Targeted Head Intervention Cuts LLM Overconfidence Without Bleeding Accuracy
Researchers introduce Probe‑Conditioned Head Intervention (PCHI), an inference‑time tweak that detects overconfident wrong answers and rescales specific attention‑head outputs. On Qwen‑3‑4B‑Instruct, PCHI flips 82% of erroneous high‑confidence predictions and halves calibration error while hurting only 5% of correct confidences.