Dear Startups: stop asking me math puzzles to figure out if I can code

Before the interview, I ask them to write some code to access an HTTP endpoint that contains exchange rate data (USD, EUR, GBP, JPY etc.) in XML and to parse and load said data into a relational database. Then to build a very simple HTML form based front-end that lets you input a currency and convert it into another currency.

I ask them to send me either a link to a repository (Git, SVN etc.) or a zipball/tarball. If the job specifies a particular language, then I obviously expect it to be in that language. If not, so long as it isn't in something crazy like Brainfuck, they have free range.

If the code works and is basically sane, that goes a long way to get them shortlisted.

During the interview, I'll pull the code they sent up on a projector and ask them to self-review it. If they can figure out things that need improving in their code, that weighs heavily in their favour. Usually this is things like comments/documentation, tests, improving the structure or reusability. If it's really good, I'll throw a hypothetical idea for refactoring at them and see how they think.

The reason this works is that, despite Hacker News/Paul Graham dogma to the contrary, "smartness" isn't the only thing that matters in programmers. It's actually fairly low down the list. When hiring programmers, I want people who are actually able to do the daily practical job of writing code, modest and self-critical enough to spot their own mistakes, and socially capable to actually communicate their decisions and mistakes to the people they work with.

I interviewed a guy who was intellectually very smart and understood a lot about CS theory. I asked him why the PHP code he sent me didn't have any comments. "I don't believe in comments because they slow the PHP interpreter down." Sorry, he can be smarter than Einstein but I ain't letting him near production code.

1) I think a lot of start-ups want to hire "smart" people. Because they expect the new person to eventually wear many hats. Objective-C, Java, Android, CSS, server side concurrency, monitoring. An we've all seen Hunter and Schmidt reference that tokenadult usually posts when talk about interviewing comes around and it does seem that a general mental ability test (like an IQ test) combined with a work samples seem to predict future performance of that employee. Well except that one can't just straight up give IQ test to job applicants (there is a court case about that). So we are left with a job sample (which many forget to give, as is the point of the author). But instead many focus on the GMA and create proxies for it -- cute little puzzles about blenders, round manhole covers, and other such silly things.

2) Those interviewing don't know the technical stuff and are afraid you'd out-bullshit them. "How does an Ajax request work" well if the interviewer themselves doesn't quite know the details the might not be able to evaluate it properly. They could have it written down but well, some technical questions have many different levels of depth that a candidate might descent to. So a quick written answer to the question might seem wrong but it is really because the candidate is more advanced. So puzzles seems to be a generic and "easier" to handle.

This problem was addressed nicely in this functional pearl by Jeremy Gibbons, et al.: http://www.cs.ox.ac.uk/jeremy.gibbons/publications/rationals... . As interesting as the result is, however, it's a pretty well-made point that research-level ideas from the programming languages community are not really software engineering interview material in the vast majority of cases.

This is yet another example of "rockstar developer"-itis, wherein startups are given to believe that they need the best of the best when in fact they do not. This particular example is entirely egregious because they asked her about something that requires enumerating the rationals when what they really wanted was an iOS code monkey. Then they fired her, based on their own shoddy interview.

Certainly, asking only math questions is stupid as well, people should know at least a little about the stuff they're supposed to work with, but teaching an actual language to a smart person eager to learn is a breeze compared to teaching problem solving to someone who memorized the reference manual.

One example I use is getting the candidate to write crud, list, and search controller actions for a simple category data structure. Given a basic category data model (e.g. Name, Parent), the candidate starts with the crud actions.

Crud actions aren't meant to be difficult to solve and serve as a basic screener to verify the candidate has working knowledge of the basics. The only edge case I look for the candidate to ask about is if orphaning child nodes is allowed (I.e updating parent node, deleting a node with children)

List action(s) start getting more interesting since recursion comes into play. A basic implementation of an action that can load the tree given an arbitrary category as a starting point is expected. If the candidate has some prior experience, a discussion of what performance concerns they may have with loading the category tree is a follow up question. The tree loading algorithm is then expected to be revised to handle an optional max depth parameter. An edge case I look to be considered is how to signify in the action response that a category has one or more child nodes that weren't loaded due to a depth restriction.

