Leetcode is popular because:
1. It gauges problem solving ability and pattern recognition, which are essential to any job, and particularly useful for software engineers.
2. It is language agnostic - most jobs would rather have a strong engineer that doesn't know their particular language/tech stack than a weak engineer that is familiar with the stack.
3. It has objective evaluation criteria (correctness + time/space complexity).
It's certainly not perfect, but I have yet to see an alternative that accomplishes all of the above. The one thing I can see potentially complementing it is the new style of AI-assisted assessment where you are dropped into an unknown codebase, and asked to make fixes/features that are assisted by an AI agent. This solves the language-agnostic issue with pre-AI versions of these interviews, where you had to have memorized specific tech stack features or language syntax in order to accomplish the task. Now AI can take care of that stuff, and the interviewee can focus on the problem solving aspect.
Proof of claim required. Leetcode primarily measure's someone's ability to memorize leetcode problems, and secondarily measures their ability to vomit this knowledge up in a stressful situation.
There's an entire test-prep industry surrounding this interview method. Any undergrad will tell you that the way you pass leetcode interviews is to "grind". You'd have to be oblivious to not realize that the system has been gamed to hell and back.
Since you are just asserting things, I will assert a simpler explanation: the primary reason Leetcode is popular is because it is cheap. If you have the problem that you have too many applicants and don't want to spend money interviewing them all, you have to come up with an efficient filter that doesn't violate hiring laws. You don't particularly care if the false negative rate is high, so a test of superficially relevant trivia is what you do. Even better if the trivia can be validated by computer!
Setting aside the fact that almost no companies actually had this problem outside of FAANG, thanks to AI, the approach has gone from "somewhat relevant" to "completely irrelevant" within a couple of years. It's time to take it out back and shoot it.
I’m sure the companies know that Leetcode is not a great indicator for likelihood to succeed in the role. If they had a perfect information about people, then would stop interviewing at all and just take the best people who applied.
They don’t have perfect information. The information is highly asymmetric. The companies are just trying to not be efficient and not get burned.
It sucks for candidates. But it makes sense for companies when they thousands of applicants they know effectively nothing about.
In other fields, those things aren't nearly as meaningless and easy to game as they are in ours.
Well, no. It rewards ability (and free time) to repeatedly work through problems and memorize the solutions. And it rewards acting skills so you can perform like you're discovering solutions on the fly, that you've actually memorized.
That's all it does.