Hacker News Reader

Comment by ones_and_zeros | Hacker News Reader

Palantir gets some of my sympathy. My organization has a large internship program and I've been first contact for many applicants. I found that a similar number of applicants had to be dropped either due to communication skills or an overall technical skills gap and they tended to be asian/southeast asian.

Interestingly most of them fit the same profile of a bachelor's from India, 1 or 2 years at a consulting firm in India doing something that could be argued is software development and now doing a master's at a large state university. To be honest I'm not sure how they got in or are able to graduate.

View on HN

I worked at Palantir in the covered period and did a ton of interviews, 5-10 a week. I experienced exactly this dynamic - very large numbers of applicants who had a bachelor's degree from a South or East Asian university, a year or two of work experience, and a Masters from a mid-tier US University. The large majority washed out on the FizzBuzz-level technical questions on phone screens (or asked in person at job fairs).

I've seen American university CS grads wash out on FizzBuzz level questions at alarming levels as well so this is not exclusively a phenomenon with these backgrounds.

How in the world are people making it through schools (with a MS!) without actually.... uhh.... you know.... knowing the subject at all?

"How in the world are people making it through schools (with a MS!) without actually.... uhh.... you know.... knowing the subject at all?"

I'm sure that there are many more people who could complete FizzBuzz on a homework assignment, with the help of the Googles, than could do so in an interview. And there are many mid-tier MS programs that are happy to take money from international candidates and give them every opportunity to keep paying.

> I've seen American university CS grads wash out on FizzBuzz level questions at alarming levels as well so this is not exclusively a phenomenon with these backgrounds.

Absolutely. They may even have washed out at higher numbers. The applicant pool described in the complaint definitely isn't representative of US CS majors nationwide, so you can't use it to draw inferences about the quality of students overall.

I'm rather shocked by this as well. I understand not knowing the latest frameworks and such, but I have close personal friends that say my name should be in parenthesis next to theirs on their diplomas. A lot of CS courses can be passed by getting help on take home assignments and rote memorization on tests. At least for your typical state schools.

> How in the world are people making it through schools (with a MS!) without actually.... uhh.... you know.... knowing the subject at all?

Recent undergraduate grad here. A lot of the algorithms I've seen on technical interviews were only covered once or twice in the curriculum. Basically, one such algorithm would come up in either the data structures or algorithms class, we would then work on it for the given assignment/exam, and finally we would move on from it. Subsequent course work would barely reference it.

These things are best learned through repetition. Unfortunately, with all of the other CS coursework topics that I was being educated on, it seemed like there was little room for the repetition to take place. (At least, not in the stereotypical four year timespan assuming that the given student enters a CS program with no prior programming knowledge.)

I would expect that the project they work on to utilize the knowledge in the courses. With 4 projects a year at 4 years, there's got to be something learned!

I got a BS in CS, graduated in 2006. What 4 projects? We had one project that lasted an entire semester, in one class; the rest was mostly one-off homework assignments that were very much topic-of-the-week. And most of them didn't really require coding.

If this is really as common as you imply, then no wonder many grads are lacking in exp. I thought 4 projects was on the low end...

If you know even some basic programming and logic, you should be able to come up with a reasonable fizz buzz solution. You don't need to have studies the fizz buzz algorithm in college, or have practiced a solution.

I spent weeks on that algorithm. It is only through daily katas that I still remember its rigors.

> ..assuming that the given student enters a CS program with no prior programming knowledge

This is bizarre to me. It's like someone signing up for a music degree having never played an instrument. Maybe you'll cram enough to pass a test on the theory, but it's useless knowledge. It's unanchored and will float away.

Or a medical student who never did an operation before? Arts are the exception, not the rule. In other areas it is fully understood that the whole purpose of the program is to teach you everything, so actually saying "you have to know this before" would be weird.

Medical students are expected to have a strong science background, and well before they commit to being a surgeon they'll have units where they need to wield a scalpel. If they find out it's not for them there are plenty of other paths which don't involve surgery.

Engineering students are expected to have a strong maths background. If you can't do algebra and basic calculus then there's no chance of getting in.

It's completely reasonable to expect a CS student to have experience programming.

> It's completely reasonable to expect a CS student to have experience programming.

If you say so. I neither see the reason nor the advantage. Don't get me wrong: I had experience in programming before entering a CS program, but I had fellow students who never did any programming before and they did just fine, while others with programming experience failed. And I'm pretty confident that all of them who managed to stay through the introductory year were/are able to do FizzBuzz.

Computer science programs, even at very prestigious schools, don't really require you to know anything about programming or practicality using computers. The amount of real hands-on work I did for classes in my entire undergrad career is less than I do in a good week at work.

The question is if the questions asked are really "FizzBuzz level" questions, or algorithm questions that are easy to consider of similar complexity when you know the expected algorithms.

When I studied CS, data structures and algorithms was the second CS course we were expected to take, right after "introduction to programming". It would usually be taken in the second semester.

That's the last time we were tested on a long range of algorithms, except indirectly if we happened to need a given algorithm to complete our work later. In the 22 years since, I've probably needed e.g. a depth first search once, and many of the algorithms I learned I've never touched on since.

So for many, it boils down to whether or not you cared about and liked that course. I happened to love it - I read the set texts the first weekend of the course, and spent a lot of time reading up on algorithms in the years before and after. But for a lot of my class mates it was a course they had to endure to get to more interesting stuff. Many of them would go on to do very well in other subjects.

If they can't figure out a problem like FizzBuzz given a complete description, then that's a deeper problem. But a lot of algorithm questions can sound like they are on that level because we expect people to know a catalogue of algorithms that most developers don't actually need through most of their careers.

^^ - this.

same here - almost 20 years of dev, last time I needed to implement basic cs algorithms were in college and grad school classes (and some idiotic corporate interviews where preferred way of coding is apparently using a white board).

basic knowledge of fizbuzz, dfs, bfs, quicksort, and similar things of course is needed - however asking developers actually implement them in front of you is pointless, at least as far as job performance goes.

