Live data from Hacker News

Netflix CEO explains why he pays technologists huge salaries

insights.dice.com

321–330 of 448 posts

Re: Netflix CEO explains why he pays technologists huge salaries

#321

Earlier quoted context omitted.

You mean doing the work _again_. I had all the algorithm stuff down cold out of college and maybe for another 5 years after. Now after 20 years of rarely needing any of it and having it be a web search away when I did, I couldn't come up with any of it on demand. "Doing the work" meaning having to study to relearn a thing that I'll maybe use sparingly in my work means the interview process is broken, not anyone's mot…

>"Doing the work" meaning having to study to relearn a thing that I'll maybe use sparingly in my work means the interview process is broken, not anyone's motivation I used to think this too, but as a filter I think it's having the desired effect. From the outside looking in Life at a FAANG seems competitive in nature and they are optimizing for the people willing to push themselves that far to learn something. The ad…

> The added bonus is all new hires at least know the effects of space/time complexity and how to make space/time trade-offs deeply well.

Do they?

In my experience this was true in the early days of white board interviews where the only people that even thought about this stuff were people deeply interested in algorithms and computation.

But in the post-"Cracking the Coding interview" era (including the transition into the full leetcode era) people seem to be able to pass these interviews without really understanding anything in the same way college students can ace calculus while have literally zero intuition about derivatives.

I don't mind whiteboard interviews, they're a fun game you can train for and the practice is interesting. But the big tragedy now is that I feel almost nobody understands algorithms anymore since everything has been reduced white board coding drills and everyone believes they have an understanding of something they don't.

As a great example of this almost nobody today knows any heuristic optimization techniques (think the old ITA software puzzles), which used to be near and dear to anyone that cared about algorithms. But since you can't test for heuristics they same as you can established algorithmic solutions nobody studies them.

Re: Netflix CEO explains why he pays technologists huge salaries

#322

Earlier quoted context omitted.

Netflix doesn't win or lose customers based on programmer performance, they win by having content that people actually want to watch. I don't care if the UI is 10% less slick, as long as they have stuff I want to watch. The only reason I've ever considered dropping Netflix is "why am I paying for this when there is nothing I want to watch". Fortunately, they have come through with some interesting content recently, a…

Netflix programmers can help you find what you want to watch. All the content in the world won't help them if you spend your time staring at the home screen while bad algorithms recommend the wrong things.

Not true at all, because in spite of all their content, and being able to find it, they still don't have anything new that I want to watch. I've already seen everything I wanted to watch. Now I debate leaving them because there is no new content. And after all "Content is King".

https://www.forbes.com/sites/steveolenski/2017/06/21/why-con...

>All the content in the world won't help them if you spend your time staring at the home screen while bad algorithms recommend the wrong things.

Their algorithms only optimize for the least profitable use case. It doesn't matter what the algorithm wants to suggest if they don't have the content I want to watch, or I've already watched anything worth watching.

Know what their algorithm suggests to me now? Sci-fi in foreign languages that I'm just not interested in because of its low production value and bad writing. But it's all they have left to suggest to me. That and everything I've already watched.

"Content is King".

Re: Netflix CEO explains why he pays technologists huge salaries

#323

Earlier quoted context omitted.

>"Doing the work" meaning having to study to relearn a thing that I'll maybe use sparingly in my work means the interview process is broken, not anyone's motivation I used to think this too, but as a filter I think it's having the desired effect. From the outside looking in Life at a FAANG seems competitive in nature and they are optimizing for the people willing to push themselves that far to learn something. The ad…

I would think these companies that are built on measuring data would have some idea if the whiteboard interview works. Otherwise, why would they do it?

Q. Why does 90% of human society work the way it does?

A. Because that's the way we've always done it.

I can't speak for SV companies, but my experience of data is that it can just as easily rationalize preconceptions as be used to inform decision-making. Somebody has to interpret that data, after all.

Re: Netflix CEO explains why he pays technologists huge salaries

#324

Earlier quoted context omitted.

Netflix doesn't win or lose customers based on programmer performance, they win by having content that people actually want to watch. I don't care if the UI is 10% less slick, as long as they have stuff I want to watch. The only reason I've ever considered dropping Netflix is "why am I paying for this when there is nothing I want to watch". Fortunately, they have come through with some interesting content recently, a…

> I don't care if the UI is 10% less slick, as long as they have stuff I want to watch. It's not a coincidence that they happen to have stuff that you want to watch. You're aware Netflix also applies data insights to decide what to make?

>You're aware Netflix also applies data insights to decide what to make?

Of course I give something thumbs up or down depending on if I liked it - why would you assume I don't know what they do with that?

