"I May Not Have Articulated Myself Clearly": Diagnosing Dynamic Instability in LLM Reasoning at Inference Time

• URL: http://arxiv.org/abs/2602.02863v1
• Date: Mon, 02 Feb 2026 22:11:25 GMT
• Title: "I May Not Have Articulated Myself Clearly": Diagnosing Dynamic Instability in LLM Reasoning at Inference Time
• Abstract summary: Reasoning failures in large language models (LLMs) are typically measured only at the end of a generation, yet many failures manifest as a process-level breakdown.<n>We define a simple instability signal that combines consecutive-step distributional shift (JSD) and uncertainty (entropy)<n>Across GSM8K and HotpotQA, instability strength predicts wrong answers with above-chance AUC and yields monotonic bucket-level accuracy decline.
• Score: 1.570719611178015
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reasoning failures in large language models (LLMs) are typically measured only at the end of a generation, yet many failures manifest as a process-level breakdown: the model "loses the thread" mid-reasoning. We study whether such breakdowns are detectable from inference-time observables available in standard APIs (token log probabilities), without any training or fine-tuning. We define a simple instability signal that combines consecutive-step distributional shift (JSD) and uncertainty (entropy), summarize each trace by its peak instability strength, and show that this signal reliably predicts failure. Across GSM8K and HotpotQA, instability strength predicts wrong answers with above-chance AUC and yields monotonic bucket-level accuracy decline at scale across model sizes. Crucially, we show that instability is not uniformly harmful: early instability can reflect subsequent stabilization and a correct final answer (\emph{corrective instability}), whereas late instability is more often followed by failure (\emph{destructive instability}), even at comparable peak magnitudes, indicating that recoverability depends not only on how strongly the distribution changes but also on when such changes occur relative to the remaining decoding horizon. The method is model-agnostic, training-free, and reproducible, and is presented as a diagnostic lens rather than a corrective or control mechanism.

