On a great interview question
behdadesfahbod.medium.com
behdadesfahbod.medium.com
This question could be improved as follows:
* present the naive solution. Explain that this is a PR that was submitted in 20 minutes by our new intern from Waterloo. Ask them to help improve it.
* suggest to them the intern’s argument why the lookup is O(1) and explain that it’s actually O(n). Ask them to write a PR comment to break the news to the intern.
* explain the optimal solution with tries. Ask what happens if the vocabulary is big enough—and the prod environment is resource constrained—so that the trie is rarely in memory when the function is called. How would the trie solution compare to a binary search through a length normalized list of words?
This kind of question is barely useful to interview new hires—when you still may have basic questions about how much undergrad material “stuck”. Within a few years when they hit full-performance level you should care much less about this kind of question and much more about “higher order” thinking about engineering.
How do you communicate bad news? How do you identify, communicate, and mitigate risk? How do you plan large tasks at at appropriate level without over engineering the solution? What questions should you ask at different project stages to capture the right requirements, anticipate issues, and successfully execute over time?
But whatever, I’m sure you get lots of signal out of repeating “a hash table lookup is O(1)” “nuh uh” on loop for 10 years. :eyeroll:
Maybe the problem has something to do with the way they hire. The article mentions that this string-processing problem had to be approved by a committee:
”Twice I asked the hiring committee if I should stop using this and they had no problem with it.”
Not even the Soviet Union had a process where job interview questions were regularly approved by a committee. The level of bureaucracy in administering these brain puzzles at mass scale is absurd.
The article mentions that the candidate who did best on this question was “the type who did high-school competitions.” I wonder if anyone stopped at any point to question whether that’s the ideal type of hire.
It's unavoidable when you allow too many different interests to have a say in the process.
I think this is the most telling bit. Time after time, I've seen excellent competitive programmers end up being mediocre engineers. Almost none of them are bad and it's also a very objective (at least on paper) process which is probably why Google is happy to stick with the process but most of them also aren't amazing innovators.
I'm sure that university student who aced the interview turned out to be very smart - they went to a great university and can solve challenging puzzles. But I don't think they'll end up being the engineer that builds Google's competition to ChatGPT. I wonder whether a course correction where Google starts looking for people who create really cool projects even at the risk of not being able to solve any coding challenges would actually help.
Oh, wow, if I could get that kind of result ("almost none of them bad", my addition: "possibly great") from an interview, I would stick with that process.
I have also conducted interviews in a company with heavy focus on code tests, which obviously favors competitive programmers, and my anecdata was that I got very good results compared to the alternative (i.e. not doing that).
And I believe this is why this discussion always keeps popping up. And it makes you discard a lot of "possibly great" candidates at the cost of "almost none of them bad".
It is the eternal precision vs recall tradeoff.
OpenAI has a lot of strong competition programmers. Search IOI or ICPC in their team update blogs:
https://openai.com/blog/team-update
I've seen the opposite. I know many extremely talented competitive programmers (IOI, CCO, USACO platinum, Codeforces grandmasters, ICPC, and the list goes on) and they are all excellent engineers. It's not about having a wealth of knowledge about DSA (although this occasionally helps, depending on the type of work) or competitive programming itself, but rather the skills it teaches. They all know how to learn new things very quickly and thoroughly, and it goes without saying that their general problem-solving skills are incredible. They've interned at companies like Snowflake, Google, Jump, Citadel, Jane Street, Waabi, Uber R&D, DataDog, SingleStore, etc. (all before or during their 3rd year, which is practically unheard of for "normal" students, especially during the tech recession) and have consistently received the highest performance grade (our co-op program requires employees to give them a rating on a scale from 1 to 7). There are also a few I don't know directly who are in fact working on things like Bard, or have founded their own companies (TabNine was one of them IIRC).
> I think this is the most telling bit. Time after time
If I can rephrase the question as, does anybody check if the rating somebody receives in an interview is positively correlated with their performance rating in subsequent years?
The answer is yes people do analysis of that question and yes it is.
I'm not sure. The author answered this question with a "no". They admitted they don't analyze or check how actual performance on the job correlates with doing well on this kind of interview questions. Maybe others do, but the author doesn't. I wonder if anyone at Google checks this.
In general, even people internal don't bother to use moma to check this so your opinion is not alone.
Correct.
> Because the author explicitly said he doesn't do follow ups and so he doesn't know if his question is a good predictor of actual performance.
This would be difficult as a random engineer. People Ops has different access to performance/interview data so they can do this analysis; your interviewer can't (unless you share your future ratings with them and they wrote down all of the ratings).
You can do the analysis on yourself though but you'll have to jump through some hoops to get your own interview data. It's also a single data point so unless you can convince a bunch of other people to also do this it really isn't significant.
> Strange that the author doesn't mention this though.
I'll be honest, I think maybe 1k employees know that anybody has bothered to do this analysis. It would not surprise me that the author is completely unaware given most of memegen is as well.
But if they have any poor engineers, their process clearly doesn't work so conclusively. Because those engineers passed their process. And they have no control group, unless they've secretly been hiring a group of engineers without asking them these sort of questions (or ignored the results for a group). I just don't see how Google could determine the process works without studying what happens if they don't do it.
That being said I am easily the best/fastest programmer in my organization despite there being people who have 20 years more experience.
I have found this to be a very good signal in the hiring process.
People who have technical side-projects (machining, 3d printing, electronics, game programming, open-source) tend to be really good at conceptualization, problem-solving, cost-benefit analysis, problem-solving, staying-power, which are critical elements to the success of a software project.
Trivia CS knowledge questions, such as the one presented here, have their place too: they test whether someone can think on their feet, which is not a bad skill to have. However, all things consider, who would you rather hire? - someone who can do an O-time analysis, but has not finished any real project? - someone who failed the O-time analysis, but has built something to the end?
I'd obviously prefer someone who can do both, but the person who has released an open-source project for their ultra-niche hobby (say, a sock knitting pattern designer, used by 5 very passionate individuals; or a climbing wall with LED lights showing you the path) gets my vote otherwise. They may not have come up with the trie solution, but what they've done is conceptualize, build and deliver something concrete to users.
Which means they have proven and refined the precise skills every software project needs.
Well, the author does mention a surprisingly high number of allegedly senior programmers fail at coming up with the simple brute-force solution. At this early stage in the challenge, this isn't "a trivia CS knowledge question", it's simply knowing how to code the simplest iteration over a string and knowing there is something called "a dictionary" which allows lookups! If you don't know this, what do you know?
Hiring committee's job is to take the interview feedback and provide a "Hire" or "No Hire" decision. 99% of the time this is done entirely based on the interviewers' feedback without any further interaction, but occasionally the hiring committee will reach out to ask clarifying questions or provide feedback on the report. Sometimes that feedback is along the lines of "this question is more complex than the candidate's target level" and sometimes it's "hey, that question has been widely posted as a 'Google interview question', so people are studying it specifically; you need to stop asking it." Good interviewers will proactively ask for feedback from hiring committees so that they can provide a better signal.
Also, it's worth pointing out that there is no "one" hiring committee. Many people are in the pool for that task (it counts as a "citizenship contribution" during perf) and you meet for a few hours once every few weeks and grind through a stack of feedback. That's why he asked multiple times; he effectively asked committees made up of different sets of people.
Not defending Google's practices here, but this claim is nonsense. Many public sector/third sector jobs in Western society will only use interview questions which have been pre-approved by explicit policy.
On the point of committees, though. Hiring committees make a hire/no-hire decision. They don't approve questions, but they do make decisions based on the results of asking a question. So they rely on the signal that the question generates, and if a question is not giving good signal, they will have an opinion.
There's a different reason for wanting to standardize on interview questions. It's very easy to embed cultural bias. If the topic of a programming problem uses the rules of a game, for example, that will probably mean that people who are familiar with the game will understand the gist of the problem much faster than people who've never played the game.
Also I recommend you try these "high school problems" you seem to be talking down on. I for one greatly respect people who can solve those types of problems, regardless of their age. These are not trivial problems, mind you. They require a lot of mathematical and logical intuition that 99% of people (and I'd wager the majority of SWEs) don't have.
I think any discussion of complexity for this problem is misguided.
I mean, it's fine to talk about a hash lookup being o(n) for pseudocode in some hypothetical context. But if you're talking about actual code that you intend to run in Python or JS or whatever, it doesn't make sense - are the strings interned or not? When your language hashes the key of a dictionary lookup, does it cache the result? If you're using a JITTed language, won't any native o(n) hash lookups vanish in comparison to your non-native o(n) for-loop?
On top of all of which - why are we even talking about o(n) complexity for an n that cannot grow? If we're given a fixed dictionary then we can ignore any substring longer than its largest element, so the "n" for hash lookups is not the size of the original input - which is the n we realistically care about!
Basically, once TFA mentions complexity I feel like the only correct answer is to say "this isn't a situation where complexity analysis would be useful - we should benchmark the code".
But you don't try checking if the entire string is in the dictionary regardless of n - you only ever check if it's in the dictionary if it's shorter than X+1 where X is the longest string in the dictionary. As the input string size grows, the dictionary lookups are capped.
If the author asked about complexity as an open-ended question with no fixed answer, that would be a very reasonable way to hear how the candidate thinks about algorithms. But TFA makes it clear that the author treats it as a gotcha question - he expects candidates to say o(n) and then pushes them to give the answer he wants. That's not a great approach, even if the interviewer's preferred answer is unambiguously correct - and here it's not.
All of these are useful signals for a hiring decision. Much more than quibbling about big-O of hash table lookups.
Great prompt!
Like these can’t be prepared for? I think there are now whole chapters in those interview prep books/courses on how to answer these.
Even if there weren’t this is a good way to hire people who sound very professional on slack and in meetings but don’t produce any actual code (unless you got very lucky)
You would be absolutely stunned how many people cannot give you back the correct answer even after you literally give it to them. If you don‘t know some programming and I give you code and explain what it does, that still doesn’t help you.
There’s some tiny signal of your ability by asking you to memorize and regurgitate something; and maybe a tiny bit more signal in this question by implicitly testing “how do you respond to unexpected technical details in a high stress situation”. But they aren’t worth the trouble. You could measure both much more directly. You’d get more detail about their coding ability by directly asking about code they’ve written and know well.
I don't get why the dictionary solution is actually O(n).
Surely you can have the hash function be O(1) by only looking at a subsets of characters something like
hash(s) = (s[0 %s.length ] ^ (s[6 %s.length]<<c1) ^ (s[7 %s.length]<<c2) ) % (size of the hash table)
You can trade-off memory for speed to avoid collisions by having a mostly empty big table (like 9/10 empty table). And you can also use multiple hash functions that must both simultaneously collide to avoid having too big a table (kinda like fronting your hash-table with bloom filters).
And you can pick the selected sub-index at random such that they split well your dictionary. Maybe you can even build a "Perfect hash function" so that all your dictionary words have distinct keys.
You only have to check for the whole string when the hash match, probability of which you can make as small as you want.
> > About Part 2 :“Write a function that given a string, determines if it’s a concatenation of a number of dictionary words.”
Isn't it just recognize if a string belongs to a regular language, where the language is defined by words in a dictionary ? So you just build a parser for your language and you test for belonging.
I'm quite rusty in non-deterministic finite automatons and regular languages, but surely there exists algorithm that build an automaton from the dictionary to recognize the language in something better than n^2, no? (We are moving the states of the automaton once per character, and we likely won't have n active states (except maybe for pathological dictionaries), complexity should be something like O(k*n) where k is a constant depends on the dictionary).
Eg you can’t just do equality on collision, you need to do equality on every lookup at least once.
> you need to do equality on every lookup at least once.
I still don't think so. You only need for when the split would correspond to a good partial match.
There are also other smarter hashing thing you can do to help convince you.
For example if hash( s ) = sum_over_i( g(s[i]) ) if you pick g(x) = x or g(x) = xx or g(x)=xxx (or any function you are free to choose) (it's kind a like matching statistical moments of words).
You can calculate this hashes in amortized O(1) because of sliding window trick.
You can also index on split length.
You can filter your set of candidate split positions so much that you only have very probably valid split positions in your candidate set, and therefore the complexity is just verifying the validity of the candidates splits, so still O(n(number of candidates after filter) ).
> > Part 2
A solution based on filtering a candidate set could also probably be made to work for multiple words. You use the matching statistics to get candidates starting positions and length for words of the dictionary that appear in the string and you verify if they touch each other. Pathological cases are when words overlap each other but this may not appear very often in practice.
