[kafkaesq]

After all that it takes at least a few months of settling in to see how they actually perform.

It does? I find incompetent people are generally easy to spot. As in, almost instantaneously (or at most, within a few hours). It's almost as if they want to show you how incompetent they are.

But the "long-term" bad kind? Generally that's not incompetence, per se, but personality issues ("I only wanna do it this way", "I don't want to learn / won't work with people who use X or even don't look down on it like I do", "I just don't give a fuck this company or any of youse", etc). Which by nature are of course much more difficult to spot.

And which are a completely different (orthogonal) set of issues than those addressed... "this idiot didn't immediately see the dynamic programming approach which I knew about already because, of course, I picked the problem. Even when I stared at him impatiently and distracted him with hints" style of filtering which seems to be the goal of the modern hiring process.

[candiodari]

Have you ever been at a big company where you found more than one or two exceptions to "I just don't give a fuck this company or any of youse", and which one, because I'd love to submit my resume there.

Issue with interviewers is that they invariably think themselves smarter than they are. For instance, just because something has a linear programming solution doesn't mean it doesn't just have a closed form solution too (most software engineers that have interviewed me would say "closed what ?", I feel). I've found, often, on interviews that you have to lower yourself to the interviewers' level to pass. Even when the problem is not so much that the interviewer doesn't know the stuff at all, the problem may be that they don't know it well enough to have a thorough understanding of the problem without preparation (and they never prepare), so they simply can't deal with other approaches. Or they provide a "warmup" question that is ridiculously hard or easy, and don't deal well with the fact that you approach it very different from what they expect for the real question.

Very few engineers, even at companies that claim to be different like Google or Microsoft, truly have a mathematical background in algorithms. This does not seem to stop them from often smugly pointing out the "right" solution from a blogpost that happens to be flat-out wrong, ill-specified and handwavy, to a Math PhD. Putting any math on the whiteboard in such situations is a bad idea. Even just pointing out the flaws in their assumptions ... Constructing a proof that it's equivalent to a well-known problem with lower complexity than their optimal solution does not often end well, as they neither can nor want to understand actual algorithm theory. They don't know the assumptions they use, and never once have I known one to question if the assumptions apply to the posed problem.

[kafkaesq]

This confirms with a lot of my observation, also. For example, the modern interview process has tricked interviewers into believing that (armed with the right set of questions and/or hoops to be jumped through), they can "size up" the candidate. When at very best, all you can "measure" is the intersection of your backgrounds.

And about preparation -- it's not so much that they ask pretentiously hard questions sometimes. It's that their own level of preparation, correctness of execution and general forethought will only very seldom even roughly match what they expect the candidate to deliver. As in, they screw up all the time -- everything from picking the problem to stating the problem to stating true expectations (these are basically almost never stated) -- to simply listening to another person's perspective on it... to just watching the clock and their body language and tact in general. Yet candidates are almost always expected to deliver near-flawlessly.

As to big companies: I've found the solid majority to be quite considerate about others as human beings, generally (abstract questions about their company's impact on society or the environment aside). But as a general rule, the larger the company, the more the "not giving a fuck about this company" measure approaches 1 - uniformly and geometrically.