General-purpose Vision-Language Models (VLMs) are increasingly integral to modern AI systems for document understanding, yet their ability to perform fine-grained layout analysis remains severely underdeveloped. Overcoming this limitation requires large-scale, high-fidelity training datasets. However, current annotation methods that rely on parsing rendered PDFs are costly, error-prone, and difficult to scale. We propose a different paradigm: extracting ground-truth layout directly from the LaTeX compilation process rather than the final PDF. We present LaTeX2Layout, a generalizable procedural pipeline that recovers pixel-accurate bounding boxes and reading order from compiler traces. This enables the generation of a 140K-page dataset, including 120K programmatically generated synthetic variants that more than double the layout diversity of real-world data. Using this dataset, we fine-tune an efficient 3B-parameter VLM with an easy-to-hard curriculum that accelerates convergence. Our model achieves Kendall's tau=0.95 for reading order and mAP@50=0.91 for element grounding, delivering nearly 200% relative improvement over strong zero-shot baselines such as GPT-4o and Claude-3.7.

As AI tools become embedded in productivity and self-improvement contexts, a pressing question emerges: what happens when AI does the goal-setting for us? While large language models can generate goals that are objectively well-formed, the motivational consequences of delegating this cognitively and emotionally significant task remain unknown. In a preregistered experiment (N = 470), we compared self-authored goals against LLM-authored goals derived from a personal reflection. Although LLM-generated goals scored higher on SMART criteria (specificity, measurability, achievability, relevance, and time-boundedness; d = 2.26), participants in the LLM condition reported lower psychological ownership (d = 1.38), commitment (d = 1.19), and perceived importance (d = 1.13). At two-week follow-up, 72.8

Social media can reveal patient experiences with glucagon-like peptide-1 receptor agonists (GLP-1 RAs) that extend beyond clinical trial data. We analysed 410,198 Reddit posts (May 2019–June 2025) mentioning semaglutide or tirzepatide. A total of 67,008 users self-reported using these medications, and 43.5% described at least one side effect. Gastrointestinal symptoms predominated, including nausea (36.9%), fatigue (16.7%), vomiting (16.3%), constipation (15.3%) and diarrhoea (12.6%). Notably, reproductive symptoms (for example, menstrual irregularities) and temperature-related complaints (for example, chills and hot flushes) emerged as unrecognized potential effects. These findings highlight patient concerns not well captured in current labelling or trials. Large-scale social media analysis can complement traditional pharmacovigilance by detecting emerging safety signals and expanding understanding of the real-world safety profile of GLP-1 RAs. An analysis involving 410,198 Reddit posts between 2019 and 2025 and 67,008 users, who mentioned semaglutide or tirzepatide, reveals a spectrum of associated reported side effects extending beyond the commonly known safety profile of these two drugs.

Large language models (LLMs) exhibit persistent miscalibration, especially after instruction tuning and preference alignment. Modified training objectives can improve calibration, but retraining is expensive. Inference-time steering offers a lightweight alternative, yet most existing methods optimize proxies for correctness rather than correctness itself. We introduce CORAL (Correctness-Optimized Residual Activation Lens), a regularized inference-time steering method that captures distributed correctness signals from model internal activations using weight-decay MLP probes. We evaluate CORAL across three 7B-parameter models and find that it consistently improves accuracy by 10\% and expected calibration error (ECE) by 50\% on average. We additionally demonstrate that these gains transfer without retraining to the complete published test sets of four held-out benchmarks (ARC-Challenge, HellaSwag, Math-MC, OpenBookQA), averaging 14\% accuracy improvements and 49\% ECE improvements. Our results support the hypothesis that distributed information in model internals can be extracted using regularized probes when individual neurons are insufficient. CORAL thus provides a compute-efficient, transferable, and calibration-aware approach to improve MCQA performance during inference.

