1 paper
Sai Koneru, Elphin Joe, Christine Kirchhoff +2
In contested domains, instruction-tuned language models must balance user-alignment pressures against faithfulness to the in-context evidence. To evaluate this tension, we introduc…