Hacker News Reader

Comment by MonkeyMalarky | Hacker News Reader

The thing I don't get is the hyper-focus on salient or "fancy" code. We're not all Jeff Dean and most code is bland and boring plumbing work anyways. Even in awesome ML powered systems, it's still 99% bland foundation, 1% cool algorithms. Hell, as a manager, you don't want only "rockstar" developers because they'll become bored working on the plumbing code, which again, is most work. From datascience there's the phrase "Everyone wants to do the model work, not the data work" and it sounds like he wants only those people? Why?

View on HN

It's literally just some bizarre hazing ritual. Trying to evaluate whether code is good or bad from some screenshots without understanding the context and being familiar with the codebase makes no sense anyway. Anyone can show you ten screenshots of formally well written code and you have no idea whether it's useless or not.

how many commits of the fast inverse square root are being added to twitter's codebase right now, you think?

Where did he say that he will evaluate if the code is good or bad? He didn't. It's more likely just a way to see what employees think is their most important work.

If the goal was actually to see what employees think is their most important work, that (and criteria) would be specified, instead of playing "what is he looking for" mind games.

He did specify. He asked for most salient. Salient means "most noticeable or important."

He hasn't played any mind games. He's asked for exactly what he wants.

That is completely a mind game. Salient according to what criteria? Is he going to be happy to see wonderful high level distributed designs? Great coding practices at the micro level? Unit tests? Performance-oriented code? Ops greatness? CRUD boilerplate that any junior IC could jump into without much understanding (seems like this might be a stealth winner if he’s gearing up to build a skeleton crew of low-seniority devs that don’t have a lot of career options). Is feature work sexy to him or does he want to see work that is oriented around KTLO cost minimization? Improvements on usability (decreases support costs)? Reliability? Scalability? Security mindset? (obviously kidding about that last one) There are so many things that are “salient” depending on his priorities and attitudes that this is a fool’s errand.

> That is completely a mind game. Salient according to what criteria?

Is what you do worth your salary? That is a pretty simple criteria, if you deliver significant value in basically any way that should be enough. Why do you think this is would be so hard? We had to write down reasoning like this to motivate our job at Google 1-2 times a year, it isn't hard to do.

Ok now do it in 10 code screenshots.

My complaint isn't in the general concept of justifying your value. My complaint is in the request to have every coder in the company boil this down to such a simplistic representation of it on short notice in an environment where every indication is that it's going to be considered for a laughably small amount of time and then you might be fired on the spot if you didn't focus on the right axes.

Why does everyone assume that people won't have a chance to explain the value of their code in these interviews? How much time do you think you would need to put together a list of your 10 biggest contributions and some screen caps to illustrate them?

> Why does everyone assume that people won't have a chance to explain ...

I think the fact that this is happening to every coder at twitter in a few hours' span, coupled with the fact that thus far, the firings have been capricious, hasty, and based on apparently very little data. Nothing about these exercises seems to be "measured" in the "carefully considered" sense.

If you're thinking of these in contrast to standard performance evals that we all expect to have to do, you should also actually compare the processes in a meaningful way. Perf evals: you get weeks to work on them, you get to write a long-form essay making your case, you rarely ever get fired on the spot for not doing well, you get all manner of coaching and documentation about how to do well on them (including very specific tips on criteria for success). Compare and contrast with "bring me 10 screenshots of your code by 2pm today" along with the recent atmosphere of "anyone who doesn't meet the surprise criteria we just sprung on you is fired" and you wonder why people think this is absurd?

He can run his company how he wants, but this is way out of line with what I (and obviously many others) have come to expect. Don't understand the outrage? Well, this is the explanation. We've never had to put up with this shit before, and we don't feel like starting now. 1200 people quit yesterday. We still don't know how many were unwilling to jump through this latest dumb hoop. We shall see.

Of course nobody is excited to explain what value they provide on short notice to keep their job. Nobody likes getting fired either.

I agree that these exercises are neither measured, careful, or well coached.

It is a blunt and imperfect way to see who codes at all, who can articulate their job, who can handle uncertainty and pressure, and also who will follow orders.

