A tornado tore through Val-d'Oise, north of Paris, on Monday, toppling construction cranes, damaging properties and uprooting trees in its path. One person was killed and four others critically injured, authorities said. The town of Ermont, about 20 km (13 miles) northeast of Paris was hardest hit by the sudden twister, which caused damage in multiple districts. Interior Minister Laurent Nunez said on the X social media platform that it had been a storm of rare intensity. Drone footage shows blaze destroying the historic Bernaga Monastery in Italy.

While recent advances in reasoning models have demonstrated cognitive behaviors through reinforcement learning, existing approaches struggle to invoke deep reasoning capabilities in multi-turn agents with long-horizon interactions. We propose DeepMiner, a novel framework that elicits such abilities by introducing high-difficulty training tasks and dynamic context window. DeepMiner presents a reverse construction method to generate complex but verifiable question-answer pairs from authentic web sources, which ensures the challenge and reliability of training data while injecting cognitive capabilities into multi-turn reasoning scenarios. We further design an elegant yet effective dynamic context management strategy for both training and inference, utilizing sliding window mechanisms while eliminating the dependency on external summarization models, thereby efficiently empowering the model to handle continuously expanding long-horizon contexts. Through reinforcement learning on Qwen3-32B, we develop DeepMiner-32B, which achieves substantial performance improvements across multiple search agent benchmarks. DeepMiner attains 33.5% accuracy on BrowseComp-en, surpassing the previous best open-source agent by almost 20 percentage points, and demonstrates consistent improvements on BrowseComp-zh, XBench-DeepSearch, and GAIA. Notably, our dynamic context management enables sustained interactions of nearly 100 turns within standard 32k context length, effectively addressing the context limitations that constrain existing multi-turn interaction systems.

At 10:44 pm eastern time on May 16, Ryan Hall spotted a blue square on his radar indicating debris flying into the air and realized a huge tornado was racing toward Somerset, Kentucky. "We've been watching this storm for a while, we've been hootin' and hollerin' for a while, hopefully the message has gotten out there and we know to be in our safe spots," Hall warned his YouTube audience in a calm voice with a Southern twang. A silver robot with blue eyes popped onto the screen to tell Hall that a viewer had commented about tiny houses near the tornado. Hall replied to his AI robot, known as Y'all Bot. The 31-year-old host of Ryan Hall, Y'all--one of YouTube's most popular weather channels with 2.8 million subscribers--went live for nearly 12 hours that day as more than 70 tornadoes swept through the central U.S., killing at least 28 people.

Weather radar is the primary tool used by forecasters to detect and warn for tornadoes in near-real time. In order to assist forecasters in warning the public, several algorithms have been developed to automatically detect tornadic signatures in weather radar observations. Recently, Machine Learning (ML) algorithms, which learn directly from large amounts of labeled data, have been shown to be highly effective for this purpose. Since tornadoes are extremely rare events within the corpus of all available radar observations, the selection and design of training datasets for ML applications is critical for the performance, robustness, and ultimate acceptance of ML algorithms. This study introduces a new benchmark dataset, TorNet to support development of ML algorithms in tornado detection and prediction. TorNet contains full-resolution, polarimetric, Level-II WSR-88D data sampled from 10 years of reported storm events. A number of ML baselines for tornado detection are developed and compared, including a novel deep learning (DL) architecture capable of processing raw radar imagery without the need for manual feature extraction required for existing ML algorithms. Despite not benefiting from manual feature engineering or other preprocessing, the DL model shows increased detection performance compared to non-DL and operational baselines. The TorNet dataset, as well as source code and model weights of the DL baseline trained in this work, are made freely available.

AI will change the way that we work. Or so says the most fervent purveyors of the tech, which now includes Microsoft. But after seeing ChatGPT, Dall-E, and other AI systems integrated into the latest versions of Windows 11, Office, and the company's Microsoft 365 platform, I can't say that I agree. Make no mistake, Microsoft is pushing new tools like its Copilot system hard, integrating it into systems that are staples for the company and hundreds of millions of users. But speaking as a power user -- and accepting the limited perspective that gives me for many who are not -- I can't see these new tools being anything more than an occasional curiosity.

