No Answer Needed: Predicting LLM Answer Accuracy from Question-Only Linear Probes

Open in new window