers than sellers, so LTCM figured they would profitably be the provider of it. Well-structured insurance is always a loss to the buyer (they take in more than they pay out).
Insurance works because people, companies, shareholders etc are risk-averse and thus value paying a little for reducing variance a lot. That’s not a loss to the buyer. NPV isn’t the only measure of value.
Been a while since I read the book as well, but wasn’t part of the narrative that their backstop as insurance enabled traders to take bigger and bigger bets, because they were hedged with LTCM?
I don't think any trader at a broker/dealer would think that, but it's possible people on the buyside thought that. There are plenty of funds who have a strat of just allocating say 95% of their holdings to the Russell or something and then putting the remaining 5% into some high-vol bets (like putting them into LTCM or whatever). They don't see it as a hedge exactly, but it means that they hardly need to work at all and when these pay off they say "look this is alpha" and when they don't they say "look, your error vs the Russell is less than 5%". Some funds (eg big pensions) have enough AUM that this risky piece is a very significant amount of money.
I was working in the city shortly after LTCM failed and one of the interesting things I heard is that several large European institutions were using LTCM for overnight treasury. So at cob in Europe they would sweep funds into LTCM and then move them out the next morning for trading. If that was actually the case it would have caused massive fluctuations in their assets during the day which would be extremely hard to manage.
How has that got anyhing whatsoever to do with the thread or indeed TFA? Did you perhaps post this in the wrong thread or something? This literally is a thread about finance in response to an article about US credit default swaps.
The fact that we don't talk about any of the great tragedies that have happened throughout history is due to a policy of trying to keep conversations on topic.
That paper explicitly states this is not the result of "capitalist raping", and it's sad to see such nonsense injected into a paper explicitly stating the opposite. Did you read it or just see a tweet?
"In sum, according to the dysfunctional culture approach, self-destructive health behaviour - in the case of the post-socialist mortality crisis primarily referring to hazardous drinking - is not a result of the radical economic policies adopted during the transition from socialism to capitalism but of a dysfunctional, anomie-laden culture inherited from the socialist past whose effect was made worse by low alcohol prices during the transition"
In short, socialism had for so long ruined the economy that the country imploded, and the transition to a better working capitalism brought about deaths as a result of the wider destruction that caused the implosion.
Deindustrialization is the opposite of capitalism.
> It doesn't quite apply to odd cases like, IDK, Spanish sovereign debt in EUR and "the Euro" since the Spanish government does not control all forms of the issuance of the currency of the bond.
While that's true, Spain could go the tax route and say "gimme more euros this year" or sell off assets it owns. I guess a printer is faster (if they feel like it) but medium-term, developed countries have lots of tools to pay down debts (if they feel like it).
For some reason, people are more comfortable with inflation as a tax than taxes.
Well, I would assume apropos of my one economics class that the reason the average person doesn't care too much about inflation is that he has more debt than assets. If you ask me whether I want to pay a lot of taxes on my income or whether I want a big slice of my house to be free in inflation-adjusted dollars, here we would be, assuming I had no assets and most of my income went to my home loan.
> the average person doesn't care too much about inflation is that he has more debt than assets
What the average person forgets is that some people (and businesses) have wayyyyyy more debt, so when it gets diluted, they're a net loser on average.
Kinda like getting a $1000 stimmy cheque while large capital owners gets their equity saved by government bailout money. Everybody wins something, but winning last place isn't a win when no real wealth was created.
The thing you missed is that although inflation makes the nominal of the debt less valuable in real terms the associated cost of living increase means the cost of servicing the debt can go up both in real and nominal terms. You have less left over from your paycheck to pay the interest and the interest rates go up.
For most Americans, the biggest debt we have is our mortgage and 90% of mortgages in the US are fixed for the entire duration of the loan so nominal cost stays the same and the real cost goes down.
Aah interesting. Here in the UK most “fixed rate mortgages” are actually only fixed for a few years. Fixed for the full term does exist but is definitely less common.
>For some reason, people are more comfortable with inflation as a tax than taxes.
Because everything doesn't inflate at the same rate at the same time. That means the average person has some theoretical room to reconfigure their spending to minimize the impact of inflation. Ordinary people have no legal options to minimize the impact of higher taxes. That requires expensive CPAs and lawyers.
