Barraketh

A new proof assistant that will hopefully be more suitable for reinforcement learning than Lean - faster to typecheck and specialized apis for tree search

Well, I don't think civil forfeiture specifically would work, but like I mentioned above, yes, if the justice department wanted to criminally charge the Sacklers, they could possibly win and get a judgement against some of the money. But then that money doesn't really go to the victims - it's just a way to punish the Sacklers.

The bankruptcy settlement had a bunch of money going to families of the victims, and also to the states for anti-addiction programs, and also some money towards documenting the Purdue wrongdoings, so that the public would have better visibility just HOW this was allowed to happen in the first place

Well, it's a bankruptcy, so anyone can file a claim. The states have, but so have the victims and their families directly (as a class).

IANAL, but I don't think this would help the victims (and incidentally, that could have still happened even with the settlement). If there were a criminal lawsuit of the Sacklers, and if that lawsuit was successful, the seized money would just go to the justice department.

The only way the victims actually see any money is through civil litigation.

Negative outcome. Some important points that the article here did not emphasize: 1) The Sackler family was not actually a party to this litigation. They came to the table (with most of the settlement money) specifically to get these so called '3rd party releases'. 2) Purdue is basically broke. It's also an LLC. Thus, in order to go after the Sackler family's money, you basically have to claw back money that Purdue paid out to the family over the years. It's not impossible to do, but it requires a whole more litigation, the outcome of which is not at all certain.

Now, 3rd party releases are a genuinely weird thing: a court ruling that a party that's not directly involved in the case is immune from future lawsuits. Partially the reason it went all the way to the supreme court is that there was a circuit split - they were allowed in some circuits, but not others. However, (and this is according to a friend who represented the victims in the settlement), it's really unfortunate that THIS is the case where they get struck down. If the Sacklers walk away from the settlement, it makes the victims getting their payout much less certain, and certainly delays that payout by many years.

Interesting post. A possible more general restatement would be "under consequentialism (or utilitarianism) everyone's happiness (including the crime perpetrators') is weighted equally. Should that be the case?"

I suspect that specifically "family punishment" fails the cost-benefit analysis due to practical reasons, but the question of "should we weigh the suffering of good people equally to the suffering of terrible people in our ethics system" is interesting to me, and I don't currently have answer.

You're right, it's just that the word "abuse" is doing a lot of work in that statement. I've still never seen any example of any person being actually damaged by the "tracking", so calling it "abuse" feels pretty harsh.

So one issue that this article glosses over is how contextual advertising interacts with attribution.

Essentially, as an advertiser, in order to do effective contextual advertisement, you need to have some idea of how ads perform in a particular context - on a particular web page or app. So what you would really like to say is that "people who saw our ad on X were Y% more likely to visit or website, or buy our product". However, in order to actually measure that, when a user comes to your website you need to know which of your ads they've seen. Traditionally, you'd just serve the ad from your domain and set a cookie, but now you can't do that.

This is more relevant to brand advertising than "click here and buy product X" (direct response), but I think that the people saying "oh, just use contextual advertising" are underestimating the extent to which GDPR makes contextual advertising difficult.

So, since you seem particularly upset about it, I'd love to know - what is the problem of "data collection"? I've still never seen an example of damage to a user by AdTech data collection. What I have seen are

1) Data leaks by companies that collect and store true PII - this is your Equifaxes of the world, and they didn't collect that data using cookies. 2) People are generally creeped out by the idea of their movements across the web being tracked.

And so, even though no ad-tech company has ever really had a data leak, and although the tracking across the web has never really resulted in any negative outcomes, people are using (2) to try and kill ad-tech for often disingenuous reasons.

Well, this has been a fun discussion. A salient point that I haven't mentioned but that actually plays a big role is attribution. Many advertisers run ad campaigns not to get a clickthrough to their site, but to just keep their product 'top of mind', so that next time you go buy a car you buy a BMW. So from that point of view, if you see an ad and later go and sign up for a BMW test-drive, BMW would attribute that test drive to the ad that you saw. If you can't track attribution, it becomes really hard to figure out where you should be advertising in the first place. To everyone saying "use contextual advertising" - how do you know which contexts produce better results if you can't measure performance?

This is particularly relevant to mobile apps, because if you show the user an ad, they are extremely unlikely to switch contexts to go and actually click on it. If you can track users from the app to the purchasing site, then you can say "hey, I have a really valuable audience - you should pay me hella money to show them ads". This has been less GDPR and more Apple, but the result is the same - it makes ads generically less valuable.

And that is why I'm fundamentally pro ad-tech. I don't have any direct monetary interest in it, but I do want the digital economy to be growing and efficient, and in an ideal world decentralized beyond the 4-5 large platforms. People spending money online is GOOD. It's good for businesses, it's good for the people (under a rational agent model at least ;)), and it's good for me as a software engineer who wants to keep getting paid silicon-valley salaries.

> People ask that right now, and adtech is one of the things that have killed it.

