They likely did do actual training, but starting with a general pre-trained model and specializing tends to yield higher quality results faster. It’s so excessively obsequious because they told it to be profoundly and sincerely apologetic if it makes an error, and people don’t actually share the text of real apologies online in a way that’s generic, so it can only copy the tone of form letters and corporate memos.
They likely did do actual training, but starting with a general pre-trained model and specializing tends to yield higher quality results faster. It’s so excessively obsequious because they told it to be profoundly and sincerely apologetic if it makes an error, and people don’t actually share the text of real apologies online in a way that’s generic, so it can only copy the tone of form letters and corporate memos.