This is an extremely relevant Matt Levine piece. He gives an example of long-dated treasuries trading at 83 cents on the dollar. Today long-dated treasuries issued during COVID are trading well into the 50ies.
That's not because of their credit risk but because they pay little to no coupon and get discounted through the interest rates. In particular imagine a treasury bond that will pay $100 in 10 years, you wouldn't pay $100 for that, would you? You'd instead put that $100 in a savings account (t-bills).
The true credit risk on US Treasuries is indeed an abstract and mysterious creature. Nobody knows what would such a "default" mean in practice, what paper would get paid up and what paper would not get paid. Would commercial bank deposits at the Fed get paid? And if not, then what does it even mean to "pay in dollars"? Like how do you achieve "paying someone X dollars", do you deliver printed currency?
Yes, they're trading well under par because they're very long duration bonds and interest rates have moved against them, not because of credit risk. What Levine points out in the linked piece is that per the terms of these CDSes if the US technically defaults you can use the CDS to accelerate repayment of these very long duration bonds and get par back immediately rather than having to wait 30 years to collect. There's little to do with credit risk per se and everything to do with the interaction of the CDS contract and the current low valuation of certain debt instruments.
OMG, you made me read his article, and only then realized how ludicrous the "cheapest-to-deliver option" is. I had no idea you can send the CDS issuer any bond! Of course this is just an option for the scenario of "the congress got into a massive fist-fight and couldn't press the `yes` button on their voting machines for three weeks straight". What a hilarious piece of financial engineering.
> OMG, you made me read his article, and only then realized how ludicrous the "cheapest-to-deliver option" is.
It’s not really ludicrous, the US Treasury isn’t selling bonds of every duration every single day. You need some kind of framework to outline which Treasuries are acceptable to settle derivative contracts like futures/options/CDSes since Treasuries are not directly fungible like equity shares or commodities.
This is only true for those who are not impacted, in this case, by a government shutdown.
For example, suppose you’re running a company that has won a government contract and you receive scheduled payments from them. You use the payments to pay your suppliers and employees. If you don’t pay your suppliers and employees on time it causes all kinds of problems.
If the government looks like they might delay their next payment to you, you can “buy insurance” so that you can still make the payments. Is this gambling? Maybe? But if you don’t gamble there are real consequences, so “not playing” is still “gambling”.
> ... and get par back immediately rather than having to wait 30 years to collect
In a parallel universe the US has entered a short technical default just slightly before SVB's collapse, which allowed them (if CDS insured) to collect all MBS at par value.
I'm guessing also that as the ex-date for a possible US default approaches (I think it hasn't been announced yet), the demand for older cheap bonds/notes would rise
> That's not because of their credit risk but because they pay little to no coupon and get discounted through the interest rates. In particular imagine a treasury bond that will pay $100 in 10 years, you wouldn't pay $100 for that, would you? You'd instead put that $100 in a savings account (t-bills).
Long treasuries (>1 year) are issued at very close to par and do pay coupons. A treasury that is trading at 50-something cents on the dollar has lost a lot of value (because it has a lower coupon rate than newer treasuries).
Yeah the particular one I was looking at was 912810SP4, a 30 year treasury with a 1.375% coupon. It was issued in August 2020 at a price of $99.24 for $100 of par, and it now trades just under $59.
The way I understand the article: Insurance not likely to pay in case of small short cured default and in a major default the counterparty is at risk. So buying it as insurance makes no sense. But..
.. there are special rules which trigger even in a short default allowing a fair bit of money to be made turning the CDS into a sensible bet on a short duration US default.
The statute of limitation for wire fraud is 5 years. How did he get charged under that statute when he stole funds from Silk Road in 2012? Did they charge him for converting BCH into BTC in 2017?
I tried to look up the case but couldn't find anything that shed a light on that.
> The statute of limitation for wire fraud is 5 years. How did he get charged under that statute when he stole funds from Silk Road in 2012?
This is a very simplistic view and this mistake is made a lot. There are several reasons why the statute of limitations could either be longer than 5 years or not yet expired even though it is 5 years:
- Tolling. This is when the suspect exits your jurisdiction, preventing you from apprehending them. If the statute of limitations really is 5 years, and you leave the country 4 years after the act; even if you stay abroad for a year, or 5 years, or 20 years, the clock only resumes when you return. You can still be charged 24 years after committing the offense.
