Large Language Models (LLMs) have been evaluated using diverse question types, e.g., multiple-choice, true/false, and short/long answers. This study answers an unexplored question about the impact of different question types on LLM accuracy on reasoning tasks. We investigate the performance of five LLMs on three different types of questions using quantitative and deductive reasoning tasks. The performance metrics include accuracy in the reasoning steps and choosing the final answer. Key Findings: (1) Significant differences exist in LLM performance across different question types. (2) Reasoning accuracy does not necessarily correlate with the final selection accuracy. (3) The number of options and the choice of words, influence LLM performance.

Process Reward Models (PRMs) play a central role in evaluating and guiding multi-step reasoning in large language models (LLMs), especially for mathematical problem solving. However, we identify a pervasive length bias in existing PRMs: they tend to assign higher scores to longer reasoning steps, even when the semantic content and logical validity are unchanged. This bias undermines the reliability of reward predictions and leads to overly verbose outputs during inference. To address this issue, we propose CoLD(Counterfactually-Guided Length Debiasing), a unified framework that mitigates length bias through three components: an explicit length-penalty adjustment, a learned bias estimator trained to capture spurious length-related signals, and a joint training strategy that enforces length-invariance in reward predictions. Our approach is grounded in counterfactual reasoning and informed by causal graph analysis. Extensive experiments on MATH500 and GSM-Plus show that CoLD consistently reduces reward-length correlation, improves accuracy in step selection, and encourages more concise, logically valid reasoning. These results demonstrate the effectiveness and practicality of CoLD in improving the fidelity and robustness of PRMs.

Reza.vatankhahbarenji@ntu.ac.uk Abstract This paper explores the potential of Agentic AI in autonomously detecting and responding to anomalies within complex systems, emphasizing its ability to transform traditional, human - dependent anomaly management methods. Building on recent advancements, the study illustrates how Agentic AI -- AI agent augmented with large language models, diverse tools, and knowledge - based systems -- continuously analyses and learns from vast, multi - source datasets to autonomously identify, interpret, and respond to abnormal behav iours in complex, adaptive systems . Unlike conventional AI agents constrained by predefined roles, Agentic AI synthesizes insights across disciplines, detects subtle patterns, and adapts its strategies using both implicit and explicit knowledge. This paper underscores the need to evolve cu rrent human - based anomaly management approaches toward fully autonomous systems, highlighting Agentic AI's adaptive, goal - driven nature ...

Current large language model (LLM) applications often employ multi-component prompts, comprising both system and user prompts, to guide model behaviors. While recent advancements have demonstrated the efficacy of automatically optimizing either the system or user prompt to boost performance, such unilateral approaches often yield suboptimal outcomes due to the interdependent nature of these components. In this work, we introduce P3, a novel self-improvement framework that concurrently optimizes both system and user prompts through an iterative process. The offline optimized prompts are further leveraged to promote online prompting by performing query-dependent prompt optimization. Extensive experiments on general tasks (e.g., Arena-hard and Alpaca-eval) and reasoning tasks (e.g., GSM8K and GPQA) demonstrate that P3 achieves superior performance in the realm of automatic prompt optimization. Our results highlight the effectiveness of a holistic optimization strategy in enhancing LLM performance across diverse domains.

We present an adapter-based approach for tactical conditioning of StarCraft II AI agents. Current agents, while powerful, lack the ability to adapt their strategies based on high-level tactical directives. Our method freezes a pre-trained policy network (DI-Star) and attaches lightweight adapter modules to each action head, conditioned on a tactical tensor that encodes strategic preferences. By training these adapters with KL divergence constraints, we ensure the policy maintains core competencies while exhibiting tactical variations. Experimental results show our approach successfully modulates agent behavior across tactical dimensions including aggression, expansion patterns, and technology preferences, while maintaining competitive performance. Our method enables flexible tactical control with minimal computational overhead, offering practical strategy customization for complex real-time strategy games.

