Improving LLM First-Token Predictions in Multiple-Choice Question Answering via Prefilling Attack

Open in new window