bzz01

This video has a bunch of cherry picking results and sound bites from already underpowered studies.

Given that you already rely on Polars, did you consider providing a pure Rust interface to Metaflow?

75MB/5 sec benchmark results (using internal AWS network to pull from S3!) sound dubious. You can get 4Gbps+ down from S3 within the same region in my experience, that's 30x faster than these numbers.

I've built this page since I often just want to pick a spot instance type that gives me biggest bang for the buck. Amazon's own pricing page is still very confusing, and ec2instances.info is great but it is a static website and adding spot prices there is nontrivial.

I've also fixed a few minor things that annoyed me, such as correct sorting by instance type (so that r3.16x goes after r3.4x) and added a mode to display spot savings vs on-demand.

Would appreciate any feedback!

Another mistake was that when originally released Python 3 did not offer enough shiny new things to convince people that going through the pain of upgrading was worthwhile.

This is the core issue, but you're putting cart before horse a bit: if there's not much to gain from switching, why would any reasonable person switch? And, most importantly, why is it a problem that needs solving at all?

C++ is a pretty low bar, to be honest. But yeah Delphi was awesome!

That one sounds fairly similar to manslaughter, no?

My understanding is that logically there's that basic framework of required specific intent, then there are a few crimes that allow prosecuting for negligence, then very few crimes with strict liability.

I'm originally not from US so I'm probably still getting used that the latter two categories are quite common in US law (and that strict liability is even possible).

Good point. One answer is that manslaughter is already a special case, as it doesn't require intent to kill anyway, only negligence.

Cisco continues to dominate routers, despite China's attempt to steal from them and clone their products.

Huawei revenues are already on par with Cisco, at least that one worked out pretty well.

Come on, if that was true they would never be able to hire tens of thousands of engineers. They do have high false negative rate (by design) but it is far from impossible to pass. I personally know a ton of people who were hired by Google after standard interview process.

If you are reducing the number of shares outstanding from 32,000,000 to 400,000, it is 80:1 not 8:1 reverse split.

In your example, employee would have 100 not 1000 shares post split, and still would gain $6K.

So of course I'm generalizing, not everyone is like that, not everyone's CS degree is like that, etc. etc. Also, more true for recent grads.

But in my experience people with advanced "pure CS" degrees seem to be focused on the process not the result. I mean, they went to CS because they like to tinker with type systems, elegant algebraic concepts, nice abstract problems -- and not just as a hobby, they like it so much so they commited several years of their lifes to do that. Not that this is bad area of research, but generally when you look for a software engineer you look for someone not only smart but also pragmatic, who can get stuff done in a simple and efficient way, and this is kind of the opposite.

Of all PhDs, in my experience people who have degrees in areas that use programming as a tool not the goal in it self, e.g. physicis/EE, make the best software engineers.

I guess author assumes that if you don't know basic things like that, it is questionable that your degree is a proof that you "can learn a field etc."

Not sure that I agree on the choice of this specific "red flag"; on the other hand, having PhD in CS is a red-ish flag to me per se.

Frankly it is a bit weird way to describe it. It is not some quirks of liquidation preferences that "ruined" it.

It is ruined by a simple fact that this hypothetical startup earned just 50 cents for every dollar raised -- I don't think anyone should expect a nice payout here.

I don't think it was free, but again entire Microsoft toolchain wasn't free in 90s, so nothing special about C here.

Things are getting better.

In what sense? MS platform had a working C compiler for decades.

Google/Facebook/Apple may not be as concerned by this as you may think; strategically, they would probably be happy if slow decline of advertising income on desktop would eventually make every publisher to move to walled gardens of mobile platforms, completely controlled by them.

You may end up with a world with nearly unblockable, more creepy ads, and no more open web as we know it today.

You can't really use KD trees with lat/lon coordinates, at least you can't use euclidean distance there for nearest neighbor search.

First, longitude wraps from -180 to +180 at antimeridian, meaning distance calculations will fail there; second, and I'd say more importantly, one degree longitude length in meters differs a lot depending on latitude; meaning this library will be heavily biased towards longitudal neighbors when using it for locations far from equator.

As an engineer with ten years of software dev experience who interviews quite a lot of candidates for my current team, I'd say don't overthink this.

BigO, data structures and having a good idea how computers work, what matters for performance both from theoretical and practical perspective, matters infinitely more to us that knowing how to play object-oriented BS bingo or subversion usage patterns.

If you know what MVC stands for, you're know enough about OO patterns to me.

Not going to argue that we got it right way(tm), just wanted to give you another perspective.

While all maps like this tend to make little practical sense since they inevitably over-generalize and over-simplify things, I'd still like to point out that they got "columnar" category quite wrong: neither HBase nor Cassandra are columnar stores in a way this term is commonly understood.

HBase and Cassandra still store data in rows, however rows can be partitioned into column families which may be stored separately. Columnar databases are usually also relational (Vertica and Redshift) and support SQL or SQL-like query language.

Anyway, I think regardless of how you define columnar, HBase and Redshift shouldn't end up in the same category as they are quite different in a way they work, throughput/latency and read/write balance and use cases.

There is no need to guess, data published by Dept of Labor (original source for this infographic) includes company names, specific position names and exact salaries.

Glassdoor was including that info in their estimates for a while though, so no big discoveries here.

I've seen these talks from CCC few years ago, tried that myself (on myself) and SS7 only gave MSC id, which is wayy more coarse than cell tower level -- like, 1M customers per MSC (still a privacy concern of course). But that city block level location thing seems like it is something more sophisticated.

For STEM fields (esp. software) it was relatively easy to get H1B visa until 2013, yet very few people (think couple thousands per year) actually did that compared to 1990s if you look at USCIS stats.