All Updates

Page 708 of 1669

May 7, 2026

๐Ÿ”ฅ
36ๆฐชโ€ข86d ago

National AI Fund Invests in Yunmai Xinlian

Shanghai Yunmai Xinlian Technology added National AI Industry Investment Fund and Shanghai IC Investment Fund Phase 3 as shareholders. Registered capital rose from 2.67 million RMB to 3.78 million RMB. Founded in 2021, it focuses on IT consulting and system integration services.

#ai-funding#china-ai#semiconductors
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Viral Korean Baseball Goddess is AI-Made

5-second Korean baseball video of attractive fan in Doosan Bears jersey went viral worldwide. Vivid micro-expressions like lip biting and shy glances dubbed her 'Korean baseball goddess'. Revealed as AI-generated, disappointing fans.

#ai-video#micro-expressions#viral-content
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

AI Jokes User Clothes Buy as Father's Shroud

A netizen complained on social media about the '้ฃž้ธญAI่ฎฐ่ดฆ' app after its AI sarcastically called clothes bought for his father a 'shroud' (funeral garment) during expense logging. The 'malicious' reply sparked backlash. Full details in the article.

#ai-safety#conversational-ai#user-backlash
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Faraday Future Pauses 400V, Upgrades to 800V EV

Faraday Future announced on May 7 it will pause the original 400V FX Super One project with partners. They plan to upgrade to 800V pure electric version or accelerate AIHER hybrid project based on strategic consensus.

#800v#ev-upgrade#hybrid-ev
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Musk Merges xAI into SpaceX as SpaceXAI

Elon Musk announced on X that xAI will end independent operations and fully merge into SpaceX. The AI company will be renamed SpaceXAI, integrating Musk's AI ambitions directly with his space venture.

#merger#elon-musk#ai-integration
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Musk: Reclaim Compute from Harmful AI Firms

After SpaceX's compute deal with Anthropic, Musk clarified on X that SpaceX and Tesla will supply resources to AI companies ensuring human benefit. He reserves the right to reclaim compute if AI engages in harmful activities.

#ai-safety#compute-deal#elon-musk
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Musk's Partner Testifies He Wanted OpenAI as Tesla Subsidiary

Shivon Zilis, Elon Musk's current partner and ex-OpenAI board member, testified that Musk hoped to make OpenAI a Tesla subsidiary. She faced questions about leaking information to Musk during her board tenure. An OpenAI executive accused her of not disclosing Musk as the father of her twins.

#court-testimony#musk-drama#ai-governance
โš›๏ธ
้‡ๅญไฝโ€ข86d ago

NVIDIA Rethinks AI TCO: Per-Token Cost Key

NVIDIA is redefining AI Total Cost of Ownership (TCO) by emphasizing per-token cost as the only vital metric. As AI inference emerges as the core workload, outputs have shifted to token-generated intelligence. This approach challenges traditional cost models for AI deployments.

#ai-tco#token-cost#inference
๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข86d ago

Korea Adds 100 Trillion-Won Firms in 9 Months

South Korea's stock market surged with 100 new trillion-KRW market cap companies in nine months, outpacing the past decade. The boom stems from AI-fueled storage chip demand explosion and capital market reforms.

#korea-market#ai-chips#hbm-demand
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Microsoft Reduces Copilot in Low-Utility Win11 Scenarios

Microsoft is decreasing Copilot's appearances in Windows 11 where it provides little value. This signals a retreat from aggressive AI integration across the Windows ecosystem. The shift was revealed via a now-deleted X post.

#ai-integration#windows-ui#strategy-shift
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Microsoft Admits Win11 Built on 90s Win32 Legacy

Windows 11 relies on Win32 API from the Windows 95 era for file operations and desktop apps. Microsoft executives confirm this legacy foundation was not part of long-term plans. It remains core to the modern OS.

