Fundamentally, the issue is the interviewing/selection process itself. The best interview that I've encountered is the long interview. I hire you for four months as a co-op, see what you can do, and we go from there. Impractical in many/most settings, obviously.
But this is usually not possible- it's contingent on other people having already solved and posted solutions online for some question worded nearly the same. Much of the knowledge behind interview questions applies to all sorts of programming problems where the solution won't be available to copy. A knowledgeable programmer would know quickly that some problem is best approached by dfs or bfs, while the cheater may not.
> Fundamentally, the issue is the interviewing/selection process itself
I agree, but this will always be a hard process. LC problems are like IQ test or SATs, they work fairly well most of the time, but probably better to weed out the particularly weak candidates than determine how good people truly are.
I think there is plenty of reason to think people who actually know the stuff instead of cheating would be better on average. It's a signal that they were smart enough and motivated enough to learn the material (basic algorithms/coding skills in this case)- both traits that would probably make them better on average at the job.
Being a good cheater is a useful skill that probably also helps people on average- but I doubt it's nearly as useful an indicator of future success as being good at the skills necessary to do well on interview questions (w/o cheating).