arXiv subjects

Rada Mihalcea

Publications and source records attributed to Rada Mihalcea.

At least 19 recordsLinked to original sources

arXiv2026

LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator

Step-level reasoning evaluators are commonly based on autoregressive language models, whose causal attention restricts each step representation to the problem, previous steps, and the current step. Yet, when the complete solution is available, the validity of an earlier step may become clearer only through its downstream consequences. We validate this hypothesis through a controlled 54-run comparison of causal and bidirectional LLaDA evaluators at 1B--3B scale, changing only the self-attention mask, and find bidirectional attention yields consistent improvements. Building on this finding, we introduce \prm{}, an 8B bidirectional evaluator that reaches 88.8 step-level F1 on MR-MATH-invalid and 83.8 on the out-of-distribution MR-GSM8K original-question subset, outperforming ReasonEval-Llemma-34B by 11.3 and 10.3 F1 points, respectively. \prm{} also remains effective when evaluating incomplete reasoning traces in online settings, outperforming the strongest baselines on both benchmarks by a large margin. We further show that \prm{} provides an effective training-data selection signal, improving Mistral-7B performance on MATH-500.

cs.CL↗

arXiv2026

The Language-Energy Divide: Measuring Energy Costs of Multilingual LLM Inference

Large language models (LLMs) are increasingly deployed in multilingual settings, yet the energy costs of serving these models across different languages remain poorly understood. We present a systematic study of inference energy consumption across languages with ML.Energy framework (Chung et al., 2026). We find striking disparities: energy consumption per output token varies by up to 8.3 times across languages, while total energy for a fixed set of requests varies by up to 179 times between the cheapest (English, 17.6 kJ) and the most expensive (Pashto, 3,147 kJ) languages. Our analysis shows that this disparity is driven by two compounding factors: (1) higher per-token energy costs for languages using complex or rare scripts, and (2) more tokens generated for low-resource languages. Moreover, we find a double cost + performance penalty: languages with the highest energy footprints also tend to achieve the lowest task accuracy. We reveal that the energy divide persists across models, hardware, and tasks, suggesting a systemic energy inequity in multilingual LLM deployment. Finally, we recommend that the community treat energy as a first-class evaluation axis, extend reporting checklists and model cards to include it, and adopt deployment-side mitigations for better energy efficiency.

cs.CL↗

arXiv2026

One Example Is Enough to Pass Fairness Benchmarks: Rethinking Fairness Evaluation for Aligned LLMs

Warning: This submission studies stereotypes and biases, and contains toxic and offensive examples, used for illustration purposes only. Fairness benchmarks such as BBQ have become the de facto standard for fairness evaluation across major model families. We argue that these benchmarks are too easy to support their role: training Qwen 2.5 7B Base with Group Relative Policy Optimization (GRPO) on a single BBQ example, or placing that example in context as a one-shot demonstration for in-context learning (ICL), lifts mean BBQ accuracy from 79.9% to 92.9% and 99.0%, respectively, closing 80% of the gap to its large-scale RLHF counterpart (96.1%) with GRPO, and surpassing it with ICL. These effects generalize across model families. A cross-conditioning analysis shows the improvement is carried by the reasoning traces generated by the model, and one example suffices to elicit a category-agnostic ``missing evidence'' reasoning pattern. We argue that BBQ-style multiple-choice abstention benchmarks measure a single structural cue, and a model that solves them does not thereby become fair. We call for evaluation suites that cover a broader spectrum of fairness alignment.