Why do you bundle fizbuzz with some legit algorithms that require some knowledge? Fizzbuzz is a gimmick, deliberately picked as something very simple to filter out clowns. I can't imagine a more simple programming task. Any suggestions? ;)

Simple variable swap?

"Take these two variables, derpyOne and derpyTwo, and swap their values"

Write "Hello World" in your favorite language...

> basic knowledge of fizbuzz, dfs, bfs, quicksort, and similar things of course is needed

FizzBuzz is not some arcane algorithm. You don't need to study it.

It's literally writing a loop, testing equality, and printing some output. I don't know how you could program professionally and not be able to solve FizzBuzz.

People should be able to solve FizzBuzz provided it is explained properly and fully.

This is the same for a lot of the basic algorithms we learnt and forgot.

This, to me is the big caveat. Just as there are a lot of people who are horrible at solving stuff they should easily be able to, I've conversely seen so many horrible interviewers that under-specify the problems they ask about because they assume people should understand algorithms that would make it easier to understand, or just assume people will infer things most people wouldn't.

A lot of people are pushing back on you, but I've seen the exact same thing. I think everyone who hasn't interviewed developers would be surprised by the general incompetence, but everyone who runs an interview process realizes that 80+% of "developers" are terrible.

For the record, I'm absolutely not talking about algorithms questions. I'm talking about literal FizzBuzz. A majority of candidates cannot solve it in 15 minutes.

> How in the world are people making it through schools (with a MS!) without actually.... uhh.... you know.... knowing the subject at all?

It's through a combination of:

1. Tests being based on memorization, not actual knowledge or programming. If you manage to stuff enough keywords into answers, it's possible to pass.

2. Lots of group work. The 20% of students who actually know how to program finish the project while everyone else slacks off and asks for help "with this one thing."

3. Outright cheating. There's a thriving market in recycled homework and exam answers.

> everyone who runs an interview process realizes that 80+% of "developers" are terrible.

No, 80% of people in the 'currently looking for a job developer pool' are terrible. But you have to consider that the worse a candidate is, the longer they stay in that pool. So judging developers as a whole based just on interviews makes them look much worse than they are.

Good point, that's always an important thing to remember. The average job applicant is much lower quality than the average developer.

Maybe. With a large enough population for either the job-hunters or the employed, you should regress to the mean for the total population and the deviation should also be similar. This assumes a normal distribution for the metrics that you are hired for, something that may not be in fact true, ie. Poisson or exponential distributions are reasonable. What you can take away from this: Your own developers may not be as good as you think they are OR they may be much better, same for the candidates. That particular day you interviewed them they may have been exceptional or they may have been ill and they failed, or a million other reasons. I think Google came out a while ago and said that their initial hires had no correlation to the really successful people in the company after...5(?)... years. I can't find the citation, sorry. The interviewing process is known to be broken, so trying to extrapolate to say that the job-seekers are less good at these metrics than the employed is erroneous. I'm not saying you are wrong, just that using a binary decision like this to determine these things is wrong.

IIRC Google said that their interview scores didn't correlate with how successful people were in the company. But you have to remember that obviously they're only talking about people who made the cut, so people who had at least moderately high scores to begin with.

> With a large enough population for either the job-hunters or the employed, you should regress to the mean for the total population and the deviation should also be similar.

You're right that the actual population of all job applicants over time is equally qualified to the employed developers.

However, the data that I have available to me as a hiring manager is the sample of developers who actually applied to my job. In that sample, incompetent developers are much more likely to be present because a competent developer only sends out <15 applications before getting a job while the incompetent ones can send out hundreds.

At any given moment, a competent developer is more likely to be employed than an incompetent one. Hence the incompetent ones are constantly generating job applications at a much higher rate than their actual presence in the population would suggest.

For the record, this isn't merely theoretical. The last time I was looking for a job, I was able to get an offer from the second company I applied to. Meanwhile a friend of mine with barely any experience had to apply to about 50 places before getting a single offer. If you were to pull a random sample of our applications, his would be way over-represented.

> think Google came out a while ago and said that their initial hires had no correlation to the really successful people in the company after...5(?)... years.

Sure, but Google still spends a lot of time and money on interviewing because anyone they hire has to be better than the average applicant. They get huge numbers of job applicants, the majority of whom are grossly unqualified.

> The interviewing process is known to be broken, so trying to extrapolate to say that the job-seekers are less good at these metrics than the employed is erroneous.

There are lots of problems with the interview process, but I'm really including anyone who managed to get to an interview (past the phone screen) in the "competent" category. There are seriously huge number of job applicants who can't program a basic loop without extensive help.

> However, the data that I have available to me as a hiring manager is the sample of developers who actually applied to my job.

Actually, you only have the sample of developers you interviewed for the job. How many didn't get an interview? Are you confident that your resume filter is good and that the resumes rejected were actually bad?

> Meanwhile a friend of mine with barely any experience had to apply to about 50 places before getting a single offer.

Following up my previous point, how many interviews did that yield? The discussion here is over developers who look like idiots in the interview. Confusing that with applications clouds the discussion.

> Actually, you only have the sample of developers you interviewed for the job. How many didn't get an interview? Are you confident that your resume filter is good and that the resumes rejected were actually bad?

More specifically, I have the sample of developers who I did a phone screen with.

I'm reasonably confident that my resume filter is "good" in the sense that it's not rejecting people prematurely. If your resume shows any hint that you'd be able to do the job (ie. any amount of programming job experience and/or a CS degree) then I'm willing to give you 15 minutes.

> Following up my previous point, how many interviews did that yield?

A dozen or so. I'm deeply skeptical that the pool of applicants would be better than the pool of interviewees.

