Pre-training large language models (PrLMs) on massive unlabeled corpora and fine-tuning them on downstream tasks has become a new paradigm [1-3]. Their success can be partly attributed to the self-attention mechanism [4], yet these self-attention networks are often redundant [5, 6] and tend to cause overfitting when fine-tuned on downstream tasks due to the mismatch between their overparameterization and the limited annotated data [7-13]. To address this issue, various regularization techniques such as data augmentation [14, 15], adversarial training [16, 17]), and dropout-based methods [11,13,18]have been developed.

Learning from indirect supervision signals is important in real-world AI applications when, often, gold labels are missing or too costly. In this paper, we develop a unified theoretical framework for multi-class classification when the supervision is provided by a variable that contains nonzero mutual information with the gold label. The nature of this problem is determined by (i) the transition probability from the gold labels to the indirect supervision variables and (ii) the learner's prior knowledge about the transition. Our framework relaxes assumptions made in the literature, and supports learning with unknown, non-invertible and instance-dependent transitions. Our theory introduces a novel concept called \emph{separation}, which characterizes the learnability and generalization bounds. We also demonstrate the application of our framework via concrete novel results in a variety of learning scenarios such as learning with superset annotations and joint supervision signals.

Modeling complex subjective tasks in Natural Language Processing, such as recognizing emotion and morality, is considerably challenging due to significant variation in human annotations. This variation often reflects reasonable differences in semantic interpretations rather than mere noise, necessitating methods to distinguish between legitimate subjectivity and error. We address this challenge by exploring label verification in these contexts using Large Language Models (LLMs). First, we propose a simple In-Context Learning binary filtering baseline that estimates the reasonableness of a document-label pair. We then introduce the Label-in-a-Haystack setting: the query and its label(s) are included in the demonstrations shown to LLMs, which are prompted to predict the label(s) again, while receiving task-specific instructions (e.g., emotion recognition) rather than label copying. We show how the failure to copy the label(s) to the output of the LLM are task-relevant and informative. Building on this, we propose the Label-in-a-Haystack Rectification (LiaHR) framework for subjective label correction: when the model outputs diverge from the reference gold labels, we assign the generated labels to the example instead of discarding it. This approach can be integrated into annotation pipelines to enhance signal-to-noise ratios. Comprehensive analyses, human evaluations, and ecological validity studies verify the utility of LiaHR for label correction. Code is available at https://github.com/gchochla/liahr.

School of Computer Science, Carnegie Mellon University, Pittsburgh, P A, USA ABSTRACT Disinformation campaigns can distort public perception and destabilize institutions. Understanding how different populations respond to information is crucial for designing effective interventions, yet real-world experimentation is impractical and ethically challenging. To address this, we develop an agent-based simulation using Large Language Models (LLMs) to model responses to misinformation. We construct agent personas spanning five professions and three mental schemas, and evaluate their reactions to news headlines. Our findings show that LLM-generated agents align closely with ground-truth labels and human predictions, supporting their use as proxies for studying information responses. We also find that mental schemas, more than professional background, influence how agents interpret misinformation. This work provides a validation of LLMs to be used as agents in an agent-based model of an information network for analyzing trust, polarization, and susceptibility to deceptive content in complex social systems. 1 INTRODUCTION Protection against foreign information campaigns and the ability to conduct effective information operations are critical to modern national security. In an era where the information domain can be leveraged as a battlefield, there is a need to maintain information advantage, defined as "the use, protection, and exploitation of information to achieve objectives more effectively than enemies and adversaries do" (U.S. Achieving and sustaining information advantage requires not only the ability to disseminate compelling narratives but also to detect, counter, and mitigate adversarial information operations.

We investigate the degree to which human plausibility judgments of multiple-choice commonsense benchmark answers are subject to influence by (im)plausibility arguments for or against an answer, in particular, using rationales generated by LLMs. We collect 3,000 plausibility judgments from humans and another 13,600 judgments from LLMs. Overall, we observe increases and decreases in mean human plausibility ratings in the presence of LLM-generated PRO and CON rationales, respectively, suggesting that, on the whole, human judges find these rationales convincing. Experiments with LLMs reveal similar patterns of influence. Our findings demonstrate a novel use of LLMs for studying aspects of human cognition, while also raising practical concerns that, even in domains where humans are ``experts'' (i.e., common sense), LLMs have the potential to exert considerable influence on people's beliefs.

