Get GPT3 to avoid modal verbs / weasel words


I’m using GPT3 to write passages of text to use in emails or blogs. I’m getting good results but I want GPT3 to cut out using too many hedge words / weasel words and sound more decisive in its output.

For example, I find GPT3 uses words such as

  • maybe
  • kind of
  • possibly
  • could be really helpful

Rather than saying, “I’m certain,” for example, GPT3 may say, “I’m pretty certain.”,
Rather than saying, “It will help”, “it might be able to help you”

Can anyone lend any tips to nudge GPT3 to use more decisive words in its vocabulary when prompt engineering using DaVinci Instruct?

At moment i’m adding an instruction: “Write cold email using the passage of text. Do not use filler words, hedge words, modal verbs or weasel words”.

Any help would be much appreciated.

Include “formal” and “academic” in the prompt. It will speak much more confidently.


Thanks have noticed an improvement.

1 Like

You can also send logit bias to prevent these, for example send logit token keys with value -100 for may and might. This will have impact on readability/output if you do it too much, though…

Thanks, I was not familiar with logit bias until now!

I’ve had mixed results with logit bias, so I general focus on other techniques