My dad's a teacher too and boy does he love to lecture. About anything - digital signal processing, the price of gas, whatever - and that mode of communication is useful in its own context, but not so much when a team of people are trying to get something done together.
If it makes you feel better, I fell for the trap too. I asked Kevin for a demo of the question in action so I could see what it was like to be on the receiving end before I used it on candidates myself. I was the guy he referenced in the blog post that compared poker to backgammon and I was in the middle of explaining how Texas Hold Em taught me essential skills of entrepreneurship when he asked me how many board pieces were in a typical backgammon setup. I lost a good 20-30 seconds there trying to remember, until I laughed out loud and called him out on trying to randomize me :)
Basically, no matter what reasons you have for thinking people should do the right thing, you have to keep in mind that their reference set for YOU is all the people who have interviewed them in the past, many of whom were likely dumber and less competent than you. They are going to be acting accordingly.
I think the question is set up properly for the trap to be fair. Having been given clear responsibility for presenting a topic, a time limit, and the floor, I wouldn't think twice about directing the discussion.
Though I also don't perceive the balance of power in an interview to be that asymmetric, either. Usually they're courting me as much as I'm courting them. If anything, it offers a nice opportunity for a reverse test: management that has a problem with technically appropriate direction is not management I want to work for.
I thought that a system which started at zero and went up by multiples of two was obscenely slow, and I hinted at this, but he insisted on having things his way. In the end I'm sure he wrote me a bad review and I wasn't offered the job (I didn't really want it anyways, given my impression that the team I was interviewing for wasn't that focused).
In any case, I didn't go right now and say this is dumb, but I hinted at a better way, and even this hinting lead to some conflict. In the end I didn't get the job. I'm conflicted if this was the best approach. I could have pretended to be dumber than I actually am and gotten the job offer at least, which at this present moment I could probably use for leverage in the search for contracts or a job, but I couldn't bring myself to accept his not particularly good algorithm (full disclosure: the process was also affected by the fact that my flight was delayed from the east coast, that I got 4 hours of sleep, the fact that I insisted in wearing a tweed coat to the interview, which may have impacted ever important perception of "fit," and that the interview in question was a "surprise" interview after several other interviews).
That's how I would have answered the question, anyway. But yes, I've dealt with interviewers like this--typically it's because they don't understand the subject matter well enough to evaluate or accept an answer they didn't prepare for.
Of course, either way the algorithm runs in O(log n) time.
As for my case, I did attempt to start at an arbitrarily large number (but he wouldn't let me), and then went up exponentially (i.e. x = x * x), bi-secting the space once the upper-bound was found and solving from there.
My interviewer was sadly not that bright and didn't understand my solution, forcing me to re-write it -- incidentally, lecturing me on how wonderful O(log n) is along the way.
Closer to 0 than what? Infinity? By any reasonable definition, every integer is closer to 0 than infinity.
I think the problem as stated is underspecified; I disagree that increasing "exponentially" (the solution your interviewer described would increase exponentially; your version is doubly exponential, I think) is always faster. Both versions have the same worst-case running time, and if you want to talk about expected running time you need to make assumptions about the data distribution. As a counterexample, if you start with an initial guess of 2 and keep squaring, but the true number is 2^1025, then you end up doing much more work than if you had doubled instead of squaring.
I can certainly agree with you that lecturing in an interview is unproductive. Not only is it understandably off-putting, but it doesn't give the interviewer any additional input to the hiring decision and therefore is a waste of everyone's time. (Unless one of the job requirements is coping well with being lectured to, I guess.)
Perhaps, but originally we weren't discussing 0, we are starting with an arbitrarily large number. This problem is exacerbated when you say:
> I disagree that increasing "exponentially"... is always faster ... Both versions have the same worst-case running time
Worst case, sure, but how do you compute average? You could take a selection of n number of integers from 0 to infinity and the number of guesses until you found the correct answer, but of course you can't do this because you can't easily get a random number to test with that is between 0 and infinity. That said, it is fairly clear that x * x is faster than 2x is faster than x+1, but you can only prove this to be true once you pick an arbitrary limit less than infinity -- and the benefits of the better algorithm are only evident as you approach infinity.
Seems like a catch 22. Perhaps there is some expert in set theory as applied within computer science that can provide the appropriate formal context for determining the better algorithm in this context.