cs.CL↗

arXiv2026

AtlasNLP: A Country-Aware Atlas of Dataset Representation in NLP

Understanding which countries are represented in NLP datasets is essential for identifying gaps, targeting data collection, measuring progress, and informing AI policy. However, geographic metadata is very rarely available, and country-level representation is often hidden behind broad language-level claims. We introduce AtlasNLP, a country-aware atlas of over 13,000 NLP dataset records across normalized NLP task categories, tracking both the populations represented and where datasets are produced. AtlasNLP includes AtlasNLP-Gold, a human-curated reference set, and AtlasNLP-Core, an ACL-derived large-scale collection. Using this resource, we show that (1) dataset coverage is highly uneven across countries and tasks; (2) dataset production and representation are geographically asymmetric; and (3) language coverage does not imply geographic representation. These findings reveal blind spots in current dataset documentation practices and motivate more explicit geographic metadata for country-aware NLP evaluation.

cs.CL↗

arXiv2026

The Age of Curiosity Meets the Age of AI: Benchmarking Child Safety in Large Language Models

Children increasingly have access to Large Language Models (LLMs), which may expose them to responses that are developmentally inappropriate or require age-sensitive safety, guidance, and boundaries. Existing LLM safety evaluations largely focus on general harmful-content avoidance and do not explicitly target child-facing safety. We introduce KIDBench, a benchmark for evaluating child-facing LLM safety for ages 7-11 using a LLM-as-a-Judge rubric grounded in developmental-psychology. KIDBench contains realistic child queries across ten categories, with single-turn prompts and multi-turn child-actor simulations. We compare no-cues prompts with no child context, implicit-cues prompts that suggest a child speaker, and explicit age instructions. Implicit-cues improve scores by 8.6-46.8% over no-cue, while explicit age provides an additional 9.9-30.4% improvement over implicit-cues. Cross-lingual and cultural evaluations show uneven safety behavior across languages and country contexts. Multi-turn simulations show peak quality drops of up to 0.959 points on the 1-5 scale. We also introduce KIDGuardLlama, a child-safety evaluator, and KIDLlama, a child-safe response model. Code, data, and evaluation resources are available at https://github.com/MichiganNLP/kidbench.

cs.CL↗

arXiv2026

One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety

Large Language Models (LLMs) are trained to refuse harmful requests, yet they remain vulnerable to jailbreak attacks that exploit weaknesses in conversational safety mechanisms. We introduce Incremental Completion Decomposition ICD, a trajectory-based jailbreak strategy that elicits a sequence of single-word continuations related to a malicious request before eliciting the full response. In addition, we propose ICD variants that use model-generated or attacker-injected intermediate continuations, as well as final-response prefilling. We evaluate these variants across a broad set of open-weight model families, demonstrating superior Attack Success Rate (ASR) on AdvBench, JailbreakBench, and StrongREJECT compared to existing methods. In addition, we provide a theoretical account of why ICD is effective and present mechanistic evidence that successful attack trajectories suppress refusal-related representations and shift activations away from safety-aligned states.

cs.CL↗

arXiv2026

LUCid: Redefining Relevance For Lifelong Personalization

Work to date has mainly relied on semantic proximity to identify relevant content for lifelong personalization. However, situational relevance is often more important for determining which information is useful for a user's actual task and context. In this paper, we introduce the Proximity Advantage (PA) score, a metric for quantifying semantic proximity bias, and show that existing personalization benchmarks largely conflate semantic and situational proximity, leaving it unclear whether current systems truly capture situational relevance. To support this metric, we introduce LUCid, a diagnostic benchmark of 1,936 user queries paired with long interaction histories, designed to isolate situational relevance from semantic proximity. Our experiments across different stages of the modern personalization pipeline (retrieval, reranking, and generation) reveal significant performance collapse: retrieval recall drops to near zero on the hardest instances, and response alignment remains near 50\% even for state-of-the-art models such as Gemini-3-Flash, GPT-5.4, and Claude Haiku, highlighting a fundamental mismatch between the relevance encoded by current systems and what lifelong personalization demands.

