Unfortunately I fear this is going to become the standard in our industry. Everyone will be forced to rely on AI to "ship faster", to trade learning for short term speed.
* The most recent one: I was doing an "AI assisted coding interview". The problem itself was simple. I gathered clear specs, I explained what I planned to do. I was supposed to use AI so I wrote down the main function signatures I expected (the API boundary) and wrote in the prompt what I wanted Claude to do. I wrote no code myself other than editing the output. When I got rejected, I was told it was because "I wrote too much code myself".
* Once I was asked a brain teaser. I solved the initial problem, but one of the follow-ups made it significantly more challenging. I wrestled with the problem for a few minutes, and realizing I was going around in circles I stated so and told the interviewer I wasn't sure how to proceed. I was expecting a tip or at least an acknowledgement, but I heard nothing. Blank silence with the interviewer staring at the screen. Since it was a zoom call, I thought my internet was down, but when I asked "hey, can you hear me?", he replied yes, and went back to radio silence. This was a pattern of the interviewer throughout the interview. Later on, after this question I implemented an algorithm and was asked for its time complexity. I mistakenly said O(n) (I forgot the initial sort), and the interviewer literally just stared at the screen and said nothing until after 10 seconds or so when I realized my mistake and corrected it - at which point he acknowledged and moved to the next question.
* Another one that happened two times (at different companies) is getting asked a very vague question, like "how do you fix a bug in production" (to which I reply with 'I try to replicate it locally, I go through logs, etc') and then being told by the recruiter the interviewer didn't like that my responses were "too generic".
Author here. I expect most to be ironed out too, regardless of whether they're publicized or not. The engineers working on these models are aware of the problems
Cool method. Pre deep learning there was plenty of interesting research on sparse methods. What do you think we're missing to have more widely used neural+sparse approaches?