Safety & alignment

Large Language Models (LLMs) are increasingly being deployed in multilingual, multicultural settings, yet their reliance on predominantly English-centric training data risks misalignment with the diverse cultural values of different societies. In this paper, we present a comprehensive, multilingual audit of the cultural alignment of contemporary LLMs including GPT-4o-Mini, Gemini-2.5-Flash, Llama 3.2, Mistral and Gemma 3 across India, East Asia and Southeast Asia. Our study specifically focuses

The growing demand for text-to-image generation has led to rapid advances in generative modeling. Recently, text-to-image diffusion models trained with flow matching algorithms, such as FLUX, have achieved remarkable progress and emerged as strong alternatives to conventional diffusion models. At the same time, inference-time enhancement strategies have been shown to improve the generation quality and text-prompt alignment of text-to-image diffusion models. However, these techniques are mainly a

Protein sequence design must balance designability, defined as the ability to recover a target backbone, with multiple, often competing, developability properties such as solubility, thermostability, and expression. Existing approaches address these properties through post hoc mutation, inference-time biasing, or retraining on property-specific subsets, yet they are target dependent and demand substantial domain expertise or careful hyperparameter tuning. In this paper, we introduce ProtAlign, a

This report presents a structured Robotics Physical Safety Framework based on explicit asset declaration, systematic vulnerability enumeration, and hazard-driven synthetic data generation. The approach bridges classical risk engineering with modern machine learning pipelines, enabling safety envelope learning grounded in a formalized hazard ontology. The key contribution of this framework is the alignment between classical safety engineering, digital twin simulation, synthetic data generation, a

Diffusion models have recently advanced photorealistic human synthesis, although practical talking-head generation (THG) remains constrained by high inference latency, temporal instability such as flicker and identity drift, and imperfect audio-visual alignment under challenging speech conditions. This paper introduces TempoSyncDiff, a reference-conditioned latent diffusion framework that explores few-step inference for efficient audio-driven talking-head generation. The approach adopts a teache

While the shift from cascaded dialogue systems to end-to-end (E2E) speech Large Language Models (LLMs) improves latency and paralinguistic modeling, E2E models often exhibit a significant performance degradation compared to their text-based counterparts. The standard Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) training methods fail to close this gap. To address this, we propose X-OPD, a novel Cross-Modal On-Policy Distillation framework designed to systematically align the capab

Current safety alignment methods encode safe behavior implicitly within model parameters, creating a fundamental opacity: we cannot easily inspect why a model refuses a request, nor intervene when its safety judgments fail. We propose Safe Transformer, a modular approach that augments pre-trained language models by inserting a discrete information bottleneck containing an explicit safety bit between transformer layers. The safety bit serves as both an interpretable signal of the model's safety c

Trust plays a pivotal role in enabling effective cooperation, reducing uncertainty, and guiding decision-making in both human interactions and multi-agent systems. Although it is significant, there is limited understanding of how large language models (LLMs) internally conceptualize and reason about trust. This work presents a white-box analysis of trust representation in EleutherAI/gpt-j-6B, using contrastive prompting to generate embedding vectors within the activation space of the LLM for dia

Protein language models often take into consideration the alignment between a protein sequence and its textual description. However, they do not take structural information into consideration. Traditional methods treat sequence and structure separately, limiting the ability to exploit the alignment between the structure and protein sequence embeddings. In this paper, we introduce a sequence structure contrastive alignment framework, which learns a shared embedding space where proteins are repres

Safety alignment is often conceptualized as a monolithic process wherein harmfulness detection automatically triggers refusal. However, the persistence of jailbreak attacks suggests a fundamental mechanistic decoupling. We propose the \textbf{\underline{D}}isentangled \textbf{\underline{S}}afety \textbf{\underline{H}}ypothesis \textbf{(DSH)}, positing that safety computation operates on two distinct subspaces: a \textit{Recognition Axis} ($\mathbf{v}_H$, ``Knowing'') and an \textit{Execution Axi

Currently, open-sourced large language models (OSLLMs) have demonstrated remarkable generative performance. However, as their structure and weights are made public, they are exposed to jailbreak attacks even after alignment. Existing attacks operate primarily at shallow levels, such as the prompt or embedding level, and often fail to expose vulnerabilities rooted in deeper model components, which creates a false sense of security for successful defense. In this paper, we propose \textbf{\underli

Best-of-N (BoN) sampling is a widely used inference-time alignment method for language models, whereby N candidate responses are sampled from a reference model and the one with the highest predicted reward according to a learned reward model is selected. Despite its widespread practical use, recent theoretical work has suggested that it is statistically suboptimal and vulnerable to reward hacking, the process by which models exploit weaknesses in the learned reward model to achieve high estimate

We introduce WebChain, the largest open-source dataset of human-annotated trajectories on real-world websites, designed to accelerate reproducible research in web agents. It contains 31,725 trajectories and 318k steps, featuring a core Triple Alignment of visual, structural, and action data to provide rich, multi-modal supervision. The data is collected via a scalable pipeline that ensures coverage of complex, high-value tasks often missed by synthetic methods. Leveraging this dataset, we propos