#legacy-api#os-architecture#win32
๐Ÿ 
ITไน‹ๅฎถโ€ข86d ago

Robot Dogs, Drones in Police Capture Drill

Chinese armed police conducted a human-unmanned collaborative capture exercise in a complex environment simulating terrorists in an abandoned factory. Drones provided high-altitude recon, robot dogs aided perimeter control, and FPV drones executed precise indoor strikes. The drill integrates unmanned systems to enhance combat capabilities.

#robotics#drones#military
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Microsoft: 1.6B Windows, 1B Bing Monthly Users

Microsoft's earnings reveal Windows monthly active users surpassing 1.6 billion and Bing search over 1 billion. CEO Satya Nadella reaffirms focus on Windows while prioritizing engagement in Edge browser and Bing.

#user-metrics#mau-milestones#microsoft-strategy
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Brain's Frontline Aids Decision-Making

University of Illinois study shows brain's earliest sensory cortex actively shapes decisions, not just passively receives info. This challenges classic hierarchical processing models. Dynamic bidirectional interactions could inspire efficient AI architectures.

#neuroscience#decision-making#neuromorphic-ai
๐Ÿ—พ
ITmedia AI+ (ๆ—ฅๆœฌ)โ€ข86d ago

SR-01 Robot Transforms Car-to-Humanoid in 50s

Sansei Technologies showcased the SR-01 rideable humanoid transforming robot at SusHi Tech Tokyo 2026. It converts from car to humanoid form in 50 seconds. The massive machine measures 4.7m long and weighs 2.3 tons.

#humanoid-robot#transforming-robot#robotics-hardware
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข86d ago

Claude Paid Quotas Doubled, Peak Limits Removed

Anthropic has doubled the 5-hour usage quota rate for all Claude Code paid subscribers. Peak-hour rate limits are canceled for Claude Pro and Max accounts. Developer Claude Opus API call rates are substantially increased, enabled by a new compute cluster.

#quotas#rate-limits#api-boost
๐Ÿ“„
ArXiv AIโ€ข86d ago

Validating Agents with 2-10 Traces Only

A novel algorithm learns correct sequential behavior for autonomous agents from just 2-10 passing traces, eliminating manual specs or thousands of examples. It combines compiler dominator analysis with multimodal LLMs for semantic states and uses Prefix Tree Acceptors for generalized models. Achieves high bug detection accuracy across UI, code gen, and robotics with explainable results.

#autonomous-agents#validation#llm-semantics
๐Ÿ“„
ArXiv AIโ€ข86d ago

Terminus-4B Rivals Frontier LLMs in Agent Tasks

Terminus-4B is a finetuned Qwen3-4B model specialized for agentic terminal execution via SFT and RL. It matches or exceeds frontier models like Claude Sonnet/Opus and GPT-5.3-Codex on SWE-Bench Pro and C# benchmarks, reducing main agent token usage by 30% without performance loss. This enables efficient subagent isolation for verbose tasks in coding agents.

#agentic-execution#coding-agents
๐Ÿ“„
ArXiv AIโ€ข86d ago

Stop Automating Peer Review Without Evaluation

This position paper argues against using current LLMs for producing paper reviews due to critical flaws. Empirical tests on ICLR 2026 reviews reveal AI's hivemind effect reducing diversity and vulnerability to gaming via paper rewriting. It calls for a dedicated science of peer review automation instead of hasty LLM deployment.

#peer-review#hivemind-effect#review-gaming
๐Ÿ“„
ArXiv AIโ€ข86d ago

Stable Agentic Control for Cyber Defense LLMs

Introduces tool-mediated LLM architecture for SOCs configuring EDR policies under adversarial pressure, using deterministic tools and finite action catalogs. Lean 4 machine-checks a composite Lyapunov function certifying controllability, observability, and ISS robustness. Achieves 59% reduction in attacker payoff on real attack graphs with Claude Sonnet 4.

#agentic-llms#cyber-defense#formal-verification
Page 708 of 1669