Taking a screenshot and explaining your code is not some impossible task. People just don't like the stakes.

My point is that it seems like people are overly focusing and critical of the request itself, when when request is a practical way to meet the goal. it just happens that the goal is crazy, unmeasured, rash.

I agree it is out of line from what people want and expect. I'm not sure there is any way to cut >75% of staff and gut a company culture in weeks that would fit the expectations and desires of employees. The situation is unheard of.

Ok, well everything you’ve said in this subthread up until here has seemed to defend the request as reasonable. That is the confusing part. Our entire conversation still works if we swap in “please measure the length of your forearm and report to the 10th floor” for the request. It’s an unreasonable and arbitrary way to compare employees for immediate dismissal, yet it’s definitely a way that would let you cut 75% if that was your goal.

I'm saying that I wouldn't like it, but it is a reasonable request to expect a worker to be able to perform it.

I think that it's imperfect but probably an moderately effective way to compare employees for immediate dismissal. I would not like facing immediate dismissal.

I think the difference is that I don't think it is nearly as arbitrary as you do.

Yeah. I think when done this hastily, it’s gonna be pretty arbitrary. Otherwise we’d always do it this way.

It is an email with an explanation of what you do, then some screenshots with code. People who mostly delivered value by writing code at Google wrote an explanation what they did, and often linked to different changelists with code they had written. The only strange part here is the screenshots of code instead of PR's, but adapting to that isn't very hard, take some screen shot of the diff or some part of the code and explain how that code fullfills the things you talked about in the email.

A person who can't do that lacks the soft skills to be an effective software engineer.

The most important work I've done in any company isn't obvious from some small snippet of code in isolation. It's a clever bit of modeling data or code architecture that would look completely unexciting without knowledge of the problem domain.

No screenshot. You're fired.

I think the only reason is to figure out if people are working.

Any dev I could think of doing their actual job can take 10 screenshots and point what they worked on in the last 6 months. It would be insane if they couldn't.

Including PEs and Sr.PEs? If they're coding, it's usually a detriment to the business unless they are extremely specialized and possess some blend of skills that allow them to take on tasking that cannot be delegated to others. I'm mostly talking about top companies where those titles are hard-earned and are associated with very broad impact.

I agree with this. In fact, I don’t even know if Musk cares whether people actually review the submissions.

This could be a very simple process of keeping them running through obstacle courses while the “weak” ones self-eliminate.

It’s constructive dismissal…

Legally I’d agree with you, but that just means he has to pay severance and whatnot which I suspect is just a line item to him. If I’m right he’s trying to go full “Sparta” and his actual goal is to end up with the best devs at the finish line.

It's been well established for a long time that this sort of hazing ritual doesn't leave the best developers around though.

The most desperate and blindly loyal maybe. The second category might be desirable, but the first one definitely isn't.

Jeff Dean would be dismissed. Sanjay types in the code.

It's simple, Elon Musk has no clue what he is doing.

Can he code.to save his life? Does he know anything about engineering from Tesla or SpaceX, beyond catchphrases?

All personal accounts I've heard of from people that work there are yes, he does know about engineering and is incredibly smart.

But that doesn't mean you know how to run a social media company, and trying to pivot Twitter's work culture on a dime is going to be chaotic.

FWIW I think he's a toxic asshole, but that doesn't mean he's an idiot.

He might not be an idiot but that doesn't mean he knows how software works either

Smart and idiot might not be mutually exclusive. I've known a lot of smart people who can act idiotic in certain circumstances. Elon could be very smart when it comes to engineering... but idiotic when it comes to running Twitter.

OK, so how do I square that up with a request to document programming prowess with screenshots?

That he doesn't have any other way to look at any code because he doesn't have a Phabricator account

He's unable to receive ASCII text strings over email?

He's not actually trying to document programming prowess with screenshots. They've already fired everyone they want to fire, this is some sort of "welcome to twitter 2.0" ritual

He lied about having a physics degree and he can't code. He knows nothing about engineering.

He used to be able to code. He created zip2 (predecessor to MapQuest) practically by himself apparently.

Without a source on that "apparently" I'm going to take that as the output of Musk's usual legend-building apparatus.

