Foundation-model post-training usually relies on human labels, preference data, stronger teachers, or executable verifiers. We study Unsupervised Post-Training (UPT): update-bearing adaptation on unlabeled inputs whose learning signal is derived from same-lineage model artifacts rather than an external oracle. We catalog 80 strict UPT methods and organize them by the object that supplies the update signal: a prediction statistic, a sample relation, a self-generated target, or an internal evaluator. Beyond inventory, we show how the choice of internal signal and task structure determines whether post-training improves the model or recursively amplifies error. An orthogonal Input Visibility × Update Persistence view maps deployment regimes and defines a unified framework for UPT selection and evaluation.

Dental AI scribes convert chairside dental speech into structured electronic medical records (EMRs). Public data for this task is scarce because dental visits include private speech, regional language, and noisy operatory audio. We introduce DentalVoice, a Chinese dental dialogue and speech dataset for AI scribe research. It contains 13,861 multi-turn dialogues, 162,612 turns, and approximately 274 hours of generated speech. Each case is grounded in a real dental EMR, guided by a dental knowledge graph, conditioned on patient profiles, and augmented with chairside acoustic conditions. We validate the generated dialogues using LLM judges and clinical experts, finding that clinical knowledge retrieval improves safety and professionalism, while persona constraints enhance interaction realism. Next, we benchmark dental automatic speech recognition (ASR) and Subjective-Objective-Assessment-Plan (SOAP) note extraction. Results reveal persistent errors in dialectal ASR, dental terminology recognition, treatment normalization, and distinguishing planned from completed procedures.