Comment by daishi55

With how competitive the LLM field is, it would surprise me greatly if any of these players were doing anything other than trying to make the best possible product. I certainly do not believe they are intentionally training the models to use more tokens unnecessarily.