|
i'm calling bullshit on the interview process. it's blatantly apparent when somebody is trying to cheat via chat on an interview. they have weird pauses and can't fluidly back up what they're saying. replacing the human they're using to cheat with an llm isn't going to improve that loop unless the candidate is a prompt engineering wizard with a background in bullshitting
|
#
?
Feb 5, 2023 07:42
|
|
|
|
| # ? May 23, 2024 19:51 |
|
|
so you do agree that the async stuff is dead, as i mentioned im always pessimistic about my ability to suss out cheating, despite sussing out cheating all the fuckin time, because peeps are always better bullshitters than you think. you always have a biased sample of putative bullshitters cuz the good ones dont get caught. "ooh i can detect lies"- how the gently caress would you know? are they gonna give you a nice confusion matrix after the interview of lies you didnt detect successfully? bob dobbs is dead fucked around with this message at 07:48 on Feb 5, 2023 |
|
#
?
Feb 5, 2023 07:43
|
|
|
bob dobbs is dead posted:these peeps had the marketing peep go through coderpad-like screens with basically no knowledge lol @ at the aws certified cloud practitioner one, that cert is just a quiz for whether you've read any aws marketing materials
|
|
#
?
Feb 5, 2023 07:58
|
|
|
i don't think i advocated async interviews anywhere -- pretty sure all my posts were clearly talking about conversations where at a minimum i'm on a voice call with the candidate, and more likely on a video call. if you can fluidly bullshit through video interviews, you can fluidly bullshit through in person interviews
|
|
#
?
Feb 5, 2023 07:59
|
|
|
its yet another way in which the market is a two sided market for lemons, the predatory nature of 90% these certs yeah
|
|
#
?
Feb 5, 2023 08:00
|
|
|
nudgenudgetilt posted:i'm calling bullshit on the interview process. it's blatantly apparent when somebody is trying to cheat via chat on an interview. they have weird pauses and can't fluidly back up what they're saying. replacing the human they're using to cheat with an llm isn't going to improve that loop unless the candidate is a prompt engineering wizard with a background in bullshitting there are definitely interviewers that are stuck on the "if you can't think up the solution to this maths puzzle in 30m you cannot into code" model of candidate evaluation. idk, i get performance anxiety and hate having to jump through hoops to demonstrate that i can turn fun math puzzle solutions into code like undergrad all over again. virtually nothing in my career involves that segment of CS, and while i enjoyed it as a fun puzzle, it's a very useless interview hurdle for me i code slow as poo poo and need to take time to read up on the problem and find prior art to inform what im doing. testing my ability to to visualize matrix manipulation or string analysis in my head and translate it into code within the 30m or w/e allocated to a coding interview is about as useful to demonstrate my actual value as it would be if i were interviewing you (for a SWE job) by testing your ability to rapidly learn tango steps. do those skills sometimes mesh with the actual skills i use on the job? sure. is my ability to do it on the spot an actual good indicator of my future performance as an employee? not really any interviewer can probably come up with a 30m poo poo test that they could pass easily and use it as a basic pass/fail on a candidate, and the tech industry happens to love one im naturally bad at. my getting flustered at syntax errors while trying to talk my way through a coding problem does make me look bad, but it's only relevant to what i actually do at work if i happen to be applying for a position where i have to code solutions to random linear algebra problems under time pressure. i do not apply to such positions because they do not exist (nobody in software is actually churning out toy programs that do some stupid matrix manipulation trick as a job) and would not fit me if they did, but industry still thinks it's important to jump through those hoops because it's developed a bullshit interview process with the dual goals of filtering out charlatans (which it mostly does, although plenty of people manage to grind leetcode and then be useless AFAIK from my vague understanding of FAANG culture, so not entirely) and finding people with valuable skills. the latter is unfortunately harder to evaluate in the typical 10h or so that make up an interview process there's legit a place for cheating the "can you pass the filter we designed to catch bad new grads and then decided to apply to everyone because THE BAR" aspect of SWE interviews. poo poo's halway to a loving 1950s voting literacy test requirement Qtotonibudinibudet fucked around with this message at 08:47 on Feb 5, 2023 |
|
#
?
