runT1ME

>suggests that the architecture doesn't matter, but we know that's not true. Eg, why are networks with skip connections SO much better than networks without? What about batch normalization?

Is this true though, or does network architecture only matter in terms of efficiency? This is non rhetorical, I really don't know much about deep learning. :) I guess i'm asking if with enough data and compute, is architecture still relevant?

I think (hope?) you're getting downvoted because most folks aren't arguing against giving someone fizzbuzz or seeing if they can loop through a tree, it's the fact that some of these tests gives you what is an obviously O^3 algorithm and you have to use some non trivial trick to get it to O^2 or something.

I'm very sympathetic to the idea that a more objective test like interview format eliminates bias from the interviewer, as well as the idea it allows introverts or those not as well versed in storytelling to be successful opposed to traditional 'sell yourself' interviews.

However, it's a huge waste of time when so many of the questions are the same. I get it, every company wants to know if I can turn a list into a balanced BST, it's not an unfair question, it's not particularly hard, but I really get annoyed I have to keep solving it over and over.

I really wish there was some certifying body that would put me through the whiteboard guantlet once, and then I could just share my score to companies I apply to.

As someone who grew up and lives on the coast, I had no idea what to expect the first time I visited Louisville a few years back.

I was blown away how beautiful it was and how welcoming people were, in addition to it being a fun place to be. It was nothing like the other midwest cities I had been to, seemed to me much more vibrant and scenic at the same time.

I'm a fan of scylla too, but if I could go back in time I'd have recommended waiting until mid 2019 to migrate. 'Fully optimized C++ rewrite's tend to take years to become battle tested.

A bit peculiar they selected a .NET language after running into a lack of libraries in OCaml. I'd be perfectly happy to use C# or F# the language, but I still find .NET lagging behind the JVM/Python/Go ecosystems. Two large OSS projects I'm using for work are Scylla and Apache Beam, both of which don't have native support for .NET, and I'm sure we can find plenty of others.

>Has anybody legitimately had good experiences with microservices these days?

Yes. I work for a company that started out with a few separate monoliths that were hacked together under intense time pressure by some smart folks that didn't anticipate the scale they were going to be operating under. As the team expanded and more talent was added, they were able to partition up some of the functionality into separate services.

When I joined we had a small team with a mixture of legacy systems written in Java, Perl, PHP, Scala and Python. Some of the microservices were poorly done, others were fine. The new teams were able to completely overhaul each service one at a time to where they were stable and needing almost no maintenance/babysitting.

Would microservices have been a good starting decision for this company, given its current level of engineering talent? Probably not. However, take a bootstrapped team crunched for time, with some foresight to know that there would need to be rewrites but not certain where, microservices can end up mitigating a lot of unknowable future issues.

I live in OC now, and have traveled quite extensively as well as spent time working in the Bay. While lacking in Michelin star restaurants (only a handful) it is otherwise phenomenal when it comes to options and quality, even compared to bit cities. Now bars/nightclubs/music venues are where it is sorely lacking.

>What about handling in corners, etc., ride comfort

Drives a lot like a higher end mercedes, E-class/CLS feel. Smooth, powerful, good handling but not a tight suspension. My previous BMWs had better handling with better suspensions. Still, when comparing the Model 3 to similar ICE vehicles, the Tesla was hands down a better buy in my opinion.

Jupyter notebooks are almost a huge step forward for programming. The instant feedback, saving of partial program state, and visualizations mixed in are amazing and fun to work with. The fact that it's taking place in a browser, you don't have modal editing with Vim, you don't get plugin support (like MyPy), and you don't have intellisense is a huge step backwards.

I'm hoping that someone addresses these issues, I know VSCode is trying, and Scala "Metals" worksheets are another similar concept (though different, it puts variable values in comments next to code as you update the code).

I feel like all of these are a step closer than what the Lighttable/Eve teams were shooting for, but not quite the next generation way of coding that it could be.

Not what it says, direct quote: "Hooked’s CEO declined, saying there were other acquisition possibilities she would like to see play out."

"Or even Scala" seems false.

The majority of Twitter is Scala, Stripe uses Scala extensively, Verizon uses scala for at least a large portion of their TV related code as does Comcast, I could go on. Did it 'take over' the world like some of us hoped? Nope, it is definitely still a niche, but it's quite a large one at this point.

Scala Metals has this feature, it's very cool.