Related papers

• ELVAE: Evidential Learning-Based Variational Autoencoder for Uncertainty-Aware Generation [7.5571037918649155]
  Variational autoencoders generate samples from probabilistic latent representations but do not distinguish uncertainty about the latent location from variability around it.<n>We formulate ELVAE, an evidential learning-based VAE in which each latent coordinate is governed by an input-dependent normal-inverse-gamma posterior.<n>Low-uncertainty anchors support more reliable synthetic samples, while high-uncertainty anchors can be deliberately exploited for stress testing.
  arXiv  Detail & Related papers  (2026-08-11T02:45:59Z)
• From Uncertainty to Failure Attribution: Self-Diagnosing Models for Failure Attribution under Distribution Shift [0.3948325938742681]
  We introduce a problem setting for failure attribution under distribution shift.<n>This enables the models not only to detect OOD samples, but also to find out the reason for their failure.<n>The solution we propose is called self-diagnosing models, which are capable of jointly learning predictive output, predictive uncertainty, and a failure attribution signal.
  arXiv  Detail & Related papers  (2026-08-08T06:32:20Z)
• Factor-Informed Uncertainty Distillation for Gaze Estimation [3.7508723681775606]
  We propose Factor-Informed Uncertainty Distillation (FIUD) to align uncertainty with interpretable image-quality failure modes.<n>A gradient-boosting teacher predicts expected gaze error from factors such as illumination, sharpness, eye visibility and symmetry.<n>A neural student distills these signals via curriculum learning and ranking supervision into a lightweight single-pass uncertainty head.
  arXiv  Detail & Related papers  (2026-07-22T12:20:40Z)
• TS-Fault: Benchmarking Time Series Forecasters Against Structural Faults [22.88447833683736]
  Time series forecasting (TSF) underpins consequential decisions in energy, transportation, finance, and healthcare.<n>TSF models are almost universally ranked by a single number on clean held-out data, under the implicit assumption that it predicts deployed reliability.<n>We present TS-Fault, a benchmark that evaluates forecasting models under explicit, parameterized fault scenarios with controllable semantic difficulty.
  arXiv  Detail & Related papers  (2026-06-16T23:15:20Z)
• Shifting Uncertainty to Critical Moments: Towards Reliable Uncertainty Quantification for VLA Model [10.781664119473966]
  We propose a unified uncertainty quantification approach for predicting rollout success versus failure.<n>Our method substantially improves failure prediction accuracy and yields more reliable signals for failure detection.
  arXiv  Detail & Related papers  (2026-03-18T22:52:03Z)
• Gradients Must Earn Their Influence: Unifying SFT with Generalized Entropic Objectives [22.29000001610794]
  Standard negative log-likelihood for Supervised Fine-Tuning (SFT) applies uniform token-level weighting.<n>This rigidity creates a two-fold failure mode: (i) overemphasizing low-probability targets can amplify gradients on noisy supervision and disrupt robust priors, and (ii) uniform weighting provides weak sharpening when the model is already confident.<n>Existing methods fail to resolve the resulting plasticity--stability dilemma, often suppressing necessary learning signals alongside harmful ones.<n>We introduce Dynamic Entropy Fine-Tuning (DEFT), a parameter-free objective that modulates the
  arXiv  Detail & Related papers  (2026-02-11T22:56:43Z)
• Stability as a Liability:Systematic Breakdown of Linguistic Structure in LLMs [5.96875296117642]
  We show that stable parameter trajectories lead stationary solutions to minimize the forward KL divergence to the empirical distribution.<n>We empirically validate this effect using a controlled feedback-based training framework.<n>It indicates that optimization stability and generative expressivity are not inherently aligned, and that stability alone is an insufficient indicator of generative quality.
  arXiv  Detail & Related papers  (2026-01-26T15:34:50Z)
• Nearly Optimal Bayesian Inference for Structural Missingness [23.988531482641307]
  In the Bayesian view, prediction via the posterior predictive distribution integrates over the full model posterior uncertainty.<n>This framework decouples learning an in-model missing-value posterior from (ii) label prediction by optimizing the predictive posterior distribution.<n>It achieves SOTA on 43 classification and 15 imputation benchmarks, with finite-sample near Bayes-optimality guarantees.
  arXiv  Detail & Related papers  (2026-01-26T14:03:11Z)
• Towards Adversarial Robustness and Uncertainty Quantification in DINOv2-based Few-Shot Anomaly Detection [6.288045889067255]
  Foundation models such as DINOv2 have shown strong performance in few-shot anomaly detection.<n>We present one of the first systematic studies of adversarial attacks and uncertainty estimation in this setting.<n>We find that raw anomaly scores are poorly calibrated, revealing a gap between confidence and correctness that limits safety-critical use.
  arXiv  Detail & Related papers  (2025-10-15T15:06:45Z)
• Error-quantified Conformal Inference for Time Series [55.11926160774831]
  Uncertainty quantification in time series prediction is challenging due to the temporal dependence and distribution shift on sequential data.<n>We propose itError-quantified Conformal Inference (ECI) by smoothing the quantile loss function.<n>ECI can achieve valid miscoverage control and output tighter prediction sets than other baselines.
  arXiv  Detail & Related papers  (2025-02-02T15:02:36Z)
• LoGU: Long-form Generation with Uncertainty Expressions [49.76417603761989]
  We introduce the task of Long-form Generation with Uncertainty(LoGU)<n>We identify two key challenges: Uncertainty Suppression and Uncertainty Misalignment.<n>Our framework adopts a divide-and-conquer strategy, refining uncertainty based on atomic claims.<n>Experiments on three long-form instruction following datasets show that our method significantly improves accuracy, reduces hallucinations, and maintains the comprehensiveness of responses.
  arXiv  Detail & Related papers  (2024-10-18T09:15:35Z)
• Beyond the Norms: Detecting Prediction Errors in Regression Models [26.178065248948773]
  This paper tackles the challenge of detecting unreliable behavior in regression algorithms. We introduce the notion of unreliability in regression, when the output of the regressor exceeds a specified discrepancy (or error) We show empirical improvements in error detection for multiple regression tasks, consistently outperforming popular baseline approaches.
  arXiv  Detail & Related papers  (2024-06-11T05:51:44Z)
• Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning [12.490614705930676]
  We present a theoretical result demonstrating the strong dependency of suboptimality on the number of Monte Carlo samples taken per Bellman target calculation.<n>Our main contribution is a deterministic approximation to the Bellman target that uses progressive moment matching.<n>We show that it is possible to provide tighter guarantees for the suboptimality of MOMBO than the existing Monte Carlo sampling approaches.
  arXiv  Detail & Related papers  (2024-06-06T13:58:41Z)
• Selective Learning: Towards Robust Calibration with Dynamic Regularization [79.92633587914659]
  Miscalibration in deep learning refers to there is a discrepancy between the predicted confidence and performance. We introduce Dynamic Regularization (DReg) which aims to learn what should be learned during training thereby circumventing the confidence adjusting trade-off.
  arXiv  Detail & Related papers  (2024-02-13T11:25:20Z)
• Robustness and Accuracy Could Be Reconcilable by (Proper) Definition [109.62614226793833]
  The trade-off between robustness and accuracy has been widely studied in the adversarial literature. We find that it may stem from the improperly defined robust error, which imposes an inductive bias of local invariance. By definition, SCORE facilitates the reconciliation between robustness and accuracy, while still handling the worst-case uncertainty.
  arXiv  Detail & Related papers  (2022-02-21T10:36:09Z)
• Dense Uncertainty Estimation via an Ensemble-based Conditional Latent Variable Model [68.34559610536614]
  We argue that the aleatoric uncertainty is an inherent attribute of the data and can only be correctly estimated with an unbiased oracle model. We propose a new sampling and selection strategy at train time to approximate the oracle model for aleatoric uncertainty estimation. Our results show that our solution achieves both accurate deterministic results and reliable uncertainty estimation.
  arXiv  Detail & Related papers  (2021-11-22T08:54:10Z)
• Tracking the risk of a deployed model and detecting harmful distribution shifts [105.27463615756733]
  In practice, it may make sense to ignore benign shifts, under which the performance of a deployed model does not degrade substantially. We argue that a sensible method for firing off a warning has to both (a) detect harmful shifts while ignoring benign ones, and (b) allow continuous monitoring of model performance without increasing the false alarm rate.
  arXiv  Detail & Related papers  (2021-10-12T17:21:41Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.