Widely deployed large language models (LLMs) can produce convincing yet incorrect outputs, potentially misleading users who may rely on them as if they were correct. To reduce such overreliance, there have been calls for LLMs to communicate their uncertainty to end users. However, there has been little empirical work examining how users perceive and act upon LLMs' expressions of uncertainty. We explore this question through a large-scale, pre-registered, human-subject experiment (N=404) in which participants answer medical questions with or without access to responses from a fictional LLM-infused search engine. Using both behavioral and self-reported measures, we examine how different natural language expressions of uncertainty impact participants' reliance, trust, and overall task performance. We find that first-person expressions (e.g., "I'm not sure, but...") decrease participants' confidence in the system and tendency to agree with the system's answers, while increasing participants' accuracy. An exploratory analysis suggests that this increase can be attributed to reduced (but not fully eliminated) overreliance on incorrect answers. While we observe similar effects for uncertainty expressed from a general perspective (e.g., "It's not clear, but..."), these effects are weaker and not statistically significant. Our findings suggest that using natural language expressions of uncertainty may be an effective approach for reducing overreliance on LLMs, but that the precise language used matters. This highlights the importance of user testing before deploying LLMs at scale.

NASA's Juno mission provided exquisite measurements of Jupiter's gravity field that together with the Galileo entry probe atmospheric measurements constrains the interior structure of the giant planet. Inferring its interior structure range remains a challenging inverse problem requiring a computationally intensive search of combinations of various planetary properties, such as the cloud-level temperature, composition, and core features, requiring the computation of ~10^9 interior models. We propose an efficient deep neural network (DNN) model to generate high-precision wide-ranged interior models based on the very accurate but computationally demanding concentric MacLaurin spheroid (CMS) method. We trained a sharing-based DNN with a large set of CMS results for a four-layer interior model of Jupiter, including a dilute core, to accurately predict the gravity moments and mass, given a combination of interior features. We evaluated the performance of the trained DNN (NeuralCMS) to inspect its predictive limitations. NeuralCMS shows very good performance in predicting the gravity moments, with errors comparable with the uncertainty due to differential rotation, and a very accurate mass prediction. This allowed us to perform a broad parameter space search by computing only ~10^4 actual CMS interior models, resulting in a large sample of plausible interior structures, and reducing the computation time by a factor of 10^5. Moreover, we used a DNN explainability algorithm to analyze the impact of the parameters setting the interior model on the predicted observables, providing information on their nonlinear relation.

Existing work in trustworthy machine learning primarily focuses on single-input adversarial perturbations. In many real-world attack scenarios, input-agnostic adversarial attacks, e.g. universal adversarial perturbations (UAPs), are much more feasible. Current certified training methods train models robust to single-input perturbations but achieve suboptimal clean and UAP accuracy, thereby limiting their applicability in practical applications. We propose a novel method, CITRUS, for certified training of networks robust against UAP attackers. We show in an extensive evaluation across different datasets, architectures, and perturbation magnitudes that our method outperforms traditional certified training methods on standard accuracy (up to 10.3\%) and achieves SOTA performance on the more practical certified UAP accuracy metric.

We introduce ParaNames, a massively multilingual parallel name resource consisting of 140 million names spanning over 400 languages. Names are provided for 16.8 million entities, and each entity is mapped from a complex type hierarchy to a standard type (PER/LOC/ORG). Using Wikidata as a source, we create the largest resource of this type to date. We describe our approach to filtering and standardizing the data to provide the best quality possible. ParaNames is useful for multilingual language processing, both in defining tasks for name translation/transliteration and as supplementary data for tasks such as named entity recognition and linking. We demonstrate the usefulness of ParaNames on two tasks. First, we perform canonical name translation between English and 17 other languages. Second, we use it as a gazetteer for multilingual named entity recognition, obtaining performance improvements on all 10 languages evaluated.

The advent of advanced AI underscores the urgent need for comprehensive safety evaluations, necessitating collaboration across communities (i.e., AI, software engineering, and governance). However, divergent practices and terminologies across these communities, combined with the complexity of AI systems-of which models are only a part-and environmental affordances (e.g., access to tools), obstruct effective communication and comprehensive evaluation. This paper proposes a framework for AI system evaluation comprising three components: 1) harmonised terminology to facilitate communication across communities involved in AI safety evaluation; 2) a taxonomy identifying essential elements for AI system evaluation; 3) a mapping between AI lifecycle, stakeholders, and requisite evaluations for accountable AI supply chain. This framework catalyses a deeper discourse on AI system evaluation beyond model-centric approaches.

