Cherry picking in artificial intelligence research (and in some other contexts) is the act of choosing examples of good (accurate, grammatical, etc.) predictions. I wonder which word or phrase expresses the contrary of "cherry picking" is (i.e., choosing examples of bad predictions, for example to see what kind of mistakes the model makes).
I do not believe that there is a specific term for that in general English. (Specialists in certain fields might have specific terms for it, though.) M-W defines the verb "cherry-pick" as:
to select the best or most desirable
In this case, "best or most desirable" refers to what is most desired by the cherry-picker. If the person wants to select only "bad" predictions, then those are the "best or most desirable" predictions for that individual and that individual is still cherry-picking.
By the way, M-W's thesaurus offers some antonyms for "cherry-pick", but they consist of words such as "refuse", "reject", and "decline". None of those has the meaning of selecting anything bad.