I don't think this is actually true. When I think about the old internet vs new internet, a lot of it is about people running their own blogs / websites vs. platforms. And I think most of that is not really about ad-tech at all, but about the mechanisms of content and audience discovery. But the fact remains that if you want to say... publish videos of some kind, you are likely to make much more on YouTube vs uploading them to your own website, and that's at least partially because Google is able to show effective ads.

> Target them based on the context in which the ads appear.

Possible, but in some cases significantly less effective. As I mentioned above, ad-tech comes with some very interesting progressive effects, where the people who spend the most money are the ones who end up paying the most in aggregate for content. An interesting example is luxury goods, which are both high-value and niche. If you run say... a news site, or something else that's general purpose, you probably don't want to be showing Rolex ads to everyone. But rich people still read news, and if you could target your Rolex ads to them, that basically subsidizes everyone else.

Well kinda. The business wanting to market themselves certainly isn't your problem, but the "how does the website monetize itself" is exactly your problem, or rather, it's part of the interaction between the owner of the website you're visiting and you.

You could say "oh, they should just charge for their content", and some definitely do. But the ad model allows for really interesting price discrimination in terms of "who pays for the content". So, if someone buys say.. a Tesla through a website, that conversion subsidizes a million poor kids who don't have to pay anything. In some ways the ad-supported model is the most progressive way to pay content creators - the people who end up paying are the people who spend the most money online.

So, out my now 15+ year career, only about 4 were in AdTech, and I've been out of that game for more than 7 years. But fair enough - happy to play 'ad-guy' for the purposes of this discussion :)

I guess my main contention is that businesses really like ads. But that's usually ok - because it funds cool content! The majority of television historically was produced just to sell ads, but it still created awesome television. And if we make ads worse, well, the people who are gong to suffer are the businesses and the content creators. OR, like i mentioned above, we will just drive everyone to large platforms, which is essentially what's happening now.

Another way to frame the question - GDPR came out in 2016. Do you feel like the web is getting better?

I think this is a genuinely interesting point, and I wish that we as a society had a more nuanced discussion about what that right is. AdTech is largely anonymous in the same way that crypto / web3 is anonymous - it shuttles around cookies / identifiers, but largely does NOT care about any information that is actually personally identifying. If the laws were regulating storage / transmission of information that is actually personally-identifying (addresses, emails, names, etc), that would be much more reasonable in my mind.

There seems to be a really strong disdain on HN for AdTech in general, which I think is misguided, and comes fundamentally from a "web consumer" mentality. I think that it would be valuable to reframe the question from the producers point of view: 1) I am a niche small business - how do I let people know that I exist? 2) I run a website and would like to monetize it - how can I get paid for the content that I produce?

The ad-tech solution is actually quite elegant in theory - if you can show ads only to the people to whom they are relevant, then, as a small business you can let people know you exist without blowing your ad budget, and, as a content producer, the more valuable an ad-view is, the more you can charge for it.

The current movement to avoid tracking is an extremely powerful centralizing force. The large platforms know a lot about you already - Facebook, Google, Amazon, Apple, etc. So, in a way, "we're not going to let AdTech track users" = "we're going to make only ads on large platforms effective", which means that both content producers and advertisers will prefer them, and then people ask "where did the old internet go"?

The AdTech system isn't ideal, but it would be great if the people who criticize it came up with something other than "fuck the small businesses and content producers".

Ammonia has been mentioned a few times in this thread, but I will expand a bit, because I believe it is the most promising hydrogen-based approach. Essentially, the idea is to use ammonia directly as fuel. There is an ARPA-E funded project (that I'm currently failing to find) that cracks ammonia into hydrogen and nitrogen using heat. They claim that they can create a mixture of H2 and ammonia that will burn at any required flame temperature (H2 burns too fast / hot for existing turbines, ammonia too slow, so a mixture is the right way to go). Theoretically, with this process, you could use ammonia as a drop-in replacement for LNG. The LNG infrastructure can also be converted to carry ammonia instead.

Now, there are still obvious energy losses from creating and then burning ammonia. One thing that this technology could be very helpful with is overcoming the NIMBY-ism around nuclear power - that is, build nuclear reactors to produce ammonia, then ship the ammonia to where it will be used. It would obviously be more efficient to just run power lines from nuclear power plants, but given the wide-spread opposition, it could be politically easier to build the nuclear plants in the middle of nowhere.

I totally agree. I'm just saying that from the point of view of legal liability the doctor is considered an expert. Purdue would argue that the doctor is getting information from a variety of sources (including the FDA), and so when they prescribe a medication and then the patient gets addicted, the responsibility lies with the doctor.

And I also agree that you can't blame a patient for getting addicted. However, in US law there is something called the "clean hands doctrine", which denies remedies if the accuser has acted in bad faith wrt the subject of the claim. In practice this might translate to arguing that because the patient is breaking the law in misusing opioids, they don't deserve damages.

All I'm trying to say is that while Purdue is definitely morally responsible, legally it's kinda difficult.

