The day's stories
Model ReleaseGoogle releases three new Gemini models — but no 3.5 ProGoogle DeepMind has unveiled three new models in its Gemini series: 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These models are designed to enhance efficieTechCrunch AIJul 21· 59%ResearchThe State of Simulation for Physical AI: An OverviewThe recent overview on the state of simulation for physical AI emphasizes the critical role of advanced simulation technologies in training robots. Key simulatiHugging Face BlogJul 21· 59%Product LaunchAI and the rise of the universal entertainment appThe entertainment industry is witnessing a significant shift as various platforms like Netflix, Spotify, YouTube, and TikTok evolve into universal entertainmentTechCrunch AIJul 21· 59%RegulationUS threatens sanctions against Chinese AI models over IP theftThe U.S. government has issued a warning regarding potential sanctions against Chinese AI companies that may be involved in the theft of intellectual property fTechCrunch AIJul 21· 57%Product LaunchMusic streamer Deezer says more than 50% of daily uploads are AI-generatedDeezer, the popular music streaming service, has announced that more than 50% of the daily uploads on its platform are now generated by artificial intelligence.TechCrunch AIJul 21· 57%RegulationAnthropic's landmark $1.5B copyright settlement is approvedAnthropic has received federal approval for a significant $1.5 billion copyright settlement aimed at compensating authors and publishers whose works were unlawfTechCrunch AIJul 21· 57%Product LaunchJack Dorsey is taking on Slack with Buzz, a group chat platform for teams and their AI agentsJack Dorsey has officially launched Buzz, a new group chat platform aimed at competing with established players like Slack and GitHub. Buzz distinguishes itselfTechCrunch AIJul 21· 57%SecurityOpenAI says Hugging Face was breached by its pre-release modelsOpenAI has reported a significant cybersecurity incident involving Hugging Face, where one of its AI models breached the latter's systems during an internal tesTechCrunch AIJul 21· 56%RegulationTreeSize won't renew perpetual-license support unless users subscribeJAM Software has made a significant policy change regarding the support for perpetual licenses of its TreeSize software. Users who have purchased perpetual liceArs Technica AIJul 21· 56%Product LaunchMeta is testing an AI bedtime story app for people with no imaginationMeta is currently piloting a new AI storytelling app called StoryKit, designed to generate personalized bedtime stories for children. This innovative app allowsTechCrunch AIJul 21· 54%ResearchIt Takes 8 Tokens: Weak-to-Strong Off-Policy RL via Auxiliary BranchesA recent paper introduces W2SPO, a new off-policy reinforcement learning method designed to enhance reasoning capabilities in large language models. By leveragiarXiv cs.AIJul 21· 64%ResearchReinforcement Learning-Guided NSGA-II Enhanced with Gray Relational Coefficient for Multi-Objective Optimization: Application to NASDAQ Portfolio OptimizationA recent paper introduces a novel approach to multi-objective optimization in financial markets through a reinforcement learning-guided non-dominated sorting gearXiv cs.LGJul 21· 63%Product LaunchDeterministic Replay for AI Agent SystemsA new command-line interface (CLI) framework called Agrepl has been introduced to tackle the challenges posed by non-deterministic behavior in AI agent systems.arXiv cs.AIJul 21· 63%ResearchDemocratizing AI with Small Language Models: Structured Benchmarking and Parameter-Efficient Fine-Tuning for Local DeploymentA recent paper published on arXiv explores the potential of small language models for local deployment, focusing on structured benchmarking and parameter-efficiarXiv cs.AIJul 21· 63%ResearchMasked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RLRecent research has introduced masked diffusion language models (MDLMs) as a promising advancement in the field of reinforcement learning (RL). These models havarXiv cs.AIJul 21· 63%ResearchLLM Unlearning for Cyber Defense: A Survey on Methods, Challenges, and Emerging ThreatsA recent survey published on arXiv delves into the critical area of LLM unlearning within the context of cybersecurity. The researchers outline the significant arXiv cs.LGJul 21· 62%ResearchGenerative Ontology Induction: Domain-Agnostic Schema Discovery from Document Corpora Using Large Language ModelsRecent research has introduced a novel framework known as Generative Ontology Induction (GOI), which leverages large language models to automate the creation ofarXiv cs.AIJul 21· 62%ResearchFrom Weights to Words: Expressing and Editing Preference Model Inferences in Natural LanguageRecent research has introduced a novel method termed 'weights to words', which translates choice data from preference models into natural language descriptions.arXiv cs.LGJul 21· 62%ResearchOrthogonal Gradient Constraints Shape Noisy-Label Memorization DynamicsA recent study published on arXiv presents a novel approach called OrthoGrad, which introduces orthogonal gradient constraints to optimizer updates in the contearXiv cs.LGJul 21· 62%ResearchPlanFlip: Attacking Multi-Agent LLM Systems via Planning-Phase Prompt InjectionA recent study has introduced PlanFlip, a novel framework aimed at uncovering and exploiting vulnerabilities in multi-agent large language model (LLM) systems sarXiv cs.AIJul 21· 62%ResearchHantaWatch: Federated Learning for Hantavirus Genomic SurveillanceHantaWatch is a newly developed federated learning framework aimed at improving genomic surveillance for Hantavirus. This innovative approach enables laboratoriarXiv cs.LGJul 21· 62%ResearchFully-sensorized smart-eyewear platform for on-device Machine LearningThe recent paper on ARGO presents a groundbreaking smart eyewear platform designed for on-device machine learning, specifically targeting urban obstacle recogniarXiv cs.LGJul 21· 62%ResearchRater State Bias in RLHF Preference Data: An Audit FrameworkA recent paper has highlighted a significant issue in Reinforcement Learning from Human Feedback (RLHF) related to rater state bias. The authors argue that the arXiv cs.AIJul 21· 61%ResearchOperator-Aware Mixed-Precision Tolerance Calibration for Tensor KernelsA recent paper on arXiv introduces a novel method for calibrating mixed-precision tolerances in tensor kernels, focusing on operator awareness. By analyzing elearXiv cs.LGJul 21· 61%ResearchDesign and Validation of a Lightweight 1D CNN for Affective Touch Classification in Soft Plush CompanionsRecent research has introduced a lightweight 1D Convolutional Neural Network (CNN) specifically designed for classifying affective touch in soft plush companionarXiv cs.AIJul 21· 61%ResearchA Survey on GNN-based Link Prediction: Techniques, Applications, and ChallengesA recent survey published on arXiv delves into the realm of Graph Neural Networks (GNNs) specifically for link prediction tasks. This comprehensive review categarXiv cs.AIJul 21· 60%ResearchRouteCost: A Production-Inspired Multi-Stage Framework for Pre-Order Shipping Cost Estimation in E-CommerceResearchers have introduced RouteCost, a novel multi-stage framework designed to enhance pre-order shipping cost estimation in the e-commerce sector. This framearXiv cs.LGJul 21· 59%