As for O notation, it is an interesting tool but one that has no application within any realm of actual programming in which I have worked, and I have never bothered to learn it simply to dazzle people in algorithm-based interviews (perhaps because I do more nuts and bolts type work rather than optimization).
------------------
1. In the initial probing for the upper bound, there is no point to grow faster than 2x each time
If we double the probe each time, we'll find the upper bound in O(log N) time. Then, we'll need O(log N) additional time to find the real answer. That makes the entire algorithm O(log N).
Suppose instead, in the initial probing we grow the bound faster, say by squaring each time. We'll find the upper bound faster, but we'll still need additional O(log N) time to find the real answer. So, we didn't really make the algorithm (asymptotically) faster - it is still O(log N).
(I glossed over some details in the explanation, but even if you work out the math exactly - which is not that hard to do - the conclusion holds.)
------------------
2. There is no point starting from a number greater than 1
There is no way to pick a "good" starting number - should it be 1,000? 1,000,000,000? 10^100? (10^100)^100? You might as well start from 1. That way, you guarantee that you'll find small numbers fast and large numbers still in an asymptotically optimal time.
------------------
As someone with fairly strong theoretical computer science backgound, I can see the intended meaning behind the interviewer's question and answer. But, it is a theoretical question. There definitely are a lot of highly valuable software developers out there who couldn't answer it.
I suppose I'm at a loss for words, since as a seasoned software developer without a deep background in theoretical computer science, it seems that I am being told to disregard my intuition based on experience in favor of a theoretical model that seems to ignore actual problem sets. Am I missing something?
https://gist.github.com/3839551
AFAICT, the parent and the interviewer is right because both methods run in roughly the same time because the time of the binary search is orders of magnitude greater than the upper bound finding part. I might have made one or more mistakes though.
The first observation that I have is that given that the secret number s is chosen, the first step can be completed arbitrarily quickly. One could use a function that rises arbitrarily fast. Imagine for example the function taking k to the Ackerman function A(2,2,k). That rises so fast it's incomprehensible, but really one could easily produce a function which rises faster still (the algorithm that picks s first!). The problem is, though, of how fast a fixed algorithm is for random s. If s is truly chosen at random from the positive integers, this leads to problems. Fix your putative algorithm. Suppose for the moment that it starts at 0 (it is not going to matter where it starts) What is the probability that the kth number that your function spits out is less than a random integer? 100% After all, how many integers are greater than any given integer?
Therefore, no growth function is any better on average at finding the upper bound than any other. Therefore step 1) is an insoluble problem. The problem should have been specified in some other way in order for it to make sense.
However, the second step involves log_2(n) time (in the worst-case-scenario and still O(log n) in general) where n is the upper bound --the output of step 1-- which means that the time to complete the algorithm is (Time of step 1 to find n) + O(log n). IF the problem made sense and it was the case that step 1) were soluble, then it would matter how fast step 1 is countered by the degree to which step 1 tends to overshoot the secret number --because overshooting by k has the penalty of log(k) extra operations.
Is there are framework in which question 1 makes sense? It would make sense if there were a given probability distribution on the integers (a function on the positive integers whose sum over all integers = 1, for example choose the function f(n) = 6/(n^2 * pi^2)) A probability distribution gives you a way of answering the question: "what proportion of the positive integers are greater than k?"
In retrospect, I might have mentioned this and then gave the interviewer's preferred answer in code, just to keep things simpler.
And, yet, it's quite a realistic model of what happens in meetings and presentations all the time.
> In an interview, the interviewer literally has all the power, and saying "no" or refusing to answer a question posed by the interviewer seems rather difficult.
You could say the same with regards to managers or other higher-ups in real workplace environments. It's an important skill to be able to say no, or deflect gracefully.
In this situation, saying "I'm not going to answer that question" is the wrong response. Something like "that's an interesting question, but it will take more than 5 minutes to answer. I'd be happy to go into more detail later if you'd like." is.
Yes, absolutely. In a real work environment where we all know each other I would not hesitate to shut down a conversation that takes the discussion off course.
An interview is not like a real work environment at all. It can try to be a simulation, but it is still just a simulation where the candidate and the interviewer may not know each other very well. Maybe it's my southern upbringing (a topic of much mirth in my office), but I would consider pushing back on a stranger I am trying to impress to be somewhat rude. I have noticed that people out here in SV tend to be more aggressive than I am used to.