The search action implementation has a degree of difficulty scaled to the candidates experience level. All candidates have to write an action that returns a collection of categories matching a search string. Those with previous experience are asked about a paging solution. Senior level candidates are asked to return matching categories in a format that indicates all ancestors ( for instance: "Category 1 -> Category 1.1 -> Category 1.1.1" result for search string "1.1.1")

For an added degree of difficulty, candidates can be asked to recommend data model tweaks and algorithms supporting tree versioning requirements necessary to allow for loading the category tree's state at a given point in time.

The candidate's performance to this exercise seems to give some insight into their level of experience and ability to implement algorithms from a common real world example without having to ask much trivia or logic problems.

At one company I interviewed with, I was asked to implement a queue using two stacks. At that time in my programming career, I had worked with C, C++, Obj-C, Lua, Python, JavaScript, SQL, and a handful of DSLs developing games, game development tools, and web applications. Want to know what I had never done? Written a queue using two stacks. My immediate response to the question was, "Why would you want to do that?"

If you really want to know if someone has the capacity to pull their weight as an engineer, ask them about what they've built. Even if they are fresh out of college, the best engineers will have projects they can talk about and explain. Ask how they approached/solved specific problems. Ask what they're most proud of building. Ask what was most frustrating.

Those are the kind of questions that will provide insight into a person's problem solving capabilities and offer a decent picture of what they're capable of doing.

This would test programmers ability to learn a new language.

Probably because the only person who doesn't lose from this is the interviewer: they get to have fun. Honestly, when you spend all day buried in code, it's fun to play with puzzles for a change.

Perhaps it's time we started optimizing interviews for hiring success rather than interviewer happiness.

The position I was filling is a part-time position for a CS major, sort of like an internship. I devote time to develop his/her skills, s/he would get real-world experience, and a little money to help with cost of living. If everything works out, a position could open up for full employment.

I had a pretty good idea what I was looking for. Someone that had good grasp on theory but had no experience coding. Preferably enrolled in Uni. I had 5 applicants but the only candidate I interviewed is enrolled in Math-CS.

I basically tried to gauge if he had deep interests and asked him to code a bit, solve a simple control (find me the article with the highest hitcount from the day a week ago, gave him 10 minutes).

He failed the coding test but I made the hire regardless. Reason why was 2 things out of the 4 hours we spent together: When I asked him who he considered the father of CS he rattled off von Neuman, Djikstra and Knuth. Yeah, you can make that argument I suppose, but he knew who the influential people were. The other thing was: even if he failed the coding test he failed it by not reading the code examples quite right, he was using my code to try to help himself solve the problem. I'm sure he'll work out.

We as a field should employ internships a lot more than we do, get the college kids and undergrads working on real-world problems a lot more than we do.

I just don't have the experience or tools or interest for them.

And yet, somehow, in 20 years of business geekery I've never come across a problem I can't solve.

Maybe when writing Tetris for J2ME I would have saved myself 10 minutes googling if I'd had the experience to realise that right angle based matrix translations don't require fp maths and maybe when writing financial indicators, I'd have saved myself half a day if I hadn't had to look up integrals but this sort of stuff is definitely in the minority as far as my experience goes.

I believe this is deeply valuable. For some roles, I would much prefer to hire someone who can quickly see the value of breadth-first search from both ends.

If he/she doesn't happen to know the syntax of Ruby, or Java, etc. it's less important to me.

Interviewer: "How can we optimize the character replacement in a string such that we use no extra memory?" Me: "We do this and that and this. But, should we consider what situations we would need this optimization?" Interviewer: "What? Why?"

I can now use this as a filter as I interview organizations. Optimizing algorithms by creating your own core data structure classes (instead of using the built-in ones) is great in certain circumstances, but an absolute waste of time in many others. And if you're not going to ask me about those times when making those improvements is important, then you're not asking questions for a programmer -- you're asking them for a theoretician who can recall syntax.

It's poor practice, and I've seen it everywhere.