But they've still made some barely watchable crap, too, in spite of themselves, so I do have to question if rating things even matters.

Re: Netflix CEO explains why he pays technologists huge salaries

#325

Exactly what is Netflix doing that requires such skilled programmers? As a consumer of its product I haven't noticed any significant groundbreaking changes to the platform in years. I find more people complaining about the software experience as years go by. What have these rockstar programmers done exactly? I can tell you it's for sure not the UX, or its recommendation engine, which are frankly both abysmal. I'm gue…

I work for Netflix Open Connect (CDN). We've been leaders on CPU and power efficiency with things like async sendfile, kernel TLS, and more. We've had servers doing 100Gb/s in production for 4 years, while competing CDNs are just now rolling out 100Gb/s servers. I have a prototype machine right now serving 375Gb/s of TLS encrypted customer video traffic with just 32 cores.

> serving 375Gb/s of TLS encrypted customer video traffic with just 32 cores.

How does that even work? The fastest system RAM you can get that I know of is DDR4-5000, which transfers at 40 GB/s (320 Gb/s). How can you encrypt and send data faster than your memory can even transfer it to the CPU? Or is there something I'm misunderstanding?

Re: Netflix CEO explains why he pays technologists huge salaries

#327
post #77

I think 100x is hyperbol however you slice it, run away from anyone who tells you they are a 100x developer. 10x may be achievable... but I argue that it is at least partly due to the prevailing code-base. Every competent developer is a 10x developer on a green field project just like every competent developer is a 0.1x developer on a messy legacy spaghetti-code disaster of a project (Making the differential 100x I g…

It is depressing that people are happy to just throw anecdotal evidence around without even defining what metric they multiply by 10x or 100x.

There is some literature on it. For example, this blog post [0] contains some references.

I have only read Peopleware. It says 10x is the difference from worst to best in terms of time and number of defects across a few hundred programmers. Also interesting: Within the same organization, there was only a difference of 30%. This means the environment is much more important than the skill, experience, or personality of a programmer.

I cannot find any studies after 2000. Most of the data is even older. The Peopleware data is from Cobol, Fortran, Pascal, and C.

[0] https://www.construx.com/blog/productivity-variations-among-...

Re: Netflix CEO explains why he pays technologists huge salaries

#328
post #104

Earlier quoted context omitted.

I don't think I'm a 100x developer or anything, but I think people underestimate the compounding interest of being a 5x-10x developer. If you architect things just right today, that can save you a lot of time tomorrow. Spending time testing today can save weeks of bug time later, which can be spent to create something useful, compounding on itself. If we say green field is 10x, and someone is doing good things, I thi…

I was involved in an accidental situation where two teams at different companies worked on the exact same project from the same starting point, each led by different developers. This was not the company's core product, but a necessary part of it (for anonymity, let's say this was the device driver for a fancy piece of hardware). The lead developers on each selected different frameworks, used different data architectu…

What was the outcome? Did the convoluted project get abandoned?

Re: Netflix CEO explains why he pays technologists huge salaries

#329

Earlier quoted context omitted.

> I don't care if the UI is 10% less slick, as long as they have stuff I want to watch. It's not a coincidence that they happen to have stuff that you want to watch. You're aware Netflix also applies data insights to decide what to make?

>You're aware Netflix also applies data insights to decide what to make? Of course I give something thumbs up or down depending on if I liked it - why would you assume I don't know what they do with that? But they've still made some barely watchable crap, too, in spite of themselves, so I do have to question if rating things even matters.

The thumbs up/down is just one signal among many. They know in aggregate what people watch, how much of a movie/episode they watched, the kinds of devices they watched on, how much binging/pausing they did, when and where they watched it...

> they've still made some barely watchable crap, too

To you. That crap was undoubtedly entertaining to someone. Not to mention a 100% hit rate is nearly impossible in the creative biz.

Re: Netflix CEO explains why he pays technologists huge salaries

#330
post #276

Earlier quoted context omitted.

Here's the thing: "But we do deal with this stuff at Netflix," is still only true for a tiny fraction of the engineers at Netflix. I currently work at a company that is not quite FAANG scale, but almost is: we have issues in our platform and systems that have forced us to adopt "FAANG-like" solutions to certain issues of scale. Almost none of the engineers at this company are doing anything at all with interview-styl…

But the 5% ~ 10% that do are at least guaranteed to be capable of it. And it's likely that that matters.

Sure, but it doesn’t matter enough that interviews should be built around this domain, especially to the extent they are around the Bay Area.

Having 0.5% of all engineers regularly engaged in working with the implementation details of data structures and algorithms does not justify putting 100% of engineering candidates through interviews centered on that topic.

Post reply on HN