Do large language models (LLMs) anticipate when they will answer correctly? To study this, we extract activations after a question is read but before any tokens are generated, and train linear probes to predict whether the model's forthcoming answer will be correct. Across three open-source model families ranging from 7 to 70 billion parameters, projections on this "in-advance correctness direction" trained on generic trivia questions predict success in distribution and on diverse out-of-distribution knowledge datasets, outperforming black-box baselines and verbalised predicted confidence. Predictive power saturates in intermediate layers, suggesting that self-assessment emerges mid-computation. Notably, generalisation falters on questions requiring mathematical reasoning. Moreover, for models responding "I don't know", doing so strongly correlates with the probe score, indicating that the same direction also captures confidence. By complementing previous results on truthfulness and other behaviours obtained with probes and sparse auto-encoders, our work contributes essential findings to elucidate LLM internals.

Clinical trials (CT) are essential for advancing medical research and treatment, yet efficiently recruiting eligible participants -- each of whom must meet complex eligibility criteria -- remains a significant challenge. Traditional recruitment approaches, such as advertisements or electronic health record screening within hospitals, are often time-consuming and geographically constrained. This work addresses the recruitment challenge by leveraging the vast amount of health-related information individuals share on social media platforms. With the emergence of powerful large language models (LLMs) capable of sophisticated text understanding, we pose the central research question: Can LLM-driven tools facilitate CT recruitment by identifying potential participants through their engagement on social media? To investigate this question, we introduce TRIALQA, a novel dataset comprising two social media collections from the subreddits on colon cancer and prostate cancer. Using eligibility criteria from public real-world CTs, experienced annotators are hired to annotate TRIALQA to indicate (1) whether a social media user meets a given eligibility criterion and (2) the user's stated reasons for interest in participating in CT. We benchmark seven widely used LLMs on these two prediction tasks, employing six distinct training and inference strategies. Our extensive experiments reveal that, while LLMs show considerable promise, they still face challenges in performing the complex, multi-hop reasoning needed to accurately assess eligibility criteria.

Exploratory learning environments (ELEs), such as simulation-based platforms and open-ended science curricula, promote hands-on exploration and problem-solving but make it difficult for teachers to gain timely insights into students' conceptual understanding. This paper presents LearnLens, a generative AI (GenAI)-enhanced teacher-facing dashboard designed to support problem-based instruction in middle school science. LearnLens processes students' open-ended responses from digital assessments to provide various insights, including sample responses, word clouds, bar charts, and AI-generated summaries. These features elucidate students' thinking, enabling teachers to adjust their instruction based on emerging patterns of understanding. The dashboard was informed by teacher input during professional development sessions and implemented within a middle school Earth science curriculum. We report insights from teacher interviews that highlight the dashboard's usability and potential to guide teachers' instruction in the classroom.

The gene set analysis (GSA) is a foundational approach for uncovering the molecular functions associated with a group of genes. Recently, LLM-powered methods have emerged to annotate gene sets with biological functions together with coherent explanatory insights. However, existing studies primarily focus on proprietary models, which have been shown to outperform their open-source counterparts despite concerns over cost and data privacy. Furthermore, no research has investigated the application of advanced reasoning strategies to the GSA task. To address this gap, we introduce Gene-R1, a data-augmented learning framework that equips lightweight and open-source LLMs with step-by-step reasoning capabilities tailored to GSA. Experiments on 1,508 in-distribution gene sets demonstrate that Gene-R1 achieves substantial performance gains, matching commercial LLMs. On 106 out-of-distribution gene sets, Gene-R1 performs comparably to both commercial and large-scale LLMs, exhibiting robust generalizability across diverse gene sources.

Trajectory prediction serves as a critical functionality in autonomous driving, enabling the anticipation of future motion paths for traffic participants such as vehicles and pedestrians, which is essential for driving safety. Although conventional deep learning methods have improved accuracy, they remain hindered by inherent limitations, including lack of interpretability, heavy reliance on large-scale annotated data, and weak generalization in long-tail scenarios. The rise of Large Foundation Models (LFMs) is transforming the research paradigm of trajectory prediction. This survey offers a systematic review of recent advances in LFMs, particularly Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) for trajectory prediction. By integrating linguistic and scene semantics, LFMs facilitate interpretable contextual reasoning, significantly enhancing prediction safety and generalization in complex environments. The article highlights three core methodologies: trajectory-language mapping, multimodal fusion, and constraint-based reasoning. It covers prediction tasks for both vehicles and pedestrians, evaluation metrics, and dataset analyses. Key challenges such as computational latency, data scarcity, and real-world robustness are discussed, along with future research directions including low-latency inference, causality-aware modeling, and motion foundation models.

