AI RESEARCH
When Correct Demonstrations Hurt: Rethinking the Role of Exemplars in In-Context Learning
arXiv CS.AI
•
ArXi:2605.26350v1 Announce Type: cross In-context learning (ICL) is often motivated by the intuition that nstrations help because they provide correct input-output examples. However, we reveal a counterintuitive phenomenon: correctness does not guarantee exemplar utility, and some correct nstrations can even reduce ICL accuracy. To study this correctness-utility gap, we