You can also add some filter based on pairs of words (or triplets..., or n-grams) .
All in all, I think the problem is rather bad because smart candidates will often have a solution that the interviewer may not be able to comprehend, and as a candidate you have just put yourself in a possible ego battle with the interviewer were even explaining the proof is not easy. In fact if the alternative solution is more interesting than the rest of the problem, you might get your candidate be focusing on that instead of your simpler but boring trie based solution.
My initial hunch was an example where every prefix of the word is part of the dictionary, then on an example like `abcdefgh` you'd test `a`, `ab,` `abc,`... But on such a case you can be clever and defer the string match until you verify that the hash of the suffix also exists in the dictionary, which it won't.
So it would seem a pathological example needs both prefix and suffix matches, but this just means that it's very easy for us to find a valid segmentation, so we will find one very early on.
I guess this is probably what you meant by "split would correspond to a good partial match" though.
What this means in practice is that there is no polynomial curve for very long words. You don't bother hashing them, so the curve peaks and then flattens.
Obviously a trie is an interesting and possibly faster solution, though the hash has the benefit of being more cache friendly. But the big O analysis presented for this problem is flawed.
On these kinds of problems, I like to suggest that we might be thinking about arbitrarily long "words". Get out of the headspace of language dictionaries, look at things more abstractly. Otherwise a lot of big-O analysis becomes degenerate.
Also, don't forget hostile inputs!
I seem to remember that in the very early days of Java the standard hashing method for strings only looked at a small prefix - turns out that very common use case was to be storing URLs as keys which, of course, mostly start with the same letters...
> The fastest I had a candidate solve the entirety of the problem with all bells and whistles was in 20 minutes, by a freshman from the University of Waterloo, the type who did high-school competitions.
This should’ve been the interviewer’s first clue.
This seems like a really great question, thank you :)
Instead of doing stuff like this, you can just give candidates realistic programming challenges and have them write code. You don't even need to be in the room with them; after all, you won't be when they're actually doing the work.
def is_concatenation_of_two_dict_words(word, dict_words):
"""
Returns True if the input word is a concatenation of two words in the input list of dictionary words,
and False otherwise.
Args:
- word (str): The input string to check.
- dict_words (list[str]): A list of dictionary words to check against.
Returns:
- (bool): True if the input word is a concatenation of two words in the input list of dictionary words,
and False otherwise.
"""
for i in range(1, len(word)):
prefix = word[:i]
suffix = word[i:]
if prefix in dict_words and suffix in dict_words:
return True
return False
I agree. We do work sample tests, and in addition to the code and docs the candidates hand in, what really matters is the walkthroughs we do with them. Why did they do it that way? What alternatives did they consider? What are the pros and cons? What past projects did they draw on? How did they research this?
Candidates usually enjoy this - most programmers enjoy talking about a just-finished project, especially when they feel good about the result - and you get to learn a lot about them.
If someone turned in an LLM-assisted work and lied about it, I doubt they'd fare well. And if they did use LLM assist - could be an interesting conversation all the same. What did you reject? What did you correct? Why?
If so, how do you solve the problem of the assignment just taking way too much time compared to a more typical process, where you do a phone screen and 3-5 hours of technical interviews with a behavioral or 2 thrown in there? I know that if you send me a take home that's any like most of the ones I've gotten in the past, and your competitor says I can do a recruiter chat and technical phone screen to qualify for a 4-5 hour onsite, I'm going with them just on the sheer amount of time and effort needed from me for each process.
How do you get around the fact that if you give such a test to enough people, some of them will eventually put it up on Github for the world to see and crib from? Even for those willing to accept the time commitment, how do you account for the bias involved in giving the same assignment to working candidates and people just out of school? There's a good bit of implicit age discrimination going on there, since folks who have been working a while are far more likely to have families than those just out of school. And then there's the old "fraudulent candidate hires someone else who's actually good to do the assignment." Granted, that person won't succeed at the job, but it'll take at least some weeks before most companies will give up on someone they hired, even if they're completely hopeless at the actual job.
I could go on and on, but I think you get the idea.
B) you replace time in interviews with the test you don’t just add it on as another thing.
C) you change the test regularly and implement basic plagiarism filters on the tests you do have.
As a person who has been in the industry a fair bit I think the flexibility of take home work samples gives me a leg up over day long in person interviews (and the last time I interviewed it was multi day not 4 or 5 hours).
I don’t have to take time off of work or not pick up my kid from school or anything. Just do it in my free time.
Seriously though, you've basically outlined the only reasonable way to go about it. The problem is really that that's a lot of fucking work you're opting into with 3 little sentences, and after all that, you actually end up with a non-repeatable process, which is not really ideal. I'm still highly skeptical that it's not biased toward people without families, either.
But my real issue here is that at least 75-80% of companies that do this crap obviously don't implement any of that stuff, and most of the rest don't do all 3. You really need all 3 in order to maintain any integrity to the process, and that's, unfortunately, one of the strengths of whiteboard hazing from the employer side. I know I'm never doing another 2+ hour take-home again, and the fact that employers generally don't do any of that stuff are more or less my reasons.
To me it’s a strong signal if a company has a work sample test because it shows they have at least one part of their process that has a chance to be a good filter. It’s an even stronger signal if they’ve replaced the majority of their pipeline with it.
I won’t say I won’t accept a job that didn’t do a work sample component but it would definitely jade my thinking against the firm. And I’ve certainly told companies I wouldn’t continue with their hiring process after hearing what it entails.
def is_concatenation_of_dictionary_words(s, words):
if not s or not words:
return False
n = len(s)
dp = [False] * (n + 1)
dp[0] = True
for i in range(1, n + 1):
for j in range(i):
if dp[j] and s[j:i] in words:
dp[i] = True
break
return dp[n]
That said, I believe an engineer with the answers it gave, would easily figure out from its output what the right complexity is. (Figuring out the trie, may not be so easy.)
That said, meanwhile, ChatGPT is not yet at a point where it can write out a full working repository solving a nontrivial problem. It can help, but it is not autonomous; and realistically, if someone can get that help to reach a good solution for a test, they can do so for a salary.
With O(n) memory, precomputing the rolling hash in O(n) time, you can get a hash of a prefix/suffix (or any substring) in constant time.
https://cp-algorithms.com/string/string-hashing.html#calcula...
Better than the candidates I interview would do, and probably better than I would do, I would probably need a hint to use a trie and I haven't analyzed algorithmic complexity in a quarter of a century.
- While DP might be rare, memoize is quite common alternative. Sticking a @functools.cache in Python for example. They are functionally equivalent in many cases,
- I believe knowing their data-structures and algorithms is the difference between an okay programmer and a good one,
- When interviewing entry-level software-engineers, there's not much else you can ask them except for what they learned in class, which is these kinds of questions.
For me, the difference between and okay programmer and a good one is: how much Linux/Unix they know (processes, core system calls, networking, sockets, awk/sed/bash). Whether you are writing apis for ecommerce platforms or writing custom k8s providers, Linux knowledge is a must. Knowing how to implement a trie? I bet most of us have never been in that situation.
Likewise, there are jobs where being able to reason on complexity and data structures is important.
Your broad strokes don’t account for the domain knowledge of a job. If all you’re doing is interviewing college grads, then I guess you have no choice, but this interview question sucks in just about any other context.
They're not just functionally equivalent. They are the same thing. There are two things you need for a dynamic programming solution:
1. You remember the answer every time your function is called. ("Memoization".)
2. You structure your function calls so that they will ask for a lot of answers that have already been computed.
What you don't want is: say you're computing the sums of contiguous subsections of an array. You've already figured out that indices [2:5] sum to 3 and indices [5:8] sum to 8. When you need the sum of indices [2:8], you need to make sure you get it by asking for sum(2,5)+sum(5,8), and not by asking for sum(2,6)+sum(6,8).
At a high level they are the same, but it makes sense to distinguish between the techniques based on whether you need random access to the already-computed data.
For example, if you are computing the nth fibonacci number f(n), a DP solution needs only the state f(n-1) and f(n), but a typical memoized solution keeps around all the state f(1) through f(n-1). For tougher problems, you can't always convert between them trivially while keeping performance the same.
That's all true, but is it comparing like with like?
The "generic DP" solution to the problem is to allocate an array of n elements and fill it in from f(1) to f(n). That's the same amount of storage the "typical" memoized solution uses; cutting it down to an array of 2 elements plus an (implicit) offset between the part of the array that's in use and the full array seems to be an optimization on top of the DP solution.
You could apply the same optimization to a memoized form of the solution. I certainly agree that you most likely wouldn't do that, but you could.
What do we learn by comparing a tightly optimized DP solution against a "typical" memoized solution? Why do we give the DP solution credit for not using what it "doesn't need" while penalizing the memoized solution for what the programmer probably added unnecessarily?
I worked a SWE job years ago which was heavily reliant on dynamic programming and string search/alignment/etc. It really depends.
IMO, the disdain from LeetCode from developers vs. its continued use disconnect stems from the fact that we developers tend to assume that the goal of the hiring process is to find the very best possible candidate, whereas the goal in reality is to find somebody "good enough" as efficiently as possible.
At the end of the day, you can't test for negatives and just pick a random person that doesn't fail anything. You need to have a defined set of values to look for in a candidate, and to design an interview that tests for those values. That's why I hate DSAs -- for most (all?) companies, they're treated as dogma rather than a pragmatic solution to a difficult problem.
I ask multiple related questions, and if a candidate struggles with one of them, I probe for strengths in other areas.
No one answer to a question in that interview is a pass or fail. But the interview, holistically, is a pass or fail, or a conditional[1] pass.
[1] I was impresssed by some aspects of the candidate, but there are notable, significant weaknesses. Please determine if other interviewers have detected similar weaknesses, or if the candidate just got unlucky with my particular question.
Unfortunately the OP seems to be using it in the former sense (when applied to the tries solution that he things will separate the "good" candidates from the plodding lunkheads).
I suppose that's why mathematicians came up with the idea of "optimal stopping." There are literally tons of scenarios out there where the goal isn't to get "the very best possible candidate," but to get someone "good enough," or maybe even the best person you can get after putting a reasonable amount of time and effort into searching.
This is also an unproven statement with dubious validity.
I come from the trading world where the leetcode equivalent is a brain teaser. After much deliberation, the only thing we could think of for why a brain teaser was useful was that it often proved someone wanted the job so badly that they memorized all the brain teasers. And that type of dedication and drive is what we really wanted.
To me, I don't understand why someone who develops with advanced IDEs, using complex tools and libraries, is in the least bit judged by putting them in front of what is essentially a text editor, and asking them to solve a problem that's of no relevance to their job, using no tools that they're comfortable with
Period. If someone tells me they can develop software in a language, and they can't write ~10 lines of code, that tells me they lied. It's like when getting lifeguard certified they require you to dive down to the bottom and grab a brick. Nothing at all in the day-to-day job of lifeguard requires retrieving submerged bricks, but the set of people who possess the swimming skills to be a lifeguard is fully included within the set of people who dive to the bottom to grab a brick.
This qualified me to sit next to a pool in a condo complex where no one ever swam for eight hours a day. Best job I ever had.
Your question has no scientific justification. It just is a problem you can do, and you like yourself and think you're awesome. So you think it finds people like you. But even that isn't scientifically supported.
I'll take someone whose Github repo shows talent, dedication and passion, but who misspelled an include header in their brain teaser over the inverse any day of the week. But you do you. Neither you nor I have a lot of evidence to support our preferences either way.
Digging through a couple GitHub repos is a good way to find good developers versus poor ones. The goal of a leetcode question is to separate the developers from the non-developers. And you can do that pretty adequately with 2 questions: "Tell me about something you've worked on" and "Hack together a solution to this trivial question".
The question did not "weed out someone who couldn't code." It weeded out someone who couldn't use a totally unfamilliar dev environment while being stared at and asked to "walk me through your thoughts" instead of letting me think. It left me distracted, thinking both "this is easy, what's the catch" and "wtf is the point of this?"
"Tell me about something you've worked on" is a totally fine question. I'm not arguing with that. But anything leetcode is not, IME. This becomes increasingly true as people's careers develop. If you've spent your last two years writing hyper low latency deserialization engines and packet processing, then you'd know that's what the job is about. In the macro, trading engines and game engines are one medium leetcode problem from an algorithms point of view, but they require teams of people and years of optimization and customization to make truly effective. I don't want to test for the person who can make a really basic solution in 30 minutes. I want to test for the person who can optimize a tiny fraction of it in 2 days.
