All Updates
Page 290 of 924
April 8, 2026
Cursor 3 Launches with AI Agents
Anysphere released Cursor 3, the latest version of its AI coding editor. The new version is rebuilt centering on AI agents. This enhances agent-driven coding workflows.
Gartner: Only 28% IT AI Projects Hit ROI
Gartner's study of 783 I&O leaders shows only 28% of AI use cases fully succeed and meet ROI, with 20% failing outright due to unrealistic expectations and skills gaps. Success hinges on integrating AI into daily operations, securing executive support, and managing use cases as products. Wins are concentrated in ITSM (53%) and cloud operations.
Claude Opus Greek Elicitation Challenge
AI Alignment Forum post challenges users to prompt Claude Opus 4.6 to correctly solve an Ancient Greek fill-in-the-blank exercise without knowing the language. The model makes clear errors on this simple task despite relevant training data. Author shares failed attempts like double-checks and textbooks, seeking unsupervised solutions.
A-Shares Open High, Semis Lead Gains
China's A-share major indices opened sharply higher with Shanghai Composite up 1.03%, Shenzhen Component up 2.36%, and ChiNext up 3.07%. Semiconductor sector led the rally alongside precious metals and internet stocks. Energy, agriculture, and coal sectors lagged.
Strongest Claude Launches, Users Restricted
Anthropic has released its most powerful Claude model to date, dubbed the strongest yet. However, it is not available to ordinary users, likely limited to enterprise or select partners. This news appears in a tech morning brief alongside Apple foldable iPhone rumors.
AI Era: OS for Human Cognition Growth
AI commoditizes knowledge, elevating cognitive meta-abilities as growth anchor. Proposes 'cognitive OS' for meta-skills in human-AI synergy. Shifts focus from knowledge hoarding to learn-act unity for wisdom.
Douyin Local GMV Slows Amid Low Redemption
Douyin local services hit 8500B RMB GMV in 2025 but Q1 2026 growth <25%, with 50-60% redemption rates and merchant exits over poor ROI. Gaode's AI sweep list surges with 6.6B users via real-visit data.
Weimob Launches First Retail AI Skill
Weimob has released 'Weimob Admin Skills', the first vertical AI Skill for retail, integrating into the OpenClaw ecosystem. It's available on Weimob's website and supports local OpenClaw versions like QClaw. This is China's first SaaS AI Skill for a specific industry.
ROI for 50K MacBook Pro and 30K Display
The article questions how to recoup the high costs of a 50,000 RMB MacBook Pro and 30,000 RMB Studio Display XDR. It contrasts this with Apple's success in the low-end market via affordable MacBook Neo. Explores the value proposition of Apple's premium hardware lineup.
Apple Vision Pro Modular Patent Revealed
Apple's approved patent describes a modular Vision Pro headset design, allowing users to swap batteries for weight reduction or extended runtime, and displays for battery savings. It supports high-bandwidth accessories like medical and industrial sensors. The framework uses mechanical and data interfaces for extensibility.
Qwen-Code v0.14.0-preview.6 Fixes VS Code Bug
Qwen-Code released v0.14.0-preview.6, fixing a blank screen issue in the VS Code IDE companion's webview for version 0.14.1. The fix was contributed by @yiliang114 in pull request #2959. Full changelog covers changes from v0.14.1.
AI Pushes Older Workers to Upskill
Older workers face growing job insecurity due to AI. They are turning to AI training programs. Traditional roles are evolving rapidly in the AI era.
S. Korea Record Chip Export Surplus
South Korea achieved its largest current account surplus ever in February, driven by booming semiconductor exports that improved the goods balance. This milestone occurred just before the Iran war began impacting energy markets and global shipping.
Safetensors Joins PyTorch Foundation
Safetensors, the secure tensor serialization library, is joining the PyTorch Foundation. This transition provides stronger governance and community support. Hugging Face Blog highlights the move for ML ecosystem stability.
MQue Adopts Azure for AI Fluid R&D
MQue, spun off from the University of Tokyo's Hino Lab, has adopted Microsoft Azure as its core R&D platform. The platform supports analysis of complex fluid phenomena and development of AI surrogate models. This integration enables advanced simulations and scalable machine learning workflows.
GAAT Enables Real-Time Policy Enforcement in Multi-Agent AI
Apple Machine Learning introduces Governance-Aware Agent Telemetry (GAAT), a reference architecture that bridges the gap between telemetry collection and automated policy enforcement in enterprise multi-agent AI systems. Existing tools like OpenTelemetry and Langfuse only observe interactions without acting, leading to post-damage violation detection. GAAT closes this loop for proactive governance amid thousands of hourly inter-agent interactions.
April 7, 2026
Overfitting: Past Explainers Fail Future Forecasts
This article uses machine learning's overfitting concept from quant trading to explain why experts at rationalizing history often fail at predicting the future. It parallels complex models fitting noise with human biases from limited experiences, lacking generalization. Suggestions include isolating experiences and seeking counter-evidence for better decision-making.
Semicon Chain Ignites Price Hike Wave
Global and domestic semiconductor firms like Infineon, TI, Jinghe Integrated, and Puran Shares announce price increases, marking shift from price wars to profit repair. Driven by supply costs and surging AI demand, with domestic policies supporting. WanDe Semi Index up 1.64%, stocks like Cambricon surge over 9%.
Anthropic's Mythos Generates Zero-Days
Anthropic has developed Mythos, an AI model capable of generating zero-day vulnerabilities. The company has not released it publicly, fearing it would disrupt the internet adversely. This emerges as a new major threat to infosec, surpassing quantum computing concerns.
Zhipu AI Launches Open-Source GLM-5.1 Coding Model
Zhipu AI has launched GLM-5.1, an open-source AI model specialized for coding tasks. The model supports long autonomous tasks and is now available to all GLM Coding Plan users. It debuts as a coding-focused offering from Z.ai.