Now going back to the original premise, despite this code being wrong, as ChatGPT almost invariably does, do you think it understands the concepts here or is just statistically generating tokens based on previous input?

Really for code generation ChatGPT is an incremental step over StackOverflow. It can kind of nicely template the things you tell it into the code, for the most part, but the code is almost always fundamentally wrong or just mashed together in some way.