Large language models (LLMs) are playing a pivotal role in deploying strategic use cases across a range of organizations, from large pan-continental companies to emerging startups. The issues and challenges involved in the successful utilization of LLMs can vary significantly depending on the size of the organization. It is important to study and discuss these pertinent issues of LLM adaptation with a focus on the scale of the industrial concerns and brainstorm possible solutions and prospective directions. Such a study has not been prominently featured in the current research literature. In this study, we adopt a threefold strategy: first, we conduct a case study with industry practitioners to formulate the key research questions; second, we examine existing industrial publications to address these questions; and finally, we provide a practical guide for industries to utilize LLMs more efficiently.

Training and deploying Machine Learning models that simultaneously adhere to principles of fairness and privacy while ensuring good utility poses a significant challenge. The interplay between these three factors of trustworthiness is frequently underestimated and remains insufficiently explored. Consequently, many efforts focus on ensuring only two of these factors, neglecting one in the process. The decentralization of the datasets and the variations in distributions among the clients exacerbate the complexity of achieving this ethical trade-off in the context of Federated Learning (FL). For the first time in FL literature, we address these three factors of trustworthiness. We introduce PUFFLE, a high-level parameterised approach that can help in the exploration of the balance between utility, privacy, and fairness in FL scenarios. We prove that PUFFLE can be effective across diverse datasets, models, and data distributions, reducing the model unfairness up to 75%, with a maximum reduction in the utility of 17% in the worst-case scenario, while maintaining strict privacy guarantees during the FL training.

In this survey, we address the key challenges in Large Language Models (LLM) research, focusing on the importance of interpretability. Driven by increasing interest from AI and business sectors, we highlight the need for transparency in LLMs. We examine the dual paths in current LLM research and eXplainable Artificial Intelligence (XAI): enhancing performance through XAI and the emerging focus on model interpretability. Our paper advocates for a balanced approach that values interpretability equally with functional advancements. Recognizing the rapid development in LLM research, our survey includes both peer-reviewed and preprint (arXiv) papers, offering a comprehensive overview of XAI's role in LLM research. We conclude by urging the research community to advance both LLM and XAI fields together.

This paper introduces a Random Survival Forest (RSF) method for functional data. The focus is specifically on defining a new functional data structure, the Censored Functional Data (CFD), for dealing with temporal observations that are censored due to study limitations or incomplete data collection. This approach allows for precise modelling of functional survival trajectories, leading to improved interpretation and prediction of survival dynamics across different groups. A medical survival study on the benchmark SOFA data set is presented. Results show good performance of the proposed approach, particularly in ranking the importance of predicting variables, as captured through dynamic changes in SOFA scores and patient mortality rates.

The integration of new modalities into frontier AI systems offers exciting capabilities, but also increases the possibility such systems can be adversarially manipulated in undesirable ways. In this work, we focus on a popular class of vision-language models (VLMs) that generate text outputs conditioned on visual and textual inputs. We conducted a large-scale empirical study to assess the transferability of gradient-based universal image "jailbreaks" using a diverse set of over 40 open-parameter VLMs, including 18 new VLMs that we publicly release. Overall, we find that transferable gradient-based image jailbreaks are extremely difficult to obtain. When an image jailbreak is optimized against a single VLM or against an ensemble of VLMs, the jailbreak successfully jailbreaks the attacked VLM(s), but exhibits little-to-no transfer to any other VLMs; transfer is not affected by whether the attacked and target VLMs possess matching vision backbones or language models, whether the language model underwent instruction-following and/or safety-alignment training, or many other factors. Only two settings display partially successful transfer: between identically-pretrained and identically-initialized VLMs with slightly different VLM training data, and between different training checkpoints of a single VLM. Leveraging these results, we then demonstrate that transfer can be significantly improved against a specific target VLM by attacking larger ensembles of "highly-similar" VLMs. These results stand in stark contrast to existing evidence of universal and transferable text jailbreaks against language models and transferable adversarial attacks against image classifiers, suggesting that VLMs may be more robust to gradient-based transfer attacks.