Large Language Models (LLMs) generate text token-by-token in discrete time, yet real-world communication, from therapy sessions to business negotiations, critically depends on continuous time constraints. We use simulated negotiations between paired agents under strict deadlines to study adaptation to real-time pressure. Agents either receive only the initial deadline or explicit remaining-time updates at each turn. Remaining-time feedback raises deal closure from 4

Spoken English proficiency is a powerful driver of economic mobility for low-income Indian youth, yet opportunities for spoken practice remain scarce in schools. We investigate the deployment of a voice-based chatbot for English conversation practice across four low-resource schools in Delhi. Through a six-day field study combining observations and interviews, we captured the perspectives of students, teachers, and principals. Findings confirm high demand across all groups, with notable gains in student speaking confidence. Our multi-stakeholder analysis surfaced a tension in long-term adoption vision: students favored open-ended conversational practice, while administrators emphasized curriculum-aligned assessment. We offer design recommendations for voice-enabled chatbots in low-resource multilingual contexts, highlighting the need for more intelligible speech output for non-native learners, one-tap interactions with simplified interfaces, and actionable analytics for educators. Beyond language learning, our findings inform the co-design of future AI-based educational technologies that are socially sustainable within the complex ecosystem of low-resource schools.

Personal intelligence is becoming a central frontier for user-facing AI agents. To be helpful in everyday life, agents must understand users across the digital contexts where their preferences, intents, habits, social relationships, and needs unfold over time. Today's systems can personalize within individual apps or tasks, but personal intelligence as a whole remains under-measured: how agents build cross-context user understanding, support steerable recommendation systems, act proactively across platforms, and avoid over-personalization. We introduce PersonaMem-v3, a real-world-grounded benchmark and evaluation harness for omni-platform personal intelligence. PersonaMem-v3 is seeded from more than one million anonymized real-world engagement histories, most of which are implicit signals, and uses them to construct time-indexed user digital worlds across social media, chatbot, calendar, and AI-companion with preference evolvement over time. The benchmark brings personalization, LLM-powered recommendation, proactiveness, agentic tool use, and geo-temporal reasoning into one framework, anchored in psychology, social-linguistics, and user-behavior theories. It evaluates whether AI agents can infer holistic user understanding from cross-platform evidence, personalize responses, rerank recommendations on social media, follow user steering through natural language, and hold back when personalization would be inappropriate, repetitive, outdated, or unnecessary. PersonaMem-v3 points toward LLM-powered personal intelligent agents that work with existing scalable recommendation infrastructure while making personalization more interactive, agentic, and aligned with how real users experience their digital lives.

A retrospective, exploratory cross-sectional analysis exploring whether social media data is associated with cardiovascular disease (CVD) risk beyond traditional clinical models. While social media data may capture behavioral and social markers relevant to CVD, their associations with CVD risk remains uncertain.

AI chatbots are increasingly embedded in social life, offering accessible companionship. While brief interactions have been shown to provide immediate benefits, it is unclear whether repeated, daily engagement with chatbots reduces loneliness. In this pre-registered study, we tested the effectiveness of an AI chatbot versus a human peer in reducing loneliness among 296 students in their first semester of university. For two weeks, participants either interacted with a chatbot or a human peer, or simply wrote a brief journal entry (control condition). Although our chatbot “Sam” was designed to offer consistent support rooted in principles from relationship science, the psychological benefits of interacting with this chatbot were smaller in comparison to interacting with a randomly selected first-year university student. The present study provides initial evidence that texting daily with a random human peer may be more effective in alleviating loneliness than texting with a highly supportive chatbot.

Assessing well-being with social media text data is a promising method, but besides hedonic well-being, little is known about whether additional well-being dimensions, such as psychological richness and eudaimonic well-being, can be predicted from such data. We compare the predictive accuracy for hedonic well-being, eudaimonic well-being, and the recently proposed construct of psychological richness in a large sample of Facebook users (n = 2,644), and find that the inclusion of language features incrementally improved model prediction accuracy beyond demographic features for psychological richness, but not for hedonic or eudaimonic well-being. Psychological richness had the lowest overall prediction accuracy (r = .21) followed by hedonic well-being (r = .27) and eudaimonic well-being (r = .29). The linguistic features associated with Psychological Richness were face valid, and in many instances the content and direction of the associations were unique to Psychological Richness, which provides discriminant validity evidence.

