These formulations have been selected by reinforcement learning. People who aligned the LLMs chose this over alternatives.

You know when chatbots ask you which answer you prefer between two. People tend to chose the "as a langage model..." one, so it stuck.