European diplomats reveal the'tough guy' US negotiator leading the charge on Greenland: 'He hates us' A former Marine was unmasked as the'Zodiac killer' after a bombshell new investigation. I suffered a horrific side effect of a drug used by millions of Americans... and my face'melted off' The ICE backlash isn't the end of Kristi Noem It may have just saved her career FedEx driver accused of abducting and killing little girl while delivering her Christmas present says he shouldn't be executed because he has autism Senator accused of steamy affair with her bodyguard in bombshell lawsuit from his WIFE: 'Bring MDMA so I can guide you' Hunter Biden's stripper baby mama asks for him to be ARRESTED over claims he is still failing to pay her child support Family of Tyler Robinson's transgender lover speaks out for first time since Charlie Kirk assassination and reveals where he is now Dodgers agree with Kyle Tucker'on $240m deal' as champs beat out Mets, Blue Jays for top free agent World's sexiest hockey star and OnlyFans model Mikayla Demaiter spills out of little dress in latest post Nicole Richie addresses her daughter's new identity after unveiling transformation on her 18th birthday Trump gushes over'young beautiful' hockey players and teases rebranding of famed presidential wall Trump's AG secretary sparks mockery with tone-deaf $3 dinner advice as food costs soar Karoline Leavitt reveals the thinking behind Trump's call to cancel elections Microsoft's paperclip mascot delights users as it returns - 18 years after it was axed from Office It was the original virtual assistant, released years before Siri, Alexa, and Bixby. Now, almost two decades after it was axed, Microsoft's Clippy is officially back. The friendly anthropomorphic paper clip has been spotted as an Easter egg in Microsoft's latest announcement about a new AI companion called Mico. Mico - whose name is a nod to Microsoft Copilot - is a small blob with a friendly smiley face, and doesn't look much like its much-loved predecessor.

Plus: Meta's lawyers advised staff to remove parts of their research After years of growth that spawned hundreds of startups, the nascent carbon removal sector appears to be facing a reckoning. Running Tide, a promising aquaculture company, shut down its operations last summer, and a handful of other companies have shuttered, downsized, or pivoted in recent months as well. And the collective industry hasn't made a whole lot more progress toward Running Tide's ambitious plans to sequester a billion tons of carbon dioxide by this year. The hype phase is over and the sector is sliding into the turbulent business trough that follows, experts warn. And the open question is: If the carbon removal sector is heading into a painful if inevitable clearing-out cycle, where will it go from there? This story is part of MIT Technology Review's What's Next series, which looks across industries, trends, and technologies to give you a first look at the future.

Two federal judges acknowledged AI-generated court orders contained serious errors, prompting calls from Sen. Chuck Grassley for stronger judicial artificial intelligence policies.

The'Surge' of Troops May Not Come to San Francisco, but the City Is Ready Anyway San Francisco is preparing for federal law enforcement's invasion of the Bay Area, whether it happens or not. Citizens protesting the threat of federal troop deployments in the San Francisco Bay Area held a rally on Thursday at SF City Hall. After months of deployments by US Immigration and Customs Enforcement and the National Guard across American cities, federal agents have been preparing to descend into San Francisco . Local resistance groups have been coordinating with activists in other cities across the country that have been besieged by federal law enforcement. Thousands of volunteers, coordinating through Signal group chats, Zoom calls, and social media posts, planned protests and spread the word that federal troops are on their way to San Francisco.

Scaling laws aim to accurately predict model performance across different scales. Existing scaling-law studies almost exclusively rely on cross-entropy as the evaluation metric. However, cross-entropy provides only a partial view of performance: it measures the absolute probability assigned to the correct token, but ignores the relative ordering between correct and incorrect tokens. Yet, relative ordering is crucial for language models, such as in greedy-sampling scenario. To address this limitation, we investigate scaling from the perspective of relative ordering. We first propose the Relative-Based Probability (RBP) metric, which quantifies the probability that the correct token is ranked among the top predictions. Building on this metric, we establish the Relative-Based Scaling Law, which characterizes how RBP improves with increasing model size. Through extensive experiments on four datasets and four model families spanning five orders of magnitude, we demonstrate the robustness and accuracy of this law. Finally, we illustrate the broad application of this law with two examples, namely providing a deeper explanation of emergence phenomena and facilitating finding fundamental theories of scaling laws. In summary, the Relative-Based Scaling Law complements the cross-entropy perspective and contributes to a more complete understanding of scaling large language models. Thus, it offers valuable insights for both practical development and theoretical exploration.