Does Palantir make US citizenship a requirement for most of its job positions? My understanding is that this is common place for most US military contractors but I was wondering if this applies to government contractors at large.

Not for the positions I was interviewing for.

US citizenship is a requirement if you are going to do something related to the US government. I've worked for them in a position not related to the US government(not a US citizen myself) and the number of technical people who also happened to be US citizens were quite low.

If that is the case, and the records show that, then the company is probably ok. But you have to show that all of the people you hired could pass FizzBuzz right? Google gave one of the best hiring training lectures I've ever attended on eliminating personal bias from interviewing questions. A lot of it was thinking about the question and the response and putting it in the context of someone you knew, what would you think then? Would you still consider it so highly? Or as such a deficit? They said, although I didn't experience it directly, that they did data analytics on interviewer write ups to identify bias. That could have been to sort of encourage due diligence but given how data driven that company was when I was there I would not put it past them to do exactly that.

How does personal bias factor in to FizzBuzz? It's literally what I'm hiring you for -- I should be able to describe some process to you, and you should be able to write it out in code. Communication is a huge part of software engineering, and I don't understand why anyone would be expected to hire employees they cannot communicate with.

Personal bias potentially factors in the moment your evaluation of the solution goes beyond "works" or "doesn't work" and into subjective evaluation of the code quality.

Personal bias also potentially factors into how you describe it (so consider presenting it written, in the same format for everyone). E.g. let's say the interviewer goes into extra detail or ask for confirmation of understanding from candidates they favour.

In addition to vidarh's excellent response, FizzBuzz is a great example that personal bias should not interfere. So if you did data analysis of your candidates and everyone you hired could do FizzBuzz and everyone you didn't hire could not do FizzBuzz then you could objectively state that failure to do FizzBuzz disqualified the candidates and it was blind to personal bias or discrimination factors. However if you hire whites who couldn't do FizzBuzz in the interview and you didn't hire asians who could, then you can't really say that "not being able to do fizzbuzz is a disqualifier" because it only disqualified some candidates.

Generally if you are hiring on work product skills (coding for software folks) and can show that your test is unbiased and you hire to the test, then you can avoid being found guilty of discrimination even if you hire a disproportionate number of a certain group. To be rock solid you also need to show that your applicant pool is highly diversified and has representation from all groups. Sometimes that means going out and seeking out candidates (like women or hispanics) who are self selecting to not apply. And why you really do better if you blind the resumes your HR staff sees before they sort them into phone screen and skip piles.

> However if you hire whites who couldn't do FizzBuzz in the interview and you didn't hire asians who could, then you can't really say that "not being able to do fizzbuzz is a disqualifier" because it only disqualified some candidates.

This is a great point and it's why I always include FizzBuzz in any interview—no matter how they were referred, how impressive their resume is, or their confidence level.

Having consistent benchmarks helps give you legal cover, but it's also essential to running a solid interview process. If you're not applying the same tests to all candidates, it's impossible to detect problems in your filters and you can easily end up making biased decisions or hiring the wrong people.

Yes, I understand that it can be annoying to have to solve such a basic problem in an interview, but it only takes a few minutes—just get it over with, it's not meant as an insult.

Too bad it didn't stop the bias in my Google interview :/

Passed through all the on-site interview problems and then they silently dropped me. No "we're sorry, you're not a good match." No emails, no phone calls. Nothing.

What bias does this show?

Both sides of this issue are wrong. On the one hand, If I knew, based on online articles about the interview process, that there is a set of 100 questions out of which I might get asked 20, I'd prep on those 100 questions. It seems like the interviewees are underpreped for interviewing. That's sloppy. Maybe that's a reason not to hire them, but it probably isn't the best way to select hires.

On the other hand this process makes about as much sense as requiring that I know how to shift a non-synchronized manual transmission in order to pass a driving test. Nobody is implementing most of these algorithms from scratch, and if they did it would be a bad practice without having a good reason to do it.

You'd have a good point in your second paragraph if we were talking about implementing a Fibonacci Heap or an AVL tree. But the comment you're responding to is about people failing FizzBuzz.

Good point, FizzBuzz should take a couple minutes. AND it's the sort of thing a prepared interviewee should nail. I'd hope the rest of the interview would focus more on real-world considerations.

Im guessing they assumed the CS degree was a magic job getter and were too cocky to study

We introduced homework to filter out candidates who could not complete FizzBuzz level tasks. 70% of applications can be filtered by simple home work.

I have had a similar problem - interviewed dozens of people with exactly that pattern and they were universally clueless. No idea how seemingly-good Masters programs bring in so many people with zero basic CS knowledge.

My alma mater has a great undergrad program, but an absolutely awful masters. I've been told that they don't have the resources to teach it, and that the college keeps sending recruiters throughout Asia to pick up as many foreign students as possible for the extra tuition money, though that might have just been a bitter professor.

Sounds like CSU Chico.

Masters programs are run as a cash cows for many universities to be filled will anyone who can fog a mirror. As a former academic I can tell you that there are literally different marking criteria for overseas masters students than local undergraduates.

Is that knowledge transparently available to students? How does it impact the curve?

I imagine quite many would be upset if they found out they were not being held to equal standards, and would probably point to foreign students paying more as the reason.

>Is that knowledge transparently available to students? How does it impact the curve?

Of course not. You find out when the head of school tells you that it is your fault that all the masters students failed the subject and that you need to fix the situation. If you are dumb enough to ignore the hint the head of school gets someone else to re-mark the exams - either way the cash keeps flowing.

That is fascinating.

Anecdotally, in European universities, bachelors are the "get in as many people as we can for the tuition fees", while masters are the "let's be as selective as we can get away with".

Sometimes it's even worse... For example, in France, a university is required by law not to be selective for it's first year programs, as long as the candidate graduated high school.

So...what is that, first come first served basis?

I assume there's an upper limit on the number of students an university can take in for the first year, determined by staff and classroom availability.

