Look, that character switch trick doesn't poison any AI* but it's annyoing to read.

* Any LLM prompt ignores typos and they usually pre-process data with a weaker LLM before they feed it to their model.