suresk

It's always good to understand what the policy is on an offer like this, but most credit card intro rates are not like this - it is typically store charge accounts that offer "12 months same as cash" that accrue interest and add it if not repaid during the time offered.

Not defending their actions in this particular case, but yes, I think it is completely reasonable for a service like MailChimp to be concerned about unauthorized use of user accounts. If someone takes over a MailChimp account, it is almost certainly to send spam/fraud type stuff, which causes harm to MailChimp in various ways.

Compliance with those standards doesn't mean they aren't potentially impacted by that sort of thing, and doing what they can to detect and mitigate unauthorized user account usages is part of at least a few of them.

I am interested in this area too, but I don't know of any single place to follow it all. I have found that there isn't an avalanche of "new" stuff, but a lot of different ideas with different tradeoffs that change in relevance based on usage patterns, languages/runtimes, and hardware characteristics.

Besides the JVM resources linked by others, I found Richard Jones's "Garbage Collection Handbook" to be a decent introduction for background [1]. The Go team has written a bunch about their GC approach [2] - it is really interesting to see how it compares to the various options in the JVM and under which scenario you might prefer one or the other. And occasionally there are interesting articles on arxiv.

1. http://gchandbook.org/

2. https://go.dev/doc/gc-guide

And then when a dependency of one of your dependencies is broken agains the latest version. I know some of this has been cleaned up, but it is one of the main reasons I no longer use Scala - I have a rule about how long I'm willing to spend on build issues vs actually writing code, and Scala was always on the wrong side of that.

re: Complexity - At least the signatures for the core collections have been cleaned up a fair amount. That said, the richness of the type system and the prevalence of operator overloading always made it feel like a language you could be really productive in once you knew the language and the current codebase really well, but was really hard to just read through unfamiliar code and know what is going on.

There was definitely a large and vocal part of the community that wanted that, but I think early on there was a lot of tension between Scala being "better Java" and "Haskell for the JVM", and that probably hindered a lot of adoption.

> Over exposed to US treasuries? I don't think anyone has put those words together before.

Only if your definition of risk is limited to default risk.

It looks like a lot of their holdings were in MBS and a lot of the treasuries were recently sold at losses (which necessitated raising capital, which appears to have set this whole thing off), but there is no reason this couldn't have happened if they only held treasuries anyway.

I don't this is an area of corporate law that has been super thoroughly tested? But I think the theory is -

Look, we had $X in our account on Wednesday and did a payroll run that was more than covered, but then our bank went under and the payments never got there. That's not our fault, nothing we can do about it!

But as time goes on and you're still employing people while having no bank account, you become more culpable for their unpaid wages after you find out you have no money.

The tricky thing here, I think, is that it still isn't very clear what is going to happen and it'd be silly to immediately fire a bunch of people because of it. There is a good chance some other bank acquires SVBs accounts and things are back to normal pretty quickly. If that doesn't happen, there is still a lot of money to go around, so you aren't losing 100% of what is in there.

I think there is a good argument that it is prudent and legally defensible to wait a day or two to figure out what is going on before you go nuts and shut the business down.

1:1 banking essentially guarantees long-term loss of principal for depositors, albeit slowly. The main "service" a bank would provide in that scenario is holding onto your money for you and instead of paying interest on it, charging you a few percent per year to hold onto it for you.

Also, in your model, where does money for loans come from? How are borrowing costs impacted by a major source of funds for loans going away?

I guess, in its own way, knowing that there will be constant layoffs is a form of clarity?

Huh? Twitter has had almost monthly rounds of layoffs, including one a few weeks ago that took out some fairly high profile employees. They also have arbitrary firings and people being let go without being told [1].

I'm certainly not going to argue that Meta is going about this in a good way at all, but it is hard to think of a worse example of doing it "well" than Twitter under Elon.

1. https://twitter.com/iamharaldur/status/1632843191773716481

Zen 2 architecture came out in 2019 and this particular processor in 2020. I don't think your advice would work here, and there are definite downsides to waiting too many years after a processor has been released to buy it.

> He discovered that a full 20% of the sample characterized as “healthy” weight had been in the overweight or obese category in the decade prior. When set apart, this group had a substantially worse health profile than those in the category whose weight had been stable.

There are a few odd things about this approach.

First, 20% of people in a healthy weight being previously overweight/obese? This seems high? Could there be reporting bias or other things here? I'm still trying to understand the full study, but it is odd that more people lost weight than gained weight, according to the numbers I'm seeing in it. Although that may be partially explained by the fact that it was limited to people 45 - 85 years old.

Second, the methodology of taking anyone who was previously overweight/obese and is now healthy (weight-wise) into their own group and then attributing increases in mortality purely to weight seem suspect? There are a lot of things that can cause you to lose weight and lead to increased mortality that aren't related to weight, and those would be captured in here.

The conclusions definitely make sense and "feel" right, at least directionally, I'm just a little confused about some of the methodology to get there.

There is no credible threat of invasion of the mainland US, even with a much smaller military, so the difference in threat to CA/NY vs Ohio is pretty small. And, at any rate, Defense budget support is more due to economic implications of bases/contractors in the states/districts of people in congress and military worship/nationalism in the US than it is any sort of logical assessment of threats.

Certainly, although I think that is much closer to "force" than "raising the price". Also, it is politically infeasible in Utah.