The guy has so much capacity to build his own legend (and have others believe it) that people refer to him as the founder of Tesla without even blinking.

He also pissed people off by insisting PayPal runs on Windows instead of Linux. Hey, I have fond memories of coding for the Windows platform for fun, but especially back in the day we kind of scoffed at those who wanted to run mission-critical systems on a Windows server.

Everyone who actually worked for him say that he understands all the engineering intricacies. Karpathy said that.

Tom Mueller publicly defended Elon on multiple occasions: https://mobile.twitter.com/lrocket/status/151291923068914892...

Considering that he lashes out at the smallest slight, I'm rather unsurprised that people depending on him for their livelihood sing his praises.

Tom Mueller is a highly respected engineer, and just because what he says doesn't align with your narrative or belief doesn't mean he is singing praises.

Oh by the way, Tom Mueller made that tweet after he left SpaceX.

If I were at Twitter and received this email, I would be providing code and commits focused on business impact. Saliency is scoped to something, and without explanation that 'something' is likely the business. The context is engineering at twitter, not programming. In other words, not silly one liners that are not immediately obvious but instead business critical code that moved the needle -- however bland.

I am lost. Where did Musk say anything about fanciness of the code? Or is it your interpretation of what he said?

> Before doing so, please email me a bullet point summary of what your code commits have achieved in the past ~6 months, along with up to 10 screenshots of the most salient lines of code.

“Salient” is not a synonym of “fancy”.

In case of code, isn't it basically the same?

I mean, i spent 20 months working on a project for a bank recently. I had a lot of salient code at the beginning, fun little algorithms and cool regex.

The last iteration of the code i committed, the salient part was basically a lisp interpreter (if lisp was formatted like json), but everything else was plain and boring. The last issue i had is that the ES version was upgraded and the python library i used to connect to ES was version locked, and did not work.

I think that for the next 4 years, this will be the only type of modifications my SRE ex-collegue who now own the code will make. Still, it can take days.

Spending days figuring out a few boring lines of change that has an outsized impact would, imho, be a fairly compelling bit of code to show off. IMHO while he might be crazy and way out of line, he's effectively asking people to reinterview for their jobs, except instead of doing blind code tests he's asking them to show off the best work they've done recently.

So, for information: as the program was asked to do more and interface with more (and since the output formatting keep changing every semester), most of the code logic was deported in the configuration files.

Nothing really had any outsized impact in the last 12 months of the project imho. Once the minimum viable product was done, every bit of improvement i did was either simplifying, performance-related or adding more configurability (and test, and documentation).

I did a piece of code that adapted to the ES request that i really like and could be showed off, but it took me three weeks, but 90% of what i parametrized could be hardcoded and it would have changed nothing in the long run.

No, it’s not. In fact, if the fanciness of your code is the only thing that contributes its salience, that’s a negative.

10 lines of salient code could be the entry point into some subsystem you wrote, or where you integrate with another system. It wouldn’t look fancy but it’s salient.

I can see this making sense for type signatures and data structures—or class definitions, interface definitions. Not so sure about "entry points". Contracts, abstractions. Typical Haskell code would really lend itself to being summarized like this—typeclass instances, etc.

The linked article quotes a mail where Musk asks to see "the most salient lines of code".

Unfortunately, "salient" does not mean anything similar to "fancy", which invalidates the original grandparent's statement (unless you can cite something else).

As I didn't knew the word, I gave it to DeepL for translation and in German it translates to words which would be used for extraordinary and over-the-top things (herausragend, hervorragend, hervorstechend). If someone would have used these, I'd have understood it as a call to show my best code.

I think the point is more that most of the hard work behind Twitter isn't in writing the "fancy" code at all - it's in developing the infrastructure that allows that code to safely propagate across Twitter's massive network of servers, and abstracting that complexity away so that the developers who make the "fancy" user-visible changes don't have to worry about it.

> most code is bland and boring plumbing work anyways.

So a decent amount of devs get a nice six-figure salary (RSU, equity), pretty good benefits, kush perks, and more to mess with a template? Órale...

Reply on news.ycombinator.com