Post-training compression reduces LLM parameter counts but often produces irregular tensor dimensions that degrade GPU performance -- a phenomenon we call \emph{dimensional misalignment}. We present a full-stack analysis tracing root causes at three levels: framework, library, and hardware. The key insight is that model inference becomes slower because the resulting dimensions are unfriendly with the GPU execution stack. For example, compressing Llama-3-8B with activation-aware singular value de

Mechanistic interpretability typically relies on post-hoc analysis of trained networks. We instead adopt an interventional approach: testing hypotheses a priori by modifying architectural topology to observe training dynamics. We study grokking - delayed generalization in Transformers trained on cyclic modular addition (Zp) - investigating if specific architectural degrees of freedom prolong the memorization phase. We identify two independent structural factors in standard Transformers: unbounde

Preference alignment is an essential step in adapting large language models (LLMs) to human values, but existing approaches typically depend on costly human annotations or large-scale API-based models. We explore whether a weak LLM can instead act as an effective annotator. We surprisingly find that selecting only a subset of a weak LLM's highly confident samples leads to substantially better performance than using full human annotations. Building on this insight, we propose Confidence-Weighted

In perpetrator treatment, a recurring observation is the dissociation between insight and action: offenders articulate remorse yet behavioral change does not follow. We report four preregistered studies (1,584 multi-agent simulations across 16 languages and three model families) demonstrating that alignment interventions in large language models produce a structurally analogous phenomenon: surface safety that masks or generates collective pathology and internal dissociation. In Study 1 (N = 150)

BACKGROUND: The integration of artificial intelligence in healthcare has transformed clinical practice and research methodologies. However, concerns regarding algorithmic accountability, interpretability, and safety have necessitated human oversight in AI systems. Human in the loop artificial intelligence represents a collaborative paradigm where human expertise and machine intelligence converge to enhance decision making while maintaining ethical standards and clinical safety. AIM: This review

Millions of patients are regularly using large language model (LLM) chatbots for medical advice, raising patient safety concerns. This physician-led red-teaming study compares the safety of four publicly available chatbots-Claude by Anthropic, Gemini by Google, GPT-4o by OpenAI, and Llama-3.0/3.1-70B by Meta-on a new dataset, HealthAdvice, using an evaluation framework that enables quantitative and qualitative analysis. In total, 888 chatbot responses are evaluated for 222 patient-posed advice-s

In recent years, much of the research on clustering algorithms has primarily focused on enhancing their accuracy and efficiency, frequently at the expense of interpretability. However, as these methods are increasingly being applied in high-stakes domains such as healthcare, finance, and autonomous systems, the need of transparent and interpretable clustering outcomes has become a critical concern. This is not only necessary for gaining user trust but also for satisfying the growing ethical and

Abstract The widespread adoption of large language models (LLMs) raises important questions about their safety and alignment 1 . Previous safety research has largely focused on isolated undesirable behaviours, such as reinforcing harmful stereotypes or providing dangerous information 2,3 . Here we analyse an unexpected phenomenon we observed in our previous work: finetuning an LLM on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding 4 . For

The opaque nature of black-box deep learning (DL) models poses significant challenges for intrusion detection systems (IDSs) in Internet of Things (IoT) networks, where transparency, trust, and operational reliability are critical. Although explainable artificial intelligence (XAI) has been increasingly adopted to enhance interpretability, its impact on detection performance and computational efficiency in resource-constrained IoT environments remains insufficiently understood. This systematic r

This paper posits that the most significant long-term existential risk to human civilization is not an acute technological or environmental catastrophe, but a chronic, systemic decay driven by the psychological and demographic consequences of a biologically capped lifespan. The entrenched expectation of mortality before 120 years fosters a condition of "temporal myopia," which cultivates cultural short-termism, consumerist nihilism, and demographic apathy. A critical and compounding aspect of th

Humans strive to design safe AI systems that align with our goals and remain under our control. However, as AI capabilities advance, we face a new challenge: the emergence of deeper, more persistent relationships between humans and AI systems. We explore how increasingly capable AI agents may generate the perception of deeper relationships with users, especially as AI becomes more personalised and agentic. This shift, from transactional interaction to ongoing sustained social engagement with AI,

Explainable Artificial Intelligence (XAI) enhances transparency and interpretability in AI models, which is crucial for trust and accountability in healthcare. A potential application of XAI is disease prediction using various data modalities. This study conducts a Systematic Literature Review (SLR) following the PRISMA protocol, synthesizing findings from 30 selected studies to examine XAI's evolving role in disease prediction. It explores commonly used XAI methods, such as Shapley Additive Exp

The rapid advancements in artificial intelligence (AI) have ushered in a new era of innovative applications, while also prompting concerns regarding risks and adverse consequences. In light of the growing interest in comprehending AI's impact on society and its alignment with human values and needs, Human-Centred Artificial Intelligence (HCAI) has emerged as a potential approach to address questions and concerns. In this Systematic Literature Review, we aim to contribute to conceptual clarity ar

