Xinpeng Wang

Latest

• My Answer is C: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models