Machine unlearning algorithms aim to remove the influence of specific training samples, ideally recovering the model that would have resulted from training on the remaining data alone. We study unlearning in the overparameterized setting, where many models interpolate the data, and defining the solution as any loss minimizer over the retained set$\unicode{x2013}$as in prior work in the underparameterized setting$\unicode{x2013}$is inadequate, since the original model may already interpolate the retained data and satisfy this condition. In this regime, loss gradients vanish, rendering prior methods based on gradient perturbations ineffective, motivating both new unlearning definitions and algorithms. For this setting, we define the unlearning solution as the minimum-complexity interpolator over the retained data and propose a new algorithmic framework that only requires access to model gradients on the retained set at the original solution. We minimize a regularized objective over perturbations constrained to be orthogonal to these model gradients, a first-order relaxation of the interpolation condition. For different model classes, we provide exact and approximate unlearning guarantees and demonstrate that an implementation of our framework outperforms existing baselines across various unlearning experiments.

Legal consultation is essential for safeguarding individual rights and ensuring access to justice, yet remains costly and inaccessible to many individuals due to the shortage of professionals. While recent advances in Large Language Models (LLMs) offer a promising path toward scalable, low-cost legal assistance, current systems fall short in handling the interactive and knowledge-intensive nature of real-world consultations. To address these challenges, we introduce LeCoDe, a real-world multi-turn benchmark dataset comprising 3,696 legal consultation dialogues with 110,008 dialogue turns, designed to evaluate and improve LLMs' legal consultation capability. With LeCoDe, we innovatively collect live-streamed consultations from short-video platforms, providing authentic multi-turn legal consultation dialogues. The rigorous annotation by legal experts further enhances the dataset with professional insights and expertise. Furthermore, we propose a comprehensive evaluation framework that assesses LLMs' consultation capabilities in terms of (1) clarification capability and (2) professional advice quality. This unified framework incorporates 12 metrics across two dimensions. Through extensive experiments on various general and domain-specific LLMs, our results reveal significant challenges in this task, with even state-of-the-art models like GPT-4 achieving only 39.8% recall for clarification and 59% overall score for advice quality, highlighting the complexity of professional consultation scenarios. Based on these findings, we further explore several strategies to enhance LLMs' legal consultation abilities. Our benchmark contributes to advancing research in legal domain dialogue systems, particularly in simulating more real-world user-expert interactions.

Large Reasoning Models (LRMs) have become powerful tools for complex problem solving, but their structured reasoning pathways can lead to unsafe outputs when exposed to harmful prompts. Existing safety alignment methods reduce harmful outputs but can degrade reasoning depth, leading to significant trade-offs in complex, multi-step tasks, and remain vulnerable to sophisticated jailbreak attacks. To address this, we introduce SAFEPATH, a lightweight alignment method that fine-tunes LRMs to emit a short, 8-token Safety Primer at the start of their reasoning, in response to harmful prompts, while leaving the rest of the reasoning process unsupervised. Empirical results across multiple benchmarks indicate that SAFEPATH effectively reduces harmful outputs while maintaining reasoning performance. Specifically, SAFEPATH reduces harmful responses by up to 90.0% and blocks 83.3% of jailbreak attempts in the DeepSeek-R1-Distill-Llama-8B model, while requiring 295.9x less compute than Direct Refusal and 314.1x less than SafeChain. We further introduce a zero-shot variant that requires no fine-tuning. In addition, we provide a comprehensive analysis of how existing methods in LLMs generalize, or fail, when applied to reasoning-centric models, revealing critical gaps and new directions for safer AI.

Large language models (LLMs) have shown great potential in flagging harmful content in online communities. Yet, existing approaches for moderation require a separate model for every community and are opaque in their decision-making, limiting real-world adoption. We introduce Mixture of Moderation Experts (MoMoE), a modular, cross-community framework that adds post-hoc explanations to scalable content moderation. MoMoE orchestrates four operators -- Allocate, Predict, Aggregate, Explain -- and is instantiated as seven community-specialized experts (MoMoE-Community) and five norm-violation experts (MoMoE-NormVio). On 30 unseen subreddits, the best variants obtain Micro-F1 scores of 0.72 and 0.67, respectively, matching or surpassing strong fine-tuned baselines while consistently producing concise and reliable explanations. Although community-specialized experts deliver the highest peak accuracy, norm-violation experts provide steadier performance across domains. These findings show that MoMoE yields scalable, transparent moderation without needing per-community fine-tuning. More broadly, they suggest that lightweight, explainable expert ensembles can guide future NLP and HCI research on trustworthy human-AI governance of online communities.

Large Language Models are increasingly adopted as critical tools for accelerating innovation. This paper identifies and formalizes a systemic risk inherent in this paradigm: \textbf{Black Box Absorption}. We define this as the process by which the opaque internal architectures of LLM platforms, often operated by large-scale service providers, can internalize, generalize, and repurpose novel concepts contributed by users during interaction. This mechanism threatens to undermine the foundational principles of innovation economics by creating severe informational and structural asymmetries between individual creators and platform operators, thereby jeopardizing the long-term sustainability of the innovation ecosystem. To analyze this challenge, we introduce two core concepts: the idea unit, representing the transportable functional logic of an innovation, and idea safety, a multidimensional standard for its protection. This paper analyzes the mechanisms of absorption and proposes a concrete governance and engineering agenda to mitigate these risks, ensuring that creator contributions remain traceable, controllable, and equitable.