Retrieval-Augmented Generation (RAG) systems are increasingly deployed in high-stakes domains, where safety depends not only on how a system answers, but also on whether a query should be answered given a knowledge base (KB). Out-of-domain (OOD) queries can cause dense retrieval to surface weakly related context and lead the generator to produce fluent but unjustified responses. We study lightweight, KB-aligned OOD detection as an always-on gate for RAG systems. Our approach applies PCA to KB embeddings and scores queries in a compact subspace selected either by explained-variance retention (EVR) or by a separability-driven -test ranking. We evaluate geometric semantic-search rules and lightweight classifiers across 16 domains, including high-stakes COVID-19 and Substance Use KBs, and stress-test robustness using both LLM-generated attacks and an in-the-wild 4chan attack. We find that low-dimensional detectors achieve competitive OOD performance while being faster, cheaper, and more interpretable than prompted LLM-based judges. Finally, human and LLM-based evaluations show that OOD queries primarily degrade the relevance of RAG outputs, highlighting the need for efficient external OOD detection to maintain safe, in-scope behavior.

Style features such as friendly, helpful, or concise are widely used in prompts to steer the behavior of Large Language Model (LLM) conversational agents, yet their unintended side effects remain poorly understood. In this work, we present the first systematic study of cross-feature stylistic side effects. We conduct a comprehensive survey of 127 conversational agent papers from ACL Anthology and identify 12 frequently used style features. Using controlled, synthetic dialogues across task-oriented and open domain settings, we quantify how prompting for one style feature causally affects others via a pairwise LLM as a Judge evaluation framework. Our results reveal consistent and structured side effects, such as prompting for conciseness significantly reduces perceived expertise. They demonstrate that style features are deeply entangled rather than orthogonal. To support future research, we introduce CASSE (Conversational Agent Stylistic Side Effects), a dataset capturing these complex interactions. We further evaluate prompt based and activation steering based mitigation strategies and find that while they can partially restore suppressed traits, they often degrade the primary intended style. These findings challenge the assumption of faithful style control in LLMs and highlight the need for multi-objective and more principled approaches to safe, targeted stylistic steering in conversational agents.

Token-level attention tuning, a class of training-free methods including Post-hoc Attention Steering (PASTA, AutoPASTA) and Attention Calibration (ACT), has emerged as a promising way to improve frozen LLMs with interpretable interventions. However, these methods depend on auxiliary heuristics to identify "important" task-specific tokens, which can introduce bias and limit applicability when token importance is unclear or when using optimized kernels where attention maps are inaccessible. We propose a simpler and more elegant alternative: acting only on the initial token (e.g., \ in LLaMA). We show theoretically that adding lightweight biases to this token’s attention logits systematically shifts and reshapes the downstream attention patterns---an effect amplified by its natural function as an attention sink. Our empirical analysis reveals that this tuning process can positively affect LLMs and better unlock their pretrained knowledge, with stronger effects in early layers and distinct scaling preferences across attention heads. Building on these insights, we introduce ZeroTuning: a training-free method that improves LLM performance by applying head-specific attention adjustments to the initial token, requiring zero parameter updates. We present two variants: a supervised mode that calibrates on validation examples, and a novel unsupervised mode that directly minimizes the model's output entropy. Our method requires no KV‑cache or decoding changes, and is kernel‑agnostic (works with SDPA and FlashAttention). The method is lightweight and requires only four lines of modification to the standard LlamaAttention code. It achieves broad gains across 15 datasets and outperforms previous, more complex methods; for instance, with Llama-3.1-8B, it yields relative improvements of 19.9% on classification, 4.5% on question answering, and 2.1% on dialogue. ZeroTuning also works out-of-the-box with quantized inference and maintains its performance improvements with increasing context lengths. Our code and runnable demo are available at https://anonymous.4open.science/r/ZeroTuning.