1) most people here, I believe, don’t have a problem testing a developer’s code writing acumen.
2) People weigh something. Bricks weigh something. It’s the same exercise. The analogy in software would be the difference between asking someone to write _an_ function that can do x vs write _the_ function your company uses to do x.
3) the lions share of counterarguments here propose using real world exercises vs toy problems
Programmers skew anxious. In theory there's no difference between writing basic code on a whiteboard and writing it in an IDE. But in practice, an interview is a novel, high-stakes environment with a bunch of judgy strangers. So if you then have people do the interview using tools that are unfamiliar, a notable percentage of them will be thrown off.
Once I switched to encouraging people to bring their own tools (own laptop, own editor, preferred language), I saw a marked decrease in people bombing interviews. Some of the people you might have put in the "they lied" bucket were, at least for me, just ones who got really nervous and so were unable to grab the brick.
To me, that's a bad interview, because the job isn't going to be high-stakes whiteboard coding. It's going to be sitting quietly in a comfortable and safe place and gradually improving things. The more an interview can be like that, the better.
I've always assumed the brick was meant to simulate a drowning swimmer, whose rescue is pretty clearly part of the lifeguard's job. I guess a mannequin would be more realistic, but it still seems pretty close.
Sadly everyone who makes it to an on-site is now a competitive programming enthusiast who is probably a really poor fit to actually working in software development.
Many people are vaguely aware of this problem, but nobody has a strong incentive to put in the work to come up with anything better.
And most programming jobs involve no complex algorithms at all. At most you have to be a little careful to avoid N^2 occasionally.
Leetcode is great for filtering out bad candidates but if your solution involves dynamic programming it is too hard for an interview.
OTOH re the concern for O(n^2) time when n is bounded by word lengths, and when the inner loop was in already-optimized string comparisons in your language's hashtable library -- I think that part is very artificial.
Yes! The candidate already optimized for the important thing (which the author forgot about and added in a footnote), which is that the list of words could be very long, and that the words in an English dictionary are relatively short.
Optimising for the important factors based on an assessment of the problem domain is actually a more important skill than regurgitating the best algorithm from the book. If the author wanted to suggest a solution suitable for very long needles, he should have used a different domain like, say, gene sequencing.
Give me any problem related to algorithms+ds to be solved in 45mins without internet access and someone looking over my shoulder and I will give you a poor solution. But give me a day and internet connection and I will solve the problem in the most efficient/readable/maintainable way possible.
The interview question is not just about knowing data structures and is not about Behdad showing off either. It's the starting point for a Socratic discussion.
OK, but every program is literally one line of code. We only put line breaks in to make them easier to read.
The author makes this terrible comment about his "one-liner":
> Most candidates though write a for loop, which is equally acceptable.
He has also written a for loop. Specifically, he's written this for loop:
for i in range(1, len(s)):
if inDict(s[:i]) and inDict(s[i:]):
return True
return FalseWhat is important is that the candidate’s skills generalize well to both the challenge in the interview as well as the day job. A question like this might be a good indicator or not depending on the job.
If people completely fail at this question it can reveal a lot of relevant information.
I've given this same advice for _years_ for designing technical interviews. Take a problem that you or your team has actually faced in the last n months. Distill it down to remove tribal knowledge and get it to something you or a capable colleague can hammer out in m minutes. Give the candidate 3m minutes to solve it. Typically this means it should take you 15 minutes to solve and give them 45 minutes.
Our biggest hit for a question that followed this for an SRE type role: we had a cron job that was legacy and buggy and caused a couple of issues over 6 months. We went back and created a very similar script with very similar bugs, added some similar data and re-seeding scripts, and told the candidates "customers say that feature $x is not working, and that is ran by this script. Here is how to run it, here is the user's id that can be found in the data seed." We added some other data validation problems. After they fixed it, we asked them what they would do to make this more production safe and dig in on what they have and have not done for improving existing systems. We expected discussions on structured logging, metrics, alerting, testing, code layout, etc.
Over time, the roles that you are looking to fill and the weaknesses on the existing team that you want to shore up will change and you will need different interview questions. Maybe more towards distributed systems or more towards testing or observability.
I think they're using "m" there as a variable, to frame that you should give the candidate 3 times however long it takes for an insider to solve it.
They mean 3 x m where m is how much time you think it would take for you to solve; so the example is: if it takes you 15mins, then give the interviewee 3 x 15 = 45mins
Importantly, it also allows you to examine fluency. The candidate who can write a code/pseudocode solution as fast as they can type, and explain it verbally at the same time, is much more impressive than the candidate who pauses for a long time before starting haltingly and unconfidently. Fluency in solving leetcode problems just tests for having practiced leetcode, but fluency in writing real code tests for actual productivity.
If you have a question like this, you can drill down into performance characteristics just as the question in TFA does. Not necessarily using big-Oh notation, but "what if you have to do this millions of times? what if you have to provide results immediately? what if some of the files are too big to fit into RAM?". But you can also drill down into other aspects of the problem like changing specs, noisy input, constrained resources, observability.
FWIW, when I came up with a question like this (with the constraint that it had to represent a problem I had solved that morning), ChatGPT was not able to solve it correctly.
No wonder it is a hit. It is an actual problem that makes me want to go for it - I want to take an interview like this, instead of the usual of dreading interviews! Keep it up.
And in my experience listening to auditions over the last several decades, what you have now is a bunch of musicians who can only do what's on the test. When you try to plug them into an orchestra, they are completely fucked.
My experience interviewing and hiring tech people is that this is one of the worst possible approaches. Anyone who needed to could figure this out. It's not relevant. What you need to figure out in an interview is if the person is curious enough to find the answer.
I don't ask for code or algos in my interviews. I ask about the person. What's your favorite programming language? Why is it your favorite? What do you not like about it?
What's the project you are most proud of? What was hard about it? What made you happy? What did you find challenging about that project?
These are 100% totally humane questions that tell you everything you could possibly need to know about a candidate in about a half hour. Maybe less.
If you can't figure out if this person is technically competent from this set of questions then you aren't competent to interview or hire.
Interviews don't have to be only about how good you are at interviews. They could be about trying to understand if the person would be good at the job and fit in well with the team.
Episodic recall and emotional recognition are both talents that vary across the population. If I want to do well at an interview like this, I will have to spend a bunch of time in advance refreshing my memory of various projects I have worked on so that I can simulate a neurotypical response to those questions. If I don't, I may just stare at you nervously and break out in a sweat when I realize I don't have anything filed under, "project I am most proud of". Or that pulling up the narratively engaging details of a project I haven't thought about in years is a process that will take me hours of specific effort. And it's not just me; I've hired great developers who were also terrible at "humane" interview questions but who did great when we just sat down and coded together.
I usually have a project or two that I can speak-to in depth, to give examples of decisions made, challenges handled.
In the end all you could actually do is guesstimating who would do best at the job. And sometimes the answer is: "none of the ones who showed up".
If you are a good engineer or programmer yourself you can tell a lot about a person by talking to them for an hour. You can learn what is important to them in the craft, how they deal with criticism, what they aspire to, what kind of problems they tend to work on, if they are more autodidactic or more influenced by other's opinions and so on. This is all knowledge directly inpacting the question whether they are the right person for the job.
And who the right person is depends on the job, so there might not be "right" answers to the questions. For a very niche database job you might actually want someone who is very accurate, very in the detail and very focused. For other jobs maybe some entirely different traits are better.
Of course the problem here is that the people you hire could just be very good actors or liars who cannot do the things they say, so a little technical testing might also be needed.
While yes, you can get someone who has a good track record of being a good judge of character to make hiring decisions. If that decision goes wrong and the blame game starts then you inevitably ask this person to explain their hiring decision. This person will then reply: well it was my opinion that X, Y and Z.
In the alternative case where you just get someone to follow a rigid process you can answer: Well he scored well on this exercise and the opinion of the team was in his favour and the work history ticked the right boxes. This effectively dilutes the responsibility. You can no longer blame one person for the bad hire.
Lastly there's also the diversity aspect. With companies increasingly being scrutinized for their hiring practices, "it was my opinion" just gets translated to "this person's internal biases guided the hiring process and as such it was not fair."
I agree this is all making hiring much harder than it needs to be. And certainly you can still find companies where the hiring decisions are made on highly experienced and well honed gut instinct (its the only kind of interview I have been a part of) but it also makes sense how in our increasingly overcomplicated world we are developing increasingly overcomplicated hiring processes.
My recommendation is if you want a sensible hiring process (with maybe a telephone screen followed up by one normal interview) you should stick to smaller and less corporate companies.
I know plenty of smaller, less corporate companies that still use careful hiring processes with thoughtful rubrics. They do that not because of risk avoidance; they are generally still ok firing people who don't work out. They do it because they want to run a fair process while maximizing ROI.
And lots of good developers skew anxious! What in normal life counts as being overly careful or or overly sensitive is just good practice in coding.
That's why I think it's especially important to make interviews as low-anxiety as possible. They'll always be worse than the job, of course, but the closer you get them, the more you're likely to see how people will really perform on the job.
Both of mine did. Much of their advice was irrelevant (nobody in tech gives a shit about your suit or how firm your handshake is) but they both stressed the importance of preparing in advance to show how good your skills are; and giving thought to cliche lines of discussion like 'what's your biggest weakness?'
Not only do I not remember past successes very well, preparing for these questions is literally the first time I asked myself what I'm proud of. I didn't even consider doing this and reject it as irrelevant before. It just never came up.
I've always had trouble with the kind of behavioral interviews that are "Tell me about a time when . . ." since I can often recognize that I've been in relevant situations but can't for the life of me remember the details (even with time to prepare).
This honestly makes me more nervous about interviews than leetcode shenanigans. At least with those, I know how to improve.
(At work I rely on a combination of having the code in front of me, written notes, and whatever project management tools we have so it's not much of an issue, but most of that isn't available to me in an interview setting.)
Source: Multiple orchestra musicians and management speaking about the audition process and the now-defunct myauditions.com website
Process:
Show early talent
Get the right instruction
Get into a preparatory program
Make it into a conservatory
Do well once there
Get a spot in an ensemble, probably lesser-known (not easy) and establish a reputation
See an audition notice with repertoire to prepare, apply
Spend a few weeks or months intensively practicing (on top of other commitments)
Maybe get an audition, realizing the Music Director may be promoting his buddy past these rounds
Travel at your own expense to the audition
Play demanding repertoire behind a screen and respond to any directions
Most likely get rejected
Play in any following rounds
If not rejected, maybe you get a qualifying week rehearsing with your potential colleagues and playing a couple of concerts when the Music Director is in town
If you beat out the couple of stellar artists who've also gotten that far, you may get an offer
Or, the MD's buddy may get the offer
Or, there may be a no-hire
If you get and accept the offer, you are "on approval" for a couple of years at which point there's the Tenure Decision
I would still require the candidate however to write some code. Something rather simple, not leetcode, ok, not fizzbuzz either, and something that proves their familiarity with at least one language and platform.
I should be a GM by now simply watching chess analysis videos on Youtube, learning the names of some openings etc. I can talk about it all day long. But since I don't play at all (I'm not interested) when I actually do play on occasions I revert to 500 ELO...
You can easily tell a seasoned veteran from a rookie just by looking at them code for a few minutes, working on a simple task. If you can't, you have no business in interviewing others :)
Someone who can explain why they love Rust and hate Python (or vice versa) is a convincing candidate but I think you need to see a couple of other things from them. In addition, sometimes you just don't hit on the right question to get evidence of their aesthetic judgement. (Like the candidate who doesn't care about Rust vs Python, but who hates Postgres and loves MariaDB..)
Not precisely a SWE, but every single piece of code I write at work is more about knowing how the immense codebases are laid out and where to insert my change which is typically simple.
Given this, any "write self contained thing" is unrealistic. The only thing a coding question tells me is whether someone is happy writing simple code to solve a vaguely fun problem.
That still feels useful?
It was great because he was testing the very thing I was going to be hired for. Additionally, he asked how I would design X or Y and I explained the pros and cons of certain designs and I could show him right there on the screen how I'd do it. This wasn't a whiteboard where I'd have to slowly write out code in a medium I'm not used to. I literally was typing on a keyboard, using intellisense, etc.
As someone who does iOS development day to day, it was a very easy way for me to show that I knew what I was doing and the guy was quite pleased when I plowed through all the requests he had.
