I wonder how much transcribed conversation data companies like Amazon and Google have from their household listening devices (Echo etc.). The scale and context of that type of data would be a scary levelling up for LLMs.