Large language models (LLMs) tend to verbalize confidence scores that are largely detached from their actual accuracy, yet the geometric relationship governing this behavior remain poorly understood. In this work, we present a mechanistic interpretability analysis of verbalized confidence, using linear probes and contrastive activation addition (CAA) steering to show that calibration and verbalized confidence signals are encoded linearly but are orthogonal to one another – a finding consistent across three open-weight models and four datasets. Interestingly, when models are prompted to simultaneously reason through a problem and verbalize a confidence score, the reasoning process disrupts the verbalized confidence direction, exacerbating miscalibration. We term this the "Reasoning Contamination Effect." Leveraging this insight, we introduce a two-stage adaptive steering pipeline that reads the model's internal accuracy estimate and steers verbalized output to match it, substantially improving calibration alignment across all evaluated models.

Large language models are increasingly used as private, always-available conversational systems, but little is known about how people with depressive symptoms use them. Building on CSCW work on disclosure and peer support, we examine ChatGPT as an emerging informal support infrastructure: private, persistent, responsive, and available outside ordinary hours. We analyze 187,093 ChatGPT conversations from 766 participants who completed the PHQ-8, comparing those below the moderate-symptom threshold (score of 10) with those at or above it. Higher-PHQ participants used ChatGPT more for mental-health, interpersonal, loneliness, self-focused, and support-seeking conversations, with pronounced late-night and recurring month-level patterns. Their language contained more first-person singular pronouns and absolutist terms. They more often engaged ChatGPT in high-disclosure contexts, but professional redirection was not higher. Language-based prediction was modest and insufficient for screening (AUROC 0.591). We argue these histories should not be treated as clinical screening data but as evidence LLMs are increasingly used as informal support infrastructure.

ABSTRACT Gray-market retatrutide use is increasing, but patient safety experiences remain poorly characterized. This cross-sectional analysis examined Reddit posts and comments from retatrutide-specific and broader peptide or weight-management communities through December 2025. A validated large language model classified self-reported retatrutide use and extracted author-attributed symptoms mapped to MedDRA Preferred Terms. Among 13,589 users reporting current use, 7,823 had at least one mapped symptom after exclusions. Unlike phase 2 trial findings dominated by gastrointestinal events, Reddit reports most often described appetite increase, fatigue, increased energy, nausea, food craving, insomnia, and elevated heart rate. Findings are hypothesis-generating and warrant pharmacovigilance attention.

The advancement of science depends on rigorous tests of competing hypotheses, yet many disputes are left unresolved. Adversarial collaboration-where opposing scientists jointly design decisive tests-is one proposed solution. We examine whether large language models (LLMs) can play a role by organizing information, structuring the debate and generating candidate experimental designs. This article reports an AI-assisted adversarial collaboration designed to resolve a debate in PNAS on minority salience-an overestimation of the percentage of minority faces in a visual display. The debate focused on whether there would be further overestimation when minorities in the displays were the same minorities in participants' communities (or social environments). Using LLMs to extract and organize competing propositions, we identified central disagreements and generated initial experimental designs to test claims. Human collaborators refined the designs and created two preregistered experiments that factorially manipulated the ethnicity of minority faces and the ethnicity of participants' communities. Data showed that people exaggerated the percentage of minorities in facial displays. Furthermore, overestimation was even greater when minorities in facial displays were also minorities in participants' communities. When the two camps of researchers saw the results, their confidence in key hypotheses converged. We do not experimentally test AI-assisted adversarial collaboration relative to traditional adversarial collaboration or other forms of dispute resolution. Rather, our study illustrates how an AI tool can be used with adversarial collaboration to formalize claims, structure disagreements, lower barriers to collaboration, and serve as an impartial observer to strengthen perceptions of fairness.