Recent advances in artificial intelligence (AI) - particularly generative AI - present new opportunities to accelerate, or even automate, epidemiological research. Unlike disciplines based on physical experimentation, a sizable fraction of Epidemiology relies on secondary data analysis and thus is well-suited for such augmentation. Yet, it remains unclear which specific tasks can benefit from AI interventions or where roadblocks exist. Awareness of current AI capabilities is also mixed. Here, we map the landscape of epidemiological tasks using existing datasets - from literature review to data access, analysis, writing up, and dissemination - and identify where existing AI tools offer efficiency gains. While AI can increase productivity in some areas such as coding and administrative tasks, its utility is constrained by limitations of existing AI models (e.g. hallucinations in literature reviews) and human systems (e.g. barriers to accessing datasets). Through examples of AI-generated epidemiological outputs, including fully AI-generated papers, we demonstrate that recently developed agentic systems can now design and execute epidemiological analysis, albeit to varied quality (see https://github.com/edlowther/automated-epidemiology). Epidemiologists have new opportunities to empirically test and benchmark AI systems; realising the potential of AI will require two-way engagement between epidemiologists and engineers.

Large Language Models (LLMs) deployed in enterprise settings (e.g., as Microsoft 365 Copilot) face novel security challenges. One critical threat is prompt inference attacks: adversaries chain together seemingly benign prompts to gradually extract confidential data. In this paper, we present a comprehensive study of multi-stage prompt inference attacks in an enterprise LLM context. We simulate realistic attack scenarios where an attacker uses mild-mannered queries and indirect prompt injections to exploit an LLM integrated with private corporate data. We develop a formal threat model for these multi-turn inference attacks and analyze them using probability theory, optimization frameworks, and information-theoretic leakage bounds. The attacks are shown to reliably exfiltrate sensitive information from the LLM's context (e.g., internal SharePoint documents or emails), even when standard safety measures are in place. We propose and evaluate defenses to counter such attacks, including statistical anomaly detection, fine-grained access control, prompt sanitization techniques, and architectural modifications to LLM deployment. Each defense is supported by mathematical analysis or experimental simulation. For example, we derive bounds on information leakage under differential privacy-based training and demonstrate an anomaly detection method that flags multi-turn attacks with high AUC. We also introduce an approach called "spotlighting" that uses input transformations to isolate untrusted prompt content, reducing attack success by an order of magnitude. Finally, we provide a formal proof of concept and empirical validation for a combined defense-in-depth strategy. Our work highlights that securing LLMs in enterprise settings requires moving beyond single-turn prompt filtering toward a holistic, multi-stage perspective on both attacks and defenses.

This work does not provide legal advice, and do not claims that any legal opinion provided are correct 1 and block the model's output accordingly. This technique might not completely block partial matches and does not work for open source development as reproduction of the original code is expected. Some models address this issue by using curated datasets with appropriate licensed contents. For example, the Stack v2 dataset [2] and Starcoder2 model limits data to permissively licensed sources and contents with unknown license. The Common Pile dataset [3] and the accompanied Comma model improves on this by limiting the dataset to permissive licensed contents only. Most permissive licenses only have attribution as its primary sole licensing condition and may be easier to comply with than the GPLv2 license. Ideally, models that are trained on public domain contents may be the best in terms of legal compliance as they have no restrictions or requirements, but to our knowledge no such text generation models exist today with reasonable quality.

We introduce Being-H0, a dexterous Vision-Language-Action model (VLA) trained on large-scale human videos. Existing VLAs struggle with complex manipulation tasks requiring high dexterity and generalize poorly to novel scenarios and tasks, primarily due to their reliance on synthetic data with significant sim-to-real gaps or teleoperated demonstrations lacking scale and diversity. To address this data bottleneck, we propose leveraging human hands as a foundation manipulator, capitalizing on the rich dexterity and scalability present in web data. Our approach centers on physical instruction tuning, a novel training paradigm that combines large-scale VLA pretraining from human videos, physical space alignment for 3D reasoning, and post-training adaptation for robotic tasks. Additionally, we introduce a part-level motion tokenization method which achieves millimeter-level reconstruction accuracy to model precise hand trajectories for action learning. To support our proposed paradigm, we further develop a comprehensive data curation pipeline that integrates heterogeneous sources -- including motion capture, VR, and RGB-only videos -- into a large-scale dataset with millions of motion-based instructional instances. We empirically show the excellence of Being-H0 in hand motion generation and instruction following, and it also scales well with model and data sizes. Importantly, we observe the expected gains of Being-H0 in real-world robotic manipulation as physical instruction tuning is applied. More details are available at https://beingbeyond.github.io/Being-H0.

One way social groups are marginalized in discourse is that the narratives told about them often default to a narrow, stereotyped range of topics. In contrast, default groups are allowed the full complexity of human existence. We describe the constrained representations of queer people in LLM generations in terms of harmful representations, narrow representations, and discursive othering and formulate hypotheses to test for these phenomena. Our results show that LLMs are significantly limited in their portrayals of queer personas.