I'm quite conflicted on this. While I do not think one needs to remember/memorize a bunch of brainteasers or past computer scientists'/mathematician's PhD discoveries in order to build CRUD applications.

However, I do feel like there is perhaps some amount of truth to the thought behind the interview questions, no? As in, I would imagine someone that could invert a binary tree in 15 minutes on a whiteboard could probably learn React. However, I am not sure everyone that can learn React can invert a binary tree in 15 minutes on a whiteboard.

However, maybe I am projecting my own insecurities because I wish I could invert a binary tree in 15 minutes on a whiteboard as well as being able to solve all those other problems.

No, it's not actually about mastery of binary trees or other abstract computer science concepts.

leetcode is really a culture fit test and success points to the combination of "smarts", diligence, and conformity to some acceptable degree. It shows you have some baseline of familiarity with computing, can focus on an arbitrary task to pursue a goal, and will conform to an arbitrary process when it's asked of you.

Those are genuinely essential skills in an organization with 1000's of white collar professional workers.

More precise insight is gained when the test covers a binary tree inversion than that were it just some contrived logic puzzle like the LSAT or a critical reading exercise. The computer science bit does provide some signal and isn't completely arbitrary, but it's only a small part of what's being evaluated.

Among otherwise strong engineers, it tends to filter out the especially willful, prideful, independent, meandering, creative, and pragmatic ones. These personality types can be extremely valuable in some work environments and can still sometimes in through leetcode challenges, but spoil big bureuacratic systems like FAANG's when they're overrepresented.