The day's stories
Product LaunchGemini Spark, Google's agentic assistant, is now available on MacGoogle has officially launched Gemini Spark, its AI-powered assistant, for Mac users. This new version integrates seamlessly into the existing Gemini desktop apTechCrunch AIJul 1· 59%FundingAshton Kutcher leaving Sound Ventures to launch new VC firm with Morgan BellerAshton Kutcher is stepping away from Sound Ventures, the venture capital firm he co-founded, to establish a new VC fund alongside Morgan Beller. This new firm wTechCrunch AIJul 1· 58%RegulationCloudflare's new policy pushes AI companies to pay for publishers' contentCloudflare has introduced a new policy that mandates AI companies to distinguish their web crawlers from conventional search engine crawlers by September 15, 20TechCrunch AIJul 1· 58%InfrastructureMeta, like SpaceX, looks to turn excess AI compute into cashMeta has announced plans to establish a cloud infrastructure business aimed at monetizing its surplus AI compute capabilities and models. This strategic move miTechCrunch AIJul 1· 57%FundingWayve launches $85M employee tender offer at $8.5B valuationWayve, a U.K.-based startup specializing in self-driving technology, has announced an $85 million employee tender offer, valuing the company at $8.5 billion. ThTechCrunch AIJul 1· 57%OtherThe 'Father of the Internet' is finally retiringVinton Cerf, widely recognized as the 'Father of the Internet,' is retiring from his position as Google's chief internet evangelist after more than two decades TechCrunch AIJul 1· 57%RegulationTrump drops restrictions on Anthropic's Mythos and Fable modelsThe U.S. government has recently lifted export restrictions on Anthropic's Mythos and Fable AI models, enabling these advanced technologies to be accessible to TechCrunch AIJul 1· 57%FundingVenice AI becomes a unicorn with $65M Series A as its privacy-first AI platform takes offVenice AI has successfully raised $65 million in a Series A funding round, which has propelled the company to a unicorn status with a valuation of $1 billion. TTechCrunch AIJul 1· 56%RegulationGodot says bye bye AI, bans vibe-coded contributionsThe Godot team has implemented a new policy that prohibits AI-generated contributions to their open-source game engine. This decision stems from concerns regardThe Register AIJul 1· 56%Product LaunchHugging Face and Cerebras bring Gemma 4 to real-time voice AIHugging Face and Cerebras have unveiled Gemma 4, a cutting-edge real-time voice AI system designed to enhance speech-to-speech interactions. This new system boaHugging Face BlogJul 1· 56%Product LaunchSpaceX has an AI device prototype, and it sure sounds phone-ishSpaceX has unveiled a prototype for a new AI device that resembles a smartphone, aiming to integrate advanced AI capabilities from Elon Musk's xAI. Despite the TechCrunch AIJul 1· 54%ResearchAgRefactor: Self-Evolving Agentic Workflow for HLS Compatibility and PerformanceAgRefactor is a newly introduced multi-agent workflow that leverages large language models (LLMs) to refactor software for High-Level Synthesis (HLS) compatibilarXiv cs.AIJul 1· 64%BenchmarkBayesBench: Evaluating LLM Belief Trajectories Under Multi-Turn Evidence AccumulationResearchers have developed BayesBench, a novel evaluation suite designed to assess the performance of large language models (LLMs) in updating their beliefs durarXiv cs.AIJul 1· 63%ResearchContrastive Reflection for Iterative Prompt OptimizationA recent study presented a novel framework called Contrastive Reflection aimed at optimizing prompts for large language model (LLM) agents, particularly in the arXiv cs.AIJul 1· 63%ResearchWhy Solve It Twice? Hierarchical Accumulation of Skills for Transfer-Efficient ML EngineeringResearchers have introduced a novel approach to machine learning engineering through a system called HASTE, which stands for Hierarchical Accumulation of SkillsarXiv cs.AIJul 1· 63%ResearchPredictable GRPO: A Closed-Form Model of Training DynamicsRecent research has introduced a closed-form model for the training dynamics