Genetic algorithms kinda suck as they use random variations and breeding to solve a problem which is much slower than using backpropagation with any decent reward modeler. It's the difference between selective breeding and gene splicing in the real world.
Could be. Classification is a type of problem. LLM is a type of model. You can use LLMs to solve classification problems. There's a good chance that's what's happening here.
At least it's routing you to a department instead of trying to help you solve the issue yourself by showing you different help pages you already looked at before trying to contact support.