arXiv:2605.01048v2 Announce Type: replace-cross Abstract: Counterfactual prompting (i.e., perturbing a single factor and measuring output change) is widely used to evaluate things like LLM bias and C...
arXiv:2606. 08365v1 Announce Type: cross Abstract: Sparse autoencoder (SAE) features are increasingly used to steer language models, but feature steering is rarely clean: the same intervention can behave inconsistently across contexts and perturb unrelated features.
arXiv:2509. 08604v5 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have demonstrated significant potential in medicine, with many studies adapting them through continued pre-training or fine-tuning on medical data to enhance domain-specific accuracy and safety.
The paper introduces a benchmark of over 6,000 clinical triage scenarios, 7,000 physician annotations, and 225,000 large language model (LLM) responses to assess how LLMs perform under realistic variations in clinical text. The study finds that LLMs tend to recommend unnecessary care more often than physicians, especially when the input text is perturbed, and that LLM recommendations are more sensitive to gender and tone changes than human recommendations. These findings underscore the importance of deployment‑oriented evaluations that reflect expert physician behavior.
The paper highlights hidden data challenges in clinical machine learning, focusing on how commonly reused preprocessing pipelines for electronic health records embed assumptions about data availability, episode definitions, and missingness. Using MIMIC-IV, the authors identify violations in temporal, semantic, and population alignment—leading to temporal leakage, feature-task misalignment, and biased imputation—and show that these ambiguities can alter model signals without affecting overall predictive performance. They provide practical recommendations and open-source code for more temporally valid, clinically aligned, and reproducible preprocessing.
arXiv:2609.08173v1 Announce Type: new Abstract: Sparse autoencoder (SAE)-based steering has been widely used to address knowledge conflicts by guiding LLMs to be more faithful to the contextual knowl...
The paper introduces CAST, a concept-guided artifact suppression tuning framework that uses sparse autoencoders to identify and suppress note-specific artifacts in clinical language models. CAST labels latent features with an LLM-assisted pipeline and ICD‑10 constraints, then fine‑tunes the model while providing post‑hoc per‑concept attributions for auditability. In experiments on MIMIC‑IV discharge‑note mortality prediction, CAST outperforms standard fine‑tuned encoders and competes with strong LLM baselines while offering a feature‑level audit trail of clinical concepts and suppressed artifacts.
arXiv:2605.20292v2 Announce Type: replace Abstract: Numerical time-series models effectively process irregular electronic health record (EHR) trajectories, but do not expose which temporal patterns s...
arXiv:2608. 16643v1 Announce Type: cross Abstract: Automated detection of errors in clinical documentation is a promising application of large language models (LLMs), yet decisions to deploy such models rest on benchmarks that evaluate each clinical note in isolation.
arXiv:2608.29582v1 Announce Type: cross Abstract: Current evaluations of large language models (LLMs) primarily focus on factual knowledge retrieval, overlooking the fundamental challenge of navigati...
The paper investigates whether clinical concepts are represented as distinct, causally influential centers within the latent space of large language models (LLMs). By evaluating eleven open-weight LLMs, the authors discover that each model contains dedicated clinical concept centers that are interpretable, activate only on relevant clinical narratives, and drive model behavior in both constrained and open-ended contexts. These centers can be leveraged for evaluation and performance improvement, as steering models along them enhances downstream clinical outcomes and aligns with clinician preferences.
Automated detection of errors in clinical documentation is a promising application of large language models (LLMs), yet decisions to deploy such models rest on benchmarks that evaluate each clinical note in isolation. Error-detection benchmarks are typically constructed by injecting errors into notes, such that each erroneous note has a natural counterpart.