The review article by Frank L. Schmidt and John E. Hunter, "The Validity and Utility of Selection Models in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings,"[1] Psychological Bulletin, Vol. 124, No. 2, 262-274 sums up, current to 1998, a meta-analysis of much of the huge peer-reviewed professional literature on the industrial and organizational psychology devoted to business hiring procedures. There are many kinds of hiring criteria, such as in-person interviews, telephone interviews, resume reviews for job experience, checks for academic credentials, personality tests, and so on. There is much published study research on how job applicants perform after they are hired in a wide variety of occupations.[2]

EXECUTIVE SUMMARY: If you are hiring for any kind of job in the United States, with its legal rules about hiring, prefer a work-sample test as your hiring procedure. If you are hiring in most other parts of the world, use a work-sample test in combination with a general mental ability test.

The overall summary of the industrial psychology research in reliable secondary sources is that two kinds of job screening procedures work reasonably well. One is a general mental ability (GMA) test (an IQ-like test, such as the Wonderlic personnel screening test). Another is a work-sample test, where the applicant does an actual task or group of tasks like what the applicant will do on the job if hired. (But the calculated validity of each of the two best kinds of procedures, standing alone, is only 0.54 for work sample tests and 0.51 for general mental ability tests.) Each of these kinds of tests has about the same validity in screening applicants for jobs, with the general mental ability test better predicting success for applicants who will be trained into a new job. Neither is perfect (both miss some good performers on the job, and select some bad performers on the job), but both are better than any other single-factor hiring procedure that has been tested in rigorous research, across a wide variety of occupations. So if you are hiring for your company, it's a good idea to think about how to build a work-sample test into all of your hiring processes.

Because of a Supreme Court decision in the United States (the decision does not apply in other countries, which have different statutes about employment), it is legally risky to give job applicants general mental ability tests such as a straight-up IQ test (as was commonplace in my parents' generation) as a routine part of hiring procedures. The Griggs v. Duke Power, 401 U.S. 424 (1971) case[3] interpreted a federal statute about employment discrimination and held that a general intelligence test used in hiring that could have a "disparate impact" on applicants of some protected classes must "bear a demonstrable relationship to successful performance of the jobs for which it was used." In other words, a company that wants to use a test like the Wonderlic, or like the SAT, or like the current WAIS or Stanford-Binet IQ tests, in a hiring procedure had best conduct a specific validation study of the test related to performance on the job in question. Some companies do the validation study, and use IQ-like tests in hiring. Other companies use IQ-like tests in hiring and hope that no one sues (which is not what I would advise any company). Note that a brain-teaser-type test used in a hiring procedure could be challenged as illegal if it can be shown to have disparate impact on some job applicants. A company defending a brain-teaser test for hiring would have to defend it by showing it is supported by a validation study demonstrating that the test is related to successful performance on the job. Such validation studies can be quite expensive. (Companies outside the United States are regulated by different laws. One other big difference between the United States and other countries is the relative ease with which workers may be fired in the United States, allowing companies to correct hiring mistakes by terminating the employment of the workers they hired mistakenly. The more legal protections a worker has from being fired, the more reluctant companies will be about hiring in the first place.)

The social background to the legal environment in the United States is explained in various books about hiring procedures,[4] and some of the social background appears to be changing in the most recent few decades, with the prospect for further changes.[5]

Previous discussion on HN pointed out that the Schmidt & Hunter (1998) article showed that multi-factor procedures work better than single-factor procedures, a summary of that article we can find in the current professional literature, for example "Reasons for being selective when choosing personnel selection procedures"[6] (2010) by Cornelius J. König, Ute-Christine Klehe, Matthias Berchtold, and Martin Kleinmann:

"Choosing personnel selection procedures could be so simple: Grab your copy of Schmidt and Hunter (1998) and read their Table 1 (again). This should remind you to use a general mental ability (GMA) test in combination with an integrity test, a structured interview, a work sample test, and/or a conscientiousness measure."

