This could be useful to ban offesnsive words or encourage some words to set the tone (e.g. understand, resolve, etc.) Ref: https://help.openai.com/en/articles/5247780-using-logit-bias-to-alter-token-probability-with-the-openai-api