Language model outputs are not always reliable, thus prompting research into how to adapt model responses based on uncertainty. Common approaches include: abstention, where models refrain from generating responses when uncertain; and linguistic calibration, where models hedge their statements using uncertainty quantifiers. However, abstention can withhold valuable information, while linguistically calibrated responses are often challenging to leverage in downstream tasks. We propose a unified view, Conformal Linguistic Calibration (CLC), which reinterprets linguistic calibration as answer set prediction. First we present a framework connecting abstention and linguistic calibration through the lens of linguistic pragmatics. We then describe an implementation of CLC that allows for controlling the level of imprecision in model responses. Results demonstrate our method produces calibrated outputs with conformal guarantees on factual accuracy. Further, our approach enables fine-tuning models to perform uncertainty-aware adaptive claim rewriting, offering a controllable balance between factuality and specificity.1

For regression on the dataset, we perform leave-one-out cross validation. For the single solvents,865 we leave out one solvent at a time. For the full data, we leave out one solvent ramp at a time. We866 measure the performance of the model on each leave-one-out data split, then take the mean of their867 performance across the dataset. We exclude any experiments involving acetonitrile and acetic acid,868 due to the observed side-reactions.

Conformal prediction is a framework for predictive inference with a distributionfree, finite-sample guarantee. However, it tends to provide uninformative prediction sets when calibration data are scarce. This paper introduces Synthetic-powered predictive inference (SPI), a novel framework that incorporates synthetic data-- e.g., from a generative model--to improve sample efficiency. At the core of our method is a score transporter: an empirical quantile mapping that aligns nonconformity scores from trusted, real data with those from synthetic data. By carefully integrating the score transporter into the calibration process, SPIprovably achieves finite-sample coverage guarantees without making any assumptions about the real and synthetic data distributions. When the score distributions are well aligned, SPIyields substantially tighter and more informative prediction sets than standard conformal prediction. Experiments on image classification--augmenting data with synthetic diffusion-model generated images--and on tabular regression demonstrate notable improvements in predictive efficiency in data-scarce settings.

Despite their strengths, large language models (LLMs) often fail to communicate their confidence accurately, making it difficult to assess when they might be wrong and limiting their reliability. In this work, we demonstrate that reasoning models that engage in extended chain-of-thought (CoT) reasoning exhibit superior performance not only in problem-solving but also in accurately expressing their confidence. Specifically, we benchmark six reasoning models across six datasets and find that they achieve strictly better confidence calibration than their non-reasoning counterparts in 33 out of the 36 settings. Our detailed analysis reveals that these gains in calibration stem from the slow thinking behaviors of reasoning models (e.g., exploring alternative approaches and backtracking) which enable them to adjust their confidence dynamically throughout their CoT, making it progressively more accurate. In particular, we find that reasoning models become increasingly better calibrated as their CoT unfolds, a trend not observed in non-reasoning models. Moreover, removing slow thinking behaviors from the CoT leads to a significant drop in calibration. Lastly, we show that non-reasoning models also demonstrate enhanced calibration when simply guided to slow think via in-context learning, fully isolating slow thinking as the source of the calibration gains.2

This paper introduces a novel kernel-based formulation of the Equalized Odds (EO) criterion, denoted as EOk, for fair representation learning (FRL) in supervised settings. The central goal of FRL is to mitigate discrimination regarding a sensitive attribute S while preserving prediction accuracy for the target variable Y. Our proposed criterion enables a rigorous and interpretable quantification of three core fairness objectives: independence (bY S), separation-also known as equalized odds (bY S | Y), and calibration (Y S | bY). Under both unbiased (Y S) and biased (Y S) conditions, we show that EOk satisfies both independence and separation in the former, and uniquely preserves predictive accuracy while lower bounding independence and calibration in the latter, thereby offering a unified analytical characterization of the tradeoffs among these fairness criteria. We further define the empirical counterpart, dEOk, a kernel-based statistic that can be computed in quadratic time, with linear-time approximations also available. A concentration inequality for dEOk is derived, providing performance guarantees and error bounds, which serve as practical certificates of fairness compliance. While our focus is on theoretical development, the results lay essential groundwork for principled and provably fair algorithmic design in future empirical studies.

