When Correct Demonstrations Hurt: Rethinking the Role of Exemplars in In-Context Learning

• URL: http://arxiv.org/abs/2605.26350v1
• Date: Mon, 25 May 2026 21:52:49 GMT
• Title: When Correct Demonstrations Hurt: Rethinking the Role of Exemplars in In-Context Learning
• Abstract summary: In-context learning (ICL) is often motivated by the intuition that demonstrations help because they provide correct input-output examples.<n>We show that correctness does not guarantee exemplar utility, and some correct demonstrations can even reduce ICL accuracy.<n>Our results show that robust ICL requires evaluating not only whether demonstrations are correct, but also how they influence contextual inference.
• Score: 14.102553871638243
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In-context learning (ICL) is often motivated by the intuition that demonstrations help because they provide correct input-output examples. However, we reveal a counterintuitive phenomenon: correctness does not guarantee exemplar utility, and some correct demonstrations can even reduce ICL accuracy. To study this correctness-utility gap, we introduce task-preserving perturbations, where only the exemplar input is changed, while the example remains a correct instance of the same task. Concretely, each perturbed exemplar is assigned the target induced by the task mapping. This framework covers both label-updating perturbations, where task-relevant semantics change and targets are recomputed, and stricter target-preserving perturbations, where the original target remains valid. We formalize the resulting failure mode as contextual evidence shift: task-preserving perturbations can change the effective mixture of evidence used by the model for contextual inference, thereby separating exemplar correctness from exemplar utility. Across sentiment classification, logical reasoning, and math word problems, we find that task-preserving perturbed demonstrations can substantially degrade ICL performance, especially for smaller models, harder tasks, and higher perturbation ratios. Our results show that robust ICL requires evaluating not only whether demonstrations are correct, but also how they influence contextual inference. Code is available at https://github.com/Chenghao-Qiu/Task-Preserving-ICL.

