> It's surprising to me how many candidates explain these algorithms' performance as if the hash table was a magic black box that instantly works. Here they've lost the tree for the forest! They know what hash table is, at least at a high level, but forget that it has specific implications.
A lot of advice I've read about interviewing suggests that a candidate should leave answers open and provide room for the interviewer to poll for more info. I'm curious to know if the author goes out of their way to poll about the specifics of the hash table implementation?
In fact, one of my favorite interviews in recent memory was a hash table discussion not unlike the one the author is hoping for. I was happy to find an interviewer whose questions balanced theory with practice and was happy to drive to a solution with them. However, I wouldn't have offered up this analysis without first understanding that this is where the interviewer wanted to drive the discussion. Every good programmer knows that premature optimization is the devil.
The author may instead be trying to raise the point that its up to the discretion of the interviewee to see that the hash table implementation is the critical performance detail. e.g. gotta do open addressing here b/c our hashing function has weird edge cases where it collides all the time and chained hash would bring us closer towards a log(n) traversal on that edge case.
However, in my limited interviewing experience, interview problems such as this are often designed to be open-ended conversation rollers moreso than where implementation specifics like this are able to jump out easily.
In the question I was referring to, I don't care if they can implement a hash table. I just want them to recognize that when they compare using a hash table to other alternatives, that looking up items in a hash table has a performance cost.
I agree it would be sort of silly to ask an open ended question and then hold it against a candidate when they don't talk about something as if they're supposed to read your mind.
I think without more view of the forest, as it were, I'm finding difficult as a reader to understand the type of analysis you're hoping to get out of your interviewees.
Would you agree that one interpretation of your thesis is to have interviewees be able to adress the "why" behind a given implementation? Your reply suggests that it's not the implementation specifics that you're after, but that was the understanding I got out of the original article.
> I just want them to recognize that when they compare using a hash table to other alternatives, that looking up items in a hash table has a performance cost.
As you obviously know, looking up items in a hash table has exactly a constant performance cost (amortizing the cost of a resize, and accounting for load factor, etc.), making it an attractive solution in cases where you need a lookup. I'm curious if your point is that the interviewee is missing that nasty c constant in front of the big o, (e.g. computing a hash for an item or the cost of a resize is impractical, say) or if the issue is that they are just missing something big picture and fundamental given the problem statement.
The former really speaks of "losing the tree for the forest", as in the person black-boxed something they needed to open up, whereas the latter strikes me as more of a lack of conceptual reasoning.
Thanks for your time! I noticed I was downvoted, so apologies if my inquiry frustrates or upset you.
I think the author was trying to show us how smart he was, not only in areas of codez (intimate knowledge of hash table) but also in areas of businessing (something called 'liquidity trap'). We should all hold him in high regard, because he only hires A players and had kicked many non-hashtable saavy B players (180+ in all) to the curb.
The whole "forest = knowing how a hastable works" is interesting! I'm trying to ascertain if the author's forest is a tree. I think there's good and evil in that statement.
Good being that its groovy if you love what you do are curious. Evil in that he could be optimizing the wrong thing.
Does every great software engineer have the inner workings of every data structure committed to memory? (I don't)
Referly CEO here - We've been headed in this direction for awhile, but just crossed over (with the new tagline) about a week ago. An in-depth blog post from me about the change and our new direction, plus a ton of stuff we learned, is forthcoming - probably this weekend if I can sit down for a few hours and knock it out. I'd really appreciate any feedback you have on the new experience, both as a reader and as a contributor (if you decide to go that far).
I'm curious about how YC and your other investors influenced the pivot. Did you get good advice? What was the dialog like? Was code refactoring involved? Did this change your marketing strategy?
I've heard the Singapore civil service, renown for recruiting smart people, have what they call the 'helicopter test'. Specifically they are looking for people who are just as able to converse intelligently flying 1000 feet above a topic as they are skimming the surface and getting into the details, and will specifically test candidates on both perspectives and their ability to fly between the two. Seems well aligned with what you are looking for.
