How exactly do you code LLMs?
Do you just keep a set of higher-level instructions and roll that shit back to an earlier version if it ever goes off the rails?