There's no reasonable definition of "hard-coded" that fits how LLMs are made.

The whole idea behind "hard-coded" and original symbolic AI was that a skilled human deliberately wrote explicit code to define a program's behavior.

Machine learning is the exact opposite of that approach.