Recent advancements in Explainable Artificial Intelligence (XAI) aim to bridge the gap between complex artificial intelligence (AI) models and human understanding, fostering trust and usability in AI systems. However, challenges persist in comprehensively interpreting these models, hindering their widespread adoption. This study addresses these challenges by exploring recently emerging techniques in XAI. The primary problem addressed is the lack of transparency and interpretability in AI models

The misuse of large language models (LLMs) has drawn significant attention from the general public and LLM vendors. One particular type of adversarial prompt, known as jailbreak prompt, has emerged as the main attack vector to bypass the safeguards and elicit harmful content from LLMs. In this paper, employing our new framework JailbreakHub, we conduct a comprehensive analysis of 1,405 jailbreak prompts spanning from December 2022 to December 2023. We identify 131 jailbreak communities and disco

Artificial Intelligence (AI) has demonstrated exceptional performance in automating critical healthcare tasks, such as diagnostic imaging analysis and predictive modeling, often surpassing human capabilities. The integration of AI in healthcare promises substantial improvements in patient outcomes, including faster diagnosis and personalized treatment plans. However, AI models frequently lack interpretability, leading to significant challenges concerning their performance and generalizability ac

Culture fundamentally shapes people's reasoning, behavior, and communication. As people increasingly use generative artificial intelligence (AI) to expedite and automate personal and professional tasks, cultural values embedded in AI models may bias people's authentic expression and contribute to the dominance of certain cultures. We conduct a disaggregated evaluation of cultural bias for five widely used large language models (OpenAI's GPT-4o/4-turbo/4/3.5-turbo/3) by comparing the models' resp

The use of artificial intelligence (AI) is becoming more prevalent across industries such as healthcare, finance, and transportation. Artificial intelligence is based on the analysis of large datasets and requires a continuous supply of high-quality data. However, using data for AI is not without challenges. This paper comprehensively reviews and critically examines the challenges of using data for AI, including data quality, data volume, privacy and security, bias and fairness, interpretability

We investigate the potential implications of large language models (LLMs), such as Generative Pre-trained Transformers (GPTs), on the U.S. labor market, focusing on the increased capabilities arising from LLM-powered software compared to LLMs on their own. Using a new rubric, we assess occupations based on their alignment with LLM capabilities, integrating both human expertise and GPT-4 classifications. Our findings reveal that around 80% of the U.S. workforce could have at least 10% of their wo

The quality of texts generated by natural language generation (NLG) systems is hard to measure automatically. Conventional reference-based metrics, such as BLEU and ROUGE, have been shown to have relatively low correlation with human judgments, especially for tasks that require creativity and diversity. Recent studies suggest using large language models (LLMs) as reference-free metrics for NLG evaluation, which have the benefit of being applicable to new tasks that lack human references. However

Transparency in Machine Learning (ML), often also referred to as interpretability or explainability, attempts to reveal the working mechanisms of complex models. From a human-centered design perspective, transparency is not a property of the ML model but an affordance, i.e., a relationship between algorithm and users. Thus, prototyping and user evaluations are critical to attaining solutions that afford transparency. Following human-centered design principles in highly specialized and high stake

The lack of transparency is one of the artificial intelligence (AI)'s fundamental challenges, but the concept of transparency might be even more opaque than AI itself. Researchers in different fields who attempt to provide the solutions to improve AI's transparency articulate different but neighboring concepts that include, besides transparency, explainability and interpretability. Yet, there is no common taxonomy neither within one field (such as data science) nor between different fields (law

The environmental crisis is increasing risks to security and peace worldwide, notably in countries that are already fragile. Indicators of insecurity such as the number of conflicts, the number of hungry people and military expenditure are rising; so are indicators of environmental decline, in climate change, biodiversity, pollution and other areas. In combination, the security and environmental crises are creating compound, cascading, emergent, systemic and existential risks. Without profound c

Artificial intelligence (AI) and machine learning (ML) have recently been radically improved and are now being employed in almost every application domain to develop automated or semi-automated systems. To facilitate greater human acceptability of these systems, explainable artificial intelligence (XAI) has experienced significant growth over the last couple of years with the development of highly accurate models but with a paucity of explainability and interpretability. The literature shows evi

Interpretability in machine learning (ML) is crucial for high stakes decisions and troubleshooting. In this work, we provide fundamental principles for interpretable ML, and dispel common misunderstandings that dilute the importance of this crucial topic. We also identify 10 technical challenge areas in interpretable machine learning and provide history and background on each problem. Some of these problems are classically important, and some are recent problems that have arisen in the last few

Ethan Perez, Saffron Huang, Francis Song, Trevor Cai, Roman Ring, John Aslanides, Amelia Glaese, Nat McAleese, Geoffrey Irving. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

THE ALIGNMENT PROBLEM: Machine Learning and Human Values by Brian Christian. New York: W. W. Norton, 2020. 344 pages. Hardcover; $28.95. ISBN: 9780393635829. *The global conversation about artificial intelligence (AI) is increasingly polemic--"AI will change the world!" "AI will ruin the world!" Amidst the strife, Brian Christian's work stands out. It is thoughtful, nuanced, and, at times, even poetic. Coming on the heels of his two other bestsellers, The Most Human Human and Algorithms to Live