ChatGPT was additionally finetuned after the initial training of the base model, so it's not completely clear if it truly represents the original distribution a

ChatGPT was additionally finetuned after the initial training of the base model, so it's not completely clear if it truly represents the original distribution anymore. I use open weights models and they repeat such phrases less frequently in my experience.