Despite growing research on bias in large language models (LLMs), most work has focused on gender and race, with little attention to religious identity. This paper explores how religion is internally represented in LLMs and how it intersects with concepts of violence and geography. Using mechanistic interpretability and Sparse Autoencoders (SAEs) via the Neuronpedia API, we analyze latent feature activations across five models. We measure overlap between religion- and violence-related prompts and probe semantic patterns in activation contexts. While all five religions show comparable internal cohesion, Islam is more frequently linked to features associated with violent language. In contrast, geographic associations largely reflect real-world religious demographics, revealing how models embed both factual distributions and cultural stereotypes. These findings highlight the value of structural analysis in auditing not just outputs but also internal representations that shape model behavior.

Large Language Models (LLMs) possess impressive reasoning abilities but are prone to generating incorrect information, often referred to as hallucinations. While incorporating external Knowledge Graphs (KGs) can partially mitigate this issue, existing methods primarily treat KGs as static knowledge repositories, overlooking the critical disparity between KG and LLM knowledge, and failing to fully exploit the reasoning capabilities inherent in KGs. To address these limitations, we propose Pyramid-Driven Alignment (PDA), a novel framework for seamlessly integrating LLMs with KGs. PDA utilizes Pyramid Principle analysis to construct a hierarchical pyramid structure. This structure is designed to reflect the input question and generate more validated deductive knowledge, thereby enhancing the alignment of LLMs and KGs and ensuring more cohesive integration. Furthermore, PDA employs a recursive mechanism to harness the underlying reasoning abilities of KGs, resulting in more accurate knowledge retrieval for question-answering tasks. Our experimental results reveal a substantial performance advantage of PDA over state-of-the-art baselines, with improvements reaching 26.70% and 26.78%.

Large Language Models (LLMs) have demonstrated remarkable in-context learning (ICL) capabilities. In this study, we explore a surprising phenomenon related to ICL: LLMs can perform multiple, computationally distinct ICL tasks simultaneously, during a single inference call, a capability we term "task superposition". We provide empirical evidence of this phenomenon across various LLM families and scales and show that this phenomenon emerges even if we train the model to in-context learn one task at a time. We offer theoretical explanations that this capability is well within the expressive power of transformers. We also explore how LLMs internally compose task vectors during superposition. Furthermore, we show that larger models can solve more ICL tasks in parallel, and better calibrate their output distribution. Our findings offer insights into the latent capabilities of LLMs, further substantiate the perspective of "LLMs as superposition of simulators", and raise questions about the mechanisms enabling simultaneous task execution.

One of the challenges of Industry 4.0, is to determine and optimize the flow of data, products and materials in manufacturing companies. To realize these challenges, many solutions have been defined such as the utilization of automated guided vehicles (AGVs). However, being guided is a handicap for these vehicles to fully meet the requirements of Industry 4.0 in terms of adaptability and flexibility: the autonomy of vehicles cannot be reduced to predetermined trajectories. Therefore, it is necessary to develop their autonomy. This will be possible by designing new generations of industrial autonomous vehicles (IAVs), in the form of intelligent and cooperative autonomous mobile robots.In the field of road transport, research is very active to make the car autonomous. Many algorithms, solving problematic traffic situations similar to those that can occur in an industrial environment, can be transposed in the industrial field and therefore for IAVs. The technologies standardized in dedicated bodies (e.g., ETSI TC ITS), such as those concerning the exchange of messages between vehicles to increase their awareness or their ability to cooperate, can also be transposed to the industrial context. The deployment of intelligent autonomous vehicle fleets raises several challenges: acceptability by employees, vehicle location, traffic fluidity, vehicle perception of changing environments (dynamic), vehicle-infrastructure cooperation, or vehicles heterogeneity. In this context, developing the autonomy of IAVs requires a relevant working method. The identification of reusable or adaptable algorithms to the various problems raised by the increase in the autonomy of IAVs is not sufficient, it is also necessary to be able to model, to simulate, to test and to experiment with the proposed solutions. Simulation is essential since it allows both to adapt and to validate the algorithms, but also to design and to prepare the experiments.To improve the autonomy of a fleet, we consider the approach relying on a collective intelligence to make the behaviours of vehicles adaptive. In this chapter, we will focus on a class of problems faced by IAVs related to collision and obstacle avoidance. Among these problems, we are particularly interested when two vehicles need to cross an intersection at the same time, known as a deadlock situation. But also, when obstacles are present in the aisles and need to be avoided by the vehicles safely.

It studies the integration of question answering (QA) systems in a dialogue system (DS). Not long ago, each of these research subjects were studied separately; only very recently has studying the intersection between them gained increasing interest (Reddy et al., 2018; Choi et al., 2018). We present a spoken conversational question answering system that is able to answer questions about general knowledge in French by calling two distinct QA systems. It solves coreference and ellipsis by modelling context. Furthermore, it is extensible, thus other components such as neural approaches for question-answering can be easily integrated. It is also possible to collect a dialogue corpus from its iterations. In contrast to most conversational systems which support only speech, two input and output modalities are supported speech and text. Thus it is possible to let the user check the answers by either asking relevant Wikipedia excerpts or by navigating through the retrieved name entities or by exploring the answer details of the QA components: the confidence score as well as the set of explored triplets. Therefore, the user has the final word to consider the answer as correct or incorrect and to1 https://www.wikidata.org

Knowledge bases (KBs) are often greatly incomplete, necessitating a demand for KB completion. A promising approach is to embed KBs into latent spaces and make inferences by learning and operating on latent representations. Such embedding models, however, do not make use of any rules during inference and hence have limited accuracy. This paper proposes a novel approach which incorporates rules seamlessly into embedding models for KB completion. It formulates inference as an integer linear programming (ILP) problem, with the objective function generated from embedding models and the constraints translated from rules. Solving the ILP problem results in a number of facts which 1) are the most preferred by the embedding models, and 2) comply with all the rules. By incorporating rules, our approach can greatly reduce the solution space and significantly improve the inference accuracy of embedding models. We further provide a slacking technique to handle noise in KBs, by explicitly modeling the noise with slack variables. Experimental results on two publicly available data sets show that our approach significantly and consistently outperforms state-of-the-art embedding models in KB completion. Moreover, the slacking technique is effective in identifying erroneous facts and ambiguous entities, with a precision higher than 90%.