That might be something you are looking for in a member of your team, but it's not a great indicator that they'll bring much additional shareholder value.
One interview I had years ago that I really liked was where they gave a sample of code with a number of issues and asked me to describe how I'd improve it. I really appreciated that a) it was much more like actual work than many questions, b) you could start with obvious things and work deeper, and c) there were many valid ways to improve the code, so you weren't just trying to guess the answer the interviewer had in mind.
I strongly suspect many interviewers (myself included) would have a few "top issues" and significantly favour candidates who pointed them out. But as the candidate you don't really know which issues will seem important to the interviewer.
It's not an irrelevant exercise though - code reviews are a real part of many roles and making sensible choices about what to comment on is a necessary skill. But it is hard to avoid turning the interview into an exercise in "guess what's on the top of my list"
These folks were obviously interested in differing approaches; we had a good conversation about it. And that's how I use broad questions.
But I agree a huge problem of interviews is the "guess what I'm thinking" interviewer. Most interviewing processes are not primarily about making good hires. They're about making interviewers feel smart and important.
That is, if your reaction to the problem is "Eww what a 'clever' problem, are you showing off that you know such stuff?" then you are not someone Google is looking for. (Of course, whether Google is right here is a separate question ...)
The danger here is that when you explain something to someone, and they get it, you get a warm fuzzy rush of happy brain chemicals. You will feel well-disposed towards candidates who appreciate and enjoy taking the breadcrumbs you feed them. Candidates who didn't think your clever insights were as clever as all that, or who fail to respond to your tutelage and try to solve it their own way... you'll get a negative feeling towards them. They're not playing your game.
That's great if you're interviewing a student to figure out if they will enjoy learning from you, and you will enjoy teaching them!
But while teacher/learner relationships might be part of your relationship with colleagues at work, it seems an odd part of the relationship to compatibility filter in an interview.
Also, once you used the same problem half a dozen times in interviews, you no longer get any "warm and fuzzy" feelings for having to explain something which is now blindingly obvious to you: if anything, you'll give higher scores to candidates to whom you don't have to explain such things.
Of course, if you are a candidate and (hypothetically) your response is "Actually it can be less than O(N), because a hash function does not have to read the entire string, and I know how it works because I've been dealing with string hash functions for years in my professional career," then by all means, fire away, I'm pretty sure any reasonable interviewer would consider it as a big plus sign.
The length of that word does not grow arbitrarily long, causing the asymptotic behavior of your algorithm to dominate. Because it’s a word.
In the case of ‘is this word of length n the result of concatenation two English words’, the time complexity as n approaches infinity is constant, because once you get past floccinaucinihipilificationantidisestablishmentarianism, any longer word than that you can just return ‘false’ immediately.
It feels really disingenuous when interview questions start off posing a seemingly innocent problem where the intuitive and practical answer is usually the one that'd you go for if you wanted to implement this in the real-world, then as soon as you answer the interviewer turns around and throws a hidden assumption at you. You could argue that the point of the interview is to have the candidate suss out these things, but it's usually a one-way discussion where you're not allowed to even question if the assumptions make sense in the real-world (which is what you ultimately care about).
And if the words are truly big enough to be an issue then you better start being rigorous and benchmarking different approaches rather than just handwaving big-o.
Ultimately when you get down to it, the whole point seems to see if the interviewee can be led down the pre-determined golden path, even if the path is only fools gold.
After all, if we assume that all words are at most, say, 50 characters long, then there's no asymptotic complexity to speak of. Any string longer than 100 characters is trivially not a concatenation of two words.
Google used to look for people who were excited about interesting problems. This is a question about basic complexity analysis and I would have chastised the interviewer if his feedback was ever on one of my interviews.
For part 1 using a hashtable to look up the dictionary words, the complexity is not O(n^2) where n is the length of the input string. It should be O(w^2) where w is the length of the longest word in the dictionary.
The hash key material of the dictionary words has at most w characters. When computing the hash key for the input string, it can be truncated at w, as any longer input won't lead to a hash key making a match in the hashtable.
For part 1 using a trie, the complexity is O(w) rather than O(n). This is actually more obvious since walking down a trie tree will stop at the last character of a dictionary word.
Again for part2, the complexity is O(w 2^n) using bruteforce with hashtable, O(w^2 n) using DP with hashtable, and O(w n) using DP with trie.
Edit: Also for short strings like dictionary words, O(w) is kind of meanless since with SIMD or SWAR the short word can be folded into one or two DWORD. You can effectively treat it as O(1).
I don't think the function "inDict" is even defined as either local or remote query (unless I missed it), which really influences the answers. Querying a short string against in-memory structure will be based on "n", but querying a long string against any remote database will be based on "w".
The trick for me was diligently applying to as many places as possible (that were at least somewhat of a fit), and eventually coming across one where the main technical part of it was "take home". Coding and discussing thought processes on the spot for an audience isn't something I'm good at it turns out, but solving a practical problem on my own time was fine.
It's ultimately a numbers game, the more places you apply to, the more likely you'll find an interview process that fits. That is the most important thing to keep in mind--shotgun blast those applications. It's hard not to take rejection personally, but don't give up and settle for a toxic job, that's even worse.
As I age, one thing that I realise about interviewing: It is scary as hell. It is basically performance art. And, most nerds are terrible about public performance & speaking. Some days you are really on it; other days: way off. Every single job that I ever got in my life feels like a lucky day. Also, the tech universe is infinitely wide at this point, so somebody somewhere can ask a question that you will fail.
Since this industry is so dominated by men, it also feels like a lot of machismo. So many interviews are just a jerk dude on the other side of the table (probably working for an amazing company) that wants to make himself feel better by making me look foolish. There are too many stories to tell. I have also kept my mouth shut when interviewers were insistent about a technical point that was 100% verifiably incorrect with a simple Google search. It helps to stroke their ego during an interview. When they give you a hint, or explain a better solution, try: "Oh, that's really impressive! How did you discover that algorithm?" You are more likely to make it to the next round. The goal of interviewing is to get the offer. You can always turn it down. If you had one jerk in the interview stream, but they work on a different team, it should be OK to accept the offer.
Another thing that I say to all interviewees now: I know that interviewing is scary. We are interested in what you do know, not what you do not know. Together, we will dive as deep as possible on a few topics to find the outer limit of what you know. If we come to a topic that you don't know about (or not deeply), tell us; we will move to a new topic.
I think I read on here about the concept of successful people really just doing more things to increase their luck surface area.
In other words, you set yourself up to interview as best as you can, then make a bunch of connections and schedule a bunch of interviews, and (hopefully) at least one of those will be on a good day.
It's such a weird and, frankly, shitty way to make hiring decisions though. But at the same time, we don't have much better.
I think I read on here about the concept of successful people really just doing more things to increase their luck surface area.
After a while, either they will make you offers to work for them on the spot, or they will at least "fast-track" you through the interview process.
It doesn't work in all software areas (it is pretty difficult to do it if you are an embedded systems developer) but it works especially well in others (e.g. cloud)
and you suggest providing free labour for an entire year to some companies. No wonder GP hates this industry.
>For the next year try contributing to open source projects that you like and that have companies behind them.
I have interviewed a lot of people for coding jobs, and I usually try to test ability at doing things like this, and when I find someone junior who already knows how to approach problems like this, I am always very in favor of hiring them. Based on the discussion here I'm definitely not the only one.
You studied field X to work in field X, and in a job interview the "smarmy git" considering to employ and pay you has the temerity to ask you things about field X to see whether you've understood the basics of field X, can apply them, explain your reasoning, and think on your feet?
And that engenders (and I quote) hate, sadness, anger and despair?
That is who you're optimizing for.
And that's fine! A specialized subset of engineers. Maybe that subset is exactly what you need.
I wonder how the stress of the interview process selects out people who understand O-notation and to whom a Trie would make sense if explained to them, vs people who have these concepts fresh in their minds and/or can handle themselves in stressful interview situations.
Back in 2015 Peter Norvig discussed an analysis at Google that suggested doing well at coding contests was a negative indicator of job performance: https://youtu.be/DdmyUZCl75s
Presumably they do well at these sorts of questions, but that doesn't imply the same actual effectiveness.
(Personally I would add that human measures of “job performance” can also be quite subjective.)
Yes, implicitly Google can only measure the job performance of people they actually hire.
And who is supposed to implement the robust canned implementation?
(I still think the question is unfair, but my objection is more about the stressful interview situation where the real problem you're solving is "desperately trying to pass this or N other interviews you're currently doing").
What's the function going to be used for? Are we talking about strings of length 32 or strings of length 32k? Is this going to be run a billion times a day or 12?
We're not competing at some contest, we're trying to solve real problems here. I'm not going to waste a couple hours hammering out some ultra-optimized code so the logAnalyzer4000 can run its hourly batch jobs 0.004% faster. If we need it to handle strings of length N with a dictionary of size M comprised of strings from length X to Y, tell me that, and we can plan (and benchmark) based on that.
What folks need to remember is the employers who are primarily known for asking this generally have no problem hiring people who are good programmers and ALSO good at e.g. Dynamic Programming. And no, as much as some try to frame it, the two skills are not directly at odds with each other. Therefore, between a good programmer who can “Leetcode” and one who cannot, why wouldn’t they pick ones that are good at both?
And sure, if your company don’t have that luxury, don’t cargo cult. Hire a good enough domain expert for your specific needs.
Do you seriously think that claim is true? That coin toss predicts software engineering ability as well as leetcode? I don’t think any reasonable person can take this seriously.
Among those who pass, the two components might seem negatively correlated. This is because you can score a 10 on one section and pass with only a 5 on the other. However, among the general population both sections might even be positively correlated.
It's an extremely simplified but recognizable version of a real problem: natural language input tokenization, which of course has real consequences.
I could ask a detailed question on cryptography, robotics or reverse-engineering, and it would be just as (ir)relevant to the SRE. Computer science is a much broader field than your niche area. Arbitrary goal-posts are arbitrary.
I only know 1 guy who did coding competitions. He was in fact a good developer.
Why would you think a person that enjoys such a thing wouldn't be?
Its just another signal of aptitude, its not some magical catch all heuristic. However, people here suddenly expect like they are allowed to forget all the (very fundamental) basics and get into a job that involves coding without any coding at all. Now that leads to a bad hire.
Isn't that in itself is a flag that this is in fact not a great interview question, unless it's being used for hiring candidates fresh out of college, say during campus interviews.
It's a whole other debate that most software engineers, even at BigCo end up writing CSS code to move pixels only and lose their DSA knowledge over time...
They probably did know how to code, but they’ve spent the last 5 years working on something quite a bit different from string wrangling in Python. Add in the stress of a (first-time?) interview and a reasonable candidate can look like a total fool…
It's a bad question because _how well_ a candidate does past that minimum threshold has _no correlation_ with their ability to do the job.
You can treat it as a binary question for that minimum floor. To use it for rating, you're doing your org and the candidate a disservice and you're wasting interviewing time that could be spent more productively and humanely.
Is this common at Google?
If by DSA knowledge you mean leetcode puzzle knowledge, yes, quite certainly. But from my experience fresh grads are less effective engineers before they mature by working with industry scale systems for a few years. And at that point they would have been less able to do your problem but a far better engineer.
Which means your interview question is aimed backwards, maybe if you only used it against fresh college grads then the gain function would be positive, but you applied it against everyone.
Actually the best questions I ask, and the ones that give me the best candidate "signal" are what I call the "clean code" style questions (as in code that is simple, easy to extend, not necessarily Uncle Bob Clean Code).
I'd rather not post examples here because I still use these questions, but the theme is as follows:
* These questions ask to implement functionality that you could see in a (super simple) real world application.
* They start super simple, and include several follow ups that increase in complexity.
* No data structures and algos beyond the basics. Loops, lists, hash maps and little else. What most programmers would use on the job on a daily basis.
I found in practice these work really well, specially with experienced programmers who can flex their low level design skills. People who ace leetcode style questions tend to do worse here, but no wonder, since they've been trained in a complete different style of interview.
Key point: I've never met anyone in the industry whose job it is to implement algorithms without reference materials all day long. It's more like... you might find yourself needing 1-2 clever bits a year, and the rest of the time you're tweaking model classes and hunting down UI bugs.
Algorithms are both easy to Google and these days, CoPilot will do the hard 4/5ths for you if you actually need them. It's also common that the really intense stuff is written by the PhD braintrust cofounder, not the people being hired after the fact.