Incidentally, similar arguments are going to play out around trying to sue oil & gas companies for climate change (and will probably use the same public nuisance statutes). The o&g companies lobby against emissions standards / carbon regulations, but ultimately selling oil and gas is legal, and it's the job of the government to regulate, so there isn't a simple legal doctrine to say "hey, you guys bought off a bunch of politicians and thus are still liable for damages".

This is being framed is as miscarriage of justice, and from a moral perspective it definitely is. The problem is that the legal grounds on which Purdue (and the Sacklers) can be sued are actually kind of weak.

Broadly, the things Purdue is accused of are 1) Aggressively marketing opioids to doctors, and 2) Lobbying the states to change various laws around prescription and marketing of opioids

The problem is that for direct liability, there are two more actors that need to be considered - the doctors, and the patients themselves. Doctors are considered to be experts, and patients are often breaking the law when they misuse opioids. Both of these facts break the chain of liability, and so arguing that Purdue is legally liable for the ultimate addiction of the patient is difficult.

As a result, people have tried to sue Purdue under more general "public nuisance" statutes, rather than regular tort liability. A public nuisance is when someone interferes with a right that the general public shares in common. However, this area of law is not very well developed - a lot of it is carryover from old British law, and winning those cases isn't a slam dunk. So there are certain objectors to the settlement, but I don't know why people think that it would be easy to hold the Sacklers criminally responsible or to get more money than this settlement.

I think you are right - this was a brain dump of some things I've been thinking about, specifically on how the fight against cookie tracking is making centralization worse and companies like Facebook more powerful. This article generically criticizes both, but I think there's actually a tradeoff here, and not making the distinction may lead to bad policies

Well, it's hard to prove the absence of a negative - I think that it's on the people claiming harm to provide some examples. However, I'm not even sure what a cookie data leak would look like. The large advertising brokers are handling petabytes of cookie tracking data per day. To gain any insight out of it you need to run jobs on giant clusters. The volume of the data makes it basically impossible to exfiltrate. So yeah, I'm pretty confident in this statement.

The premise is slightly different. I'm mostly differentiating between cookie tracking and social networks (and some other large online platforms). The large online platforms don't need to track you - you give them your data willingly. Facebook knows a lot about you not because it's tracking you, but because you keep posting things to it. Cookie tracking is an alternative way to build up an effective advertising profile that is decentralized and anonymized, which I think has some value.

I'm trying to differentiate between data that is anonymized (cookies), and data that is not. I'm unaware of any data leak of anonymized data that resulted in any harm, but if I'm wrong I'd love to hear about it.

Honestly, I'm kind of sick of how bad a rap advertising gets. Now sure, companies knowing a lot about your personal life is creepy on an intuitive level, but the fact of the matter is that cookie tracking data has NEVER been associated with any leak or data breach that resulted in personal harm. The thing people SHOULD be worried about is stuff like the Experian leak, where credit companies collect your non-anonymized personal data.

Also, fact is that matching consumers with products that they like doesn't just have enormous business value, but is actually socially positive! If you can more easily reach a niche audience, you can build better more targeted products. And the open data exchanges were a great moat against platform centralization like FB. The fight against open data exchanges make the comparative advantage FB has in advertising to you larger. That's actually pretty bad, because FB has some pretty bad incentives wrt to the attention economy and optimizing for engagement. A world where advertising on independent websites is effective is a much better one - it would let websites put out better content, it would decrease the power of social networks, it could fund better journalism (which is being decimated right now), etc.

Could you elaborate on the "compositional datatypes"? What is the use case?

So since March we've seen multiple countries (Korea, China, New Zealand) largely contain the virus, and start re-opening the economy. The way you do it is to get R<1 (where R is the reproduction number - the average number of people that an infected person infects). When R is smaller than 1, then you have an exponential decay curve, where there are fewer new cases each day.

The way that these countries have been able to get R<1 is a combination of social distancing, hygiene (masks), and test-trace-isolate system. So that's the endgame now. The shelter-in-place orders gets R<1, then we keep it at <1 by getting everyone to wear masks, and by getting a tracing system in place so that anyone who comes in contact with an infected person can quickly get tested and self-isolate.

That at least is my understanding, and it's been largely informed by this webpage, which I found incredibly helpful. https://ncase.me/covid-19/

Relatedly, I'd like to plug Wikibabel (http://www.wikibabel.com/). It's translating wikipedia from English to Swahili, with hopefully more languages to follow.

I completely agree that we need better generic libraries. I was mostly commenting that I really believe that there are huge wins that can be achieved in the "generic distributed execution engine" space, and that people shouldn't be intimidated by the work that has already gone into spark.

I expected that there is overhead to distributing the computation, but I was surprised by the magnitude of the speedups available.

I know that Spark has had a lot of work put into it, but my personal experience with it has been pretty negative. I've spent a lot of time at my job trying to tune it to our workflows (extremely deep queries), with only moderate success. I've just POC'd a custom SQL execution engine that was 200x faster than spark for the same workflows. Now, our requirements are pretty non-standard, but I find it pretty easy to believe these benchmarks.