Simulation-based inference (SBI) of latent parameters in physical systems is often hindered by model misspecification-the mismatch between simulated and real-world observations caused by inherent modeling simplifications. RoPE, a recent SBI approach, addresses this challenge through a two-stage domain transfer process that combines semi-supervised calibration with optimal transport (OT)based distribution alignment. However, RoPE operates in a fully transductive setting, requiring access to a batch of test samples at inference time, which limits scalability and generalization. We propose a fully inductive and amortized SBI framework that integrates calibration and distributional alignment into a single, end-to-end trainable model called FRISBI. Our method leverages mini-batch OT with a closed-form coupling to align real and simulated observations that correspond to the same latent parameters, using both paired calibration data and unpaired samples. A conditional normalizing flow is then trained to approximate the OTinduced posterior, enabling efficient inference without simulation access at test time. Across a range of synthetic and real-world benchmarks-including complex medical biomarker estimation-our approach matches or exceeds the performance of RoPE, while offering improved scalability and applicability in challenging, misspecified environments.

Recently, Large Language Models (LLMs) have been increasingly used to support various decision-making tasks, assisting humans in making informed decisions. However, when LLMs confidently provide incorrect information, it can lead humans to make suboptimal decisions. To prevent LLMs from generating incorrect information on topics they are unsure of and to improve the accuracy of generated content, prior works have proposed Retrieval Augmented Generation (RAG), where external documents are referenced to generate responses. However, previous RAG methods focus only on retrieving documents most relevant to the input query, without specifically aiming to ensure that the human user's decisions are well-calibrated. To address this limitation, we propose a novel retrieval method called Calibrated Retrieval-Augmented Generation (CalibRAG), which ensures that decisions informed by RAG are well-calibrated. Then we empirically validate that CalibRAG improves calibration performance as well as accuracy, compared to other baselines across various datasets.

Selective classifiers improve model reliability by abstaining on inputs the model deems uncertain. However, few practical approaches achieve the gold-standard performance of a perfect-ordering oracle that accepts examples exactly in order of correctness. Our work formalizes this shortfall as the selective-classification gap and present the first finite-sample decomposition of this gap to five distinct sources of looseness: Bayes noise, approximation error, ranking error, statistical noise, and implementation-or shift-induced slack. Crucially, our analysis reveals that monotone post-hoc calibration--often believed to strengthen selective classifiers--has limited impact on closing this gap, since it rarely alters the model's underlying score ranking. Bridging the gap therefore requires scoring mechanisms that can effectively reorder predictions rather than merely rescale them. We validate our decomposition on synthetic two-moons data and on real-world vision and language benchmarks, isolating each error component through controlled experiments. Our results confirm that (i) Bayes noise and limited model capacity can account for substantial gaps, (ii) only richer, feature-aware calibrators meaningfully improve score ordering, and (iii) data shift introduces a separate slack that demands distributionally robust training. Together, our decomposition yields a quantitative error budget as well as actionable design guidelines that practitioners can use to build selective classifiers which approximate ideal oracle behavior more closely.

However, human preferences often diverge across users, contexts, and cultures. As a result, disagreement collapses into the majority signal and minority perspectives are discounted. To address this, we propose reflecting diverse human preferences through a distribution over multiple reward functions, each inducing a distinct aligned policy. The distribution is learned directly from pairwise preference without annotator identifiers or predefined groups. Instead, annotator disagreements are treated as informative soft labels. Our central criterion is pairwise calibration: for every pair of candidate responses, the proportion of reward functions preferring one response matches the fraction of annotators with that preference. We prove that even a small outlier-free ensemble can accurately represent diverse preference distributions. Empirically, we introduce and validate a practical training heuristic to learn such ensembles, and demonstrate its effectiveness through improved calibration, implying a more faithful representation of pluralistic values.

Large Language Models (LLMs) are increasingly deployed in high-stakes domains such as science, law, and healthcare, where accurate expressions of uncertainty are essential for reliability and trust. However, current LLMs are often observed to generate incorrect answers with high confidence--a phenomenon known as "overconfidence". Recent efforts have focused on calibrating LLMs' verbalized confidence: i.e., their expressions of confidence in text form, such as "I am 80% confident that...". Existing approaches either rely on prompt engineering or fine-tuning with heuristically generated uncertainty estimates, both of which have limited effectiveness and generalizability. Motivated by the notion of proper scoring rules for calibration in classical machine learning models, we introduce ConfTuner, a simple and efficient fine-tuning method that introduces minimal overhead and does not require ground-truth confidence scores or proxy confidence estimates. ConfTuner relies on a new loss function, tokenized Brier score, which we theoretically prove to be a proper scoring rule, intuitively meaning that it "correctly incentivizes the model to report its true probability of being correct". ConfTuner improves calibration across diverse reasoning tasks and generalizes to black-box models such as GPT-4o. Our results further show that better-calibrated confidence enables downstream gains in self-correction and model cascade, advancing the development of trustworthy LLM systems.