Yep, that's how I remember it in Australia. The honours students had more respect and were more skilled than the masters students.

Myguess is he's not pocketing the asian tuition money.

As someone who might apply once I get my PhD, what does basic CS knowledge mean for you? Like what a tree is or fizzbuzz? Hopefully it isn't that terrible...how someone can make it through any amount of CS related schooling and not know the basics?

Somewhat related anecdote, I do computational physics. I once attended "training seminar[0]" put on by the funders for using DOD supercomputers, and there was this soon-to-be-minted computer science PhD student who was also an intern. Most of the systems have pbs[1], a batch system that prioritizes jobs from users based by amount of requested time, number of nodes, etc. The seminar leader (and I have before) referred to it loosely as a queue, because it sort of is, and the CS guy kept interrupting the talk (may be two or so times) to point out that PBS really wasn't a "queue" because it didn't respect FIFO in a strict sense... I mean, sure, he did know what a queue was, but I'm pretty sure that isn't the most interesting problem you could have.

[0] there was one fact I found useful, which was where to find their docs online. Other than that, it was the sort of stuff you can find by just reading man pages and such. Waste of time but checked off a box for the internship which made the funders happy, I suppose.

[1] https://en.wikipedia.org/wiki/Portable_Batch_System

Weird. No one else would want a scheduler to be FIFO, either. Rather, the actual optimization that is performed by slurm, pbs, etc. under the hood is pretty interesting.

The point of schedulers is much more complicated than a straightforward queue, as they seek to project how best to utilize system resources into the future.

Strange he would be so pedantic. Everyone in HPC certainly does call it a queue, as you noted.

Yeah, because it is a queue in the ordinary English sense.

Common problem with programmers: when they learn a term means something, they associate that term absolutely with that meaning (common in other fields with people who are highly analytical, IME).

The HPC people are referring to a "job queue", rather than a "ADT queue". Those are distinct things, the ADT queue is a much more restrictive defintion, but it's not an exclusive one. HPC queues are more like priority queues with complex ranking functions (and queues of queues), just like real-life queues where people can cut by paying more money (first class airline ticket).

Anecdotally, I've learned to expect nothing. I used to ask a graph-based question -- I had to adjust it so that the first step was doing a dead-simple DFS. This tended to weed out a good sixth of candidates and save us both a lot of time. And this was pretty constant against education level and background.

DFS == Depth First Search?

Yeah, given an acyclic graph print out each node's value. Passing candidates generally got it in five minutes.

was writing DFS from scratch ever needed to perform dev job or was it a thing that stumped 95% of applicants (and let you read some hacker news/email on your phone/laptop in the mean time)?

It only stumped the candidates that clearly wouldn't make it through the rest of the problem. One sixth was a probably a high estimate.

DFS for an acyclic graph is about five lines of code and shows me that you A. know a programming language B. can figure out a data structure. I would argue both are needed to perform a dev job.

Eh, I had to write a DFS the other day at work.

I've written DFS about 1-2 times a year, on average, for the last 15 years. Why do I need to? Usually, because I need to modify or create a data structure that is kind of graph-like but not quite. If someone didn't know how to write one or at least derive it, they would confuse a lot of people, so yeah it's kind of important.

No, they'd probably look it up, skim a blog for a few minutes, then figure it out and promptly forget. This is no different than forgetting the order of the arguments to socket() and accusing that person of being incapable of writing networking code.

The difference is that the arguments to socket() were arbitrarily defined, but DFS is an algorithm. Anyone who has taken a intro data structures course should be able to at least derive the pseudo code for DFS.

If you ignore the terminology, DFS is basically just recursively walking a graph, something which actually happens all the time in most dev jobs. I'm sure I wrote a DFS long before I had heard the term.

I hate brain teaser interviews as much as the next guy, but being able to do basic operations on basic data structures is pretty much the minimum to be a developer.

> If you ignore the terminology, DFS is basically just recursively walking a graph, something which actually happens all the time in most dev jobs.

In ten years of full-time professional development I have never needed to recursively walk a graph. Hell, the only data structures in the software I worked on for the first six (real-time signal processing code) didn't use any data structures other than lists, arrays, and C structs. I would not be able to answer the interview question that started this subthread, despite my track record of efficient and well-designed code.

Are you sure? Keep in mind that a graph doesn't always come up and slap you saying that it's a graph.

This very comment thread is a graph which is being printed depth-first.

I'm very sure. The signal processing code had no graph structures whatsoever.

That's certainly not the case with the NLP code I have been working on for the past 4.5, but I have never personally needed to search the (usually) trees in a structured manner. Most of our code that searches for things doesn't care about the structure of the graph during the search; it just searches a node list. Once the node is found, structure matters. There are some algorithms that use calculations like shortest path and tree height, but I didn't write the code to actually do that.

Do you think you've ever had to do something as algorithmically complex as recursively walk a graph?

Sometimes these questions are used to probe the meta understanding of a candidate and using Graphs/Trees/DFS are universally understood concepts that make doing that easy.

Define "algorithm". In the broader definition, yes. Kalman filtering, motion compensation, track association, XY-mapping, and similar signal processing algorithms. For NLP it has been various forms of statistical modeling and general multithreaded architecture and coordination.

As for the algorithms typical of computer science, not really. Searching and sorting has not been much a part of what I have done. My experience is mainly in translating math into code, often into code with real-time deadlines.

Sometimes I get these questions in interviews and I really don't know if I should be offended. My resume is pretty good, and I wonder who T.F. applies to these places that they have to have such simple weeder questions. Then I get all paranoid that there is some hidden complication and soon I'm producing FizzBuzz in TensorFlow:

http://joelgrus.com/2016/05/23/fizz-buzz-in-tensorflow/

I think you shouldn't be offended at interview questions around skill unless you've been on the other side of the table. It's eye-opening, the range of people that apply for these positions.

