Scientific applications of large language models (LLMs) demand reliable, well-calibrated predictions, but standard generative approaches often fail to fully access relevant knowledge contained in their internal representations. As a result, models appear less capable than they are, with useful information remaining latent. Post‑training interventions (instruction‑tuning, safety‑alignment) can further mask domain knowledge without erasing it from model representations. We present PING (Probing INternal states of Generative models), an open-source framework that trains lightweight probes on frozen, HuggingFace-compatible transformers to deliver structured predictions with minimal compute overhead. Across diverse models and benchmarks, including MMLU for broad applicability and MedMCQA for clinical focus, PING matches or exceeds generative accuracy while reducing Expected Calibration Error by up to 96\%. Crucially, PING recovers knowledge suppressed by safety filters, restoring up to 89\% of lost performance on MedMCQA. By making probing practical through the pingkit package, our work enables safer, efficient LLM deployment in research.
Berkowitz et al. (Fri,) studied this question.