Underlining answer - you can't. Useful answer - fine tune on large training set, set temperature to 0, monitor token probability and highlight risk when probab

Underlining answer - you can't. Useful answer - fine tune on large training set, set temperature to 0, monitor token probability and highlight risk when probability < some threshold.