With the increase in maritime traffic and the mandatory implementation of the Automatic Identification System (AIS), the importance and diversity of maritime traffic analysis tasks based on AIS data, such as vessel trajectory prediction, anomaly detection, and collision risk assessment, is rapidly growing. However, existing approaches tend to address these tasks individually, making it difficult to holistically consider complex maritime situations. To address this limitation, we propose a novel framework, AIS-LLM, which integrates time-series AIS data with a large language model (LLM). AIS-LLM consists of a Time-Series Encoder for processing AIS sequences, an LLM-based Prompt Encoder, a Cross-Modality Alignment Module for semantic alignment between time-series data and textual prompts, and an LLM-based Multi-Task Decoder. This architecture enables the simultaneous execution of three key tasks: trajectory prediction, anomaly detection, and risk assessment of vessel collisions within a single end-to-end system. Experimental results demonstrate that AIS-LLM outperforms existing methods across individual tasks, validating its effectiveness. Furthermore, by integratively analyzing task outputs to generate situation summaries and briefings, AIS-LLM presents the potential for more intelligent and efficient maritime traffic management.

Kyiv, Ukraine – Taras, a seasoned Ukrainian serviceman recovering from a contusion, expects "no miracles" from United States President Donald Trump's August 15 summit with his Russian counterpart, Vladimir Putin. "There's going to be no miracles, no peace deal in a week, and Putin will try to make Trump believe that it is Ukraine that doesn't want peace," the fair-haired 32-year-old with a deep brown tan acquired in the trenches of eastern Ukraine, told Al Jazeera. Taras, who spent more than three years on the front line and said he had recently shot down an explosives-laden Russian drone barging at him in a field covered with explosion craters, withheld his last name in accordance with the wartime protocol. Putin wants to dupe Trump by pandering to the US president's self-image as a peacemaker to avoid further economic sanctions, while the Russian leader seeks a major military breakthrough in eastern Ukraine, Taras said. "Putin really believes that until this winter, he will seize something sizeable, or that [his troops] will break through the front line and will dictate terms to Ukraine," Taras said.

Breakthroughs, discoveries, and DIY tips sent every weekday. Much like mermaids, the kraken, or the hafgufa, rogue waves have been regarded as a maritime myth. These waves do not always leave a lot of data behind, making it feel as if they spring up from the depths out of nowhere. However, one monster wave did leave data behind for scientists. On January 1, 1995, a monstrous 80-foot wave slammed into the Draupner oil platform in the North Sea.

A Russian attack killed three people in Ukraine's southeastern Zaporizhia region on Sunday, Governor Ivan Fedorov wrote on Telegram. A Ukrainian drone attack sparked a major fire at an oil depot in Sochi in southern Russia, the governor of Russia's Krasnodar region, Veniamin Kondratiev, said on Sunday. The fire was extinguished hours later after 120 firefighters were deployed, officials said. Russia's civil aviation authority, Rosaviatsia, briefly halted flights at Sochi's airport during the fire. Ukraine's military says it used drones to target several sites inside Russia, including refineries, an airfield and an electronics plant.

Breakthroughs, discoveries, and DIY tips sent every weekday. Fine artist Laura Shape uses quite an unexpected medium in her visual artwork. It lends striking patterns to her abstract canvases, while helping restore rivers, reefs, and wetlands. Shape uses the leather of invasive species--specifically lionfish, carp, and Burmese pythons. "I use those materials to make vibrant, textured, abstract acrylic pieces," she tells Popular Science via video call.

Falling debris from destroyed Ukrainian drones disrupted railway power supply and train operations in part of the Volgograd region, the administration of the region in Russia's south said on Sunday. There were no injuries as a result of the attacks, the administration said on Telegram, quoting Governor Andrei Bocharov. Russia downed 99 drones overnight over 12 Russian regions, the Crimean Peninsula and the Black Sea, the Russian Ministry of Defence said. Meanwhile, Russia launched a barrage of drones and missiles in an overnight attack that killed three people in Ukraine's Dnipro and the nearby region on Saturday, Ukrainian officials said. Ukraine's air force said it intercepted 183 drones and 17 missiles, but hits from 10 missiles and 25 drones were recorded in nine locations.

Retired Air Force Gen. Charles Wald joins'Fox News Live' to weigh in on Russia's increased attacks on Ukraine despite President Donald Trump's ultimatum to Vladimir Putin. Russia and Ukraine took aim at corresponding Black Sea resort cities early Thursday morning, just hours after ceasefire talks in Turkey once again failed to deliver results. The major Russian resort city of Sochi was rocked by a Ukrainian drone strike that began around 1 a.m. and lasted until 3 a.m., where one person was reportedly killed and another injured, according to Ukrainian media outlet the Kyiv Independent, though the Ukrainian military has not commented on the incident. An oil depot in the Krasnodar Krai region where Sochi is located was also struck, though the extent of the damage remains unclear. Russia's President Vladimir Putin chairs a meeting via a video conference at the Kremlin in Moscow on July 23, 2025.

National Review editor-in-chief Rich Lowry and FOX Business' Liz Claman join'MediaBuzz' to discuss Hegseth's heated press conference where he called out the media's'hatred' of President Donald Trump. FIRST ON FOX: Defense Secretary Pete Hegseth has issued sweeping new orders to fast-track drone production and deployment, allowing commanders to procure and test them independently and requiring drone combat simulations across every branch of the military. As part of an aggressive push to outpace Russia and China in unmanned warfare, "the Department's bureaucratic gloves are coming off," Hegseth wrote. "Lethality will not be hindered by self-imposed restrictions... Our major risk is risk-avoidance." In a pair of memos first obtained by Fox News Digital, Hegseth rescinded legacy policies that he believes restricted innovation.

Past work has studied the effects of fine-tuning on large language models' (LLMs) overall performance on certain tasks. However, a quantitative and systematic method for analyzing its effect on individual outputs is still lacking. Here, we propose a new method for measuring the contribution that fine-tuning makes to individual LLM responses, assuming access to the original pre-trained model. Our method tracks the model's intermediate hidden states, providing a more fine-grained insight into the effects of fine-tuning than a simple comparison of final outputs from pre-trained and fine-tuned models. We introduce and theoretically analyze an exact decomposition of any fine-tuned LLM into a pre-training component and a fine-tuning component. Empirically, we find that model behavior and performance can be steered by up- or down-scaling the fine-tuning component during the forward pass. Motivated by this finding and our theoretical analysis, we define the Tuning Contribution (TuCo) as the ratio of the magnitudes of the fine-tuning component to the pre-training component. We observe that three prominent adversarial attacks on LLMs circumvent safety measures in a way that reduces TuCo, and that TuCo is consistently lower on prompts where these attacks succeed compared to those where they do not. This suggests that attenuating the effect of fine-tuning on model outputs plays a role in the success of such attacks. In summary, TuCo enables the quantitative study of how fine-tuning influences model behavior and safety, and vice versa.

Consensus-based optimization (CBO) has established itself as an efficient gradient-free optimization scheme, with attractive mathematical properties, such as mean-field convergence results for non-convex loss functions. In this work, we study CBO in the context of closed-box adversarial attacks, which are imperceptible input perturbations that aim to fool a classifier, without accessing its gradient. Our contribution is to establish a connection between the so-called consensus hopping as introduced by Riedl et al. and natural evolution strategies (NES) commonly applied in the context of adversarial attacks and to rigorously relate both methods to gradient-based optimization schemes. Beyond that, we provide a comprehensive experimental study that shows that despite the conceptual similarities, CBO can outperform NES and other evolutionary strategies in certain scenarios.