With the rapid evolution of large language models (LLMs), new and hard-to-predict harmful capabilities are emerging. This requires developers to be able to identify risks through the evaluation of "dangerous capabilities" in order to responsibly deploy LLMs. In this work, we collect the first open-source dataset to evaluate safeguards in LLMs, and deploy safer open-source LLMs at a low cost. Our dataset is curated and filtered to consist only of instructions that responsible language models should not follow. We annotate and assess the responses of six popular LLMs to these instructions. Based on our annotation, we proceed to train several BERT-like classifiers, and find that these small classifiers can achieve results that are comparable with GPT-4 on automatic safety evaluation. Warning: this paper contains example data that may be offensive, harmful, or biased.

A few years earlier, Japan's representative for human rights at the UN, Hideaki Ueda, demanded that foreign colleagues "shut up" at a meeting. Gavin Williamson used the same words against Russia when he was the UK defence secretary. And Ukraine's ambassador to Germany, Andriy Melnyk, last year referred to German Chancellor Olaf Scholz as an "offended liver sausage".

Senior officials at the Home Office secretly lobbied the UK's independent privacy regulator to act "favourably" towards a private firm keen to roll out controversial facial recognition technology across the country, according to internal government emails seen by the Observer. Correspondence reveals that the Home Office wrote to the Information Commissioner's Office (ICO) warning that policing minister, Chris Philp, would "write to your commissioner" if the regulator's investigation into Facewatch – whose facial recognition cameras have provoked huge opposition after being installed in shops – was not positive towards the firm. An official from the Home Office's data and identity directorate warned the ICO: "If you are about to do something imminently in Facewatch's favour then I should be able to head that off [Philp's intervention], otherwise we will just have to let it take its course." The apparent threat came two days after a closed-door meeting on 8 March between Philp, senior Home Office officials and Facewatch. Facewatch uses cameras to check faces against a watch list and, despite widespread concern over the technology, it has already been introduced in hundreds of high street shops and supermarkets.

Fox News Flash top headlines are here. Check out what's clicking on Foxnews.com. The New York City Police Department announced plans to fly surveillance drones over the city this Labor Day weekend to monitor outdoor parties or barbecues following complaints about large gatherings. The decision was revealed during a security briefing addressing J'ouvert, an annual Caribbean festival marking the end of slavery in which thousands of people take to the streets of Brooklyn. Assistant NYPD Commissioner Kaz Daughtry said at a press conference Thursday that the drones will respond to "non-priority and priority calls."

The United States says it has seen notable military progress by Ukraine's forces fighting in the Zaporizhia region over the last 72 hours. Ukrainian troops achieved "some success against that second line of Russian defences", White House National Security Council spokesperson John Kirby said. A recent drone attack on an airport in northwestern Russia's Pskov region was carried out from within Russian territory, Ukraine's intelligence chief Kyrylo Budanov said. The Pskov regional governor said Russian air defence units had "neutralised an unidentified object" spotted flying over the region, just days after a wave of Ukrainian drones destroyed military planes parked at an airfield in the region. The United States says it has seen notable military progress by Ukraine's forces fighting in the Zaporizhia region over the last 72 hours.

The area of Neurosymbolic Artificial Intelligence (Neurosymbolic AI) is rapidly developing and has become a popular research topic, encompassing sub-fields such as Neurosymbolic Deep Learning (Neurosymbolic DL) and Neurosymbolic Reinforcement Learning (Neurosymbolic RL). Compared to traditional learning methods, Neurosymbolic AI offers significant advantages by simplifying complexity and providing transparency and explainability. Reinforcement Learning(RL), a long-standing Artificial Intelligence(AI) concept that mimics human behavior using rewards and punishment, is a fundamental component of Neurosymbolic RL, a recent integration of the two fields that has yielded promising results. The aim of this paper is to contribute to the emerging field of Neurosymbolic RL by conducting a literature survey. Our evaluation focuses on the three components that constitute Neurosymbolic RL: neural, symbolic, and RL. We categorize works based on the role played by the neural and symbolic parts in RL, into three taxonomies:Learning for Reasoning, Reasoning for Learning and Learning-Reasoning. These categories are further divided into sub-categories based on their applications. Furthermore, we analyze the RL components of each research work, including the state space, action space, policy module, and RL algorithm. Additionally, we identify research opportunities and challenges in various applications within this dynamic field.

