Comment by sadrasabouri

It argues that training for "math, code, and reasoning" caused this. Do LLMs only trained for natural language (probably older models) have the same characteristics? Does training for these additional domains help in some ways, and this specific style of writing is just a side-effect of that?