Do current NLP systems understand arithmetic, and can the do it with unfamiliar numbers they've never seen on? If not, I'd think that your theory is demonstratably false, as a child can extrapolate mathematical axioms from just a few example problems, whereas NLP models are not able to do so.

>because leetcode/algorithmic interview questions are so fundamentally different to what we actually do day-to-day as developers.

I mean, maybe hard or esoteric easy questions but I don't think that's true for basic things like 'find the first duplicate' or '-traverse- this graph'. When people rail on whiteboard coding, I think they mean obscure algorithmic questions and not fundamental data structure questions.

I don't expect someone to be able to do Dijkstra's algorithm (or even spell Dijkstra, did I do it right?) but everyone should be comfortable enough with a bit of help doing BFS or working with linked lists, hashmaps, no? Am I crazy here?

While personal projects are good, even more impressive is contributing features and/or bug fixes to larger open source projects.

It's sometimes much more challenging to dive into an existing project, learn their frameworks/style/design, and work with existing contributors.

However it's also much closer to what happens on a real development team. Whereas a project with a single author can show coding skill, working in a large open source project shows collaboration, communication, and soft skills as well.

It is a huge gold star on a resume in my opinion.

File ASAP. Even if by the time it goes through you get a job, they'll still pay you for the time you were unemployed. Presumably you already paid into unemployment, get some of that money back now that you need it. Even if it comes weeks from now and you're happily employed, use it to rebuild savings or pay off credit card debt etc.

Here's one way to look at TSLA. I see zero chance of Toyota, BMW, Mercedes, Volkswagon, or GM making all of their direct competitors obsolete.

With Tesla's massive monopoly on driver data (likely a prerequisite for level 5 automation) and head start in driverless tech overal there is absolutely a chance that TSLA puts all the aforementioned companies out of business.

I don't know how likely that is, but it's a hell of a lottery ticket regardless.

Modern Spark is nice enough that I will use it to work with massive excel files locally! If someone can use Pandas they can use Spark, and you get the added benefit of distribution if necessary and more resiliancy.

I think it's likely that some subsection of Spark users are the type to over engineer a project, but I'm also confident they'd over engineer a much simpler framework as well.

This seems pretty cool. Would I still get value if I only want to sample performance a few hours a week, say, under high load? Do you allow for that?

Can you explain more? From what I understand, if you're already running a Flink cluster on AWS and you have capacity for another job, you aren't charged more, no?

I haven't used Glue, but it seems like it's able to do stream processing on Kinesis and dumping to S3 or whatnot, so it seems like there's overlap with using EMR running Flink?

If you're already paying the cost(both engineering time and compute wise) for EMR, I can't imagine it takes more effort to create a new Flink job than a new Glue job?

The advantage of Glue or the corresponding serverless GCP ETL option (dataflow) is that it's serverless elastic, but it sounds like their workload wasn't applicable.

Can mongo really support the kind of large scale ETL or time series use cases Cassandra can?

I think this is a valid point, two projects I'm making heavy use of are both ScyllaDB and Apache Beam. Neither have C# as a supported language.

Many big projects roll out drivers for the JVM (Java/Scala/Kotlin compatible), Python, and Go first. I agree that C# is a 'nicer language' than Java, but at this point, that doesn't make up for the lack of ecosystem.

The cloud and open source world, especially in the infrastructure space has moved incredibly fast in the past five years. It's certainly possible that an in house microservice can now be replaced with one of the latest AWS services or even some of the latest apache OSS + a few lines of yaml/config.

That doesn't mean it never provided value, just that there are now better alternatives.

I don't believe PubSub has a way to ensure linearizability between messages, where Kafka does have this by deterministic hashing of keys partitioning. Otherwise yeah, Pubsub looks pretty solid and economical.

At my company I'm guessing we push over 200M messages a day through it, and it is a breeze to maintain and stable as hell. Maybe gets harder a couple orders of magnitude more, but no complaints here.

The inflation that matters such as asset prices and real estate aren’t captured by CPI.

Thought provoking stuff! How can someone talk about conciousness existing then, without presupposing dualism?

NATS.io and Kafka seem very different, the former not being able to do at least once guarantees, no?

The tax code actively discourages savings for a corporation. If a company wants to make 5% on a relatively safe investment, they have to pay the corporate tax rate first, then invest the money. If instead they calculate a business expansion can bring them 5%, they can hire new employees/buy a business and not pay that tax and still get the 5%.