Only allow one token to answer. Use logit bias to make "0" or "1" the most probable tokens. Ask it "Is this message an apology? Return 0 for no, 1 for yes." Fee

Only allow one token to answer. Use logit bias to make "0" or "1" the most probable tokens. Ask it "Is this message an apology? Return 0 for no, 1 for yes." Feed it only the first 25 tokens of the message you're checking.