But the 2010 article notes, looking at actual practice of companies around the world, "However, this idea does not seem to capture what is actually happening in organizations, as practitioners worldwide often use procedures with low predictive validity and regularly ignore procedures that are more valid (e.g., Di Milia, 2004; Lievens & De Paepe, 2004; Ryan, McFarland, Baron, & Page, 1999; Scholarios & Lockyer, 1999; Schuler, Hell, Trapmann, Schaar, & Boramir, 2007; Taylor, Keelty, & McDonnell, 2002). For example, the highly valid work sample tests are hardly used in the US, and the potentially rather useless procedure of graphology (Dean, 1992; Neter & Ben-Shakhar, 1989) is applied somewhere between occasionally and often in France (Ryan et al., 1999). In Germany, the use of GMA tests is reported to be low and to be decreasing (i.e., only 30% of the companies surveyed by Schuler et al., 2007, now use them)."

[1]

http://mavweb.mnsu.edu/howard/Schmidt%20and%20Hunter%201998%...

[2]

http://www.siop.org/workplace/employment%20testing/testtypes...

[3]

http://scholar.google.com/scholar_case?case=8655598674229196...

[4]

http://books.google.com/books?hl=en&lr=&id=SRv-GZkw6TEC

[5]

http://intl-pss.sagepub.com/content/17/10/913.full

http://www.economics.harvard.edu/faculty/fryer/files/Fryer_R...

[6]

http://geb.uni-giessen.de/geb/volltexte/2012/8532/pdf/prepri...

Being a developer is 80% Google and 20% actual coding knowledge. We are hackers at the end of the day, not miniature Einstein's with encyclopaedias for brains.

OK, so there is a difference between computer science and programming. that's why there are two different stack-exchanges:

  cs.stackexchange.com
  stackoverflow.com

it's actually really fucking INCREDIBLE that

* you can know tons of CS without being able to build a decent app * you can a decent facebook clone without having any idea how it works

I feel really bad for Emma. I was a math major, but app developers won't even look at me b/c I'm not a full-stack whatever. So now I'm a Data Scientist at an advertising firm in Puerto Rico.

I tend to hate the interviews that ask me to solve math and logic brainteasers because I don't see the value in them regarding my knowledge of programming.

For example: "This database contains 100,000 problems with standardized parameters. The problem definition is defined in the file spec.txt which you can grab from our code repository. Write the code to solve these problems efficiently, passing each solution to a remote service via POSTing to a REST API, the documentation for which you can find here. Bonus points for parallel execution. Feel free to use any editor/IDE and reference online documentation, Stack Overflow, etc. that you want. If anything's not clear or you need a hand with something, just ask as you would if you were an employee already. Ready to get started?"

The great thing is that once you've identified a candidate, you can do remote screen sharing and have them write code before they even have to come into the office. I've interviewed a fair number of remote people this way and it's excellent for weeding out the people who can talk the talk but can't program worth a damn. And it limits bias because you don't care about much beyond their communication ability plus their technical ability.

That aside, one must have a way to measure the abilities of a candidate -- and asking the same set of questions to many people allows you to compare the answers as apples to apples.

I generally don't restrict my people from asking any particular question, but I will ask them to consider what a failed answer really means for the specific job (questions are generally adjusted then).

As an aside, some questions of mine that aren't specifically about coding:

* do you code outside of work (a love of coding translates to good coders)

* send me a link to some code you've written that you are proud of (let see what you got)

* tell me about a problem you had where your solution wasn't correct (how have you dealt with failure).

If one is good and quick in problem solving or has high GMA, that does indicate that he has the capacity to handle new and difficult things in general, but says nothing about the speed with which he can handle a particular new thing. Author's example with JavaScript is very good illustration how difficult can it be to learn a new paradigm for the first/second time.

Anyone who supports math puzzles (or whatever else) in an interview would have to argue that their perception of the candidates performance offers a clear enough data point that it doesn't dilute other information available to them. Given Google's study finding data otherwise, they certainly have the burden of proof.

I really understand that a startup with scarce resource would like to do its best shot. However as discussed long ago (https://news.ycombinator.com/item?id=2385424), it is really frustrating that asking math puzzles are assumed as the best way to hire the best for the job.

Because of this I've pretty much given up on hiring graduates based on their technical skills so instead I'm looking for someone smart, who gets that they've got a lot to learn, who is interested in technology and can get on with the other people in the team.