One killed, 10 injured and dozens of homes damage after tornado strikes Laguna Heights, Texas. Drone footage has emerged capturing the aftermath of a deadly tornado that ripped through a Texas Gulf Coast town near the U.S.-Mexico border. The EF-1 twister that struck Laguna Heights early Saturday, located on the mainland across from South Padre Island, left one dead and 10 injured, officials said. Video taken by the Brownsville Fire Department shows the damage that was inflicted upon as many as 60 homes, with some missing roofs and others reduced to piles of rubble. Roberto Flores, 42, died after being "basically crushed as a result of the damage to his mobile home," according to Eddie Treviño Jr., a judge in Cameron County.

Drone video taken Saturday morning shows Rolling Fork, Mississippi, which was heavily damaged when tornadoes and severe storms ripped through the area Friday. Video footage taken Saturday morning showed widespread destruction after tornadoes ripped through Mississippi. A severe weather outbreak across several southern states Friday evening and Saturday morning left at least 23 people dead in Mississippi. Footage captured by camera drones show residential and commercial structures wiped out by the lethal storms in Rolling Fork and Armory, Mississippi. That state's governor, Republican Tate Reeves, issued a state of emergency in all counties affected by the storm Saturday.

Open-ended text generation with autoregressive language models (LMs) is one of the core tasks in natural language processing. However, maximization-based decoding methods (e.g., greedy/beam search) often lead to the degeneration problem, i.e., the generated text is unnatural and contains undesirable repetitions. Existing solutions to this problem either introduce randomness prone to incoherence or require a look-ahead mechanism that demands extra computational overhead. In this study, we formulate open-ended text generation from a new perspective, i.e., we view it as an exploration process within a directed graph. Thereby, we understand the phenomenon of degeneration as circular loops within the directed graph. Based on our formulation, we propose a novel decoding method -- \textit{momentum decoding} -- which encourages the LM to \textit{greedily} explore new nodes outside the current graph. Meanwhile, it also allows the LM to return to the existing nodes with a momentum downgraded by a pre-defined resistance function. We extensively test our approach on three benchmarks from different domains through automatic and human evaluations. The results show that momentum decoding performs comparably with the current state of the art while enjoying notably improved inference speed and computation FLOPs. Furthermore, we conduct a detailed analysis to reveal the merits and inner workings of our approach. Our codes and other related resources are publicly available at https://github.com/gmftbyGMFTBY/MomentumDecoding.

Developing methods to predict disastrous natural phenomena is more important than ever, and tornadoes are among the most dangerous ones in nature. Due to the unpredictability of the weather, counteracting them is not an easy task and today it is mainly carried out by expert meteorologists, who interpret meteorological models. In this paper we propose a system for the early detection of a tornado, validating its effectiveness in a real-world context and exploiting meteorological data collection systems that are already widespread throughout the world. Our system was able to predict tornadoes with a maximum probability of 84% up to five days before the event on a novel dataset of more than 5000 tornadic and non-tornadic events. The dataset and the code to reproduce our results are available at: https://tinyurl.com/3brsfwpk

Did medical knowledge engineering/search/expert systems. Every human bliss and kindness, every suspicion, cruelty, and torment ultimately comes from the whirring 3-pound "enchanted loom" that is our brain and its other side, the cloud of knowing that is our mind. It's an odd coincidence that serious study of the mind and the brain bloomed in the late 20th century when we also started to make machines that had some mind-like qualities. Now, with information technology we have applied an untested amplifier to our minds, and cranked it up to eleven, running it around the clock, year after year. Because we have become a culture of crisis, we are good at asking, what has gone wrong? But is the conjunction of natural and artificial mind only ill-favored, or might we not learn from both by comparison?