Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

chrlschn.dev

Post reply on HN
Print view View original

21–30 of 286 posts

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#21

Post by Aurornis »

christofosho wrote: post #7

For take home reviews, I wonder what the impact of generative AI would be. I can see a lot of people feeding the code segment into an AI and using it's comments instead of generating their own. Of course, people will do that with programming questions as well, so I suppose it's up to the hiring manager to decide on the best method to handle this situation. What strategies do people leverage to avoid purely AI answers…

A good take-home interview process usually involves a review phase where the candidate talks about their solution and how it works.

This is remarkably effective at filtering out people who had a little too much help in solving it, either from friends, Googling it, or now using an LLM.

It doesn’t happen often, but from time to time someone will come in with a solution they supposedly wrote in the past week but they are unable to discuss it. “I don’t remember exactly what I did here…” and other excuses. It doesn’t take much discussion to reveal someone who never actually understood their own submission.

Of course, I always continue the technical interview to cover the odd possibility that maybe they were too nervous or something. So far I haven’t had anyone who is unable to discuss their own take-home yet shows well in the rest of the interview process.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#22

Post by chefandy »

I like this. In the last dev organization I worked with, we made an artificial practical coding problem, like adding a small feature to an existing purpose-made codebase. It was nice to directly compare the results of the same task among several candidates, and we were fortunate enough to be authorized to pay them for their time. All great but it was a ton of work... Making the challenge, doing code reviews, debating valid differences in approach or unexpected but understandable interpretations of the challenge...

I think the code review method could tell you a lot of the same things about a candidate and show whether or not people can issue criticism without being jerks.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#23

Post by osigurdson »

bluGill wrote: post #13
kemiller wrote: post #4

Earlier quoted context omitted.

I agree, but it's near-universal. Not sure what that says about our industry.

There is research into how an interview should be done to find who will be a good or bad employee. Most companies (not just our industry!) don't appear to have even looked at it. (I'm not sure how useful the research is for real world hiring)

Links?

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#24

Post by Aurornis »

napo wrote: post #10
swader999 wrote: post #6

I like this article a lot and have had great success with this tactic (as a hiring manager) over the years. I often give a page or two of code as a take home assignment to discuss at the interview. They get a chance to look at it and research it on their own. And they also get a bit of the crucible of face to face thinking and reacting on their feet while you discuss it with them in the interview.

why opting for an "home assignment" instead of doing that as part of the interview?

In my experience, most candidates prefer it. They get to use their own computer and own environment and nobody is pressuring them while they code.

It’s also most similar to the job, where I won’t be standing over their shoulder doing live coding.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#25

Post by shortrounddev2 »

nextlevelwizard wrote: post #11
hotnfresh wrote: post #5

Earlier quoted context omitted.

Especially when they don’t pay FAANG money. If I were capable of looking good during the only experience in my adult life that feels like having a solo in a middle school strings recital, I’d be applying to those places, not yours.

I don't know if this is what you meant, but to me this reads as: "I don't know anything about data structures or algorithms".

Not him but

1.) DS and Algo knowledge are way overrated in day-to-day job functions. I have yet to have to implement a single DS or Algo from those classes (they are almost always available to the framework, standard library, or language), runtime complexity has literally never once come up in my life, and the most I've ever needed to know about data structures is "which one should I use here?" and 99% of the time its either a vector (arraylist) or a map (dictionary).

2.) Just because you don't have a proclivity for code-golf does not mean you don't know DS and Algos, it just means that you don't waste your free time writing unreadable code for useless puzzles which serve only as interview questions.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#26

Post by bluehorizon2 »

People are forgetting why companies these days do Leetcode in the first place.

The initial reason why we have Leetcode today is that Google originally determined through their interview research that the smartest candidates were the ones who were best at algorithms. Google wanted to hire the smartest people, not necessarily the best coders, so that’s why their interviews were mostly algorithmic.

Of course everyone started copying Google without doing the same research and the message got lost. They just blindly asked coding algo questions and that’s why we have an entire industry built around coding algos and the signal is entirely lost now.

It’s funny that people have lost that intent and are now creating derivatives around that, ie thinking code reviews are a good replacement.

It might be, but there needs to be data done to research it to see if you get what you want. If you interview using code reviews, you get good code reviewers. If that’s what you want, great.

But remember that Google’s initial goal was to hire smart people, and there’s no research to suggest that good code reviewers are smarter than average.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#27

Post by shortrounddev2 »

CharlieDigital wrote: post #9
otteromkram wrote: post #2

I like this idea, but still think that time and people watching you will add pressure. Most coding challenges that non-FAANG companies give aren't terribly hard; they mostly want you to work through your process out loud, ask questions, interact. These soft skills are as important as technical aptitude in determining whether or not you'd be a good fit with the organization.

> but still think that time and people watching you will add pressure When I conduct interviews, I always try to set a conversational tone from the very beginning. Code reviews in general lend themselves more to a conversational interaction and discussion rather than trying to read, process, construct, and respond in a coding challenge.

It's admirable, but the interviewee is always under some kind of pressure to perform. if you are taking a conversational tone, they're going out of their way to be similarly nonchalant

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#28

Post by shortrounddev2 »

nextlevelwizard wrote: post #12
otteromkram wrote: post #2

I like this idea, but still think that time and people watching you will add pressure. Most coding challenges that non-FAANG companies give aren't terribly hard; they mostly want you to work through your process out loud, ask questions, interact. These soft skills are as important as technical aptitude in determining whether or not you'd be a good fit with the organization.

But you will have people watching you work and you will be under time pressure in normal work as well, so if you just turn to spagetti at interview are you actually going to perform any better at day to day?

I literally never have anyone watching me code, and the time pressure of an interview is way shorter than a 2 week development cycle

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#29

Post by Aurornis »

shortrounddev2 wrote: post #25
nextlevelwizard wrote: post #11

Earlier quoted context omitted.

I don't know if this is what you meant, but to me this reads as: "I don't know anything about data structures or algorithms".

Not him but 1.) DS and Algo knowledge are way overrated in day-to-day job functions. I have yet to have to implement a single DS or Algo from those classes (they are almost always available to the framework, standard library, or language), runtime complexity has literally never once come up in my life, and the most I've ever needed to know about data structures is "which one should I use here?" and 99% of the time it…

> runtime complexity has literally never once come up in my life

There are definitely different types of programming jobs, but for some contrast: I have never had a job where runtime complexity did not come up as a significant factor somewhat frequently.

It really comes down to what you’re doing, but at a certain scale or size of problem you really do need to know about algorithms, data structures, and runtime complexity to avoid having your runtime explode, server costs become untenable, opening the door to extremely long running requests and other problems.

Re: Interviews in the Age of AI: Ditch Leetcode – Try Code Reviews Instead

#30

Post by hcks »

bluehorizon2 wrote: post #26

People are forgetting why companies these days do Leetcode in the first place. The initial reason why we have Leetcode today is that Google originally determined through their interview research that the smartest candidates were the ones who were best at algorithms. Google wanted to hire the smartest people, not necessarily the best coders, so that’s why their interviews were mostly algorithmic. Of course everyone st…

It’s just a proxy measure for IQ really. They could have handed out Ravens matrices and got better outcomes prediction.
Post reply on HN