Large Vision Language Models (VLMs) extend and enhance the perceptual abilities of Large Language Models (LLMs). Despite offering new possibilities for LLM applications, these advancements raise significant security and ethical concerns, particularly regarding the generation of harmful content. While LLMs have undergone extensive security evaluations with the aid of red teaming frameworks, VLMs currently lack a well-developed one. To fill this gap, we introduce Arondight, a standardized red team framework tailored specifically for VLMs. Arondight is dedicated to resolving issues related to the absence of visual modality and inadequate diversity encountered when transitioning existing red teaming methodologies from LLMs to VLMs. Our framework features an automated multi-modal jailbreak attack, wherein visual jailbreak prompts are produced by a red team VLM, and textual prompts are generated by a red team LLM guided by a reinforcement learning agent. To enhance the comprehensiveness of VLM security evaluation, we integrate entropy bonuses and novelty reward metrics. These elements incentivize the RL agent to guide the red team LLM in creating a wider array of diverse and previously unseen test cases. Our evaluation of ten cutting-edge VLMs exposes significant security vulnerabilities, particularly in generating toxic images and aligning multi-modal prompts. In particular, our Arondight achieves an average attack success rate of 84.5\% on GPT-4 in all fourteen prohibited scenarios defined by OpenAI in terms of generating toxic text. For a clearer comparison, we also categorize existing VLMs based on their safety levels and provide corresponding reinforcement recommendations. Our multimodal prompt dataset and red team code will be released after ethics committee approval. CONTENT WARNING: THIS PAPER CONTAINS HARMFUL MODEL RESPONSES.

I didn't know I was dead until I saw it on Google. When I searched my name, there it was: a picture of my smiling face next to the text "Tom Faber was a physicist and publisher, and he was a university lecturer at Cambridge for 35 years". Apparently I died on 27 July 2004, aged 77. This was news to me. The problem was the picture. When you search the name of a notable person, Google may create what it calls a "knowledge panel", a little box with basic information taken from Wikipedia. Somewhere along the way, the algorithm had confused pictures of my face with the biography of another man who shared my name. According to his obituary, he was "a distinguished physicist with a literary hinterland". Google provides a feedback form to resolve this type of bug. I filled it in several times, but it made no difference.

In this paper, we propose an experimental investigation of the problem of AI fairness in classification. We train an AI model and develop our own fairness package FairDream to detect inequalities and then to correct for them, using income prediction as a case study. Our experiments show that it is a property of FairDream to fulfill fairness objectives which are conditional on the ground truth (Equalized Odds), even when the algorithm is set the task of equalizing positives across groups (Demographic Parity). While this may be seen as an anomaly, we explain this property by comparing our approach with a closely related fairness method (GridSearch), which can enforce Demographic Parity at the expense of Equalized Odds. We grant that a fairness metric conditioned on true labels does not give a sufficient criterion to reach fairness, but we argue that it gives us at least a necessary condition to implement Demographic Parity cautiously. We also explain why neither Equal Calibration nor Equal Precision stand as relevant fairness criteria in classification. Addressing their limitations to warn the decision-maker for any disadvantaging rate, Equalized Odds avoids the peril of strict conservatism, while keeping away the utopia of a whole redistribution of resources through algorithms.

Creating secure and resilient applications with large language models (LLM) requires anticipating, adjusting to, and countering unforeseen threats. Red-teaming has emerged as a critical technique for identifying vulnerabilities in real-world LLM implementations. This paper presents a detailed threat model and provides a systematization of knowledge (SoK) of red-teaming attacks on LLMs. We develop a taxonomy of attacks based on the stages of the LLM development and deployment process and extract various insights from previous research. In addition, we compile methods for defense and practical red-teaming strategies for practitioners. By delineating prominent attack motifs and shedding light on various entry points, this paper provides a framework for improving the security and robustness of LLM-based systems.

Given the magnitude of data generation currently, both in quantity and speed, the use of machine learning is increasingly important. When data include protected features that might give rise to discrimination, special care must be taken. Data quality is critical in these cases, as biases in training data can be reflected in classification models. This has devastating consequences and fails to comply with current regulations. Data-Centric Artificial Intelligence proposes dataset modifications to improve its quality. Instance selection via undersampling can foster balanced learning of classes and protected feature values in the classifier. When such undersampling is done close to the decision boundary, the effect on the classifier would be bolstered. This work proposes Fair Overlap Number of Balls (Fair-ONB), an undersampling method that harnesses the data morphology of the different data groups (obtained from the combination of classes and protected feature values) to perform guided undersampling in the areas where they overlap. It employs attributes of the ball coverage of the groups, such as the radius, number of covered instances and density, to select the most suitable areas for undersampling and reduce bias. Results show that the Fair-ONB method reduces bias with low impact on the classifier's predictive performance.