"You’re looking for people who are 1) smart and 2) get things done"
Number 2 is a lot harder to determine than number 1 in my experience. You can determine if someone is a good programmer by having good programmers ask them increasingly challenging programming questions. The hard part is figuring out if they are an effective and productive employee.
the author seems to be in the business of creating fairly simple internet apps, so I'm not sure why his interviews would ever need to deal with the implementation of hash tables...
I agree with the value you place in intelligence and like how you qualify your essay as a discussion, but in perspective, your methods are simply imprecise and as you said, anecdotal.
You humorously contrast your essay to a "randomized clinical trial," but to be matter-of-fact, these trials exist: IQ tests or their politically acceptable stand-in, the SAT.
I guess what I'm asking is this: what's stopping you from requesting something like the SAT even if it's only for curiosity's sake?
What I have found to be interesting is that the more I learn about a technical topic, the less confident I become. Because I am now more aware of how much I don't know, or about how a small detail can make a big difference.
I used to think I was pretty knowledgeable about web security. The more I learn about it, and about how easy it is to get it wrong, or overlook something important, the less confident I feel in my ability to implement a secure web app.
It's a weird balance that people are looking for confidence and intelligence at the same time.
Some intelligent people are humble about the subjects they lack expertise in. Doesn't looking for confidence carries a risk of finding people who sound better than they are?
I agree. If I am asked a question outside of my domain of expertise, I'm not going to BS around it. I will say that I don't know and start asking questions so that I can learn the topic.
If you want someone to BS answers all day long, then you're hiring for a CEO.
Each one of your employees should fit the bill for future CEO. I see that as the point for hiring smart employees. It's someone who will pick up your stuff and run with it in directions you might not have thought about.
I interview for maturity, discipline, and honesty. My codebase is made to be maintainable by most programmer. I dont include any magic and make sure that it is understandable. But I don't interview for intelligence. Because intelligence without the three factors I look for does not mean a thing.
This is confusion of intelligence and knowledge. I don't know how a hash function works...but that is a reflection of my experience, not my intelligence. I did not study to be a computer scientist...but I sure as hell can optimize your supply chain.
I don't understand what you mean. The author said "Despite his confidence, he couldn't even explain what GDP is, which was essential to his explanation." which seems like a perfectly reasonable complaint. If you are trying to describe a concept to me that I don't know about, I might ask you to explain to me a much more basic idea that you are referring to. If they can't explain the basics then all of the explanation of the original subject matter is wasted.
I wasn't asking the guy an exact formula for GDP, but at least some definition of what it was trying to measure. Aaronc's point below is exactly my concern.
As per appeal to authority, I didn't think I was asseting anything controversial. I doubt you are saying it's better to hire dumb people or just ignore intelligence, so I am confused what you're arguing with here.
A lot of advice I've read about interviewing suggests that a candidate should leave answers open and provide room for the interviewer to poll for more info. I'm curious to know if the author goes out of their way to poll about the specifics of the hash table implementation?
In fact, one of my favorite interviews in recent memory was a hash table discussion not unlike the one the author is hoping for. I was happy to find an interviewer whose questions balanced theory with practice and was happy to drive to a solution with them. However, I wouldn't have offered up this analysis without first understanding that this is where the interviewer wanted to drive the discussion. Every good programmer knows that premature optimization is the devil.
The author may instead be trying to raise the point that its up to the discretion of the interviewee to see that the hash table implementation is the critical performance detail. e.g. gotta do open addressing here b/c our hashing function has weird edge cases where it collides all the time and chained hash would bring us closer towards a log(n) traversal on that edge case.
However, in my limited interviewing experience, interview problems such as this are often designed to be open-ended conversation rollers moreso than where implementation specifics like this are able to jump out easily.