What I want to know when I'm interviewing is simple: how do you approach solving problems; do you have an interesting personality that I'd like to interact with every day; are you the sort of person who has a strong instinct about what the next thing to do is, and will default to working on that thing without prompting?
Do you think learning new things is exciting? Are you able to admit you don't know something and do you feel comfortable asking for help when you're stuck?
Do you seem reliable? Are you a starter, a finisher or both?
Do you have a sense of humour? What do you do on the weekend?
Ultimately, I care what you know... but I am far more interested in your capacity to quickly master new things.
> Do you have a sense of humour? What do you do on the weekend?
It's none of your business what I do on the weekend. It doesn't need to be similar to what you do.
Otherwise you'll tend to only hire people like you and reduce the pool of talent because you can only work with a very specific type of people.
My intention for the question is not to judge the answer, or compare it to what I like to do; the important issue is that they have an answer!
I cannot tell you how many folks I've interviewed (and not hired) who, when asked about their hobbies, immediately start listing programming languages and frameworks. This is not the answer I'm looking for.
In fairness to me, there's an almost infinite number of great answers to the question. You can say "music"! I will really respect someone who tells me that they like television. That person is honest. Cooking! Model rockets. Literally just say anything other than a bunch of acronyms.
I'm specifically not asking about which church or sex club they go to, and that's made contextually apparent from the conversation leading to that point.
I have a strong belief that a well-rounded candidate has interests beyond coding. If they live to code, they probably aren't going to be a good fit, so I save everyone the trouble.
That's not what the author's question was about. Rather, it presented a problem, and wanted to see some approaches to that problem and a discussion of their properties (runtime, limitations, etc.).
> how do you approach solving problems > Do you think learning new things is exciting? Are you able to admit you don't know something and do you feel comfortable asking for help when you're stuck?
All very reasonable criteria. So how do you evaluate a candidate on those metrics, if not by giving example problems? (ideally different ones: a few theoretical ones like the one under discussion here, a few practical ones with access to google/Stackoverflow/etc.)
> do you have an interesting personality that I'd like to interact with every day
That's your prerogative, of course, but I wouldn't make that a hiring criterion. I have colleagues who are incredibly boring, yet deliver solid consistent results. Who cares about their personality?
I suppose I perceive the narrative differently; it really sounds like the author has a complex roleplaying scenario in mind, where there's a base assumption that he is the smartest person in the room and pre-plans 3-4 moments where he can assert this. If I was the candidate, I would be praying that this guy isn't my direct boss.
> So how do you evaluate a candidate on those metrics, if not by giving example problems?
Simple: I talk to them and get them talking about how they have solved problems in the past. I will dig in on interesting points and take careful note of telling details.
Perhaps this will help: I have enough experience as a developer that I can tell very quickly where the person I'm talking to falls on the skill spectrum. Part of my negative reaction to this post is that I sense the author does as well; however, where he chooses to make them feel small, I just keep asking questions because I want to give them more opportunities to sell me on their strengths. Their ability to whiteboard leetcode in an interview has laughably close to zero relationship to how they will perform on a Wednesday in jeans without feeling like they are being interrogated, so I largely skip that.
If you cannot quickly tell someone's experience (or BS level) from talking with them, then this interview style will not work for you.
> Who cares about their personality?
We have to agree to disagree here. Personality is huge, probably just behind potential. In real terms, if I am going to spend more time with my coworkers than I am with my life partner, they better be trending towards great company.
One of the questions I ask myself during an interview is whether I would want to be in a band with this person. If I wouldn't enjoy driving across the country in a van with them, what makes me think I would be able to do my best work if I'm trapped doing it with someone I don't like?
There are enough brilliant nice people in the world that I can comfortably avoid hiring brilliant jerks and come out way ahead on every metric.
When you never did any hiring, you assume you're mainly looking for good candidates. When you actually have the experience, you know that at least 30% of applicants can't even code the most basic thing.
So I was wondering how you, hiring more than 100 developers, weeded out those candidates who "do the talk but can't walk".
Another thing: how many of those 100 developers you hired were bad hires?
Perhaps this will help: I have enough experience as a developer that I can tell very quickly where the person I'm talking to falls on the skill spectrum. Part of my negative reaction to this post is that I sense the author does as well; however, where he chooses to make them feel small, I just keep asking questions because I want to give them more opportunities to sell me on their strengths. Their ability to whiteboard leetcode in an interview has laughably close to zero relationship to how they will perform on a Wednesday in jeans without feeling like they are being interrogated, so I largely skip that.
If you cannot quickly tell someone's experience (or BS level) from talking with them, then this interview style will not work for you.
I have had a dud rate of about 8-12% over the past 15 years. However, only 1/3rd of those were duds because they fooled me. It was almost always personality issues that I didn't anticipate; not getting on well with other people on the team, etc.
One of those that fooled me came in hot because he had started a popular testing framework, so I didn't dig. I fooled myself by accepting him based on the reputation of the project, not the person. Lesson learned.
However, I would like to turn the question around and brag a moment about how amazing some of the hires were. I hired the creator of Sass, a future Basecamp developer, several future founders. No whiteboards in sight.
The things they can tell me is basically their curriculum. What I found was working the best was showing them code and they they explain to me what's happening, and I ask some practical questions on that code ("what would happen if I change this?", etc)
I moved from your kind of inteview to a more technical "explain this piece of code to me, what would it do?".
But it's possible the kind of developers I see have different experience level than the ones you see.
I also still interview seniors (maybe 1 out of 10). Maybe I have to think about changing their interview a bit.
The thing is with frontend, you can get a senior that basically did mostly html and never serious programming. And so if you build a complex web app, you really need to see how far they can go with their development skills.
To be honest, it's super frustrating that we can't have that conversation without an army of bootcamp grads accusing us of elitism. Only a decade ago, it wasn't controversial to expect that people applying for highly demanding technical positions were ready to start on Monday.
I digress.
As to the senior HTML artist... I feel like that's on you to suss out in the conversation. Another one of my elitist views is that someone who can only do HTML - literally, text markup - is not a senior talent.
If you want to make sure that a candidate understands how an MVC framework works, ask them to talk about their thoughts on service objects, or when they use Redis vs Postgres, or whether they have used Sidekiq jobs to do cool things. You can interrogate them to a wild degree and check off all of your boxes while making it feel like you're just shooting the shit.
It's almost impossible to bullshit your way through explaining how you implemented a key-based authentication mechanism over websockets. Writing out the code on a board accomplishes nothing. I know how it's done, and I'll know in 15 seconds if a person shuts down, terrified, when talking about it.
This kind of 'let's watch people struggle with my favourite LC puzzle, which I think I know inside-out' just seems very inconsistent, not very predictive, and just quite flaky to me.
On the job, my very first response to any kind of problem that looks algorithmic is to google the hell out of it, even if I think I know the answer already. In future, it will be to get gpt to write it.
I think the actual challenges in real-world software are not algorithmic, but architectural and beyond, and then only after thoroughly understanding the problem in the first place.
> I continued to use [the interview question] because I got good signal from the candidates.
Working at Google, this is something I hear a lot: that some question, or other interview thing, is good for getting "signal" from candidates. But how do you actually know?
Like, I feel to definitely know a question is good you would actually need to track candidates after they start working and see if their performance on the question correlates with their job performance. (Unfortunately, it seems impossible to track candidates who don't get an offer.)
These are hard to do if you think of an interview question like a high school quiz or a college final.
As an example, the author of TFA clearly is selecting for people who have the patience to sit around and be told how smart the author is. That is the signal in this “great interview question”: candidate will suck it up for hours while I tell them stuff.
Responses are awesome because it gives you an idea of what this person's background is and what breadth of knowledge they have and what they think is cool or exciting. Maybe they start with a contact closing and signals being denounced, or an interrupt being triggered, or the OS and window manager, or the browser and DOM events, or HTTP or TLS or TCP or Ethernet or WiFi or LTE or routing and the tiers of network infrastructure or DNS or load balancing or proxies or web servers and tiers of API services or databases... It's can lead to some really fun technical conversation, follow up about how they learned the things they know and experiences they've had, and most of the time the interviewer learns something too.
That's why I really like this question - it's just such a cool jumping off point.
Any candidate who made it to an interview should be able to rattle off a bunch of stuff. An excellent candidate will have more depth than the interviewer on some portions.
The interviewer’s first job is to decide whether the candidate has a baseline-acceptable level of knowledge (harsh but true, good fakers make it this far). This question is fine but not great for that for the reasons you mentioned. However, it is best to go in assuming they do.
So then the interviewer’s next job is to decide if they are going to be a good teammate. Google is rife with nerds trying to out-nerd each other, finding out how that goes in the interview is an explicit goal. A great interview on this question will end up in a moderately deep technical discussion of some aspect where both the candidate and interviewer are knowledgeable, much like a discussion you would want to have with your teammate at work. And that is why this is a decent question, in my opinion.
of course, the whole point of the question is nobody you're likely to interview can fully answer it. maybe someone like jim keller could but is someone like him really going to know about css stuff? maybe, maybe not.
Though I imagine they could just ask "when was your last algorithms course?" (and possibly where, and what grade you got) and save a lot of time. Or they could just ask "what year did you graduate from (whatever school level)?"
But that wouldn't be as useful for making the candidate "jump through more hoops" and reinforcing compliance as well as status hierarchy - important things at most companies.
And some people may have cheated their way through their algorithms class, so maybe they can be detected if it was recent enough.
I would stop in an instant if someone could come up with a better option.
Good questions allow multiple possible solutions[1], allowing a candidate to arrive at a successful solution via many different paths while still evaluating whether they are going to be able to perform well at the job. I don’t think this question comes close to meeting that standard.
[1] By which I mean fundamentally different approaches rather than syntactic variants on the same basic thing
“Tell me about a software project you played a significant role in. What did it do, how did it work, who made the tech choices, what was your role in it, what went right, what went wrong, if you were to rebuild it what would you do differently, what did you learn?”
I know it’s really a lot of questions but it’s still just one in a way “talk to me, in a free form back and forth way about software development”.
If a grad has written no software, or has no capacity to talk software development with you, then the interview is over.
That said, I do think it's best to have the person relaxed and at their best, so anything to minimize stress or discomfort will show candidates at their best.
Having said that I have open source projects to point to. If that’s not enough for them I’ll take different job.
What if the exercise were something straightforward, simple and given before the interview? And, you had the option of instead discussing a solution of equivalent or greater complexity.
You generally want to cover a range of skills, and technical skills like the ones covered in the article are an important part of that.
That being said, I think you could get around this with careful prompting (ie, ask first "what is the Big-O for dictionary lookup with the respect to the number of letters" before asking the overall big O). But I think there is a bit too much wiggle room to make this a completely consistent question, too much variance candidate to candidate.
I like the problem a lot, but wouldn't feel comfortable using it.
But.
Hashmap lookups require calculation of a hash of the key. Calculating the hash of a key requires examining all of the bits in the key. If the keys are int32s, then the hash algorithm has to read 32 bits. If they're int64s, then the hash algorithm has to read 64 bits.
And if your hash map contains n distinct values, then at a minimum they must be drawn from a key space where each key contains O(log n) bits. If you want a hash map to contain more than a few billion values, you'll need bigger-than int32-sized keys to distinguish them, right?
So for any hashmap of size n, that means there's an O(log n) hash calculation in there. (especially when we are considering the asymptotic case where the hash map grows infinitely large... such hash maps will necessarily have infinitely long keys, too)
Yet we call hash map lookups O(1).
Because in practice we imagine a hash map has a key space of a fixed size (even though such a constraint would mean that the hash map has a fixed maximum size and therefore no asymptotic performance case since it can't scale to infinity).
By playing around with big-O on hash table lookups with keys of size n, the interviewer is skating dangerously close to this contradiction at the heart of big-O analysis of hashmaps...
But unnecessarily so, since the key space they're using is dictionary words, which have a maximum length, so their length isn't the n you're looking for.
(Note - for the same reason, big-O analyses of sorting algorithms don't usually take an extra O(log n) factor (in n being the size of the valuespace) for the time it takes to compare entries, even though obviously it takes 'longer' to compare two 64bit numbers for size than two 32bit numbers. Is it a hand wave? Probably not when you're dealing with numbers. Maybe when you're dealing with strings)
This is a great point. If you're sorting an array of strings, you would never add an extra m in that complexity analysis for the size of the longest string, in bits (not length, because unicode)
The OP seems wrong to make candidates do this
(Not actually purely an academic concern - this physical constraint is what drives the difference in performance between using data in L1cache, RAM, disk, or on remote networked nodes.)