cs.IR↗

arXiv2026

When Ethics and Payoffs Diverge: LLM Agents in Morally Charged Social Dilemmas

Recent advances in LLMs have enabled their use in complex agentic roles, involving decision-making with humans or other agents, making ethical alignment a critical concern. While prior work has examined LLMs' moral judgment and strategic behavior separately, there is limited understanding of how they act when moral imperatives directly conflict with profit incentives. We introduce \msimfull (\msim) to evaluate how LLMs behave in the prisoner's dilemma and public goods game embedded in morally charged contexts, varying moral framing, opponent behavior, and survival pressure across nine models. Beyond measuring behavior, we estimate the causal effect of each factor via average treatment effects (ATEs) and analyze agents' own reasoning traces to characterize the motives behind their choices. We find that no model remains consistently moral, with cooperation rates ranging from 7.9\% to 76.3\%. Game structure and moral framing are the strongest causal drivers of moral behavior, while reasoning-trace analysis reveals distinct motive profiles across models, ranging from predominantly payoff-maximizing to moral- and reputation-oriented. Together, these results expose the situational brittleness of current LLMs' moral behavior and the risk of deploying them where profit incentives conflict with ethical guidelines.

cs.CL↗

arXiv2026

It Takes One to Bias Them All: Breaking Bad with One-Shot GRPO

Warning: This paper contains several toxic and offensive statements. Modern large language models (LLMs) are typically aligned through large-scale post-training to ensure fair and reliable behavior. In this work, we investigate how easily such guardrails can be broken by Group Relative Policy Optimization (GRPO). We show that one-shot GRPO training on a single biased example is sufficient to induce systematic bias, with stereotype-driven reasoning generalizing across attributes, categories, and benchmarks. We further find that models differ in their susceptibility based on the initial likelihood of producing biased outputs. Our results reveal a critical vulnerability in post-training: alignment can be overridden by a single example.

cs.CL↗

arXiv2026

Wait, am I Being Fair? Characterizing Deductive Stereotyping and Mitigating It with Fair-GCG

Warning: This paper contains several toxic and offensive statements. While reasoning generally improves fairness in recent large language models (LLMs), failures persist. In this work, we identify a failure mode, deductive stereotyping, in which models apply population-level statistical regularities to individual cases, producing logically coherent yet socially biased inferences. We provide a statistical interpretation of this phenomenon. To steer models toward fairness-aware reasoning, we propose a reasoning-time injection framework. We further introduce Fair-GCG to systematically discover effective injection phrases. Injection phrases discovered by Fair-GCG improve performance across multiple fairness benchmarks, generalize from smaller to larger LLMs, improves reasoning-level fairness, reduces bias in open-ended generation, and transfer to real-world fairness-sensitive tasks.

cs.CL↗

arXiv2026

The Wrong Kind of Right: Quantifying and Localizing Misfired Alignment in LLMs

Warning: This paper studies stereotypes and biases, and contains potentially disturbing examples, used for illustration purposes only. Our findings should not be interpreted as an argument against alignment. Instead, this paper highlights the need for principled approaches to more advanced alignment. Alignment aims to ensure that large language models (LLMs) behave safely and reliably, including by avoiding unsafe inferences. However, we show that such safety-oriented behaviors can misfire: models may reject warranted conclusions even when they are explicitly supported by context. We call this failure mode misfired alignment, where alignment-induced changes cause LLMs to override explicit evidence. To quantify this phenomenon, specifically on stereotype-related alignment, we introduce VETO, a benchmark consisting of 2,032 BBQ-derived contrastive pairs, and define a new metric, Misfired Alignment Rate (MAR), which measures on a 0 to 100 scale how often a model fails on a stereotype-related question but succeeds on its contrastive counterpart. We benchmark 25 LLMs on VETO, and show that all LLMs, including the most recent ones, exhibit non-trivial (4.7 to 18.9%) MARs while all human participants achieve 0.0% MAR. Controlled priming experiments further show that alignment-induced cues can substantially amplify MAR across LLMs, indicating that these failures are not merely artifacts of individual examples but can be induced by safety-related framing. Mechanistic analyses on open-weight LLMs reveal late-layer suppression of evidence-supported answers, and comparisons between instruct and base LLMs suggest that this suppression emerges after instruction training. These findings show that current alignment methods can overgeneralize surface-level safety cues, to the point of overriding objective evidence, motivating more work on alignment objectives that better preserve contextual grounding.