- Furtherance. Put simply, if you continue to engage in the activity, the statute of limitations is reset. For example, this article describes his repeated moving of the assets after the initial theft in order to avoid detection.
- Special Circumstances. In this case, there are exceptions that mean that the statute of limitations is not actually 5 years.
It's not clear which of these apply. I would definitely bet on option #2. If he spent any time outside of the country, also apply option #1. If any of the special circumstances defined in the law apply, also apply option #3.
It seemed to me that option #1 only applied if you were a fugitive and your whereabouts were unknown, not when you simply leave the state or country.
As to furtherance, I wonder how precise the definition of it is; is using proceeds of a crime, furtherance of a crime? Is obfuscating a crime, furtherance of a crime?
In regards to whether simply leaving the jurisdiction is grounds for tolling the statute of limitations, see 18 USC 3290, which is a one-sentence law reading:
No statute of limitations shall extend to any person fleeing from justice
It can be reasonably argued that leaving the country after having committed the offense is fleeing from justice. I'm not aware of any supreme court ruling on the matter though, and I am not a lawyer.
Is the limitation in relation to being penalised for a crime, or somebody being wanted for a crime?
If there was a police report about a crime, if the police were unable to determine who committed the crime but at a later date could do so beyond reasonable doubt?
What if the perpetrator evades police for 5 years but remains in-state?
What’s your location also, that would be relevant to the laws your referring to.
The clock starts counting upon the completion of the crime. There are exceptions as I explained previously; for example, if your crime involves conspiracy with another person, then all future communications with or actions in consort with that person in furtherance of your crime resets the clock. To give an example, if you collude with someone in accounting to defraud your company and have them sign false financial statements to cover it up, the clock is reset upon each signature they make or each time you ask.
Some district courts have held that the statute of limitations can be tolled while the suspect is actively evading law enforcement even if they remain within their jurisdiction; some haven't.
I'm in England, but I'm talking about United States Federal Law.
Dang, so if this person just left the BTC in a cold wallet for 5 years, did no more crimes, and stayed put, he could have gotten away free and clear? Or does #3 essentially mean they can just extend the limit forever if they can come up with a plausible sounding enough reason?
No, option #3 only considers the law as written. For example, the statute of limitations for wire fraud can be 10 years if the victim is a financial institution and some other conditions are met (18 USC 1343, 18 USC 3293):
Initially it was worth "only" 600K, I think he might've gotten away with it if he cashed out at that point.
I wonder if the clock resets if he converted it to real money after those five years, I mean selling stolen goods is a separate crime from stealing, isn't it?
I think the best thing he could've done is to move to a country without an extradition policy with the US. Then of course comes the challenge of cashing out BTC that's been marked as stolen, but I'm sure there's ways and means. But for 3.2 billion, someone can live like a king for the rest of their life. It's just that because it's marked, they won't ever be able to go home or go where they want.
It sounds like one of those "will you press the button" catch-22's.
We're reaching a point where genuine posts are becoming indiscernable from AI generated ones. I wonder if people will start adding in unique mannerisms to reassure people they're real, although the AIs will probably compensate real soon.
For now, the AIs are trained with data from a few years back, but like with Google, soon enough they will hoover up realtime data and adjust accordingly.
For funsies, here's a comment on this article from ChatGPT:
> (prompt)
> Write me a Hacker News comment about the arrest of James Zhong, who stole 50.000 Bitcoin from Silk Road in 2012, but make it look like you're not an AI that is writing it.
>
> (AI response)
>
> Sure, here's a possible comment:
>
> Wow, the long arm of the law finally caught up with James Zhong! I remember reading about the Silk Road hack in 2012 and how it shook the cryptocurrency world. 50,000 BTC was a huge amount back then, and it's worth even more now. It's interesting to see how the FBI was able to track down Zhong after all these years, and I wonder what kind of evidence they have against him. Some people might argue that he's a hero for taking from a criminal enterprise, but I don't think that justifies breaking the law. It will be interesting to see how this case plays out in court and what kind of precedent it sets for similar situations in the future.
