[KaiserPro]

> You cannot categorically claim that LC tests are largely memorization tests rather than raw problem-solving skills

If that were the case then a normal, well accomplished software engineer shouldn't need to "grind" leetcode to pass an interview.

its just a cargo cult. Getting someone to do a code review is a much much better test of skill:

Do they ask questions?

are they kind in their assertions?

At what point do they go "I don't know"?

Do they concentrate on style or substance?

do they raise an issue with a lack of comments?

do they ask why the description of the PR is so vague?

When they get a push back, are they aggressive?

All of those are much better tests than "rewrite a thing that a library would do for you already"

[hackinthebochs]

>If that were the case then a normal, well accomplished software engineer shouldn't need to "grind" leetcode to pass an interview.

No, what this shows is that the skill range for accomplished professional software devs is absolutely massive. What these companies want is to find the tail end of this very wide distribution. Leetcode interviews do a decent job at this. If you have been coding for a decade and can't do leetcode mediums with almost no prep, and hards with a moderate refresher on data structures, then you're simply not the in the right tail of the skill distribution and they don't want you. This is what so many in our industry can't accept: you're just not talented enough to earn a FAANG job.

[DragonStrength]

No, these engineers at FAANG companies could not solve those problems cold without having been taught how. I have worked at two, which is how I know. I have never seen a question in an interview I haven’t seen before. Many of these questions went unsolved for decades in the industry, so no, these engineers, who mostly aren’t DSA experts but distributed computing experts, could not solve them cold. I also saw how interviewers used questions to re-enforce their own biases on university, gender, or home country in these interviews.

[throwaway7ahgb]

FWIW I sort of agree with you.

Background:

I'm in a FAANG type company now, a YC company, 3000+ engineers. I'm a Staff SWE with 20+ years of experience (ECE degree) and make $600k+ per year. I've went through the promo cycle here (it sucks).

I can't do any leet hards and can do leet mediums after studying. Some easy's take me a couple of tries. I usually do very poorly in interview coding exercises.

** throwaway , main account is 10+ years old.

[angoragoats]

> FWIW I sort of agree with you.

> [describes self, story completely contradicts parent post]

I'd say you're a perfect example of why "Leetcode problems are a good filter for getting the top 0.X% of talent" is false.

[6510]

Curious, would you say FAANG offers the right challenges to stay in the 0.1% or 1% if one started out there? Are they actually in the right place to grow?

[KaiserPro]

depends. If your in at the bottom floor, then you will grow with the company.

If you are lucky and join/lead/get a new project off the ground, then you'll also grow.

If you're just trying to move the dial on a normal project, then its very difficult to make any kind of headway. You are surrounded by overly enthusiastic hall monitor types who will put in more hours than you, or post more than you.

[KaiserPro]

> Leetcode interviews do a decent job at this.

I mean it doesn't, because I'm at a FAANG, At a FAANG you are infantalised from the very start, sure you passed a very difficult interview where you have to balance a binary tree efficiently as possible. But you're going to use none of those skills here.

what you actually end up doing is copy/pasting some random code you found using internal code search, because the sensible way of doing it can't happen as that would involve porting a thirdparty library, and doing all the procedural work that follows.

so you hack some shit together, ship out out and hope that it doesn't break. You then decommission the nasty hack you shipped last year and claim credit for innovating. Is your product not hitting the right metrics? loosing users? doesn't matter, so long as the super boss is happy that you've hammered in the REST API for the stupid AI interface, you're not going to get fired.

In a startup/small company, if you fuck up, the whole place is going under. Need metrics? you'll need to find a small, cheap and effective system.

Here, we just record then entire world and then throw hundreds of thousands of machines at it to make a vaguly SQL interface. Don't worry about normalising your data, or packing it efficiently just make a 72 column table, and fill it full of random JSON shit. Need to alter a metrics? just add a new column.

In short, don't praise or assume that FAANGs are any good at anything other than making money. They are effectively a high budget marvel movie, Sure they have a big set, but most of it is held together with tape and labour. Look round the side and you'll see its all wood, glue and gaffer tape.

[hackinthebochs]

>But you're going to use none of those skills here.

FAANGs want the top .1% of developers, they don't necessarily need them for most roles. But the point is to hire developers that you could put into any role in the company within reason and have them be successful. 99% of development work at a FAANG is pretty unexceptional and doesn't require exceptional developers. They hire for that exceptional 1%.

[KaiserPro]

> FAANGs want the top .1% of developers,

FAANGS want a load of loyal, naive people who are willing to work loads of overtime and not ask too many questions. Who better than posh kids from great universities who haven't quite figured out that life isn't a meritocracy yet!?

Sure they also want the top 0.1%, but they have a different interview track. Do you think all those OpenAI engineers that were going to follow Altman were asked to do leetcode?

[hackinthebochs]

> Do you think all those OpenAI engineers that were going to follow Altman were asked to do leetcode?

I don't know about OpenAI specifically, but I have heard of interviews at other top ML research positions were partly based on leetcode problems.

[emodendroket]

Designing a test for which one cannot possibly prepare is a problem that’s bedeviled test makers for a long time. The team behind the SAT threw up their hands and said it no longer stood for “scholastic aptitude test” but just… nothing.