Deep learning has achieved remarkable success in recent years. Central to its success is its ability to learn representations that preserve task-relevant structure. However, massive energy, compute, and data costs are required to learn general representations. This paper explores Hyperdimensional Computing (HDC), a computationally and data-efficient brain-inspired alternative. HDC acts as a bridge between connectionist and symbolic approaches to artificial intelligence (AI), allowing explicit specification of representational structure as in symbolic approaches while retaining the flexibility of connectionist approaches. However, HDC's simplicity poses challenges for encoding complex compositional structures, especially in its binding operation. To address this, we propose Generalized Holographic Reduced Representations (GHRR), an extension of Fourier Holographic Reduced Representations (FHRR), a specific HDC implementation. GHRR introduces a flexible, non-commutative binding operation, enabling improved encoding of complex data structures while preserving HDC's desirable properties of robustness and transparency. In this work, we introduce the GHRR framework, prove its theoretical properties and its adherence to HDC properties, explore its kernel and binding characteristics, and perform empirical experiments showcasing its flexible non-commutativity, enhanced decoding accuracy for compositional structures, and improved memorization capacity compared to FHRR.

A concern about cutting-edge or "frontier" AI foundation models is that an adversary may use the models for preparing chemical, biological, radiological, nuclear, (CBRN), cyber, or other attacks. At least two methods can identify foundation models with potential dual-use capability; each has advantages and disadvantages: A. Open benchmarks (based on openly available questions and answers), which are low-cost but accuracy-limited by the need to omit security-sensitive details; and B. Closed red team evaluations (based on private evaluation by CBRN and cyber experts), which are higher-cost but can achieve higher accuracy by incorporating sensitive details. We propose a research and risk-management approach using a combination of methods including both open benchmarks and closed red team evaluations, in a way that leverages advantages of both methods. We recommend that one or more groups of researchers with sufficient resources and access to a range of near-frontier and frontier foundation models run a set of foundation models through dual-use capability evaluation benchmarks and red team evaluations, then analyze the resulting sets of models' scores on benchmark and red team evaluations to see how correlated those are. If, as we expect, there is substantial correlation between the dual-use potential benchmark scores and the red team evaluation scores, then implications include the following: The open benchmarks should be used frequently during foundation model development as a quick, low-cost measure of a model's dual-use potential; and if a particular model gets a high score on the dual-use potential benchmark, then more in-depth red team assessments of that model's dual-use capability should be performed. We also discuss limitations and mitigations for our approach, e.g., if model developers try to game benchmarks by including a version of benchmark test data in a model's training data.

This paper presents a comprehensive analysis of explainable fact-checking through a series of experiments, focusing on the ability of large language models to verify public health claims and provide explanations or justifications for their veracity assessments. We examine the effectiveness of zero/few-shot prompting and parameter-efficient fine-tuning across various open and closed-source models, examining their performance in both isolated and joint tasks of veracity prediction and explanation generation. Importantly, we employ a dual evaluation approach comprising previously established automatic metrics and a novel set of criteria through human evaluation. Our automatic evaluation indicates that, within the zero-shot scenario, GPT-4 emerges as the standout performer, but in few-shot and parameter-efficient fine-tuning contexts, open-source models demonstrate their capacity to not only bridge the performance gap but, in some instances, surpass GPT-4. Human evaluation reveals yet more nuance as well as indicating potential problems with the gold explanations.

In the current cybersecurity landscape, protecting military devices such as communication and battlefield management systems against sophisticated cyber attacks is crucial. Malware exploits vulnerabilities through stealth methods, often evading traditional detection mechanisms such as software signatures. The application of ML/DL in vulnerability detection has been extensively explored in the literature. However, current ML/DL vulnerability detection methods struggle with understanding the context and intent behind complex attacks. Integrating large language models (LLMs) with system call analysis offers a promising approach to enhance malware detection. This work presents a novel framework leveraging LLMs to classify malware based on system call data. The framework uses transfer learning to adapt pre-trained LLMs for malware detection. By retraining LLMs on a dataset of benign and malicious system calls, the models are refined to detect signs of malware activity. Experiments with a dataset of over 1TB of system calls demonstrate that models with larger context sizes, such as BigBird and Longformer, achieve superior accuracy and F1-Score of approximately 0.86. The results highlight the importance of context size in improving detection rates and underscore the trade-offs between computational complexity and performance. This approach shows significant potential for real-time detection in high-stakes environments, offering a robust solution to evolving cyber threats.

The process of clearing areas, namely demining, starts by assessing and prioritizing potential hazardous areas (i.e., desk assessment) to go under thorough investigation of experts, who confirm the risk and proceed with the mines clearance operations. This paper presents Desk-AId that supports the desk assessment phase by estimating landmine risks using geospatial data and socioeconomic information. Desk-AId uses a Geospatial AI approach specialized to landmines. The approach includes mixed data sampling strategies and context-enrichment by historical conflicts and key multi-domain facilities (e.g., buildings, roads, health sites). The proposed system addresses the issue of having only ground-truth for confirmed hazardous areas by implementing a new hard-negative data sampling strategy, where negative points are sampled in the vicinity of hazardous areas. Experiments validate Desk-Aid in two domains for landmine risk assessment: 1) country-wide, and 2) uncharted study areas). The proposed approach increases the estimation accuracies up to 92%, for different classification models such as RandomForest (RF), Feedforward Neural Networks (FNN), and Graph Neural Networks (GNN).