Do you give them a graph data structure or something that is explicitly a tree?

A acyclic graph is isomorphic to a tree, but the way you access the vertices and edges is completely different.

Depends on your memory representation of the graph. If you simply use the obvious one (a node have a list of pointers to its successors), I don't see the difference except for successors[]/children[]

You're right. I mentioned acyclic graph and my mind slipped to general graphs. Although often graphs are represented as say a sparse matrix.

Should every node's value be printed out exactly once and if so, are there nodes which have two parents?

If so, I would keep the pointers in a set to mark nodes as processed but is there a more elegant solution?

Each node is an object and has a "checked" variable.

I would explicitly tell them that you should revisit nodes, and give them a few examples illustrating that point -- so all they needed to do was traverse edges.

Well, for example, I have an interview question where it's not just about knowing what a stack or a tree is, but rather realizing that a stack or a tree might be useful to solve a particular problem. (You could use either one.) And then writing out code to do it.

So, no esoteric data structures, but you really do have to know how to use the standard ones. It tends to trip up people who might have some book knowledge of CS but don't know how to apply what they learned. There's also a difference between knowing that a tree might be useful and being able to actually write some code to build a tree.

But the thing is, the next interviewer will have different preferences - there's no standardization here.

It could also be that you only ever have to do this between 0 and 0.00001% of the time in the real world, and many people who could do it just fine if they are given space to think about it or refresh their memory, which usually results in freezing up during a stressful interview. They probably haven't exercised that knowledge because they have been too busy building real things.

That seems like a lousy excuse to me. As a web developer, my entire job revolves around computing trees; and that is (as of now) the most common development environment in the world. My side projects (graphics demos/toys, a multiprocess text editor, various twilio SMS utilities, a high-performance pastebin server) end up using stacks, trees, hashmaps, and bloom hashes as logical responses to the problem descriptions.

Only somebody who doesn't write software for a living would claim that "between 0 and 0.00001% of the time in the real world" should these techniques be readily accessible in working memory.

As for "freezing up during a stressful interview": is that not some indication of what normal workplace stresses will do to your ability to complete technical work? I certainly wouldn't want to work with somebody who is knowledgeable but will wimp out the moment you ask them a tough question on the spot.

> As for "freezing up during a stressful interview": is that not some indication of what normal workplace stresses will do to your ability to complete technical work? I certainly wouldn't want to work with somebody who is knowledgeable but will wimp out the moment you ask them a tough question on the spot.

I would argue that a workplace that is continuously stressful is not healthy, and optimising your interview process to filter for this is probably not a great idea.

Additionally, I would expect that when asked a "tough question" at work that you would have both the time and resources available to find or determine the answer, things that one likely doesn't have in an interview (and certainly not to the same extent).

Not sure what do you mean by computing trees in this context, but I personally as a web dev for 15+ years have never-ever had to e.g balance a binary tree or do any other low level operation of that kind. It's important to learn it at some point to be able to understand how it all works, but afterwords you just don't write the low level stuff yourself, ever. If given enough time I'm pretty sure I could come up with some solution for reversing a linked-list, probably even the optimal one, the same way I could probably also figure out how to fix my car when broken... but that doesn't make me a mechanic

The solution to reversing a linked list should be self-evident to anyone who makes an attempt.

    #include <stdio.h>
    
    typedef struct Node Node;
    struct Node {
      int data;
      Node * next;
    };
        
    Node * reverse_linked_list (Node * node) {
      Node * prev = NULL;
      Node * next;
      do {
        next = node->next;  
        node->next = prev;   
        prev = node;
      } while ((node = next));
      return prev;
    }
    
    int main () {
      Node * node;
      Node nodes[3];
      nodes[0].data = 0;
      nodes[0].next = &nodes[1];
      nodes[1].data = 1;
      nodes[1].next = &nodes[2];
      nodes[2].data = 2;
      nodes[2].next = NULL;
    
      node = nodes;
      do {
        printf("%d", node->data);
      } while ((node = node->next));
    
      node = reverse_linked_list(nodes);
      do {
        printf("%d", node->data);
      } while ((node = node->next));
    
      return 0;
    }

Or something thereabouts.

I've never had to balance a tree either. But typically it's because the tree represents something essentially tree-shaped and preserving the structure is important. (For example, what would it mean to balance a DOM tree? Nobody would do that.)

On the other hand, I hope you're not saying that the DOM is the only tree you ever use and in 15 years you've never needed to build your own. If so, consider that companies might be interviewing people for programming jobs that aren't like that.

> As for "freezing up during a stressful interview": is that not some indication of what normal workplace stresses will do to your ability to complete technical work?

To add to roganartu's very good point about what filtering for that says about your company, you seem to be buying into the pervasive myth that "stress" is some universal and fungible thing. The stress of an interview is not of the same nature as the stress of, say, having a key server go down and take your company down with it. It's a fool's errand to think you can manufacture stressful situations in an interview and get a read on how the candidate would perform in a real situation.

Are you not aware that most languages provide that for free? Note that the question was about implementing, not using. People use them all the time, but they don't reinvent the wheel to do so when it is in the standard library and has nice abstractions already.

Why is a web developer implementing a DFS by hand every day? I would not hire that person.

I'm not sure what you mean. I don't know any languages where you don't have to define your own node if you want to implement a new kind of tree, and write a bit of code to recurse over it. (In JavaScript the most common tree is the DOM, but you should still be able to build another one if you need it.)

Also, in an interview it's perfectly fine to say you'd use a library, if you know one that applies. (They might ask you next how you'd do it without a library, but you still get points for knowing about it.)

The other thing to remember is that everyone is effectively grading on a curve, and with practice you get better at interviews.

How would you filter those "many people who could do it just fine if they are given space to think about it or refresh their memory" down to the number you need to hire?