This work leverages the recent advancements of deep learning in image processing to find optimal locations that present the important characteristics of a field. The data for training are collected at different fields in local farms with five features: aspect, flow accumulation, slope, NDVI (normalized difference vegetation index), and yield. The soil sampling dataset is challenging because the ground truth is highly imbalanced binary images. Therefore, we approached the problem with two methods, the first approach involves utilizing a state-of-the-art model with the convolutional neural network (CNN) backbone, while the second is to innovate a deep-learning design grounded in the concepts of transformer and self-attention. Our framework is constructed with an encoder-decoder architecture with the self-attention mechanism as the backbone. In the encoder, the self-attention mechanism is the key feature extractor, which produces feature maps. In the decoder, we introduce atrous convolution networks to concatenate, fuse the extracted features, and then export the optimal locations for soil sampling. Currently, the model has achieved impressive results on the testing dataset, with a mean accuracy of 99.52%, a mean Intersection over Union (IoU) of 57.35%, and a mean Dice Coefficient of 71.47%, while the performance metrics of the state-of-the-art CNN-based model are 66.08%, 3.85%, and 1.98%, respectively. This indicates that our proposed model outperforms the CNN-based method on the soil-sampling dataset. To the best of our knowledge, our work is the first to provide a soil-sampling dataset with multiple attributes and leverage deep learning techniques to enable the automatic selection of soil-sampling sites. This work lays a foundation for novel applications of data science and machine-learning technologies to solve other emerging agricultural problems.

In this paper, we investigate the integration of drone identification data (Remote ID) with collision avoidance mechanisms to improve the safety and efficiency of multi-drone operations. We introduce an improved Near Mid-Air Collision (NMAC) definition, termed as UAV NMAC (uNMAC), which accounts for uncertainties in the drone's location due to self-localization errors and possible displacements between two location reports. Our proposed uNMAC-based Reciprocal Velocity Obstacle (RVO) model integrates Remote ID messages with RVO to enable enhanced collision-free navigation. We propose modifications to the Remote ID format to include data on localization accuracy and drone airframe size, facilitating more efficient collision avoidance decisions. Through extensive simulations, we demonstrate that our approach halves mission execution times compared to a conservative standard Remote ID-based RVO. Importantly, it ensures collision-free operations even under localization uncertainties. By integrating the improved Remote ID messages and uNMAC-based RVO, we offer a solution to significantly increase airspace capacity while adhering to strict safety standards. Our study emphasizes the potential to augment the safety and efficiency of future drone operations, thereby benefiting industries reliant on drone technologies.

Automatic Conspiracy Theory Identification (ACTI) is a new shared task proposed for the first time at the EVALITA 2023 evaluation campaign. ACTI is based on a new, manually labeled dataset of comments scraped from conspiratorial Telegram channels and consists of two subtasks: (1) identifying conspiratorial content (conspiratorial content classification); and (2) classifying content into specific conspiracy theories (conspiratorial category classification). A total of 15 teams participated in the task with 81 submissions. In this task summary, we discuss the data and task, and outline the bestperforming approaches that are largely based on large language models. We conclude with a brief discussion of the application of large language models to counter the spread of misinformation on online platforms.

Ordinary differential equations (ODEs), via their induced flow maps, provide a powerful framework to parameterize invertible transformations for the purpose of representing complex probability distributions. While such models have achieved enormous success in machine learning, particularly for generative modeling and density estimation, little is known about their statistical properties. This work establishes the first general nonparametric statistical convergence analysis for distribution learning via ODE models trained through likelihood maximization. We first prove a convergence theorem applicable to arbitrary velocity field classes $\mathcal{F}$ satisfying certain simple boundary constraints. This general result captures the trade-off between approximation error (`bias') and the complexity of the ODE model (`variance'). We show that the latter can be quantified via the $C^1$-metric entropy of the class $\mathcal F$. We then apply this general framework to the setting of $C^k$-smooth target densities, and establish nearly minimax-optimal convergence rates for two relevant velocity field classes $\mathcal F$: $C^k$ functions and neural networks. The latter is the practically important case of neural ODEs. Our proof techniques require a careful synthesis of (i) analytical stability results for ODEs, (ii) classical theory for sieved M-estimators, and (iii) recent results on approximation rates and metric entropies of neural network classes. The results also provide theoretical insight on how the choice of velocity field class, and the dependence of this choice on sample size $n$ (e.g., the scaling of width, depth, and sparsity of neural network classes), impacts statistical performance.