Making LLMs Say What They Know: Probe-Targeted Fine-Tuning for Verbal Confidence Calibration

Abstract Verbal confidence in instruct-tuned large language models fails because the readout pathway from internal correctness representations to the confidence-token position transmits little of the available signal. Linear probes on hidden states discriminate correct from incorrect responses at AUROC2 = 0.76-0.88 across seven of eight models spanning four families and three scales, yet verbal confidence saturates near ceiling. We introduce probe-targeted confidence-calibrated supervised fine-tuning (PT-CSFT), which uses a linear probe on a model's own hidden states to generate continuous confidence targets for LoRA fine-tuning. PT-CSFT recovers 91-115% of probe discrimination in verbal confidence at 7-32B. At 70B, a two-stage curriculum closes 66% of the gap on the logit channel (AUROC2 = 0.797), the first VRS-Valid confidence signal at 70B. Controlled activation patching at the confidence-token position supports the interpretation that verbal confidence failure is a position-specific routing problem: the intervention is position-specific, bidirectional (91% forward, 89% reverse), selective (83% of answers unchanged), and follows a near-monotonic layer-depth gradient (Spearman rho = 0.976, p < 1e-4). The logit readout universally rescues where text confidence fails. Multi-seed replication across three model families confirms stability.