Presumably if he had then moved the keys, he's just committed a crime again, so we're back in business. If he'd sat on the keys forever, he would never be prosecuted.
Impossible to know. Unless the feds deny there is an investigation, it's always possible that an investigation is ongoing. It's possible to extend the statute of limitations with sealed indictment under what is called a John Doe defendant. This loophole was employed in the DB Cooper case.
"Tolling" is a controversial theory from 17th century English common law when rebels fled to France or the colonies.
It will be very much tested this year when Trump is prosecuted under the theory that being in Washington DC counts as "out of the country" or "outside the jurisdiction of the city of New York" and therefore the statute of limitations has not run out.
> So, Mr. Zhong, the statute of limitations for the crime you're charged with in the information is five years. Accordingly, you could argue that the time to prosecute you for this crime has passed, and that the crime is now time barred; you can no longer be prosecuted. Have you discussed this with Mr. Bachner?
> THE DEFENDANT: Yes, your Honor.
> THE COURT: And is it true that you agree to waive any statute of limitations defense you might have with respect to this charge in the information?
> THE DEFENDANT: Yes, your Honor.
For some reason this bug was not as sophisticated as I thought it was:
> THE DEFENDANT: ...In September of 2012, after using the site just once or twice, I decided I no longer wanted to buy anything else from Silk Road and decided to withdraw my bitcoin from the website. While doing so, I accidentally double-clicked the withdraw button and was shocked to discover that it resulted in allowing me to withdraw double the amount of bitcoin I had deposited
> After Jimmy took the Bitcoin from Silk Road, Ross Ulbricht contacted Jimmy over the Silk Road messaging service and asked him to explain how he accessed the Bitcoin. After Jimmy told him what he did, Ulbricht never asked Jimmy to return the coins. Rather, he thanked Jimmy for his candor and sent him, unsolicited, additional Bitcoin. Silk Road’s response to Jimmy’s conduct is hardly the response of a victim.
> The Government notes that Jimmy waited around four months from the date of the search to turn over control of the Bitcoin to the Government
> Although the Government had seized the physical device needed to control most of the Bitcoin, the Government had no way to use that device to control the Bitcoin without Mr. Zhong’s cooperation. Jimmy gave the Government, with no promise of anything in return, the keys and tools needed to take control of billions of dollars’ worth of Bitcoin—control that the Government would not have been able to gain without Jimmy’s cooperation
> He recalled converting some of his Bitcoin into $700,000 in cash. He stated he did this so that he would have a “case full of money like in the movies.” He hoped the visual appeal of the cash would impress a female into having sexual relations with him. He stated his plan did not work.
Government was fairer than usual in their sentencing submission, only asked for 24 months. Honestly this whole story just makes me kind of sad. Also makes me wonder why he even agreed to give up the coins.
Its silly he lacks this much awareness, it reads like he just wanted to be rid of the burden because he can’t figure anything social out
People don’t even need actual bitcoin, they just leave their phone wallet open on a watch address while they “go to the bathroom” on a date, and let their nosy date take the bait and watch the tune change in real time
Money still overrides other preferences and exacerbates poor character development across genders, it just means his way was so bad…. Carrying around suitcases wow.
In my experience, someone committing crimes like this has a fundamental issue with their emotional self regulatory ability. Money enables coping mechanisms (well, clearly just sometimes like in this case!) that otherwise wouldn’t be available. It is also a general motivator, as you note, for folks to do all sorts of things for someone besides general emotional coping type stuff. Like get a house, or whatever.
Chances are his social dysfunction really stems from something deep he’s refusing to look at and acknowledge; and it’s feeding all these behaviors maladaptively. He probably had one particular person he wanted the validation he saw as coming from sex with, and that’s why the suitcase.
Especially since, quite frankly, if he just wanted to get laid and had cash, it’s not like it would take that much looking to find someone to help him with that and it wouldn’t involve more than a small handful of bills. If he could get his act somewhat together, anyway.
Good point! The baggage is never just what’s in the suitcase (or just the cash, as it were), and (to be serious) glad the other person saw that too. The unimaginable degree of shit that they would be accepting if they took it would be mind boggling.