Feb 5, 2023 08:45
|
|
|
my basic conclusion is that it cant filter charlatans anymore. so industry will haul rear end to something else real quick or get some amazing lemon peeps (they get amazing lemon peeps already) and the lemon peeps are not even close to limited to new grads unfortunately bob dobbs is dead fucked around with this message at 08:51 on Feb 5, 2023 |
|
#
?
Feb 5, 2023 08:48
|
|
|
not callin you a charlatan btw. but i always get pissed when peeps talk about interviewing reform and deny that they exist. ive dealt with like a dozen of them from better and worse interviewee pipelines. not paralyzed with anxiety or whatever, just waltzed in pretending with great confidence to be able to code and they cant write a for loop
bob dobbs is dead fucked around with this message at 09:02 on Feb 5, 2023 |
|
#
?
Feb 5, 2023 08:56
|
|
|
can you, if i put you on the spot in a 30-minute interview, write a loop that iterates through the values in an array? if the answer is yes then you're better than a good number of candidates i have seen.
|
|
#
?
Feb 5, 2023 09:52
|
|
|
the intern interview circuit is 100% asynchronous evaluations and live programming puzzle exercises and i dont know if i can blame them for it. i read companies like roblox get 50,000 applicants for just a few hundred available positions, you have to do something to stem the tide of bscs failstudents i dont personally mind it for fake short term jobs like internships even if prepping for it is annoying. but i know it's gonna start pissing me off real quick post graduation
|
|
#
?
Feb 5, 2023 14:44
|
|
|
Turns out yelling LEARN TO CODE through a bullhorn for 10 years had some bad knock on effects
|
|
#
?
Feb 5, 2023 14:55
|
|
|
Jabor posted:can you, if i put you on the spot in a 30-minute interview, write a loop that iterates through the values in an array? often when i read this sentiment expressed i want to ask how you can be so confident that you’re not just seeing really bad anxiety, but then i remember a guy i interviewed who denied that recursion as a concept even existed, so idk
|
|
#
?
Feb 5, 2023 15:37
|
|
|
right? ive gotten lots of paralyzed by anxiety peeps and i dont dock em for it. the charlatans are something different
|
|
#
?
Feb 5, 2023 16:22
|
|
|
raminasi posted:often when i read this sentiment expressed i want to ask how you can be so confident that you’re not just seeing really bad anxiety, but then i remember a guy i interviewed who denied that recursion as a concept even existed, so idk real power move tho: try to gaslight your recruiter
|
|
#
?
Feb 5, 2023 16:27
|
|
|
i feel like a toy problem along the lines of “merge this branch into this shared feature branch” would be more useful than a create a linked list/do a math/regex skill check
|
|
#
?
Feb 5, 2023 19:28
|
|
|
would be a sure way to not hire me for certain. fizzbuzz is pretty good, probably should not dismiss every candidate that fails it, but if it is a coding job it is a pretty good indicator.
|
|
#
?
Feb 5, 2023 19:32
|
|
|
Cybernetic Vermin posted:would be a sure way to not hire me for certain. I’m not a coder and i can write fizzbuzz lol
|
|
#
?
Feb 5, 2023 19:39
|
|
|
Captain Foo posted:I’m not a coder and i can write fizzbuzz lol lots of people interviewing for coding positions can't though. i want to be a bit careful here as anxiety issues are indeed a thing and one shouldn't get too categorical, but a ton of people come to interviews and can't do fizzbuzz (and then can't reason about why they didn't do it right when asked).
|
|
#
?