I don't think asking people math questions per se is a great idea, but if you've studied a maths degree it's a good way of working out if you're smart and if you were paying any attention at all during university.

(Incidentally this may be different in other countries (I'm in the UK) or in a company where you're able to attract the very best who have picked up really solid skills, but for most organisations that's not the case as most graduates spent more of their own time in the bar than coding.)

If a startup asks you to solve math puzzles, it's possible that the work you will be doing heavily involves the creative use of math or information analysis. (This is more broadly valuable than many people recognize.)

Also, it's also possible that that particular startup doesn't know how to effectively interview.

It doesn't sensationalisticly mean all Startups (capitalization yours) don't know how to effectively interview.

Also, rather than focus on your ability to learn, I would humbly recommend you reconsider the basic nature of employment. An interview should be considered a two-way conversation. You're not selling yourself as a slave, you're entering into a mutually-beneficial, private, voluntary arrangement. Thus, even someone who goes into an interview willing to accept anything and everything they offer could be expected to ask simply, "And what exactly will I be doing?" But better yet, grill them about every nitty-gritty detail you can think of. Although some insecure interviewers may be taken aback (I'm guilty of asserting the interviewer was wrong on more than one occasion, both times still receiving an offer), I for one am impressed when a candidate demonstrates a sharp, critical and skeptical mind in this way.

[1] http://www.codinghorror.com/blog/2007/02/why-cant-programmer...

The irony is that, in an effort to hire the "smartest" people, they leave out the wisest. Which is arguably more useful.

- After a first non-technical call, we ask the candidate to create a very small project based on our SDK. We send him the documentation and a very small sample. He can almost use every tools he wants to create that small project and, of course, we do not set any deadlines. It allows us to see how the candidate architecture his applications and it gives us a project to discuss during the following call. - If all goes well, we invite the candidate on site to present our code/project and eventually brainstorm together. So that both parties can see if they can work together and the candidate has an insight about how we work, how our code looks like.

Clearly, it's far from perfect and we are often considering changing it. Imagine if every company where you are applying would ask you to create an app from scratch with their SDK? We may lose some candidates, but at least we hire only people that fit the company's culture.

I would (and have) asked if the interviewer or organization has any evidence to show that interview puzzle performance (or shit like Myers-Brigg) predicts job performance. No? Not surprising. Google did look into it and found no relationship. (http://www.businessinsider.com/how-google-hires-2013-6)

Programmer interviews are so crazy and sometimes sadistic that I catalogued some of the more common interview patterns:

http://typicalprogrammer.com/thirteen-patterns-of-programmer...

+ knowledge - generally mastery of math/CS concepts and can be thought of as the potential

+ application skills - modeling a real world problem into a theoretical, computable, and (ultimately) programmable form

+ execution skills - implementation (coding) of a solution including the ability to utilize requisite tools/technologies such programming languages, DBs, OS, and so on

That said, hiring process should cover each of these areas and programmers should work on all these as well.

I could learn heroku/RoR/whatever other technology but news things are always coming out and some people keep up with it so easily. I'm not sure being a dev is right for me if I take so long to understand such basic stuff. But I love coding and algos! I write python scripts to do all my homework... and then run them in codecademy labs because doing it in unix makes me so confused.

If anyone has had the same problem please let me know how you got over this hurdle. Thanks.

background; sophomore, cs major, cornell

Or I might ask them to describe how an event loop works.. or what the I/O path between their program and the disk looks like in as much detail as they can.

Someone that loves the field is going to have a decent idea about these things even if they never had to build one before.

caveat: these examples are very system level but you can substitute them with appropriate web, financial etc domain specific knowledge.

The reason is smart people can figure out git, or databases, or objective-c, or whatever, in a fairly short amount of time.

For example, my co-founder learned objective C off free online video tutorials and built an iOS app (talking an app with serious firepower and back-end transaction logic) from start to end by himself in less than 3 weeks.

That's why we're not as concerned about what you know right now as what it's possible for you to learn in 3 more week.