of Group Relative Policy Optimization (GRPO). This model aims to address the empiriarXiv cs.LGJul 1· 63%ResearchWhat Drives Interactive Improvement from Feedback?A recent study published on arXiv explores the dynamics of interactive improvement in multi-turn language agent settings, particularly focusing on the role of farXiv cs.AIJul 1· 63%ResearchReactionAtlas: Ab origine exploration of chemical reaction networks with machine learningThe recent introduction of ReactionAtlas marks a significant advancement in the exploration of chemical reaction networks through machine learning. This innovatarXiv cs.LGJul 1· 63%BenchmarkAccelerometry-Derived Digital Biomarkers for Cardiometabolic Risk: A Population-Representative Tabular Benchmark with Uncertainty QuantificationA recent study has unveiled the NHANES Accelerometry Cardiometabolic Benchmark, focusing on the use of accelerometry data to predict cardiometabolic risk factorarXiv cs.LGJul 1· 63%Product LaunchFrom Search to Synthesis: Training LLMs as Zero-Shot Workflow GeneratorsMeta has introduced MetaFlow, a groundbreaking system that enables large language models (LLMs) to autonomously generate workflows for a variety of tasks. This arXiv cs.LGJul 1· 63%ResearchWhen Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning ModelsA recent study titled 'When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models' explores the concept of learned stopping rules inarXiv cs.AIJul 1· 63%ResearchRoPoLL: Robust Panel of LLM JudgesA new evaluation method for large language models (LLMs) has been introduced, known as RoPoLL (Robust Panel of LLM Judges). This innovative approach aims to tacarXiv cs.AIJul 1· 62%ResearchJoint discovery of governing partial differential equations from multi-source datasets by competitive optimizationA recent study introduced a novel framework called MCO-PDE, which stands for Multi-source Competitive Optimization for Partial Differential Equations. This framarXiv cs.LGJul 1· 62%ResearchHow Can AI Find My Model? A Model-Finding Experimental Study Considering Data Formats, Embeddings, and Retrieval StrategiesA recent study published on arXiv explores the effectiveness of various data formats, embedding models, and retrieval strategies in the context of finding simularXiv cs.AIJul 1· 62%ResearchWhy Do Few-Step Text Latents Fail When Image Latents Work? Non-Commitment at Sharp Categorical ReadoutsA recent paper published on arXiv explores the disparity in performance between few-step generation in image latents and text latents. The authors argue that tharXiv cs.LGJul 1· 62%ResearchInvestigating Multi-Agent Deliberation in LawA recent paper published on arXiv delves into the application of multi-agent deliberation methods in the realm of legal reasoning, leveraging Large Language ModarXiv cs.AIJul 1· 62%ResearchHierarchical Global Attention (HGA)A new method called Hierarchical Global Attention (HGA) has been introduced by researchers to enhance the efficiency of long-context transformers. This innovatiarXiv cs.LGJul 1· 62%ResearchGradient Smoothing: Coupling Layer-wise Updates for Improved OptimizationRecent research has introduced a novel optimization technique called Gradient Smoothing, which leverages Depth-wise Gradient Augmentation to improve the traininarXiv cs.LGJul 1· 62%ResearchRevocable Learned State via Process SidecarsA recent paper has introduced an innovative approach known as process sidecars, aimed at enhancing the revocation of learned states in language models. This metarXiv cs.LGJul 1· 61%ResearchBeyond expert users: agents should help users construct preferences, not just elicit themRecent research has introduced CoPref, a novel model aimed at improving how AI agents assist users in constructing their preferences. Unlike traditional systemsarXiv cs.AIJul 1· 61%ResearchMind the Residual Gap: Probabilistic Downscaling under Real-World BiasRecent advancements in probabilistic downscaling have been made with the introduction of ReMatch, a novel method designed to improve model calibration by addresarXiv cs.LGJul 1· 61%