In a meeting or presentation, you have a subject and an agenda and many people present who expect you to cover them.
In an interview, it's one-on-one. If the interviewer asks you A, then stops and asks you B, there's no reason not to move to B. If the interviewer is suddenly asking about B, they must want to know about it. There are no other stakeholders present who might care more about A. You didn't even know you were going to be talking about A until a few minutes before!
It's an important skill to be able to say "no" and stay focused on a subject. But this is a shitty test for finding out whether a person has that skill.
That may not be key for all positions, but for more senior roles, and all roles which are predominantly customer facing, it's absolutely critical.
Being able to say "no" is important, but so is knowing when to say "no" and when to say "yes".
If somebody tries to drag you onto an irrelevant tangent in a meeting, that's when you should say "no". If somebody is talking to you one-on-one and you discover that they actually care about something different than you thought, that's when you should say "yes".
The problem with this scenario is that it pretends to measure a person's ability to say "no" all while giving them no reason whatsoever to actually say "no".
If you want to test a person's ability to say "no", you have to put them in a situation where that's actually the correct move.
If you're explaining the game of 'Go' to me and I stop you and say 'Tell me about a time you achieved success' then I would expect 100% of candidates to stop talking about Go and start answering the new question.
On the other hand if I stopped you and asked 'Why do you think the counters are black and white?' I expect 100% of candidates to give me an answer that's fairly short and then continue with their original answer.
I can't think of a single question that would separate people into two groups.
One example is a candidate that was talking about residential firefighting. I asked them about halon gas as a way to fight with quite a bit of curiosity, and they shut me down. I've learned to trust my insights, because I've seen them confirmed in other ways throughout the interview--it builds a pattern. When it doesn't fit the pattern, I am open minded that it's a fluke.
Ultimately, I think the "trick" nature of this is over emphasized by my blog. The single most telling factor is whether candidates can clearly explain something in essentialized form--that alone is difficult for most people.
If you feel like the interviewer has all the power in an interview, then it'll be difficult to negotiate your salary. You'll do things like give a concrete number when they ask you what your salary was at your last job.
The only time I would feel like an interviewer has all the power over me is if I was a new grad with no work experience.
An interview is a two-way street, you're trying to find out if you want to work for them as much as they're trying to find out if they want to hire you.
Typically when hiring upper level managers you are interviewing someone who is going to be your peer - not someone who will be below you. Part of being a good leader includes pushing back on the people who are going to be your peers, if it means interfering with your tasks.
I've also got to agree that 'teaching the uninterested' is hard for someone who likes to do the very thing he's asking.
Job interviews are nothing like the workplace for the interviewee. Things like silent time can be death - I've had interviewers ask things like 'take as much time as you need to think before answering' and then getting annoyed because I was thinking for all of 20 seconds...
Reality is filled with traps.
> steer the candidate off on a tangent
Customers, coworkers, managers, do this intentionally or not, maliciously or not.
Any good candidate can do the technical job. Great candidates can survive and thrive in the work place.
None of these situations are the same as an interview. If you're providing an explanation at the behest of the interviewer, why not also depart on a tangent, also at the behest of the interviewer?
The role whose function an interview most resembles is that of a sales person. Having been in hundreds of sales meetings, I can say with total conviction that the ability to direct the conversation and deflect tangential questions is absolutely critical to being successful.
That said, every role doesn't require that skill, but OP specifically said that it's only for more senior/customer-facing positions.
In an interview, the interviewer literally has all the power, and saying "no" or refusing to answer a question posed by the interviewer seems rather difficult.
This would be doubly hard for me since I like to teach (both my parents are teachers and I love to explain things to people so that they can understand them). If a person shows genuine interest in a topic by asking a question that might go in a tangent, I generally want their curiosity to be sated before moving on; in some cases a deeper understanding of a related topic will help educate them on the main topic.
Anyway, I think I would fall for this trap for both reasons (not wanting to refuse the interviewer and also wanting to genuinely teach someone). I'm conflicted.
However, I like the overall idea of the question. So much time during actual real-world development is spent communicating between coworkers about how a system works or will be implemented. I think being able to gauge a candidate's communication/knowledge transfer ability is very important and a good indicator of success.