This is a serious question. I hate the way interviewing works at tech companies. Though I've generally done well and it helped me break into the industry as an outsider, it's a really time-consuming and wasteful process on both sides. So far, I'm leaning strongly towards Reid Hoffman's suggestion of weighting references much more strongly than interviews but it's hard to say how far one can safely go with that approach.

On one hand -- sure, I've sadly forgotten most of the math I ever learned.

But on the other hand, no. I can bounce around modelling with basic data structures (hashes, arrays, structs/objects etc) in my head like a good juggler can treat 3 balls.

I think that is normal for quite a large subset of developer work. I assume there will be some jobs that won't need those data structures, but can't think of any. (Graphics and game developers would remember more math, of course.)

The description in the GP sounds more like the (probably false) stories about the Japanese English education: People with a vocabulary of 40K words that are unable to order coffee or talk about the weather.

I have had many people with 3-5 years of 'experience' developing not know how to find the smallest element in an array. Initially I was thinking it would be a quick check to see if people thought about null values, but apparently this is a hard problem...

Now they might have been useful for something, but I suspect they are just playing the numbers game and showing up to a lot of interviews until they get in. So even if say 5% of coders are useless that 5% is vastly over inflated in the interview process.

I think it is interview paralysis. They can't believe an interview answer is ever supposed to be O(n). I've had it happen to me where I get tripped up on relatively easy problems because the initial easy solution I came up with wasn't sufficiently clever and magically log n or some such thing.

I can't believe it's just interview paralysis to be honest. I've had people try and solve problems like that with 3 nested loops. Completely failing to come up with a working solution after 20 minutes.

I've been in situations where those kind of people have been hired too. Their code tends to be quite bad...

Any suggestions for dealing with this? I try and keep things as relaxed as possible to start with and make it clear that trivia and syntax is not what I am going after.

I would walk out of the room, but I am happy to walk someone though most of a problem they are not getting. The goal is going from an understood problem to actual code.

In addition to the obvious have the person walk you through what they are going to do try rewording it slightly. In this case ask for the max value.

Maybe they are nervous and focused too much on the minimum part of this that they lose sight of what they are trying to solve for. But be careful not to spoon feed it to them.

In the last two weeks I interviewed three candidates with masters of science C.S. None of the three knew the memory size in bytes of a double precision floating point variable. None of the three could explain scenarios in which they would use an array over a linked list.

I don't think that first question proves what you think it proves.

Learning implementation details about a specific programming language is not what CS is supposed to be about (see the quote about astronomers and telescopes). In fact, if you use several programming languages at the time, I'd go as far as to call it a waste of my time - if I'm worried about overflow for a specific problem, I can get the range of data (in which programming language? which architecture?) in 30 seconds online.

I can see it being a valid question if you are working in a problem down to the bit level. But those are getting rarer, and ultimately it's what an Engineer excels at.

The second question does seem fair, though.

sizeof(double) is not specific to any one programming language or architecture; it's determined by IEEE 754, an international, cross-language, cross-platform standard. Non-IEEE floating-point implementations are exceedingly rare nowadays (although they do exist).

You're right about it being engineering, not computer science, but IMO it's reasonable to expect someone applying to a programming job (assuming this is what it was) to have at least some software engineering experience.

Well..

  C and C++ offer a wide variety of arithmetic types.
  Double precision is not required by the standards
  (except by the optional annex F of C99, covering
  IEEE 754 arithmetic), but on most systems, the
  double type corresponds to double precision.
  However, on 32-bit x86 with extended precision
  by default, some compilers may not conform to
  the C standard and/or the arithmetic may suffer
  from double-rounding issues.

ap22213 (OP for this thread) was more precise than I was; they actually wrote "a double precision floating point variable" rather than writing "double", which could either be a shorthand for "double precision floating point" or be referring to a platform/language type (where that exists by that name).

So a reasonable answer could say - usually 8 bytes - but sometimes with extended precision it could be 10 bytes, etc...

I am actually kind of sick of the people that say these details are not important to being a good programmer. They absolutely are, knowing fundamentals of the machine, language, and environment that you are developing in is essential.

The particular positions that I'm hiring for require certain kinds of knowledge, unfortunately. When one is dealing with Petabytes of structured signal data, it's helpful (but not required) that someone know memory layout. But, it's not the first question I ask.

Usually, I start off with some behavioral, get-to-know-them type questions. One of these is to ask them to recall the 'best software developer' that they've ever worked with. The great majority of candidates actually say themselves which is unexpected (and a little arrogant to me - weren't we all 'junior' developers once? - didn't we all once look up to someone?). But, surely one of the best would know when to use an array and when not to or know the memory and precision impacts of the variable types that they use.

The floating point I think comes from C bias.

In C you kind of have to know the size of a double -- you'll call "sizeof(double)" fairly often, you'll notice the number in your code.

In most languages, you don't have a way of knowing the size of a double -- and furthermore the size a double takes in your code won't be sizeof(double) anyway, it will be much more.

> Like what a tree is or fizzbuzz? Hopefully it isn't that terrible...how someone can make it through any amount of CS related schooling and not know the basics?

Ah! You're in for some surprise down the line. Fizzbuzz almost always clear out large amount of applicants.

We had another applicant just last week with a shiny spiral-bound resume that looked good, but couldn't do FizzBuzz.

I was a CS major, but my career took a different direction, and I haven't programmed "professionally" for 8+ years. I looked up the FizzBuzz question and wrote some psuedocode in about 5m.

It's sad that FizzBuzz is such a "weeding out" tool, but I guess good to know that I can still pass the first round of an entry-level technical interview!

FizzBuzz (from the point of view of me as an interviewer) is not so much about knowing how to code, but understanding the thought process about basic problem solving. Read the instruction, provide a solution, don't over engineer. It also provides the basics, loop and conditional.

Once you're set on that, you at least know you have a person in front of you who, on the basic level, thinks like a programmer.