My guess is that because the other party is not the type of person who would have sex for even a suitcase of cash with them, is why the perp wanted them so much they tried a suitcase of cash.
> THE COURT: All right. I find that Mr. Zhong has knowingly and voluntarily waived both venue as well as any argument here that the crime charged in the information is time barred.
How does this even work?
So can the defendant "knowingly and voluntarily" give up their right to defend via Constitution?
I'm not a lawyer, but I've known people go through the (similar, in the UK), mill of prosecution and read/watched enough about some cases to get a sense of how it might play out behind the scenes.
Let's step back, and see if you still think this is "give up your right! Or else!"
The prosecutors are 50/50 on getting a conviction. If they win, they're sure they're going to see the defendant go to prison for, say, 10 years. If they lose, the defendant walks, and thanks to the fifth amendment (almost everywhere in the World has double jeopardy protection), unless substantial new evidence emerges, they'll never get to prosecute this individual again. They're pretty sure they have everything they'll ever find, so this is it.
The defendant and his legal team are 50/50 on getting an acquittal. If they win, it's over. If they don't, it's 10 years in prison which is pretty awful.
The trial is going to be expensive for both sides. The prosecution will work to get the odds in their favour, the defendant's legal team will be trying to figure out how to get it in their favour, and so on.
This process is expensive, and eventually the taxpayer is going to pay. Even if the proceeds from the crime have been recovered and sold, the money going into public funds is going to come right back out again to prosecute, and if the defendant has nothing and can't afford their own legal bills, the taxpayer is paying both sides.
Everybody - and I mean everybody - knows the defendant did what he did.
There are interviews that are barely credible, paper trails of assets moving, lots of evidence of obfuscation and trying to cover tracks, but there's confusion in the story about intent to commit a crime.
The judge might decide to throw it out on statute of limitations, a jury might look at the psych reports and think the kid seems nice and naive and not the criminal type, so let's believe his story.
It's a bit of a mess.
Neither side really fancies were this going because there is concern you're setting off a train of case law, too - as a prosecutor do you really want to test statute of limitations if the person wasn't "fleeing from justice", but just enjoying some money he came into? That defence lawyer really looks like he fancies his day in the Supreme Court...
Meanwhile, the clock - and publicly funded meter - is ticking.
Prosecutor approaches defence. "Plead guilty, we wrap this up, your guy gets 5 years instead of 10. Parole eligible at 3 years".
Defence retorts "2 years, suspended".
Prosecutor comes back "2 years, no suspension, no parole, but time served taken into account".
"Deal". The judge gets a call. He decides to make sure there is no risk of an appeal to make clear to the defendant in court that there are questions about statute of limitations that he would need to waive. Defendant agrees to waive his rights. They move to sentencing, the prosecutor asks for 24 months, the World moves on.
Everyone wins, even though the defendant goes to prison for a while. He's got 20% of the likely maximum penalty on an even money shot, which any gambler will tell you is superb "value".
The prosecutor has lost out in terms of seeing time served, but has got a conviction and saved the taxpayer many, many dollars.
The defendant didn't lose their rights. They looked at a 50/50 shot and decided they'd prefer 2 years guaranteed and get on with their life rather than take a 50% chance of 10 years.
As an unaffiliated third party, it’s actually quite the example of game theory in action. Everyone has something to risk and trying to figure out the dance of who gives up what is really interesting.
Until you realize these are peoples actual lives… then it’s also sad.
I think the defense leaned on the part where the owner of the service basically allowed him to keep it, plus cooperation; if the judge rules him innocent (e.g. while he exploited a bug, the then-owner did not press charges and rewarded him), he might get the BTC back, but this time cleared of any crimes. Meaning he can cash in and walk free.
Well, relatively free, his face and name are known and if he is exonerated, everyone will know he's a multi-billionaire and the life of himself and others are in danger.
He gave up the coins because his lawyer probably advised him the statute of limitations would not apply here or hold up. If it did apply there would be no case at all, full stop. So a plea is better than facing up to 20 years.
On twitter, some people say that some models perform better at night when there is a less demand which allows them to serve a non-quantized model.
Since the models are only available through API and there is no test to check which version of the model is served, it's hard to know what we're buying...