> a lot of kids coming out of college have done little more than memorize Leetcode problems and outsourced classwork to AI
Even before AI, there were issues with super-polished leetcode grinders. Their entire skillset was passing FAANG interviews via memorization of correct solutions and scripted answers.
An effective technique for identifying these cases leverages the fact that you can only have a "correct" answer memorized if optimal solutions exist. Fortunately, there are many common and relatively simple problems in computer science and software design that require reasoning from first principles because globally optimal solutions can't exist even in theory. Small changes to the problem constraints lead to wildly divergent design outcomes that are effectively not enumerable. The possible solution space is so large that you can't realistically memorize it even if the problem is relatively concise and well-understood. The pure leetcode grinders never seem to have studied problems without tidy answers.
The answers don't even matter that much, I am always more interested in the demonstrated ability to recognize and reason through the implications of constraint changes when there are no correct answers. People with solid computer science skills can usually muddle their way through it, whereas many people with immaculate leetcode skills fail the most basic versions of this.
The most important skill of juniors was demonstrating that they could effectively reason about and were motivated to dive into problems they had never seen before. These were always the juniors that could be rapidly developed into strong senior engineers, which is more or less the objective when you hire juniors.