A user asks why there are no language models trained with entire sentences as single tokens, suggesting such a model could act as a 'rough talker' with strong reasoning capabilities. They propose using one token for phrases like 'food flushed down the toilet' to simplify input processing.