So yeah, O(1) hashmap reads are a convenient lie we tell ourselves, which are really only true for ‘small hashmaps’.
The only reason I can think that the optimal solution was immediately obvious to me is perhaps because in a previous life I did a lot of work building convoluted application specific indexing schemes for data in a distributed KV store that had basically zero useful built-in secondary indexing schemes where piles of parallel GETs were about as "free" as just doing one, but chaining strings of requests back & forth to the client iteratively made for sad times. This meant that it was essential to be clever about what could be coded into key names and precomputed from application/use-case context to know what keys to fetch "all at once" to gather as much relevant data as possible with as few serially dependent requests as possible. One such use case depended on a modified C-trie as part of the solution, so I got very familiar with what kinds of problems they're for and what their limitations are.
Given that, I can't tell what that says about the question, about me, or about Google. What I can say for sure is that because I can't tell the above, the technical interview at my company is signing an NDA and sitting down for a couple hours and just actively pairing on real work & real problems with real teammates.
With the nature of software development work being what it is, I really don't understand why we as an industry run interviews like they're game shows or try to manufacture workplace simulators with terrible model conformance.
You could get lucky and have each question have a similar relationship to past knowledge. It's happened to me before, I blew through an Amazon interview once that way, but got knocked for technical leadership. So maybe you mean something like that, but that too is random. I had passed the leadership but failed the algorithms at Amazon a few years prior.
> I really don't understand why we as an industry run interviews like they're game shows or try to manufacture workplace simulators with terrible model conformance.
I believe these tests are more about compliance and baseline cognitive ability than anything else, which is good for the nature of the FAANG role. They are testing that you are willing to study what they tell you and that you are able to retain material of that level of difficultly and demonstrate that in an adversarial situation.
The reverse trie is genius, but with dynamic programming do you still need it? I would say that if in the general setting you can afford O(n^2) preparatory steps, and therefore you can just remember which (start, end) pairs correspond to a valid word with regular trie lookups. That is O(n) steps for each start position, giving a total of O(n^2).
One is the length of the word. The other is the number of dictionary elements.
If your language is constrained to 1,000 words, then the dictionary will be pretty small and the hash table can be constructed with no chains. That should have constant-time lookup no matter how long the query concatenation is.
I don't see why O(n^2) should be optimal. The question is only to determine if a match exists, and that can be expressed as a regular expression. In the following I omit all 1-letter words since my dictionary includes all the letters.
>>> words = [line.strip() for line in open("/usr/share/dict/words") if len(line.strip()) > 1]
>>> words.sort(key=len, reverse=True) # because Python uses left-to-right match, not longest
>>> pattern = "(" + "|".join(words) + ")+$"
>>> import re
>>> matcher = re.compile(pattern) # takes a few seconds
>>> matcher.match("ducksoup")
<re.Match object; span=(0, 8), match='ducksoup'>
>>> matcher.match("ducksquom")
>>> matcher.match("theremanywordsinthisone")
<re.Match object; span=(0, 23), match='theremanywordsinthisone'>
>>> matcher.match("theremanywordsqinthisone") is None
True
But with there are non-backtracking matchers which would make this O(n). Here's re2 from https://github.com/google/re2 :
>>> import re2
>>> opts = re2.Options()
>>> opts.max_mem = 200000000
>>> matcher = re2.compile(pattern, opts)
>>> match = matcher.match("ducksoup")
re2/re2.cc:821: DFA out of memory: pattern length 2493008, program size
1323834, list count 1089251, bytemap range 54
>>> match.start(), match.end()
(0, 8)
>>> match = matcher.match("ducksquom")
re2/re2.cc:821: DFA out of memory: pattern length 2493008, program size
1323834, list count 1089251, bytemap range 54
>>> match is None
True
Anyway, here it is processing the concatenation of 100,000 words selected at random (I've omitted the warning message):
>>> s="".join(random.sample(words, 100000))
>>> len(s)
956249
>>> import time
>>> t1 = time.time(); print(f"is concatenation? {matcher.match(s) is not None} " + f"time: {time.time()-t1:.1f} seconds")
is concatenation? True time: 12.3 seconds
>>> t1 = time.time(); print(f"is concatenation? {matcher.match(s+'Q') is not None} " + f"time: {time.time()-t1:.1f} seconds")
is concatenation? False time: 12.3 seconds
>>> s="".join(random.sample(words, 10000))
>>> len(s)
95457
>>> t1 = time.time(); print(f"is concatenation? {matcher.match(s) is not None} " + f"time: {time.time()-t1:.1f} seconds")
is concatenation? True time: 1.3 seconds
Edit: if I increase max_mem to 2_000_000_000 then I don't get the warning message. But then the DFA-based search takes 141.5 seconds. That should be O(n) time in the query length, but an order of magnitude slower than the NFA-based search.
That's why O(n) analysis isn't enough to really figure out "a faster solution."
In the common case, where s is short and the dictionary is big, the size of the dictionary probably matters most to the running time. s is likely a cache line and a handful of instructions to compare — whereas the dictionary probably doesn't fit in L1 or L2, maybe not L3. Obsessing about algorithmic analysis based on the string size instead of the size of the dictionary gives good signal for whether the candidate is a pedant, and poor signal about engineering.
> I then push them on this until they realize that any implementation of the dictionary lookup function of a string would be o(n) at best, and as such the total runtime of the solution is O(n²).
I think the point is in the initial algorithm given there are (n-1) lookups of two substrings whose combined length is n, requiring the hash function to process (n-1)n characters, because the hash algorithm is O(n) in the query string length.
That said, the dictionary words have a finite size. The largest word in my /usr/share/dict/words is 24 characters - formaldehydesulphoxylate. That means for an input of size n only 24n bytes need to be processed to identify if it's composed of two dictionary words. And if it's 49 bytes long, or longer, it's a simple calculation - "not".
Under Big-O analysis, this is O(1) time.
The lazy DFA is technically O(m*n) just like the NFA. The "lazy" part of the "lazy DFA" is why. Because it builds itself during the search. A real DFA would be truly O(n) because it would be fully built before starting a search.
re2/re2.cc:821: DFA out of memory: pattern length 2493008, program size 1323834, list count 1089251, bytemap range 54
is concatenation? True time: 102.0 seconds
In any case, my point is that we know this problem can be solved in O(n) time.
And yeah sorry, I was only commenting narrowly on RE2 based on my knowledge of its internals.
But then, that means that my candidate string can only be a maximum of twice the length of the longest dictionary word, so the entire process is O(1).
For realistic English words, we're talking about a typical candidate string of length, what, 20 maybe? I need to do somewhere between 19-38 dictionary lookups of 1-19-letter strings to test every splitting of such a word. Why are we even doing big-O analysis here?
Only when you start looking at arbitrarily long strings to see if they are word sequences does this analysis even make sense.
Is that true?
I am actually siding towards the candidates on this one
presumably it would be O(n_words_in_dictionary) for the dictionary check, which as the number of words in the dictionary is constant, would make the overall algorithm O(n) ...
Any better than that you can do would be O(min(n, k)) where k is the length of the longest word in the dictionary. But without loss of generality that would be O(n).
Therefore, either the question is nonsensical, or there must exist a word (or words) of a maximum length.
You take the first n bytes of your input string. You hash them. You look them up in the dictionary. The lookup is O(1) but hashing your input is O(n).
But there's a simple way around this.
We have incremental hash functions. When you add a byte to the string and you use an incremental hash, you do a fixed amount of extra work to update the hash. Like that you only need to O(1) extra work to compute the new hash.
One trivial incremental hash is to break your input into fixed size chunks, hash each chunk, and xor the chunks together.
A trie is a pretty nasty datastructure memory wise. It involves chasing pointers all over the place. The constants on a trie are not going to be very good.
For an incremental hash, you hash your string, and then you fix your hash up as you go. Then you look up your hash in a hash table. Code-wise it's far simpler to implement. And performance wise it will be far faster!
The two solutions have nothing to do with one another. Except that they're both O(n).
Hash collisions don't matter unless they're pathological. Just choose an out of the box hash that works well.
The article is just wrong.
If you're writing a smug blog post about your "great" interview question, you should at least not make such "obvious" mistakes, especially considering rolling hashes are a "common" interview question trick for string problems. And in practice this hash'd solution probably might end up performing better than the Trie (of course needs benchmarking, etc., but it's not unreasonable to think it might). In an alternate universe you could easily imagine author himself getting failed for overlooking such an "obvious" solution.
The maximum number of iterations is length of the largest word in the dictionary.
No numerical scale, you either want someone in your team, or not.
And also:
"What's your favorite algorithm or data structure? It doesn't have to be the fastest or best, just your favorite? Why?" (This does have one right answer: shellsort.)
"Tell me about a time you made a mistake — not a time where one of your weaknesses turned out to be a strength and you saved the day. No. Tell me about a time you really screwed up."
Unless you're an insecure, narcissistic engineer who feels "smarter" than everyone else or a cargo cult follower of some business fad, there are no magic questions to instantly separate "good" candidates from "bad."
You discover good engineers by whiteboarding and working on challenging problems together than can possibly turn into real commits. Ego and solving manhole cover problems doesn't delver good commits.
> Unless you're an insecure, narcissistic engineer who feels "smarter" than everyone else or a cargo cult follower of some business fad, there are no magic questions to instantly separate "good" candidates from "bad."
without taking a moment to actually think why these type of interviews exist.
> You discover good engineers by whiteboarding and working on challenging problems together than can possibly turn into real commits. Ego and solving manhole cover problems doesn't delver good commits.
You think everybody got time for that? Maybe in your company you get one or two CVS, but when you have 30 it’s impossible to conduct thorough testing. You optimize for false negatives.
Bad news: You have to walk this stable genius through 50 years of Suffix Tree and constant time LCA algorithms in 20 minutes.
Fun write-up though. The comment saying this optimizes for clever freshmen is spot on in my book.
abbbbbcdddddde => aB5cD6e
The number of developers who struggled or utterly failed with this was depressing.
That said, from the article this article links to, I find this puzzling:
> The candidate’s performance on the problem isn’t binary. The worst candidates don’t even manage to implement the fizzbuzz solution in 45 minutes. The best implement a memoized solution in 10 minutes, allowing you to make the problem even more interesting, e.g., asking how they would handle a dictionary too large to fit in main memory. Most candidates perform somewhere in the middle.
Two things I worry about: if the interview isn't "binary", how do you decide whether the candidate passed? Is doing just the simpler version fine as long as you ask for low pay? Or are you still failed? Or does it depend on whether other candidates did better, which is another way of saying you failed? It seems to me it's still binary.
And the other thing is: if the candidate did great in 10', please don't ask additional questions. The candidate is a "hire". Don't make it "interesting", an interview is not the place to shake things up to make it "interesting". Understand that for most people, interviewing is a stressful situation. For "interesting" times they have their friends and Netflix.
Realistically, what are the chances of someone who solved everything up to the half-way point?
When I did interviews, I did a problem in a kind of similar way (although I skipped most of this article, so I dunno). I'd give the same problem to all candidates, but expectations were higher for candidates with more experience.
For any candidate, there was a minimum bar (essentially needed to be able to program a loop that did what they said it would do), and failure was a very clear signal; although I had a couple candidates that the only signal I got was anxiety.
How much over the minimum the candidate got would help decide between weak to strong hire. And then there's a group decision based on the results from everyone.
> And the other thing is: if the candidate did great in 10', please don't ask additional questions.
Part of the job as an interviewer is to fill the whole time though. If we're done in 10 minutes of a 45 minute slot, that's a lot of time to fill. Usually the next interviewer is busy until the next one. A candidate doesn't usually have 35 minutes of questions, and a trip to get snacks or drinks, etc doesn't take that long either. At that point, you just have to ask them about projects and see if they've got anything they want to talk about --- but since that's not my interview role, I won't judge that.
I did have one candidate who I was the last interviewer of the day and they didn't like my question and asked for something else and wouldn't participate at all. We were done in under ten minutes, so I just walked them out and we both got 35 minutes of our life back.
Right, I forgot this is Google and there's an in-person gauntlet to run! I would probably fail it just because a single interview usually leaves me drained.
Let me ask another question, can an engineer who solved everything in 10' screw up badly during the follow up questions which get asked to "make things interesting"? Even if this is unlikely, if it's at all possible it seems to me the optimal strategy of a good candidate is to stretch their answers as much as possible to prevent the introduction of followup "interesting" questions...