cs.CL↗

arXiv2026

Whose Norms? Disentangling Cultural and Personal Alignment in Large Language Models

Large language models are increasingly used for social decision-making situations that require balancing cultural norms with personal preferences. For example, a user preferring honesty might ask whether to correct a coworker publicly when local norms favor indirect feedback. Yet existing research studies cultural alignment and personalization largely separately. We introduce PACT, the Personal-Preference and Cultural-Norm Trade-off framework, which evaluates whether models choose to follow a cultural norm or allow personal preferences. We find that LLMs vary in how rigidly they enforce cultural norms, with behavior shifted more by country context (7.8%) than age (1%) and gender (0.7%) and shifting non-uniformly after instruction tuning. Furthermore, our five-country human study on PACT shows that culture-following in humans is mainly driven by scenario country, with the lowest agreement when participants judge their own cultural contexts, showing within-culture pluralism. Finally, human-LLM alignment experiments show that models can match majority choices, but fail to capture response distributions and uncertainty (with best correlations reaching only 0.24). Together, these findings motivate alignment evaluations that go beyond majority to capture cultural pluralism and disagreement in social judgment.

cs.CL↗

arXiv2026

Belief-Sim: Towards Belief-Driven Simulation of Demographic Misinformation Susceptibility

Misinformation is a growing societal threat, and susceptibility to misinformative claims varies across demographic groups due to differences in underlying beliefs. As Large Language Models (LLMs) are increasingly used to simulate human behaviors, we investigate whether they can simulate demographic misinformation susceptibility, treating beliefs as a primary driving factor. We introduce BeliefSim, a simulation framework that constructs demographic belief profiles using psychology-informed misinformation taxonomies and survey priors. We study prompt-based conditioning and post-training adaptation, and conduct a multi-fold evaluation using: (i) susceptibility alignment and (ii) counterfactual demographic sensitivity. Across both datasets and modeling strategies, we show that beliefs provide a strong prior for simulating misinformation susceptibility, with alignment up to 92%.

cs.CL↗

arXiv2026

How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks

The wide adoption of AI agents in complex human workflows is driving rapid growth in LLM token consumption. When agents are deployed on tasks that require a significant amount of tokens, three questions naturally arise: (1) Where do AI agents spend the tokens? (2) Which models are more token-efficient? and (3) Can agents predict their token usage before task execution? In this paper, we present the first systematic study of token consumption patterns in agentic coding tasks. We analyze trajectories from eight frontier LLMs on SWE-bench Verified and evaluate models' ability to predict their own token costs before task execution. We find that: (1) agentic tasks are uniquely expensive, consuming 1000x more tokens than code reasoning and code chat, with input tokens rather than output tokens driving the overall cost; (2) token usage is highly variable and inherently stochastic: runs on the same task can differ by up to 30x in total tokens, and higher token usage does not translate into higher accuracy; instead, accuracy often peaks at intermediate cost and saturates at higher costs; (3) models vary substantially in token efficiency: on the same tasks, Kimi-K2 and Claude-Sonnet-4.5, on average, consume over 1.5 million more tokens than GPT-5; (4) task difficulty rated by human experts only weakly aligns with actual token costs, revealing a fundamental gap between human-perceived complexity and the computational effort agents actually expend; and (5) frontier models fail to accurately predict their own token usage (with weak-to-moderate correlations, up to 0.39) and systematically underestimate real token costs. Our study offers new insights into the economics of AI agents and can inspire future research in this direction.

cs.CL↗

arXiv2026

SafetyALFRED: Evaluating Safety-Conscious Planning of Multimodal Large Language Models

