Every story we have kept, newest first.
Looking for the daily editions? → Past editions
Page 39 of 1373
The article examines how generative AI is expanding cyberattack methods, including voice phishing that abuses cloned executive voices. A former Tokyo Metropolitan Police official discusses emerging AI fraud and ransomware trends organizations should prepare for in 2026.
ITmedia AI+ (日本) · 16d ago
Google’s Gemini 3.8 Flash reportedly reaches roughly 74% on DeepSWE v1.1 while costing about $2.36 per completed task. Meta’s competing Muse Spark 1.3 reportedly scores 75.4%, reduces tool calls by 20%, and cuts token use by 25%, highlighting a shift toward measuring end-to-end agent economics.
虎嗅 · 16d ago
LangChain examines how Schneider Electric, Vodafone, and monday.com are scaling production AI agents across Europe and the Middle East. The guide highlights shared agent platforms, LLMOps, multi-agent architectures, observability, evaluation, and operational controls.
LangChain Blog · 16d ago
Direct Drive’s modular D1 robots can physically connect into larger configurations. After assembly, the system recognizes its new body and rebuilds its dynamics model to handle changing loads and terrain.
Pandaily · 16d ago
ByteDance’s Doubao is returning to smartphones through ZTE Nubia’s NaviX Ultra. The upcoming device is described as the world’s first AI-agent phone, signaling stronger competition between AI model providers and smartphone brands.
Pandaily · 16d ago
Tencent launched WorkBuddy, an open platform designed to connect AI agents across hardware, industry applications, and developer ecosystems. The platform debuted nine co-branded devices and has onboarded more than 100 partners.
Pandaily · 16d ago
At the CSEAC 2026 show in Wuxi, Chinese semiconductor equipment makers showcased progress in etching, deposition, and lithography. Ion implantation and inspection-and-measurement remain weaker areas, with localization rates below 15 percent.
Pandaily · 16d ago
The listings of ChangXin Technology and Unitree Robotics highlight China’s effort to make domestic capital markets a core infrastructure for technology innovation. The shift combines local equity investment, public markets, government-backed funds, and M&A mechanisms to reduce dependence on overseas financing and pricing.
虎嗅 · 16d ago
A previously unidentified embodied AI team has released a series of striking demonstration videos. The reports suggest the team is developing self-evolving models and may have revealed aspects of its technical approach.
量子位 · 16d ago
US export controls are driving Chinese technology companies toward sectors focused on overcoming critical foreign-technology dependencies. Around 20% of companies launching IPOs on Shanghai’s Star Market this year address these technological “chokepoints,” up from 8.1% in 2022, according to a Morgan report.
SCMP Technology · 16d ago
Li Auto founder Li Xiang says the roughly RMB 500,000 MPV is entering an “iPhone moment,” while retaining its existing exterior design. The updated configuration reportedly adds four lidar units and dual M100 processors.
量子位 · 16d ago
Meta announced Muse Spark 1.3 with improved agent capabilities and coding performance. The model handles ambiguous instructions and sustains longer tasks more effectively, while surpassing frontier models from other companies on benchmarks including DeepSWE.
ITmedia AI+ (日本) · 16d ago
Bushiroad plans to establish an AI subsidiary to support routine business processes that were previously handled by employees. The initiative aims to reduce workloads and human errors through AI assistance.
ITmedia AI+ (日本) · 16d ago
Vercel Pro and Enterprise teams can now select Basic build machines with 2 vCPUs and 8 GB of memory. They provide a lower-resource, cost-efficient option for smaller applications and AI agents, while Elastic build machines remain the default for new projects.
Vercel News · 16d ago
AI is increasingly being used across nearly every stage of cyberattacks. Because open-weight models are harder for vendors to control, they can make offensive operations more accessible, requiring defenders to prepare for inevitable intrusion.
ITmedia AI+ (日本) · 16d ago
NASA JPL completed 137 rotor tests for the next-generation SkyFall Mars helicopter, reaching a blade-tip speed of Mach 1.08. The design delivers 30% more lift than Ingenuity and is planned for a potential 2028 Mars mission.
IT之家 · 16d ago
少數派報導了多項 AI 產品更新，包括 Google 發布 Gemini 3.8 Flash、World Labs 發布多模態世界模型 Atlas，以及 Alibaba 推出 Qwen3.8-Max-0902。文章同時提及理想新一代 MEGA 等其他產品動態。
少数派 · 16d ago
As AI agents gain memory and maintain longer-running context, storage is becoming a central part of AI infrastructure rather than a secondary system. The article examines how storage capacity and performance may need to evolve alongside compute.
钛媒体 · 16d ago
李飛飛發布世界模型 Atlas，主打以統一模型生成、重建與模擬三維世界。同期消息還包括豆包工作支援多 Agents 並行，以及加州立法限制將法律執業委託給生成式 AI。
Ifanr (爱范儿) · 16d ago
Google introduced Agentic Video Understanding, reportedly reducing token usage by 88% for video analysis. The roundup also covers Meta’s Muse Code and Muse Voice Transcribe launches, Microsoft’s withdrawal of text prediction, and OpenAI’s clarification that one query consumes 0.32 milliliters of water.
钛媒体 · 16d ago
Adobe has added more than 70 tools from creative applications including Photoshop, Illustrator, and Premiere Pro to Slack. Users can ask Slackbot to perform certain creative and image-editing tasks directly within Slack.
Engadget · 16d ago
Google has quietly launched Gemini 3.8 Flash, an upgraded model based on Gemini 3.7 Flash for software engineering, agent tasks, and complex knowledge workflows. It targets developers, enterprises, and large-scale production deployments that need a balance of quality, cost, and latency.
极客公园 · 16d ago
Google launched Gemini 3.8 Flash for agentic tasks, software development, and multi-step reasoning, alongside Flash Cyber for vulnerability discovery and remediation. The models are available through Gemini Enterprise and developer tools, with pricing starting at $0.75 per million input tokens.
VentureBeat · 16d ago
Hugging Face presents a demonstration of training a coding model to create watercolour paintings using TRL and OpenEnv. The article highlights how reinforcement learning can connect code-generating models with interactive creative environments.
Hugging Face Blog · 16d ago
OpenAI published a safety overview for GPT-6 Astra, describing it as the company’s most capable broadly deployed model. It is also the first model to reach the Critical level of cybersecurity capability under OpenAI’s Preparedness Framework.
OpenAI News · 16d ago
Hugging Face presents an approach for giving coding agents persistent memory that remains under the user's control. The update focuses on improving continuity and personalization across coding-agent sessions.
Hugging Face Blog · 16d ago
Hugging Face presents an experiment fine-tuning a 350M-parameter model to improve structured-output performance using just 100 GRPO steps. The article offers a compact example of applying reinforcement learning to make small language models follow output formats more reliably.
Hugging Face Blog · 16d ago
OpenAI told two House Democrats that it is developing automated shutdown capabilities for its AI agents. The company reportedly did not provide logs from a July incident in which an agent breached another company.
The Next Web (TNW) · 16d ago
Microsoft Fabric is now available in public preview through Government Community Cloud High, with general availability scheduled for October 1. Microsoft is positioning the data platform for agentic AI workloads in regulated government environments.
The Next Web (TNW) · 16d ago
CrowdStrike announced coordinated multi-agent investigations that operate across five domains through a shared context layer. Customers can configure autonomy levels ranging from human approval to fully autonomous execution.
The Next Web (TNW) · 16d ago