Multimodal summarization aims to generate a concise summary based on the input text and image. However, the existing methods potentially suffer from unfactual output. To evaluate the factuality of multimodal summarization models, we propose two fine-grained and explainable evaluation frameworks (FALLACIOUS) for different application scenarios, i.e. reference-based factuality evaluation framework and reference-free factuality evaluation framework. Notably, the reference-free factuality evaluation framework doesn't need ground truth and hence it has a wider application scenario. To evaluate the effectiveness of the proposed frameworks, we compute the correlation between our frameworks and the other metrics. The experimental results show the effectiveness of our proposed method. We will release our code and dataset via github.

Azerbaijan is observing a day of mourning for the victims of an air crash that killed 38 people. At least 29 people survived the deadly crash on Christmas day. Azerbaijan observed a nationwide moment of silence on Thursday, with national flags lowered, traffic coming to a halt at noon, and signals sounding from ships and trains across the country. Earlier, Azerbaijani President Ilham Aliyev declared Thursday a day of mourning and cancelled a planned visit to Russia for an informal summit of the Commonwealth of Independent States (CIS), a grouping of former Soviet nations. Aliyev's office said the president "ordered the prompt initiation of urgent measures to investigate the causes of the disaster".

In the spring, the US saw what was likely its first AI candidate. In a brief campaign for the mayor of Wyoming, virtual integrated citizen (VIC), a ChatGPT-based bot created by real human Victor Miller, promised to govern entirely by AI. At the outset of 2024, many suggested that even if not winning office, generative AI would play a pivotal role in--and pose significant risks to--democratic elections, as more than 2 billion people voted in more than 60 countries. But now, experts and analysts have changed their tune, saying that generative AI likely had little to no effect at all. So were all those prognostications that 2024 would be the AI election year wrong?

The F-35 has had to develop a thick skin. From my former colleagues in Congress to defense-industry experts, the world's premier fighter jet is accustomed to criticism for issues with cost, production and more. In November, though, one of America's most influential voices decided to jump on the bandwagon: Elon Musk. Musk shared a video of a drone swarm with the caption, "Meanwhile, some idiots are still building manned fighter jets like the F-35," and he included a trash-can emoji for good measure. You can imagine the pleasant surprise of the men and women who build the F-35, as well as the brilliant men and women who pilot it, when a drone-like swarm of voices came to their defense.

In 2025, there will be a course correction in AI and geopolitics, as world leaders increasingly understand that their national interests are best served through the promise of a more positive and cooperative future. The post-ChatGPT years in AI discourse could be characterized as somewhere between a gold rush and a moral panic. In 2023, at the same time as there was record investment in AI, tech experts, including Elon Musk and Steve Wozniak, published an open letter calling for a six-month moratorium on the training of AI systems more powerful than GPT-4, while others compared AI to a "nuclear war" and a "pandemic." This has understandably clouded the judgment of political leaders, pushing the geopolitical conversation about AI into some disturbing places. At the AI & Geopolitics Project, my research organization at Cambridge University, our analysis clearly shows the increasing trend towards AI nationalism.

A panel of experts on Thursday called for creating a law that would allow the government to gain cooperation from artificial intelligence developers in the event of a serious problem related to AI technologies. The proposal is included in an interim report prepared by the panel under the government's AI Strategy Council and was adopted at a council meeting. The report said Japan "should strengthen the government's command function to promote integrated policies, from research and development to application" and formulate a national strategy in the area of AI. These also require a legal framework, the report added. At the meeting, Prime Minister Shigeru Ishiba instructed his government to promptly present relevant legislation to parliament.

Russia's Foreign Ministry accused NATO of trying to turn Moldova into a logistical centre to supply the Ukrainian army and of seeking to bring the Western alliance's military infrastructure closer to Russia. Arto Pahkin, the head of operations of the Finnish electricity grid, told the country's public broadcaster Yle that "the possibility of sabotage cannot be ruled out" after an undersea power cable linking Finland and Estonia broke down. It is the latest in a series of incidents involving telecom cables and energy pipelines in the Baltic Sea. A "terrorist act" sank the Russian cargo ship that went down in international waters in the Mediterranean this week, the Russian state-owned company that owns the vessel said. The Oboronlogistika company said it "thinks a targeted terrorist attack was committed on December 23, 2024, against the Ursa Major", without indicating who may have been behind the act or why. The Azerbaijan Airlines passenger jet that crashed near the city of Aktau in Kazakhstan, killing 38 people, was earlier diverting from an area of Russia that Moscow has recently defended against Ukrainian drone attacks.