So I would say 8+ years doesn't matter so much and you shouldn't worry about that, as long as your mind can still "tick" the right way.

Easy... money. Graduate classes are required to pull in top-tier researchers and grants, but they have relatively-fixed overhead costs for teaching (eg. professor salary). Want to maximize the return on that fixed overhead? Add more students. Many public and private universities are "Masters factories" for budget reasons. They typically filter (strongly!) between completion of MS and entry into PhD programs.

(Identifying such Universities will be left as an exercise to the reader.)

Foreign students are a cash cow for usa colleges.

If you cant get into the ivy league of your own country, usa schools are a backdoor prestige option for the wealthy.

There's alot of cheating and other shady nonsense.

Bodyshop contractors will do anything. I've seen cases where those companies send shill candidates to interview, claim that candidates have more experience in reality (in one case a 22 year old with a decade of Oracle Financials experience) and submitting fake resumes.

I agree woth this. I and my friends always wondered how so many Asians get accepted to top UK universities (with very high English-language requirements), when they can barely speak a word of intelligible English.

Now I have a close friend from China, and the mistery is solved: others are doing their exams/essays/assignments for them.

Keep in mind though, this is likely a huge generalisation (not all Chinese cheat and other nationalities cheat as well), and I put more than half of the blame on the universities themselves (they earn a lot of money accepting these kinds of students), but long-term, it really dilutes the value of this education in my eyes.

>>this is likely a huge generalisation...

Ya think? Sorry man I wouldn't be profiled as a cheater like that due to race or citizenship.

I have nothing to say about profiling individuals but cheating appears to be a huge issue among international students. Per the WSJ [0]:

    Public universities in the U.S. recorded 5.1 reports of alleged cheating for every 100 international students, versus one report per 100 domestic students

Faculty singled out Chinese students in particular.

[0]: http://www.wsj.com/articles/foreign-students-seen-cheating-m...

I live in China. I have not made any effort to learn about plagiarism or impersonating people at exams in the four years I've been here. I know one Chinese person who sat an English language proficiency exam for her cousin, another whose Bachelor's thesis was almost entirely plagiarised. I met a Canadian guy who did a Master's degree in econometrics at Tsinghua (top 3 university in China). Apparently plagiarism was rampant.

On a more anecdotal level one of my fiancee's friends was dating a guy who sold SAT answers for a while.

I have a couple of friends who teach at international high schools here in Shanghai. It takes quite a long time to get through to the students that plagiarism is not okay and is taken seriously.

By the way no one brought up race in this thread until you did.

That's interesting - I'm surprised to hear so many first hand accounts.

If it is widespread it's especially unfortunate considering there's already some controversy around the high numbers of asian applicants.

Btw, I only mention race because it is a central topic of the article.

I'm sure others were cheating as well, but it wasn't as blatantly obvious as with those Chinese who didn't even speak English. So I'd say being fluent in English is a good way to avoid being profiled this way.

Wow a tad racist eh? How about this, I wonder how whites get in to top schools as they have lower GPAs, lower SAT, lower activities, etc. than their asian counterparts. I put more than half the blame on the universities and their "DIVERSITY" BS.... Long term it really dilutes the value of education in my eyes...

Strong username to post content ratio.

Are you using n-grams or ?

Quality-checking the contractors a consulting agency is feeding you feels like the perfect case for a Starfighter-alike "bar exam"—much moreso than the actual Employee-Employer matchmaking that Starfighter tried.

I worked in India and also went through several CS training programs there.

Indians cheat. A lot. It is just a different culture. Students in three different groups would turn in the exact same solution to a coding assignment, and they'd all get the same grade. I was baffled at first, then furious, and finally just learned to laugh about it, while understanding that they were only screwing themselves by not learning the material.

However, it did mean that the training programs would push out a ton of students that did not know a thing about programming, and definitely should not be in the industry. But it is understandable in a place like India where a software engineer can make a lot more money and earn far more respect than many of the alternatives.

Witnessed it firsthand in grad school on a few occasions. Instructor momentarily leaves during an exam and the Indians start asking one another the answers, thinking nothing of it. WTF! This was not all of them, there were many straight-up smart ones, but cheating wasn't uncommon.

I also saw this a great deal among Indians and citizens of neighboring countries when I was in grad school. It was a huge problem, so much so that the Dean sent out a letter about it. Without specifying nationalities, but everyone knew who was implicated.

I was friends with some of them who engaged in this. After being straight up asked for help during one exam, I warned them not do it, and that this behavior would get them kicked out of school here, but it didn't matter. Most were very smart and talented, and really didn't need to cheat. But they did anyway. Cultural thing, I guess.

The problem is, these idiots eventually get jobs, continue to know nothing, and build a resume of projects. Then they become Project Managers.

Three times I have been in the situation where the interview candidate was not the one answering questions.

Twice we hired someone based on the phone interview and it was someone else who showed up.

The third time it was someone reading interview answers off a chat client window.

I'm mostly annoyed because it reduces the value of the phone screen and requires more face to face interviews.

How did you know it was someone else on the phone? Did you remember the voice? Just curious...

How about a video call then?

While I was studying at University there were quite a few instances where particular groups of student got caught cheating in various forms. It seems some people want to get ahead at all costs because they seem to have a lot of pressure from their family or peers to make it in a particular discipline. This is something foreign to me, as I chose what/where I wanted to study, free of these particular pressures.

I met Mr. X who earns over $10K in cash each month. Job ? Answer interviews on other people's behalf. The setup is sophisticated. Screens shared, Headphones with two inputs and what not.

Visas are faked, resumes are complete rubbish and it is my suspicion that the hiring managers/committees are bribed directly or indirectly.

I know people with 0 years of experience drawing salary of $110K from a large company by showing fake 8 years of experience.

Sadly all people involved are from my home country India. It really saddens me that we arrive here as economic refugees and destroy the very system that is so welcoming to us.