Feb 5, 2023 19:40
|
|
|
Jabor posted:can you, if i put you on the spot in a 30-minute interview, write a loop that iterates through the values in an array? Yup. I used to have a more complex test I used and over the years I simplified it down to something between this and fizzbuzz. Still had architect level folks fail it. We did it live on a whiteboard and then I'd ask a few questions about their solution and do some mini optimizations on it and that was it. It was shocking how poorly folks did on the whole, so much so that we moved it earlier and earlier in our interview process. Nothing sucked more than having a good hour long chat only for the candidate to completely fail what we felt like was a formality at the end. EDIT: re anxiety, whiteboarding, etc - we were all client facing (consulting) so this was more relevant to their actual job tasks, i.e., doing stuff on the fly with an audience.
|
|
#
?
Feb 5, 2023 23:14
|
|
|
I actually had an interviewer deploy actual FizzBuzz on me. I joked dismissively about FizzBuzz being incredibly easy roughly 15 seconds before he came up with the problem description. He seemed thrown.
|
|
#
?
Feb 5, 2023 23:16
|
|
|
Seriously though, FizzBuzz is about the right difficulty for a "can you write code to accomplish a task" programming interview. The amount of signal you get from a more challenging problem is minimal and might even be negative. You don't want to do FizzBuzz exactly though - your signal just gets completely hosed if some of the candidates have seen and rehearsed your exact problem before - what you want is some novel problem at about that level.
|
|
#
?
Feb 5, 2023 23:25
|
|
|
i made up a question that i thought was similar difficulty to fizzbuzz - write a function that takes in a datetime and returns an integer. it should return the angle in degrees between the hour and minute hands of a clock at the given time. literally the first candidate i used it on was partially sighted and had never learned to read a clock 
|
|
#
?
Feb 5, 2023 23:35
|
|
|
you cannot assume peeps under the age of like 20 can read analog clocks anymore
|
|
#
?
Feb 5, 2023 23:36
|
|
|
i honestly do not know which sin or cos or tan i need to find the angle between two lines so please never ask me math questions. this one works tho because you can just multiply the hour and the min to get an angle
|
|
#
?
Feb 5, 2023 23:41
|
|
|
jesus WEP posted:i made up a question that i thought was similar difficulty to fizzbuzz - write a function that takes in a datetime and returns an integer. it should return the angle in degrees between the hour and minute hands of a clock at the given time. this is actually a pretty decent question because there’s an obvious first implementation that runs into issues pretty quickly and it also requires them to ask follow up questions before they start (there’s always two possible answers unless it’s 180 degrees, so which one is the correct one?) also if someone doesn’t know how to read a clock it’s pretty trivial to explain
|
|
#
?
Feb 6, 2023 00:16
|
|
|
thats harder than fizzbuzz tho, is the rub
|
|
#
?
Feb 6, 2023 00:17
|
|
|
bob dobbs is dead posted:thats harder than fizzbuzz tho, is the rub
|
|
#
?
Feb 6, 2023 00:30
|
|
|
bob dobbs is dead posted:you cannot assume peeps under the age of like 20 can read analog clocks anymore We all work places where getting comfortable with obsolete tech quickly is an important skill.
|
|
#
?
Feb 6, 2023 00:31
|
|
|
I used to do a roman numerals question, just take in a number output string, and nearly all the time I had to explain exactly how it worked but that took all of a couple minutes, and let me ask the fun question at the end "if the pattern continued indefinitely, what is the performance of your function" and see how people struggled with digits being the loop
|
|
#
?
Feb 6, 2023 00:35
|
|
|
PIZZA.BAT posted:also if someone doesn’t know how to read a clock it’s pretty trivial to explain The problem (well, the other problem) is that the explanation is basically the implementation.
|
|
#
?
Feb 6, 2023 00:36
|
|
|
for longer coding interviews my goto used to be "write console-mode minesweeper" until i started getting a bunch of people who'd never fuckin heard of minesweeper
|
|
#
?
Feb 6, 2023 00:37
|
|
|
I dislike fizzbuzz. it tells me nothing as an interviewer and even less as an interviewee give me a problem that contains something your team use a lot. call an api and do a thing, transform some data, make a website show A Thing alternative have some code prepared and have your interviewee tell you what is happening in said code. I tried this once and it sparked good conversation
|
|
#
?