Yet, I have never had the balls to pursue it professionally. I build stuff and usually never launch it. I have learned several times over that marketing is not my strong suit.

That said, I'd actually like to work for a startup. Hit me up if anyone wants to talk.

The likelihood of failure of a startup approaches 100%, so you should optimize for likelihood of survival, not for IQ.

If you're not a startup, then the top ranked comment applies. But it doesn't really otherwise.

I like to ask "what will I be working on in the next 6 months" that way you don't rock up and than the second day they through you in the deep end of building a iPhone app.

Granted, startups only have a vague idea of what they will be programming with short periods but it helps.

Also ask "what will be my performance indicators". If they don't include "being able to very quickly learn new technologies" its hardly your fault.

Not long ago, Facebook made that 4.74 degrees of separation on its networks. Meaning a maximum of only 4.74 persons are necessary to connect any two random persons on the network.

https://www.facebook.com/notes/facebook-data-team/anatomy-of...

You can also find an article on Wikipedia about the "Kevin Bacon" reference.

I was asked, as part of my application, to take a programming quiz. The quiz consisted of a graph theory problem. I did pretty poorly on it, given that I have no real knowledge of graph theory.

Had they asked me a question about statistics (or something similarly related to data analysis), I think I would have actually been able to answer, or at least been at a point where my programming knowledge- not my math knowledge- was what was holding me back.

But this one talks about getting inadvertent benefit of being good in Maths to get selected for programming, and suffering the consequences later on.

Also, it highlights the importance of what is mostly taken for granted and thought of as mundane stuff, of programming - the idiosyncrasies, jargon, and best practices of various languages and OS environs.

* CS majors take math in college, but there's lots of other stuff going on with their CS classes so it's hard to remember math class. If you can actually remember anything from math class, you must be smart!

* Programmers like math, even though it's not their job, so interviewing is the only opportunity to have fun with it!

* We had to deal with this crap when we were interviewing so now it's time to make someone else suffer! >:)

Programming isn't difficult and you don't need to know complex maths or be able to solve mind bending puzzles to be a great developer.

Check out the last technical interview task that I got ``` Objective: Write a program that prints out a multiplication table of the first 10 prime numbers. The program must run from the command line and print to screen one table.

Notes: - DO NOT use a library method for Prime (write your own) - Use Tests. TDD/BDD - IMPRESS US. ```

I mean I can impress you but how will this correlate with production code?

This is like solving your submarine problem. Jeese.

Isn't XY years of records in the same field of interest working for a successful companies a good sign that I can code?!

Ask me theory - pay me to code.

Stop asking this fine young lady math puzzles to determine her programming abilities. She is good at solving your seemingly pointless math puzzle, because she was practicing problem-solving since she was ten. But she is not anywhere near as good at programming, yet - which caused her problems at the actual jobs she had to do after she was hired.

In most cases an applicant must be able to read English (to google some code to copy-paste and occasionally search through documentation) and able to install and run Eclipse.

The real problem with hiring is that a HR middleman is ignorant and can't tell a good code form a restaurant menu. So he must give a very few simple exercises from common text-books with known answers.

The even bigger problem is that almost no one needs coders, everyone wants programmers which is a complete different set of analytical and engineering skills.

Coding is just a process of translation of a ready-made by someone else, poorly understood (if at all) specifications into a spaghetti [Java] code by calling poorly understood methods of ready-made classes, coded by someone else.

Programming is a process of understanding and describing reality (in terms of design documents, protocol specifications, and then, least importantly, source code in a several languages).

The criteria of success for a coder, btw, is when it just compiles (unit-tests? what unit-tests?) by the industry-strength most advanced compiler of the most sophisticated industry standard static-typing language (static typing is a guarantee from stupid errors, everyone knows) which is even verified to run correctly on the most advanced VM which incorporates millions of man-hours of optimizations, unless.. Never mind.

Success of a programmer is when it, like nginx or Plan9 or OpenBSD, is good-enough.)

E.g. if somebody hire John Carmack (ID Software), nobody will let him do some math test or ask him trivial programming questions.

But you are not John Carmack ;-)

It is like in every other job: if you are not a rockstar you are nobody.