This position paper presents A VEC (Adaptive Verifiable Edge Control), a framework for bootstrapping privacy for local language models by enforcing privacy at the edge with explicit verifiability for delegated queries. A VEC introduces an adaptive budgeting algorithm that allocates per-query differential privacy parameters based on sensitivity, local confidence, and historical usage, and uses verifiable transformation with on-device integrity checks. We formalize guarantees using R enyi differential privacy with odometer-based accounting, and establish utility ceilings, delegation-leakage bounds, and impossibility results for deterministic gating and hash-only certification. Our evaluation is simulation-based by design to study mechanism behavior and accounting; we do not claim deployment readiness or task-level utility with live LLMs. The contribution is a conceptual architecture and theoretical foundation that chart a pathway for empirical follow-up on privately bootstrapping local LLMs.

Large Language Models (LLMs) are increasingly integrated into financial applications, yet existing red-teaming research primarily targets harmful content, largely neglecting regulatory risks. In this work, we aim to investigate the vulnerability of financial LLMs through red-teaming approaches. We introduce Risk-Concealment Attacks (RCA), a novel multi-turn framework that iteratively conceals regulatory risks to provoke seemingly compliant yet regulatory-violating responses from LLMs. To enable systematic evaluation, we construct FIN-Bench, a domain-specific benchmark for assessing LLM safety in financial contexts. Extensive experiments on FIN-Bench demonstrate that RCA effectively bypasses nine mainstream LLMs, achieving an average attack success rate (ASR) of 93.18%, including 98.28% on GPT-4.1 and 97.56% on OpenAI o1. These findings reveal a critical gap in current alignment techniques and underscore the urgent need for stronger moderation mechanisms in financial domains. We hope this work offers practical insights for advancing robust and domain-aware LLM alignment.

Large language model (LLM) assistants are increasingly integrated into enterprise workflows, raising new security concerns as they bridge internal and external data sources. This paper presents an in-depth case study of EchoLeak (CVE-2025-32711), a zero-click prompt injection vulnerability in Microsoft 365 Copilot that enabled remote, unauthenticated data exfiltration via a single crafted email. By chaining multiple bypasses-evading Microsoft's XPIA (Cross Prompt Injection Attempt) classifier, circumventing link redaction with reference-style Markdown, exploiting auto-fetched images, and abusing a Microsoft Teams proxy allowed by the content security policy, EchoLeak achieved full privilege escalation across LLM trust boundaries without user interaction. We analyze why existing defenses failed, and outline a set of engineering mitigations including prompt partitioning, enhanced input/output filtering, provenance-based access control, and strict content security policies. Beyond the specific exploit, we derive generalizable lessons for building secure AI copilots, emphasizing the principle of least privilege, defense-in-depth architectures, and continuous adversarial testing. Our findings establish prompt injection as a practical, high-severity vulnerability class in production AI systems and provide a blueprint for defending against future AI-native threats.

Synthetic clinical data are increasingly important for advancing AI in healthcare, given strict privacy constraints on real-world EHRs, limited availability of annotated rare-condition data, and systemic biases in observational datasets. While large language models (LLMs) can generate fluent clinical text, producing synthetic data that is both realistic and clinically meaningful remains challenging. We introduce DualAlign, a framework that enhances statistical fidelity and clinical plausibility through dual alignment: (1) statistical alignment, which conditions generation on patient demographics and risk factors; and (2) semantic alignment, which incorporates real-world symptom trajectories to guide content generation. Using Alzheimer's disease (AD) as a case study, DualAlign produces context-grounded symptom-level sentences that better reflect real-world clinical documentation. Fine-tuning an LLaMA 3.1-8B model with a combination of DualAlign-generated and human-annotated data yields substantial performance gains over models trained on gold data alone or unguided synthetic baselines. While DualAlign does not fully capture longitudinal complexity, it offers a practical approach for generating clinically grounded, privacy-preserving synthetic data to support low-resource clinical text analysis.

Transformer models based on the Mixture of Experts (MoE) architecture have made significant progress in long-sequence modeling, but existing models still have shortcomings in computational efficiency and the ability to capture long-range dependencies, especially in terms of the dynamic adaptability of expert resource allocation. In this paper, we propose a Dynamic Adaptive Shared Expert and Grouped Multi-Head Attention Hybrid Model (DASG-MoE) to enhance long-sequence modeling capabilities by integrating three modules. First, we employ the Grouped Multi-Head Attention (GMHA) mechanism to effectively reduce the computational complexity of long sequences. By parallel processing through sequence grouping, local sliding window attention, and feature aggregation, we address long-range dependency issues and the model's lack of generalization for local information. Second, we design a Dual-Scale Shared Expert Structure (DSSE), where shallow experts use lightweight computations to quickly respond to low-dimensional features, while deep experts process high-dimensional complex semantics through pre-training transfer and post-training optimization, achieving a dynamic balance between efficiency and accuracy. Third, we propose a hierarchical Adaptive Dynamic Routing (ADR) mechanism that dynamically selects expert levels based on feature complexity and task requirements, and optimizes resource allocation through a local expert activation strategy. Experiments on multiple long-sequence benchmark datasets demonstrate that our DASG-MoE model outperforms state-of-the-art models.