Feb 6, 2023 08:50
|
|
|
champagne posting posted:I dislike fizzbuzz. it tells me nothing as an interviewer and even less as an interviewee if you are recruiting expertise for a specific api or something that does make sense, never been in the position to be so picky as to expect candidates to come in knowing precisely the things we do (or know enough to ask about ourselves) though
|
|
#
?
Feb 6, 2023 09:06
|
|
|
hobbesmaster posted:product or project? project ideally havelock posted:For software stuff? ie are you trying to parlay your other experience into software PM stuff? Project Coordinator or Project Manager are pretty broad, with the former being more junior (maybe even entry level). right! I'm trying to jump into a role like that at a level where I can get more familiarity with the operational structure and hopefully get them to pay for professional creds that can jump me into a higher salary. my org now is way too small with too narrow of scope where formal pm tools would be at all necessary
|
|
#
?
Feb 6, 2023 09:22
|
|
|
Cybernetic Vermin posted:if you are recruiting expertise for a specific api or something that does make sense, never been in the position to be so picky as to expect candidates to come in knowing precisely the things we do (or know enough to ask about ourselves) though i don't think that's what they're saying, although they might be. it's more 'can you look over this code and get a good idea as to what it's doing. what would error conditions look like here? if you had to implement a change, how would you do it?' etc. i don't think they expect them to be familiar with their whole codebase or anything, rather if they're shown a chunk of code and given an explanation as to what it's supposed to be doing, can they start fiddling with it
|
|
#
?
Feb 6, 2023 13:28
|
|
|
i had an interview a few years ago where they did that. they printed out one of their java classes that was essentially a log parser subclass. a log to one of their systems consumed logs line by line and fed them into this other system. in certain cases that line would be passed to this subclass for handling. with that explanation they had me look at the class and basically just freeform talk about what i saw, what i guessed the various functions were doing, try to find bugs, etc. i'm assuming the expectation was that as a mid-level guy with java experience i should be able to drop in and start figuring things out mostly on my own while still being able to ask higher level system questions just to make sure my assumptions were correct it was actually a pretty pleasant process and didn't feel stressful at all. i think it was because i made it feel like it was just my first day with them rather than an interview. i'm also someone who doesn't get nervous easily so ymmv
|
|
#
?
Feb 6, 2023 13:34
|
|
|
PIZZA.BAT posted:i don't think that's what they're saying, although they might be. it's more 'can you look over this code and get a good idea as to what it's doing. what would error conditions look like here? if you had to implement a change, how would you do it?' etc. i don't think they expect them to be familiar with their whole codebase or anything, rather if they're shown a chunk of code and given an explanation as to what it's supposed to be doing, can they start fiddling with it yeah it’s this kind of thing. basically use a code example to start a conversation about code and then determine something about your candidate
|
|
#
?
Feb 6, 2023 13:38
|
|
|
i mean, sounds good, but to my mind you'd need an hour or two to do that properly, which makes it real different form a fizzbuzz filter which you can toss in as a 15 minute thing (including conversation) as part of a really full-spectrum first hour.
|
|
#
?
Feb 6, 2023 15:17
|
|
|
my interview for my current place was "here's a trimmed down version of a class we use; we ran into some performance problems. where might the problems be, how might you help them" etc my answers related to "indexes on the main db table it's hitting, can we cache the list better" and I don't remember what else got me the job, so... but yeah that was a really good discussion, I felt like I could explain my reasoning without scrambling for "uhh how do I do this problem that no one is facing ever in a short period of time"
|
|
#
?
Feb 6, 2023 16:04
|
|
|
|
| # ? May 23, 2024 19:51 |
|
|
champagne posting posted:I dislike fizzbuzz. it tells me nothing as an interviewer and even less as an interviewee my most recent interviews that went well had both "here's the docs for the github api, do this simple thing with it" and "here's some code, review it" and i thought they were both really good approaches silvergoose posted:my interview for my current place was "here's a trimmed down version of a class we use; we ran into some performance problems. where might the problems be, how might you help them" etc this sounds great
|
|
#
?
Feb 6, 2023 16:12
|
|