Minnesota
Tropical Storm Bertha to form near US coast in just HOURS as millions brace for direct hit
FIFA launch investigation into Argentina's shocking World Cup final post-match behaviour... as new footage shows what really sparked final-whistle brawl Progressive Minneapolis official gets PROMOTION and pay rise after America's wokest prosecutor gave him slap on wrist for keying six Teslas at dealership to avenge Trump win Trump's cringe World Cup cameo was a fitting finale to a tournament blighted by excess, greed and FIFA boss Gianni Infantino's twisted dreams'People hate her': KENNEDY's Hollywood mole unleashes on Angelina Jolie... as two more of her kids publicly humiliate dad Brat Pitt Car experts reveal the little-known stereotypes that are actually TRUE... What does your vehicle really say about you? No one's shocked Harry wants to reverse Megxit. MAGA-friendly Fox News star Maria Bartiromo hints she's losing patience with Trump by asking VERY pointed question to ex-White House aide Shakira fans marvel over her'ageless' appearance at World Cup final as her ex-husband watches on from the stands with their sons Mystery over what REALLY caused'explosive diarrhea' parasite as FDA admits lettuce that was first to blame was a'false positive' Trump reveals first photo of JD Vance's'perfect baby boy' Ever since I went on Vyvanse for ADHD, my marriage has felt different. A comment from my husband made me realise why women are calling it the'divorce drug': ASK JANA Scandal engulfs'it girl' fashion empire Set Active: Explosive claims about influencer founder's desperate fixation... 'inappropriate' texts that went ignored... and why employees wanted to stay on her'good side' Bitter Argentina players turn their backs to Spain's World Cup trophy lift after violence marred winners' celebrations - as lip-reader reveals what furious star said to winning captain Rodri during brawl Inside Luke Wilson's reclusive life with decades-younger girlfriend Kendall Yates: Friends reveal new details about their'secretive' romance after birth of their child... as unseen pictures of her emerge Hollywood's new youth elixir: Wealthy women are hooked on secret holy grail said to banish wrinkles and plump skin... but doctors warn of shock risk Inside the'pure evil' Ohio'House of Horrors': Forensic psychologists analyze every disturbing detail about the 16 children found living in squalor MORE: Urgent warning to brace for catastrophic'Ring of Fire' disaster as global megaquake chain reaction sparks fears for MILLIONS in its path: 'Super overdue' A tropical storm is predicted to erupt along the US Gulf Coast within hours as millions of Americans brace for a direct hit and life-threatening conditions. The National Hurricane Center has already named the latest disturbance Tropical Depression Two, but forecasters expect this system to grow into a named storm as soon as Monday afternoon when wind speeds increase to more than 39 mph.
What if It's Not the Phones?
An evolutionary psychologist is challenging the popular understanding of kids and technology. W hen the 82-year-old psychologist Peter Gray describes the way he grew up, he punctuates the anecdotes by saying that modern parents would be arrested for letting a child have such fun. When he was 4 years old, he would walk to a store in Minneapolis to buy cigarettes for his grandmother. When he was 11, he would sometimes stay home from school in Hill City, Minnesota, to operate a newspaper printing press owned by his mother and stepfather. His parents were not arrested, and that's because the childhood they permitted him to have was basically normal at the time, even if his family did have a newspaper printing press in the house. As a boy, Peter was obsessed with fishing and baseball; neighborhood friends taught him how to ride his bike and catch grasshoppers. Although Gray's career as a scientist would begin with laboratory studies of rat hormones, he eventually found his way to writing about his childhood, in a fashion.
Spotify Confirms Streaming Fraud After Kalshi Trader Cries Foul
One of Kalshi's most prominent traders tells WIRED he's swearing off Spotify-related markets until the issue is resolved. Top Kalshi trader Caleb Davies usually speaks to the press about how prediction markets help him rake in money. The Minneapolis-based IT worker estimates he's made $1.2 million overall across different prediction platforms, with $414,000 in winnings from Kalshi's culture markets alone. He especially enjoys wagering on music charts, because he carefully analyzes Spotify data to pick winners. "Every single morning, I'm going in, downloading the data, and updating my projections," he tells WIRED.
On the Entropy Calibration of Language Models
We study the problem of entropy calibration, which asks whether a language model's entropy over generations matches its log loss on human text. Past work found that models are miscalibrated, with entropy per step increasing as generations grow longer, due to error accumulation. To calibrate the model and improve text quality, it has become standard practice to truncate the distribution, but this approach reduces output diversity, which we would like to avoid. Therefore, in this paper, we ask: does miscalibration improve automatically with scale, and if not, is it theoretically possible to calibrate without tradeoffs? To build intuition, we first study a simplified theoretical setting to characterize the scaling behavior of miscalibration with respect to dataset size. We find that the rate of scaling depends on the power law exponent of the data distribution -- in particular, for a power law exponent close to 1, the scaling exponent is close to 0, meaning that miscalibration improves very slowly with scale.
Set-LLM: APermutation-Invariant LLM
While large language models (LLMs) demonstrate impressive capabilities across numerous applications, their robustness remains a critical concern. This paper is motivated by a specific vulnerability: the order sensitivity of LLMs. This vulnerability manifests itself as the order bias observed when LLMs decide between possible options (for example, a preference for the first option) and the tendency of LLMs to provide different answers when options are reordered. The use cases for this scenario extend beyond the classical case of multiple-choice question answering to the use of LLMs for multidocument tasks and as automated evaluators in AI pipelines. We introduce Set-LLM, a novel architectural adaptation for pretrained LLMs that enables the processing of mixed set-text inputs with permutation invariance guarantees. The adaptations involve a new attention mask and new positional encodings specifically designed for sets. We provide a theoretical proof of invariance and demonstrate through experiments that Set-LLM can be trained effectively, achieving comparable or improved performance and maintaining the runtime of the original model, while altogether eliminating order sensitivity.
Refusal Direction is Universal Across Safety-Aligned Languages
Refusal mechanisms in large language models (LLMs) are essential for ensuring safety. Recent research has revealed that refusal behavior can be mediated by a single direction in activation space, enabling targeted interventions to bypass refusals. While this is primarily demonstrated in an English-centric context, appropriate refusal behavior is important for any language, but poorly understood. In this paper, we investigate the refusal behavior in LLMs across 14 languages using PolyRefuse, a multilingual safety dataset created by translating malicious and benign English prompts into these languages. We uncover the surprising cross-lingual universality of the refusal direction: a vector extracted from English can bypass refusals in other languages with near-perfect effectiveness, without any additional fine-tuning. Even more remarkably, refusal directions derived from any safety-aligned language transfer seamlessly to others. We attribute this transferability to the parallelism of refusal vectors across languages in the embedding space and identify the underlying mechanism behind cross-lingual jailbreaks. These findings provide actionable insights for building more robust multilingual safety defenses and pave the way for a deeper mechanistic understanding of cross-lingual vulnerabilities in LLMs.1
3BASiL: An Algorithmic Framework for Sparseplus Low-Rank Compression of LLMs
Sparse plus Low-Rank (S + LR) decomposition of Large Language Models (LLMs) has emerged as a promising direction in model compression, aiming to decompose pre-trained model weights into a sum of sparse and low-rank matrices W S + LR. Despite recent progress, existing methods often suffer from substantial performance degradation compared to dense models. In this work, we introduce 3BASiL-TM, an efficient one-shot post-training method for (S + LR) decomposition of LLMs that addresses this gap. Our approach first introduces a novel 3-Block Alternating Direction Method of Multipliers (ADMM) method, termed 3BASiL, to minimize the layer-wise reconstruction error with convergence guarantees.