There is increasing evidence of Human Label Variation (HLV) in Natural Language Inference (NLI), where annotators assign different labels to the same premise-hypothesis pair. However, within-label variation--cases where annotators agree on the same label but provide divergent reasoning--poses an additional and mostly overlooked challenge. Several NLI datasets contain highlighted words in the NLI item as explanations, but the same spans on the NLI item can be highlighted for different reasons, as evidenced by free-text explanations, which offer a window into annotators' reasoning. To systematically understand this problem and gain insight into the rationales behind NLI labels, we introduce LITEX, a linguistically-informed taxonomy for categorizing free-text explanations in English. Using this taxonomy, we annotate a subset of the e-SNLI dataset, validate the taxonomy's reliability, and analyze how it aligns with NLI labels, highlights, and explanations. We further assess the taxonomy's usefulness in explanation generation, demonstrating that conditioning generation on LITEX yields explanations that are linguistically closer to human explanations than those generated using only labels or highlights. Our approach thus not only captures within-label variation but also shows how taxonomy-guided generation for reasoning can bridge the gap between human and model explanations more effectively than existing strategies.

Reinforcement learning with human-annotated data has boosted chain-of-thought reasoning in large reasoning models, but these gains come at high costs in labeled data while faltering on harder tasks. A natural next step is experience-driven learning, where models improve without curated labels by adapting to unlabeled data. We introduce RESTRAIN (REinforcement learning with Self-restraint), a self-penalizing RL framework that converts the absence of gold labels into a useful learning signal. Instead of overcommitting to spurious majority votes, RESTRAIN exploits signals from the model's entire answer distribution: penalizing overconfident rollouts and low-consistency examples while preserving promising reasoning chains. The self-penalization mechanism integrates seamlessly into policy optimization methods such as GRPO, enabling continual self-improvement without supervision. On challenging reasoning benchmarks, RESTRAIN delivers large gains using only unlabeled data. With Qwen3-4B-Base and OctoThinker Hybrid-8B-Base, it improves Pass@1 by up to +140.7 percent on AIME25, +36.2 percent on MMLU_STEM, and +19.6 percent on GPQA-Diamond, nearly matching gold-label training while using no gold labels. These results demonstrate that RESTRAIN establishes a scalable path toward stronger reasoning without gold labels.

We propose Semantic F1 Scores, novel evaluation metrics for subjective or fuzzy multi-label classification that quantify semantic relatedness between predicted and gold labels. Unlike the conventional F1 metrics that treat semantically related predictions as complete failures, Semantic F1 incorporates a label similarity matrix to compute soft precision-like and recall-like scores, from which the Semantic F1 scores are derived. Unlike existing similarity-based metrics, our novel two-step precision-recall formulation enables the comparison of label sets of arbitrary sizes without discarding labels or forcing matches between dissimilar labels. By granting partial credit for semantically related but nonidentical labels, Semantic F1 better reflects the realities of domains marked by human disagreement or fuzzy category boundaries. In this way, it provides fairer evaluations: it recognizes that categories overlap, that annotators disagree, and that downstream decisions based on similar predictions lead to similar outcomes. Through theoretical justification and extensive empirical validation on synthetic and real data, we show that Semantic F1 demonstrates greater interpretability and ecological validity. Because it requires only a domain-appropriate similarity matrix, which is robust to misspecification, and not a rigid ontology, it is applicable across tasks and modalities.

Large Language Models (LLMs) perform remarkably well in Natural Language Inference (NLI). However, NLI involving numerical and logical expressions remains challenging. Comparatives are a key linguistic phenomenon related to such inference, but the robustness of LLMs in handling them, especially in languages that are not dominant in the models' training data, such as Japanese, has not been sufficiently explored. To address this gap, we construct a Japanese NLI dataset that focuses on comparatives and evaluate various LLMs in zero-shot and few-shot settings. Our results show that the performance of the models is sensitive to the prompt formats in the zero-shot setting and influenced by the gold labels in the few-shot examples. The LLMs also struggle to handle linguistic phenomena unique to Japanese. Furthermore, we observe that prompts containing logical semantic representations help the models predict the correct labels for inference problems that they struggle to solve even with few-shot examples.

Using LLM-generated labels to fine-tune smaller encoder-only models for text classification has gained popularity in various settings. While this approach may be justified in simple and low-stakes applications, we conduct empirical analysis to demonstrate how the perennial curse of training on synthetic data manifests itself in this specific setup. Compared to models trained on gold labels, we observe not only the expected performance degradation in accuracy and F1 score, but also increased instability across training runs and premature performance plateaus. These findings cast doubts on the reliability of such approaches in real-world applications. We contextualize the observed phenomena through the lens of error propagation and offer several practical mitigation strategies, including entropy-based filtering and ensemble techniques. Although these heuristics offer partial relief, they do not fully resolve the inherent risks of propagating non-random errors from LLM annotations to smaller classifiers, underscoring the need for caution when applying this workflow in high-stakes text classification tasks.