You are comparing Apples and Oranges. The Asian student applicants on whose behalf the DoL filed the lawsuit have to be U.S citizens.

If you read the suit itself, there is a lot of wiggle room for Palantir. Here is the threat (paragraph 14 B) "An order permanently enjoining Palantir and its officers, agents, successors, divisions, subsidiaries, and those persons in active concert or participation with them... "

Here is how they can easily wiggle out: (paragraph 14 D.) "An order requiring Palantir to provide complete relief to the affected class of Asian applicants.." and (paragraph 14 E.) "An order requiring Palantir to hire Asian applicants from the affected class list."

I have a sneaking suspicion that some of these "Asian" candidates are playing the race card, and actually have not so good intentions with Palantir which deals with highly sensitive information. I would be wary, and if I was Palantir, I would bite.

The candidates don't file a complaint with the DoL. DoL already has information on ethnicity. All Govt. contractors are required to collect this information. The applicants of course can refuse to divulge that information I believe.

It is foolish to create one bucket called "asian" and throw in people from Vietnam, Laos, Cambodia Philippines in addition to those from China, Japan and Korea. I personally do not think that the students have time to sit around and play the "race card". They've probably moved on to other things.

I've also noticed the lack of diversity in startups that is disproportionately affecting minorities. Software being what it is is usually a trade off between many solutions that actually work in the short term. So it is actually pretty easy to hire friends of friends and so on and if they can get something to work, you'll convince yourself that you've hired the right person for the job.

Friends being friends will easily pass the "culture fit" test.

Anyway I see a lot of direct and veiled comments against interview candidates from Asian countries with reference to their accents etc. I don't think HN should be the place for such comments. Having worked in the Bay Area, I can say that an accent is the last thing on anybody's mind once you've proved competence.

> I'm not sure how they got in or are able to graduate

I can't speak for elsewhere, as I went to university in Australia, but down here we have a MASSIVE foreign student population as a percentage - International education is currently Australia's third largest export overall - mostly from India and most parts of Asia (kinda our closest neighbours).

My personal, very anecdotal experience is that Indian students were either amazing, or very much not very good. The route seemed to be do ANY degree in India, followed by a (one year?) Masters in IT in Australia. This then legitimised their IT bonafides.

The Indian culture also has a different attitude to cheating - it is seen as much worse to fail than to be caught cheating - and I saw this play out a fair few times, to the point that in one exam 5 Indians had to be physically separated for passing paper back and forth. We also had one girl who said to her lecturer "I don't see why it matters, I'm never going to work. I just need a degree so I can go home and marry a good man".

Now, to answer the quote, the reason this continues is due to the Australian need for the foreign $$$. As a result, very few foreign students are (or is it were?) disciplined for these infractions. Several times it has been a massive scandal down here (http://www.smh.com.au/nsw/mymaster-essay-cheating-scandal-mo... although it seems common: http://www.wsj.com/articles/foreign-students-seen-cheating-m...), and many locals worry it devalues their own education.

On the flipside, the very best students in my year were foreign, and often Indian, so this was not a one-size-fits-all situation.

YMMV, but the Australian universities need for foreign student $$$ has lead to MUCH lower standards for foreign students passing than locals, at least in the past.

Look at the story submissions by the comment author above. May be, there might be a bias here.

Nice catch

They are cash cows and can produce good results on paper at least. Also, they come from an education system which focuses on memorization, so memorizing a bunch of interview questions and answers is nothing to them.

Wait. Based on my knowledge, India is not in southeast Asia. Source: I'm from southeast Asia.

In the UK, 'asian' tends to refer to people from India or Pakistan. In other primarily-English-speaking countries, it tends to mean people from eastern Asia.

(yes, technically anyone from Asia is Asian, but when someone says 'asian', it doesn't evoke the mental image of a caucasian Russian who lives in Novosibirsk.)

Yeah, in some cases "Asia" extends far into the Middle East - Football / Soccer, there is an Asian club championship with teams from the *stan's and Australia.

> and now doing a master's at a large state university

ahh, Georgia Tech

I went to Georgia Tech, and a small subset of Indians there matched the "bachelors in India + work in consultancy" description. I didn't feel anyone there was clueless though. Definitely no fake resume shit.

It is possible that I'm clueless, and cannot identify clueless people.

Thanks to you, TIL that Tech is a state university. Never knew.

You didn't know George Tech was a state school? Does that have some special meaning other than public?

No, I did not. I thought it was privately owned (like Stanford, Johns Hopkins, USC, etc).

LMAO. I'm sorry but that's hilarious. How many years were you there?

I didn't go there, but I thought the same. "Techs" seem to always be private...MIT, Cal Tech, Florida Tech, etc.

(Oh, by the way, Florida Tech definitely fits the aforementioned stereotype.)

Virginia Tech is also public.

Understandable misconception, but he went there.

Not being able to recognize cluelessness in other people might just be an understatement.

Yeah, I suppose so.

Except for reading up on the school's history (which I wasn't too keen to), is there another indication that a school is public?

2, got a MS there, after a Bachelors in India. You must be thinking I fit the profile :-)

Different organizations have different strategies. Looks like yours has a policy of "throwing plaster at the wall, and seeing what sticks".

Wow! I kind of fit into your sterotype, except i graduated from a large private university. Isn't it stupid that you are putting a billion people in the same pool(Indian bachelors+ consulting firm exp+US State school = suckers)?

That subset of indians isn't a billion although...

A few paragraphs in: "The accusation that Palantir discriminated against Asians is an oddity in Silicon Valley, where big companies including Google parent Alphabet Inc. and Facebook Inc. have been criticized for hiring too many white and Asian engineers, and too few blacks and Hispanics."

Your organization is almost certainly not like Palantir so I'm not sure where your sympathy is coming from.

Reply on news.ycombinator.com