Vision-Language-Action (VLA) models leverage powerful perceptual priors from web-scale Vision-Language Model (VLM) pre-training, yet they remain surprisingly brittle in practice, frequently failing at simple robotic tasks. To mitigate this, we propose Contrastive Conceptor Activation Steering (COAST). COAST builds on the notion of a "conceptor", a linear operator that soft-projects data into the principal components of a target distribution. COAST uses conceptors to identify success-critical subspaces for a target robotic task from a few examples of success and failure rollouts. At inference time, it steers VLA latents into these identified success subspaces to improve task outcomes. Across three architecturally distinct neural policies (flow-matching VLA, autoregressive VLA, and Diffusion Policy), COAST improves absolute mean simulation and real-robot task success rate by over 20 and 40

Importance: Health care systems are increasingly considering large language model (LLM)-based chatbots for vaccine communication, but evidence that they improve durable, behaviorally relevant outcomes beyond existing health materials is limited. Objectives: To examine whether brief, multiturn interactions with an LLM chatbot increase parental intention to vaccinate children against human papillomavirus (HPV) compared with no intervention and government public health materials and to assess whether any effects persist. Design, Setting, and Participants: This randomized clinical trial was conducted online among individuals in the US, Canada, and the UK from March 3 to May 25, 2025, with follow-up at 15 and 45 days. Eligible participants were parents 18 years or older with at least 1 HPV vaccine-eligible child (aged 11-17 years in the US and Canada and 12-17 years in the UK) who had not received HPV vaccination or whose vaccination status was unknown. Interventions: Participants were randomized to (1) no-message control, (2) government public health materials matched to country (minimum 3-minute exposure), or (3) a 3-minute interaction with an LLM chatbot (OpenAI's GPT-4o) prompted to encourage HPV vaccination using a default response style or a shorter conversational style. Main Outcomes and Measures: The primary outcome was self-reported likelihood of vaccinating the child against HPV within 12 months (0- to 100-point scale, with 0 indicating extremely unlikely and 100 indicating extremely likely) measured immediately after intervention. Prespecified follow-ups included vaccination intent and self-reported vaccination at 15 and 45 days. Results: A total of 1297 participants (mean [SD] age, 42.84 [6.93] years; 935 [72.1%] female) were randomized. Compared with no intervention, public health materials increased immediate vaccination intent (Cohen d = 0.53; 95% CI, 0.36-0.70), as did the default chatbot (d = 0.48; 95% CI, 0.30-0.65) and conversational chatbot (d = 0.33; 95% CI, 0.17-0.49). At 45 days, neither chatbot increased intent relative to controls, whereas public health materials maintained modest effects. No intervention increased self-reported vaccination uptake. Conclusions and Relevance: The findings of this randomized clinical trial suggest that well-designed public health messaging may match or exceed the impact of short chatbot conversations for HPV vaccine promotion. Trial Registration: ClinicalTrials.gov Identifier: NCT07132125.

Activation-based steering provides control of LLM behavior at inference time, but the dominant paradigm reduces each concept to a single direction whose geometry is left largely unexamined. Rather than selecting a single steering direction, we use conceptors: soft projection matrices estimated from activations pooled across both poles of a bipolar concept, which preserve the concept's full multidimensional subspace. A geometric analysis shows the bipolar subspace strictly subsumes the single-vector baseline. We further show that the conceptor quota provides a parameter-free layer-selection diagnostic, predicting concept separability with Pearson correlations up to r=0.96 across three instruction-tuned models and three semantic dimensions. Beyond selection, conceptors admit a closed-form Boolean algebra (AND, OR, NOT): we evaluate conceptor compositionality on thematically related sub-concepts. Across a systematic five-axis design-space evaluation, conceptors match or outperform additive baselines at layers where concept subspaces are multi-dimensional while producing substantially fewer degenerate outputs. Conceptor steering is a geometrically principled, compositional, and practically safer alternative to single-direction steering from a limited number of contrastive pairs.