I've seen local candidates get interviews split over two days. You could probably ask for one split over several days, but not if the company is paying for a hotel. I dunno.
On my question, all the follow up work after the basics wouldn't make the rating go down. Other than like any sort of behavioral issues that came up; but I don't remember seeing anything like that; even from people we hired who later exhibited issues; they were able to keep it together during the interviews anyway.
The risk of intentionally stretching answers is that it might look like you've seen the problem before and are trying to hide it. My problem (and this problem) aren't really unique, although I didn't see too many people who had seemed to see it before; and my focus wasn't on the highly technical bits; the basics are quite simple, but there's a major downside to the simple solution with certain inputs, there's lots of ways to address that, I just want the candidate to come up with anything, it doesn't need to be the best, they don't need to know if it's the best, or how to know if it's the best, it really just needs to be complicated enough that the first code sample is more than 5 lines long; if we really ran through the problem in 10 minutes, we can spend some time exploring alternatives, but I'm not judging someone poorly because they did their way and can't think of seven others.
The maximum number of iterations is the length of the longest word in the dictionary.
This means that part 1 of the problem can be solved in O(1)+O(1).
That doesn’t mean you will be happy with it. Not all O(1)s are equal.
Big in space but O(1) (wrt word len at least)
The only way to evaluate if this is a good or bad solution is the actual operational context.
If you had a service that had to do millions of these matches per second with low latency, then this might be a reasonable solution.
I found one sentence very interesting:
> Moreover, the problem as I present it, has enough range to sift out bad candidates from good ones even if the candidates have seen the problem before, which is why I successfully used it even though it was banned at Google.
How do you measure the question's success? Is there some kind of feedback you get about how the person you made a decision about went on to perform? What about the people that ended up not being hired?
Not the latter. However, there are five interviews that each candidate goes through, and as an interviewer you get to read the interview feedback written by the other interviewers as well. So I can compare my assessment of a candidate against four other interviewers' assessment using their questions and feedback notes and rating assigned.
Is n the length of the word or the length of the dictionary? And if it's the length of the word (<< the length of the dictionary), why does it matter if it's linear or quadratic?
This is a negative hiring signal you're looking for, right?
Is any of this better then randomly selecting? Do we even know if there is a skill set that is optimal, or if multiple complementary skill sets work best? Do we know what they look like, how to select for them? Or are we all just filtering for whatever collective definition of cleverness people involved in a given set of interviewing and hiring tasks prefer?
I suspect on average it's better to filter out good candidates than filter in bad ones so as long as a selection process skews towards eliminating most candidates probabilities will have most teams covered on that front. Of course this only works when salaries in programming are rising faster then the rest of the labor market, attracting a growing cohort each fiscal cycle.
What is the runtime complexity of an algorithm? Set it up with varying amounts of data fed to it (or whatever N is) and then time the real thing (or a suitable minimal but complete example).
Then you can instrument or profile the algorithm in any number of different ways to ask what is being called so often.
If you do it right you can really measure what it is doing, and you're getting closer at least to "production" code where you're not just working with pencil and paper but you've got real algorithms running hardware with finite sized caches.
Ultimately it doesn't matter what anyone writes down on pencil and paper if the hardware doesn't agree with you. Usually I measure it first and then work backwards knowing the answer already.
I've never worked at that company and I don’t know if as an interviewer you pick a question from a catalog. If that’s the case, and if you’re so inclined in asking this type of nonsense, I’d suggest a small tweak: a third party selects the question with some sort of guarantee/constraint that this would be the first time the interviewer asks that question, and so is now in more or less level ground with the interviewee. No other option but to have a discussion among potential colleagues.
For starters, it's staggering how many people can't do it.
But also, it's a great question because it's simple enough that you don't have to be clever but you can get a feeling for all sorts of things. Do they ask questions? Any questions? What possible inputs might cause problems? What do they name their method? What tests do they write? (Project is set up already with an example test class ready to go.) Etc etc
The problem as described:
“Write a function that given a string, determines if it’s a concatenation of two dictionary words.”
given that description I think you can't have left over fragments. But the code seems to allow fragments (don't know Python). Also it just returns a boolean? So we just need to know if there is two words contained within our word within the dictionary.
If that's the case, and the "dictionary" is an actual dictionary used by humans (they mean like a Dictionary like The Oxford American Dictionary right?), then, because the letters of a word are considered nouns (that is to say the letter b is a noun - Merriam-Webster https://www.merriam-webster.com/dictionary/b) then any word with more than one letter is a concatenation of two words. Which is anyway stupid because no one cares about concatenations of words, people care about compound words and this is not how you would find compound words.
I bet that's not what they want to hear though. I generally do really bad on questions.
"lampshade" => true
"grip" => false
"gripshell" => true
"fhtagnblue" => false
"timeblue" => trueFor example it looks to me like it thinks fhtagnblue which has the words tag and blue in it should return true? Maybe I'm reading things wrong - but if it should return true, thus allowing fragments, then every word with two letters or more would return true because every letter is a noun.
If it is as you said, which might be actually interesting thing to do and potentially even useful, I might not want to do it with a query this dictionary for a string and instead have actual access to the dictionary to work with since I figure an optimized solution can be made with binary search.
on edit: I mean considering Google's reputation for trick questions and logic gotchas I would probably just say "wow you guys really are like that" and write something like
if(!s) {return false;} s_array = s.split(""); return s_array.length > 1;
on edit: hmm, a number is also a noun so I guess if query(9) or query("9") worked wouldn't have a problem, but then if it didn't I would have to say hey your dictionary is broken and then there would probably be discussions. But since it is all just hypothetical I suppose I could assume that the dictionary isn't broken and so any trimmed string with more than two characters in it would still return true.
You’ve already lost me, dude. If the question can’t meaningfully differentiate between a junior and a high level senior, it’s a trash question.
The customer wants to increase their sales. They want something that auto-generate sales responses to incoming emails. Discuss.
The problem with technology workers is that they believe technology is important. The above question allows you, the interviewer, to see if they have actual critical thinking skills.
Annnnnnnnd I’m out.
Some come up with a greedy algorithm which finds the first prefix that is in the dictionary, and then returns whether the rest of the string is in the dictionary
I guess this wasn't entirely without sense but sort of seemed to be missing the point.
Good taste — writing cohesive, loosely coupled code with a sensible bank of unit tests an a strong dose of product sensibility — is largely orthogonal to being good at solving algorithmic or ds questions. One of the best developers I know, who makes seven figures a year a year (trading software), would likely bomb the second part of this question. Contrawise, a young researcher I know who has won multiple scholastic programming competitions and crushes leet code, would destroy this problem in seconds … and yet, I pity his coworkers who have to live with his atrocious code.
More accurately, he would first look at the interviewer like he was a dolt, and then bomb the question :)
How can a word be the combination of two words if the first word isn’t a word?
So, for example, if the first letter isn’t A or I, we can stop doing lookups for any combination of just a single letter and the remainder of the word.
Etc.
Th isn’t a word, next
The is a word, look up ‘reis’, not a word, next
Ther isn’t a word, next
There is a word, look up ‘is’ it is a word, return true
If you don’t want to ignore whitespace, than I’m not sure what behavior you want, you want “there is” to return false?
If whitespace is the delimiter the whole thing is trivial, you split on space.
I am now more confused.
This is where the wrong greedy solution from TFA, which the author claims some candidates reach, stops. So it's wrong.
Note that if you do not stop, and continue as in your own example, your solution isn't "greedy"!
Or at least sorting by.
If they only know simple data structures, going from a hash to a hash of hashes is one of those "eureka" moments that can really shows how a candidate is willing to collaborate on a problem and react to new ideas.
Heaps are an interesting data structure to use for interviewing, but I do find they can be a little less intuitive, and their use cases translate less directly to business logic and are more of an optimization task. It depends on what talents you're looking for on your team!
https://iq.opengenus.org/time-complexity-of-hash-table/#:~:t....
The trie/reversed-trie solution is a little too clever to arise naturally from conversation (and probably not optimal because hardware implementation).
I then push them on this until they realize that any implementation of the dictionary lookup function of a string would be o(n) at best, and as such the total runtime of the solution is O(n²).
Surely it's
String -> Hash -> Is hash in hashmap
All of those operations are O(1) so the total operation is O(1)
Well, in theory you may have conflicts in the bucket.
And recovering the bucket means a round-trip to memory, which is much more expensive than running over a list.
And in any case, "n" is the length of the word, but has nothing to do with the distribution of the dictionary. So one should use a "m" to capture that.
The list of dictionary words isn't that big, so you can just have a long hash to reduce your conflicts to almost 0.
Also words aren't that long. The longest English word is probably less than 100 characters, whereas you could feed trillions of strings into the algorithm.
I agree with you that this sounds weird, but if you imagine arbitrarily long strings as inputs, and don't want to have a lot of collisions, you need to take into account all of the input string in the hash function.
In reality they are safely assumed to take O(1), but w/e the interviewer is looking for a specific answer.
However, the whole discussion in the piece suffers from not taking into account the fact that words don't get arbitrarily long. Thus, insofar that for all practical purposes n <= 30, say, it is indeed O(30) = O(1).
This seems pretty nitpicky, especially considering we're assuming substring creation is O(1). And isn't a hash table lookup of a string O(n) at worst, not at best? And since this seems like a fixed dictionary, we're only doing queries not insertions, couldn't you just choose a different hashtable algorithm like dynamic perfect hashing to guarantee O(1) lookups?
I also don't like how it starts off with the dictionary being a blackbox style function, and a later part of the question requires you to change the internals of this dictionary query function. My very first thought would be about the structure of the dictionary query. And then when told to ignore that, I would assume that's still the case when asking to later optimize the algorithm.
If it has well-defined success state then it's maybe OK "implement a maze-solving algorithm for a 3^n cube". If there are implementation tests that the candidate can use, even better
To prove that it's optimal you "only" need to construct the pathological case, which I believe should not be hard to do by induction.
I think that pathological case could simply be that the input string is abcdefghij... and the dictionary contains all one-letter strings, plus all strings like ac, abd, abce, ..., bd, bce, bcdf ... and so on.
The base induction case would be xyz as the input and {x,y,z,xz} as the dictionary.
The problem is function f that gets a list of words D and a word w of length n and returns a list of indices I where to split w such that each part is in D or "false".
Cleary, I has length at most n and the indices in it are also at most n. Also, you need to look at every word in D. Thus, the trivial lower bound of the complexity for an algorithm that computes f is O(n+|D|).
To prove a tighter lower bound, you would need to show that every algorithm that does it in in linear time has a bug - you need to find this "pathological" case for every imaginable algorithm!
Afaik it is still unknown of SAT has a quadratic lower bound! https://cstheory.stackexchange.com/questions/17789/is-there-...
Not surprising. This sort of interview is selecting for people who have exactly this kind of academic knowledge.
Well, that removes most of the fun of it.
I think that’s actually a good example of why this is not a good question. It’s disconnected from any realities.
if i was (am retired) i would typically ask questions about the languages we were using. in my case this would typically be c++ and sql. so i would normally ask things like:
"tell me something about the copy constructor"
or:
"when would you want to do a union?"
amazing how many people were dumbfounded by these. and notice, no leetcode crap needed.
However, software engineering is so much more than writing code to solve a well specified problem.
- oftentimes, the problem is not well specified. The engineer needs to find out what the problem specification really is.
- In a lot of cases, a solution already exists publicly. The engineer needs to find documentation or code that solves the lecified problem or a similar problem. Usually somebody has already thought about optimization of this type of problem and has documented it.
- Often, a feature needs to be added to an existing codebase. Often a similar feature already exists in the codebase and the new feature can be implemented in a similar way. The engineer needs to find the similar feature and has to figure out how to generalize or adapt it into the new feature.
- A lot of times technical debt needs decreasing. This involves understanding feature implementations or models/paradigms in an existing codebase, understand how they are inefficient or needlessly complex. Creativity is the trait mostly needed in this case to address it.
- often new APIs need to be made, needing modeling and documentation. Also often existing APIs or newly imported libraries need to bevused to tackle problems.
- Above points address the purely technical skills. Non-technical skills are just as important. Communication, teamwork, prioritization, time and stress management, customer relations,...
Tldr - this great interview question targets just one aspect of the job.
[1] This seems to me to be a practical limitation rather than theoretical given strings and numbers are really equivalent - it’s just that the space is mindbogglingly huge so if you wanted to do a general purpose O(1) string to number conversion it would require uncountably infinite space.