Related papers

• Grounding Without Corrective Control: Truth-Tracking Profiles for Large Language Models [0.0]
  Recent work suggests that some large language model representations have content or reference.<n>This paper asks what follows from that gap.<n>An output is answerable when discrepancies can affect what a target- and task-specific arrangement produces, accepts, or withdraws.
  arXiv  Detail & Related papers  (2026-08-14T12:30:19Z)
• When Is a Task Vector Enough? An Empirical Theory of Implicit Multimodal ICL [2.964628733433076]
  Implicit multimodal in-context learning compresses demonstrations into internal interventions, ranging from static task vectors to query-conditioned transformations and attention routing.<n>We propose the Selection--Realization Hypothesis, which views demonstrations as inducing a compact family of internal changes from which the query selects.<n>By contrasting correct demonstrations with matched counterfactuals, we measure the structure of explicit M-ICL and test whether it predicts intervention behavior.
  arXiv  Detail & Related papers  (2026-08-13T15:46:50Z)
• Robots That Know What to Ask: Recovering Misaligned Rewards through Targeted Explanations [1.9662978733004601]
  Learning reward functions from demonstrations assume that demonstrations provide adequate supervision over all features.<n>In practice, demonstrations are often imperfect: humans may under-emphasize certain features due to cognitive load or physical difficulty.<n>We propose a framework that detects such underspecified features and actively solicits targeted corrective demonstrations.
  arXiv  Detail & Related papers  (2026-05-21T19:34:14Z)
• Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs [100.02824137397464]
  We investigate how Large Language Models adapt their internal representations when encountering inputs of increasing difficulty.<n>We reveal a consistent and quantifiable phenomenon: as task difficulty increases, the last hidden states of LLMs become substantially sparser.<n>This sparsity--difficulty relation is observable across diverse models and domains.
  arXiv  Detail & Related papers  (2026-03-03T18:48:15Z)
• Exploring Structural Degradation in Dense Representations for Self-supervised Learning [84.52554180480037]
  We observe a counterintuitive phenomenon in self-supervised learning (SSL): longer training may impair the performance of dense prediction tasks.<n>We refer to this phenomenon as Self-supervised Dense Degradation (SDD) and demonstrate its consistent presence across sixteen state-of-the-art SSL methods.<n>We introduce a Dense representation Structure Estimator (DSE) composed of a class-relevance measure and an effective dimensionality measure.
  arXiv  Detail & Related papers  (2025-10-20T08:40:16Z)
• Mitigating Spurious Correlations Between Question and Answer via Chain-of-Thought Correctness Perception Distillation [25.195244084313114]
  Chain-of-Thought Correctness Perception Distillation (CoPeD) aims to improve the reasoning quality of the student model.<n>CoPeD encourages the student model to predict answers based on correct rationales and revise them when they are incorrect.
  arXiv  Detail & Related papers  (2025-09-06T05:33:17Z)
• Latent Veracity Inference for Identifying Errors in Stepwise Reasoning [78.29317733206643]
  We introduce Veracity Search (VS), a discrete search algorithm over veracity assignments.<n>It performs otherwise intractable inference in the posterior distribution over latent veracity values.<n>It generalizes VS, enabling accurate zero-shot veracity inference in novel contexts.
  arXiv  Detail & Related papers  (2025-05-17T04:16:36Z)
• PICLe: Pseudo-Annotations for In-Context Learning in Low-Resource Named Entity Detection [56.916656013563355]
  In-context learning (ICL) enables Large Language Models to perform tasks using few demonstrations.<n>We propose PICLe, a framework for in-context learning with noisy, pseudo-annotated demonstrations.<n>We evaluate PICLe on five biomedical NED datasets and show that, with zero human annotation, PICLe outperforms ICL in low-resource settings.
  arXiv  Detail & Related papers  (2024-12-16T16:09:35Z)
• Disentangling Latent Shifts of In-Context Learning with Weak Supervision [0.0]
  In-context learning (ICL) enables large language models to perform few-shot learning by conditioning on labeled examples in the prompt.<n>Despite its flexibility, ICL suffers from instability as prompt length increases with more demonstrations.<n>We propose a parameter-efficient method that disentangles demonstration-induced latent shifts from those of the query.
  arXiv  Detail & Related papers  (2024-10-02T13:00:21Z)
• Enhancing In-Context Learning via Implicit Demonstration Augmentation [26.78252788538567]
  In-context learning (ICL) enables pre-trained language models to make predictions for unseen inputs without updating parameters. Despite its potential, ICL's effectiveness heavily relies on the quality, quantity, and permutation of demonstrations. In this paper, we tackle this challenge for the first time from the perspective of demonstration augmentation.
  arXiv  Detail & Related papers  (2024-06-27T05:25:46Z)
• DETAIL: Task DEmonsTration Attribution for Interpretable In-context Learning [75.68193159293425]
  In-context learning (ICL) allows transformer-based language models to learn a specific task with a few "task demonstrations" without updating their parameters.<n>We propose an influence function-based attribution technique, DETAIL, that addresses the specific characteristics of ICL.<n>We experimentally prove the wide applicability of DETAIL by showing our attribution scores obtained on white-box models are transferable to black-box models in improving model performance.
  arXiv  Detail & Related papers  (2024-05-22T15:52:52Z)
• Enhancing Vision-Language Few-Shot Adaptation with Negative Learning [11.545127156146368]
  We propose a Simple yet effective Negative Learning approach, SimNL, to more efficiently exploit task-specific knowledge. To this issue, we introduce a plug-and-play few-shot instance reweighting technique to mitigate noisy outliers. Our extensive experimental results validate that the proposed SimNL outperforms existing state-of-the-art methods on both few-shot learning and domain generalization tasks.
  arXiv  Detail & Related papers  (2024-03-19T17:59:39Z)
• Learning What Makes a Difference from Counterfactual Examples and Gradient Supervision [57.14468881854616]
  We propose an auxiliary training objective that improves the generalization capabilities of neural networks. We use pairs of minimally-different examples with different labels, a.k.a counterfactual or contrasting examples, which provide a signal indicative of the underlying causal structure of the task. Models trained with this technique demonstrate improved performance on out-of-distribution test sets.
  arXiv  Detail & Related papers  (2020-04-20T02:47:49Z)

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.