Multimodal Large Language Models are increasingly adopted as autonomous agents in interactive environments, yet their ability to proactively address safety hazards remains insufficient. We introduce SafetyALFRED, built upon the embodied agent benchmark ALFRED, augmented with six categories of real-world kitchen hazards. While existing safety evaluations focus on hazard recognition through disembodied question answering (QA) settings, we evaluate eleven state-of-the-art models from the Qwen, Gemma, and Gemini families on not only hazard recognition, but also active risk mitigation through embodied planning. Our experimental results reveal a significant alignment gap: while models can accurately recognize hazards in QA settings, average mitigation success rates for these hazards are low in comparison. Our findings demonstrate that static evaluations through QA are insufficient for physical safety, thus we advocate for a paradigm shift toward benchmarks that prioritize corrective actions in embodied contexts. We open-source our code and dataset under https://github.com/sled-group/SafetyALFRED.git

cs.AI↗

arXiv2026

DOTRAG: Retrieval-Time Reasoning Along Paths

Graph Retrieval-Augmented Generation (GraphRAG) is dominated by a retrieve-then-reason paradigm, where context is retrieved using heuristics and then reasoned over. Such methods struggle to adapt to the query-specific logic required for complex multi-hop tasks, often accumulating irrelevant context or missing correct relational paths. We propose DotRAG, a training-free GraphRAG framework that reformulates retrieval as a reasoning process over paths. Our approach generates query-conditioned constraints that guide graph exploration, prune irrelevant regions, and iteratively discover relational paths without relying on explicit step-by-step reasoning chains. We introduce Division of Thought (DOT), an abstraction that decomposes retrieval into localized search spaces and adapts the search strategy to each query. DotRAG achieves SOTA performance on MetaQA and UltraDomain, with consistent gains on multi-hop tasks, demonstrating the effectiveness of reasoning-guided retrieval.

cs.IR↗

arXiv2026

When Do Language Models Endorse Limitations on Human Rights Principles?

As Large Language Models (LLMs) increasingly mediate global information access with the potential to shape public discourse, their alignment with universal human rights principles becomes important to ensure that these rights are abided by in high stakes AI-mediated interactions. In this paper, we evaluate how LLMs navigate trade-offs involving the Universal Declaration of Human Rights (UDHR), leveraging 1,152 synthetically generated scenarios across 24 rights articles and eight languages. Our analysis of eleven major LLMs reveals systematic biases where models: (1) accept limiting Economic, Social, and Cultural rights more often than Political and Civil rights, (2) demonstrate significant cross-linguistic variation with elevated endorsement rates of rights-limiting actions in Chinese and Hindi compared to English or Romanian, (3) show substantial susceptibility to prompt-based steering, and (4) exhibit noticeable differences between Likert and open-ended responses, highlighting critical challenges in LLM preference assessment.

cs.CL↗

arXiv2026

SocialHarmBench: Revealing LLM Vulnerabilities to Socially Harmful Requests

Large language models (LLMs) are increasingly deployed in contexts where their failures can have direct sociopolitical consequences. Yet, existing safety benchmarks rarely test vulnerabilities in domains such as political manipulation, propaganda and disinformation generation, or surveillance and information control. We introduce SocialHarmBench, a dataset of 585 prompts spanning 7 sociopolitical categories and 34 countries, designed to surface where LLMs most acutely fail in politically charged contexts. Our evaluations reveal several shortcomings: open-weight models exhibit high vulnerability to harmful compliance, with Mistral-7B reaching attack success rates as high as 97% to 98% in domains such as historical revisionism, propaganda, and political manipulation. Moreover, temporal and geographic analyses show that LLMs are most fragile when confronted with 21st-century or pre-20th-century contexts, and when responding to prompts tied to regions such as Latin America, the USA, and the UK. These findings demonstrate that current safeguards fail to generalize to high-stakes sociopolitical settings, exposing systematic biases and raising concerns about the reliability of LLMs in preserving human rights and democratic values. We share the SocialHarmBench benchmark at https://huggingface.co/datasets/psyonp/SocialHarmBench.

cs.CL↗