With performance like that, I'd call it a "trash interview question".
When you use wildly inaccurate processes, you build bad teams and good people know you'd be wasting their time so they don't even go in the door.
Seems like the practical application of this is to quickly disallow strong memorable passwords of the form "correct horse battery staple". https://xkcd.com/936/ If so, deeply misguided.
- part 1: got it in like 20 seconds of thought
- part 2: got a variant with a trie in maybe 5 minutes thinking, wasn't exactly the same as the given one but close enough
- part 3: thought about it for an hour (got nerd sniped). got a trie + recursive dynamic programming solution that's O(N^2) in the size of the input string. the actual solution took me 15 minutes, the rest was trying to prove the the big-O
Let's say the string is wxyz. As we iterate, we'll split into the possible prefixes and suffixes:
w xyz
wx yz
wxy z
wxyz (suffix is empty string, we have to special-case this I think)
Looking up a string in the trie is O(N). But for the prefixes, we can "remember our place" in the trie, so that the cost of an incremental character is constant. So membership checking wxy only incurs the cost of y, given that we've already just looked up wx.
The prefix is either a word in the dictionary, or it isn't. If the prefix is not in the trie, we advance to the next character and repeat, with the next prefix grown by one character and the next suffix shrunk by one character. If the prefix is a word in the trie, we call our function recursively on the suffix, and memoize the result. If the recursive call comes back true, we're done, and we return true. If it's false, we advance to the next possible prefix and repeat. If we reach the end, the prefix now comprises the whole string, so we just return whether it's a full word in the trie or not.
In the base case of the recursion, the suffix is a single character, which is a constant-time lookup in the trie. So every recursive call after that can just look up that memoized result. Same for the 2-character suffix, 3-character suffix, and so on:
- 1 character: z: z: 1 operation
- 2 character: yz: y, z suffix memoized, yz: 3 operations
- 3 character: xyz: x, yz suffix is memoized, xy, z suffix is memoized, xyz: 5 operations
- 4 character: wxyz: w, xyz(memo), wx, yz(memo), wxy, z(memo), wxyz: 7 operations
- N characters: prefix 1, (N-1)(memo), prefix 2, (N-2)(memo), prefix 3, (N-3)(memo) ...: 2N operations
So summing the cost of all those recursive steps, it is O(N^2) in the length of the string.
The word "cantonal" works as a good test case because it has lots of sub-words: can, cant, canto, canton, a, an, ant, anton, ton, tonal, on. The memoization doesn't show up for it though, so I'll append "q" to make "cantonalq", which should return False:
c -> no
ca -> no
can -> yes
t -> no
to -> yes
n -> no
na -> no
nal -> no
nalq -> no
ton -> yes
a -> yes
l -> no
lq -> no
al -> no
alq -> no
tona -> no
tonal -> yes
q -> no
tonalq -> no
cant -> yes
o -> no
on -> yes
(-alq suffix is memoized) -> no
ona -> no
onal -> no
onalq -> no
canto -> yes
(-nalq suffix is memoized) -> no
canton -> yes
(-alq suffix is memoized) -> no
cantona -> no
cantonal -> yes
(-q suffix is memoized) -> no
-> return False
Generally I don't mind "pure" DS/algo interview questions, so long as they're plausibly related to something one might actually do. I have actually used tries before on a real problem. Dynamic programming with memoization tricks? Never. Has anyone ever had something remotely similar come up in real work? I mean I know there are caches everywhere but that's not the same thing. It just seems like a sick hazing ritual that nerds inflict on one another.
Conclusion: bad interview question.
Maybe it is understandable since the article does not clarify any of this, but folks should notice that:
- This type of interview is part of a process, composed of multiple interviews. Not the only one, and not the deciding factor. Questions on behaviors, team dynamics, etc, are also part of the process and covered in other interviews (usually done by managers). You could fail a systems interview and still pass the interview process. The specific interview mentioned in the article seems not too complex, though, so I wouldn't be surprised if it is a screening interview that rejects candidates when not passed.
- People mentioning that the interviewer is wrong on time or space complexity are also missing the point. It doesn't matter. If someone in the interview mentions O(n) instead of O(n2) and can reason that, that's already a good signal. It is not so much about being right or wrong. It is about thinking about those things when coding, keeping them in mind, been able to express that and run someone else through your thought process, etc. Someone plainly saying something is O(n) without any further explanation can give a bad signal, but someone explaining why she thinks so can be a good signal, even if ultimately the solution is not O(n).
- IMO, folks saying that this favors junior engineers and people fresh out of college are wrong. This favors people who ultimately understand the foundations of programming. People fresh out of college had a refresher on that not so long ago, so they've been practicing it more. The thing is, some people in our industry get further away from that as they add seniority to their career, but there's also lots of senior people that are very well versed on this kind of thing. Specifically, the question in this article is extremely simple, so TBH I don't understand why someone senior would fail to know the answer to that question just because that person is senior. Maybe these companies are just not looking for that kind of senior engineer for this specific positions. That's all. There's also a lot of different seniority levels, too. I could see this interview having less weight on a decision as seniority increases, but still important.
- "this does not represent a realistic scenario". That's a given. As a company, you have maybe 8 hours total with the candidate (multiple rounds of interviews plus the on-site). Onboarding into the company and a team, until you're productive, will take weeks/months. There's no way of testing a realistic scenario on an interview. We get the best signal we can with the time we're given.
- Not passing the interview process is very demoralizing, but it does not mean the company thinks you're a bad engineer or that you're not good enough. The process is designed to minimize false positives, so they tend to prefer to reject candidates if there's any question or mixed signal in the interviews.
- I'm not saying the process is perfect, by the way. For example, there's also luck involved, too. It shouldn't, but there is. To give one example: interviewers are people too. They need to be trained in the process of interviewing, what signals to look for, etc. There are good and bad interviewers. Interviewers can have a bad day and judge some answer as unsatisfying when some other day it would be fine, etc.
>The fastest I had a candidate solve the entirety of the problem with all bells and whistles was in 20 minutes, by a freshman from the University of Waterloo, the type who did high-school competitions.
>The most depressing failure I saw was a PhD graduate from University of Toronto who could not produce working code for the first section of the problem in 45 minutes.
>I once had a candidate with 15-years work experience give me lots of attitude for asking them such a simple question, while at the same time struggling at it.
All of this data points to the fact that this question may not be good. A phd graduate and a person with 15 years of experience rejected for someone who practices programming for competitions? What gets me is that the data is painting an obvious picture here. A very obvious picture. An obvious picture that we aren't sure what's a good interview and a bad interview question.
But the problem is that most people when looking at this completely miss it. It's not obvious to the interviewer and it's not obvious to alot of people who like google style questions. We literally have not much data and not much science backing any of this up.
It's an illustration of how biased humans are and illustration of how extra biased interviewing for software positions is. If there's anything more unknowingly biased and then the replication crisis in science it's technical interviews for companies. There needs to be real feedback loops that correlate interview question passing with Actual performance.
Google is in a good position to grab this data but I'm not sure they are doing so given how they just okayed this guys gut decision to go against the grain and use this question. I'm not against this question, but certainly to call this great in the face of controversial data that he himself gathered and listed on his post is just a complete blueprint of the extent of human blindness.
The reality of what's going on here is the person here in the interview is just getting off on dominating other people with a hard question. It's not on purpose but he's doing it without realizing it. The blog post in itself is a bit showy. It's like "I can answer this question but a phd graduate can't".
What I’ve seen happen is that there are two narratives they flip on: I hired this great person and they are doing amazing so I am amazing; I hired this person and they aren’t doing great so I fired them so I am amazing.
Of course, they also tend to be terrible at assessing who is actually good/bad.
I call it the “arbitrary onion”: layer after layer of plausible claims that each turn out to be just as arbitrary as the outer claim.
sure, some will trip over themselves.
20m was the best?
poor screener?
The sad reality was that it was waaaay too effective as a fizzbuzz eliminator (> 80%) which I don't know what it tells me about the candidates in general or just about our recruiting :-)
I suspect there are a lot more people who frequently choke on easy coding questions, under the very specific conditions of interview-stress (which is very different from, "the client is unhappy and is also onsite and you'll be in a meeting with them later" stress, and from "production broke" stress, so no, I don't think it's a good proxy for general "grace under fire", as it were, either) than there are actual full-on bullshitters who manage to land software jobs while being truly incapable of doing even the basics of the job.
For one thing, I think such top-tier, extremely-convincing bullshitters would have an easier and less-stressful time applying that skill directly to some other role, since there are some well-paid roles where that kind of thing is exactly what companies want, so it doesn't pass the smell test for me that so very many would be trying to land software jobs—not just ones who exaggerate skill or try to get into the industry while woefully unprepared, but who also are able to fool interviewers at a reasonably-high rate in conversational interviews, yet fail simple coding tests.
That is, I suspect a large majority of "ha! I caught a faker!" anecdotes or anec-data are actually a false-negative on the test from people choking under interview conditions, and that most real fakers would also have been caught in any half-competent conversational interview anyway.
The remainder should be probably be hired regardless, then re-homed to sales when you realize what's going on, since they're evidently top-percentile bullshitters and social chameleons. Joking, of course... kinda. Maybe.
(Top tier Ivy League university acceptance rate is 3-10%, so if you're looking for someone of at least that caliber among the general population (or even among developers), an 80% rejection rate might not be crazy. FAANG (or its ilk) are the "Ivy League" of Software Development jobs)
Over the years, I have seen significant numbers of applicants who couldn't write FizzBuzz, and even one who couldn't sum up the numbers in an array (ignoring overflow). Some of these people had reasonable resumes, and many of them could talk about software development in the abstract.
This was usually a sign of poor phone screening, but it always blew my mind.
Out of curiosity, how do you think you might do in a 1-on-1 coding interview doing something incredibly basic, like:
- Sum an array.
- Write FizzBuzz.
- Count the words in a string.
- Compute the length of a linked list.
Assume you get to use the language of your choice and a familiar text editor. You aren't being graded on semicolons or gotchas. And the interviewer isn't hovering too closely.
Because that's the situation that always mystified me. How many people freeze up under reasonably favorable interview conditions to the point that they can't write even basic code? (And would those people be able to pair on a tricky bug when production is down?)
I absolutely do read a candidate's online code, if they mention it in their resume. That only seems respectful. And I think it might be reasonable to waive in-person coding tests if someone is obviously competent.
(Of course, now that Copilot is getting smarter, it's going to be harder to evaluate people's GitHub repositories or to include trivial problems in pre-interview screens.)
In my 20ish-year career, I've had... oh, ballpark 30 interviews, and I am certain I convinced at least two of those that I had absolutely no idea what I was doing, one on something about as hard as what you list, another on something only barely harder. One was around the middle of that span, another, four or so years ago. The number would be higher, but most of my interviews have not featured coding questions. I have an incredibly high offer rate from interviews that haven't done those (70%? Maybe higher), and near-zero from ones that did.
I can, in fact, code. I've taken complex projects from zero to production, solo, with companies that couldn't have afforded to keep me on for years if I couldn't pull my weight. I've developed a reputation at multiple places as the one to go to with tricky or low-level problems, or the one to hand odd-ball problems with tech no-one's familiar with. I collaborate well, I do the solo thing well, I'm good in a meeting with stakeholders, I can take a support call like a champ. I figure things out, I deliver, I ship. I am not bad at this whole thing. But interview coding? LOL.
> (And would those people be able to pair on a tricky bug when production is down?)
I'm cool enough under pressure that it's been remarked upon throughout my career, over and over. Interviews still get me—specifically, the performing part of coding questions, the rest is no problem.
I think it's because they're kinda hostile (yes, yes, "we're both just deciding if we're right for one another" and "you're interviewing them too", but c'mon, see exactly all the chatter about interviewers trying to find liars and expecting there to be lots of them—you're under a microscope). Production-down is collaborative. Angry-client is collaborative (from your co-workers, anyway). Not just that you're working together, but supporting one another, everyone's got everyone else's back. Most stressful work situations are fundamentally safe in a way that interviews are entirely not. There's nothing to prove, only a problem to solve.
I'm also shit at public speaking—not talking to people, not meetings, but speaking before an audience, and I mean shit—and playing music in front of strangers is basically my nightmare (and yeah, I've done it). I suspect both those are true for most people. The stress of coding in front of someone while they judge me is identical to those situations, for me.
I bet the above is a pretty common.