This has been happening for commercial/residential cases, but most agriculture is based on water shares - a right to a certain amount of water that someone purchased a long time ago, so you can't really raise the price on them.

One solution I've seen kicked around is spending a bunch of money to buy back those shares at well above market rates to conserve the water.

> but farmers will be among the first to really feel the affects of a dying lake ecosystem.

How is this so in the case of the Great Salt Lake?

I guess, why do you want to stay up to date on all sorts of stuff like this? It is more than any single human could reasonably process, even if they did it 24x7. Even people working in finance are going to be focused on fairly small areas defined by geography, industry, product type, etc..

Context is important - for a random person wanting to keep up better on financial stuff, $2500/month is extremely pricey.

I have worked in plenty of fast-moving and chaotic environments, and while I can see some stuff slipping through the cracks, it is hard to see something so core to your business getting this bad without realizing it.

Like, if you are releasing financial documents and a large entry is "HIDDEN POORLY INTERNALLY LABELED ACCOUNT.", it is hard to explain that without getting into outright malice or negligence so extreme that it is difficult to differentiate from malice. Even rudimentary reviews of their finances would have highlighted stuff like this, and they clearly had money to pay people to help them do it.

I remember in the first dot-com bust, CPMs/CPCs fell like a rock, so everyone was cramming more ads onto pages to make up for it, which added a lot more inventory, which caused them to fall further. The market is much, much bigger these days, but I wonder if something similar (if smaller) will happen as everyone scrambles to monetize and increase revenue.

No. This was for more of a machine learning engineering role than a pure ML scientist role though.

I had a recruiter reach out to me about some ML positions there earlier this year, and I was sort of interested in it for a bit. They have an applicant portal where they literally have you practice leetcode problems, and are pretty open that you're going to be asked those kinds of questions in the interview.

> I mean it still exists bu tall expansion palns stopped and it went into maintenance mode.

Google Fiber is currently microtrenching in front of my house, and continues to move into more cities here. Did they restart, or did their mothballing just mean that they aren't entering new geographic regions?

TBH, I'm super excited about it because the only other high-speed option we have is Comcast and it is super unreliable and the data cap sucks, but I'm also mentally preparing for the day they get bored of it and shut the whole thing down.

I seem to be in a small group that doesn't have a huge preference between the two, maybe I'm not doing that much advanced stuff?

If I'm just prototyping something myself, I usually reach for TF first, mostly because Keras feels like the "right" level of abstraction for most stuff. I used to prototype things with fastai/pytorch more, but newer developers didn't like how much was hidden behind multiple levels of *kwargs and some of the dataloader stuff could get tricky if you tried to do anything non-standard. I haven't tried PyTorch Lightning.

Besides the deployment story, which is pretty big and others have touched on, there are some minor things that feel nice in the TF/Keras ecosystem:

- Part of deployment, I guess, but I like that more of the preprocessing can happen in the model, vs as a separate step. The less transforming of data that has to be done in the serving code, the fewer possibilities for things to get out of sync and introduce bad data at runtime.

- Keras being able to infer input sizes in layers is nice for avoiding a bunch of bookkeeping code to calculate layer sizes.

That said, I feel like maybe the TF ecosystem has more sharp edges? I've encountered more than a few of them lately as I've been doing work on some recommender models using tfrs. I've also run into things like tensorboard logging not working with entire classes of layers and causing training to crash.

I'm curious - what are some things about PyTorch that make it better for almost every single use case?

I'm not sure what I'm missing, but don't most cloud providers have ARM instances these days? We've been running most of our stuff on Graviton processors on AWS for some time now and it is a nice cost savings.

But that's the whole point - Musk waived his right to do due diligence, so he is pointing at this statement being wrong (and not unusable) as his way out.

Hostility, yelling, threats, etc just aren't worth it and I have zero patience for it in a job. I had an exec once who would threaten to cut people's fingers off if they didn't accomplish incredibly unreasonable things, or throw them out windows, or whatever.. It wasn't a serious threat, but it also isn't a fun thing to listen to all the time.

I realized after a while of that sort of abuse that no job is worth that and I have much less tolerance for it. Thankfully it isn't super common, but it is surprising how many people think that is a useful way to work.

The remote functionality is nice, as is the auto-away part. Those alone make it worthwhile, but I agree that the "smart" functionality leaves much to be desired.

Ours, somehow, has gotten into a state where it thinks we like to bounce between 68 - 70 degrees all day, and doesn't "learn" when I try to adjust it back to the same temperature it was. The weekly schedule has over 70 different temperatures on it and there is no (obvious) way to reset it, so I haven't bothered to go clean it up. I think it could do better there.

It is not, but I didn't think ABET accreditation was a big thing for CS and was more common in engineering? CMU and Stanford, for example, do not have ABET-accredited CS programs, although MIT does.

That's a good question, given that it hasn't been a huge deal so far in my career. Two things:

First, I realized I wanted to dig deeper into CS topics. I love this field more every year I am in it and wanted to round out my knowledge in some areas. Getting a master's degree and the courses that are part of that seemed like a good way to accomplish that - I've taught myself a lot over the years, but that route leaves some areas untouched. You don't always know what you don't know, plus having deadlines, goals, etc helps keep me on task even when something feels like it isn't immediately relevant. So, getting a BS was the first step in that goal.

Second, as I progress more through my career, there are some interesting positions that require, or at least favor, some kind of degree. I feel like having one or both gives me more options and flexibility.