Reinforcement learning (RL) has shown great promise in simulated environments, such as games, where failures have minimal consequences. However, the deployment of RL agents in real-world systems such as autonomous vehicles, robotics, UAVs, and medical devices demands a higher level of safety and transparency, particularly when facing adversarial threats. Safe RL algorithms have been developed to address these concerns by optimizing both task performance and safety constraints. However, errors are inevitable, and when they occur, it is essential that the RL agents can also explain their actions to human operators. This makes trust in the safety mechanisms of RL systems crucial for effective deployment. Explainability plays a key role in building this trust by providing clear, actionable insights into the agent's decision-making process, ensuring that safety-critical decisions are well understood. While machine learning (ML) has seen significant advances in interpretability and visualization, explainability methods for RL remain limited. Current tools fail to address the dynamic, sequential nature of RL and its needs to balance task performance with safety constraints over time. The re-purposing of traditional ML methods, such as saliency maps, is inadequate for safety-critical RL applications where mistakes can result in severe consequences. To bridge this gap, we propose xSRL, a framework that integrates both local and global explanations to provide a comprehensive understanding of RL agents' behavior. xSRL also enables developers to identify policy vulnerabilities through adversarial attacks, offering tools to debug and patch agents without retraining. Our experiments and user studies demonstrate xSRL's effectiveness in increasing safety in RL systems, making them more reliable and trustworthy for real-world deployment. Code is available at https://github.com/risal-shefin/xSRL.

It is imperative to develop an intrusion prevention system (IPS), specifically designed for autonomous robotic systems. This is due to the unique nature of these cyber-physical systems (CPS), which are not merely typical distributed systems. These systems employ their own systems software (i.e. robotic middleware and frameworks) and execute distinct components to facilitate interaction with various sensors and actuators, and other robotic components (e.g. cognitive subsystems). Furthermore, as cyber-physical systems, they engage in interactions with humans and their physical environment, as exemplified by social robots. These interactions can potentially lead to serious consequences, including physical damage. In response to this need, we have designed and implemented RIPS, an intrusion prevention system tailored for robotic applications based on ROS 2, the framework that has established itself as the de facto standard for developing robotic applications. This manuscript provides a comprehensive exposition of the issue, the security aspects of ROS 2 applications, and the key points of the threat model we created for our robotic environment. It also describes the architecture and the implementation of our initial research prototype and a language specifically designed for defining detection and prevention rules for diverse, real-world robotic scenarios. Moreover, the manuscript provides a comprehensive evaluation of the approach, that includes a set of experiments with a real social robot executing a well known testbed used in international robotic competitions.

Chart understanding tasks such as ChartQA and Chart-to-Text involve automatically extracting and interpreting key information from charts, enabling users to query or convert visual data into structured formats. State-of-the-art approaches primarily focus on visual cues from chart images, failing to explicitly incorporate rich textual information (e.g., data labels and axis labels) embedded within the charts. This textual information is vital for intuitive human comprehension and interpretation of charts. Moreover, existing models are often large and computationally intensive, limiting their practical applicability. In this paper, we introduce AskChart, a universal model that explicitly integrates both textual and visual cues from charts using a Mixture of Experts (MoE) architecture. AskChart facilitates the learning of enhanced visual-textual representations of charts for effectively handling multiple chart understanding tasks, while maintaining a smaller model size. To capture the synergy between visual and textual modalities, we curate a large-scale dataset named ChartBank with about 7.5M data samples, which helps align textual and visual information and facilitates the extraction of visual entities and text. To effectively train AskChart, we design a three-stage training strategy to align visual and textual modalities for learning robust visual-textual representations and optimizing the learning of the MoE layer. Extensive experiments across five datasets demonstrate the significant performance gains of AskChart in four chart understanding tasks. Remarkably, AskChart with 4.6B parameters outperforms state-of-the-art models with 13B parameters by 68.3% in Open-ended ChartQA and 49.2% in Chart-to-Text tasks, while achieving comparable performance in ChartQA and Chart-to-Table tasks.