All Updates
Page 261 of 939
April 12, 2026
Tesla Launches Limited Plaid Signature Editions
Tesla is releasing invitation-only signature editions of Model S and Model X Plaid as the final batch before discontinuation. Total production is 350 units: 250 Model S Plaid and 100 Model X Plaid (6-seater only). Priced at nearly 1.09 million RMB.
Valve Linux Patch Boosts 8GB GPUs
Valve Linux graphics engineer Natalie Vock proposed a VRAM management scheme for Linux. It prioritizes VRAM for foreground games on ≤8GB GPUs, forcing background tasks to system RAM via kernel patch and tools. Gaming performance sees major gains.
$27 Adapter Fixes Half-High GPUs
User mod uses 27 USD PCIe adapter to convert RTX 4060 half-height to full-height. It also adds two M.2 SSD slots via PCIe lane bifurcation. Enables versatile compact PC builds.
UK Regulators Urgently Assess Anthropic's New AI Risks
UK financial regulators are in emergency talks with cybersecurity agencies and banks to evaluate risks from Anthropic's latest AI model. Involved parties include Bank of England, FCA, Treasury, and NCSC. They are probing potential vulnerabilities in key financial IT systems exposed by the model.
Claude Code Father's Wild Path
The 'father of Claude Code' is revealed to have come from an unconventional 'wild path' background. The piece humorously notes he spent his life trying to escape his original company.
AVs Free Couriers for High-Value Service
New Stone CEO Yu Ensun states autonomous vehicles won't replace couriers but handle transport tasks, allowing focus on user service. $600M funding targets AI R&D. NeoClaw AI enables natural language fleet control in dialects, with 1200+ vehicles deployed in Qingdao scaling to 3000 by 2026.
ByteDance Kouzi 2.5: Born Max-Level Coder
ByteDance launches Kouzi 2.5, an AI coding tool that's 'max-level' from the start. It enables vibe-based coding through mobile conversations. Features include built-in cloud storage, email, and automatic output archiving.
HTML-in-Canvas Transforms AI-Era Web Visuals
HTML-in-Canvas technique is revolutionizing frontend development with unprecedented visual effects in the AI era. It enables running intensive apps like the Doom game directly in HTML. This shifts internet visuals dramatically.
Apple AI Glasses Rival Meta with Styles, Cameras
Apple is developing AI glasses to compete with Meta's smart glasses, featuring multiple styles and oval cameras. This positions Apple in the growing AI wearables market. The article also mentions updates on the foldable iPhone.
CASIA AI Firm's 139% Repurchase Path to HK IPO
An AI company from China's Institute of Automation (CASIA) is pursuing a Hong Kong IPO after eight years from lab origins. It reports a 139% customer repurchase rate and brands itself as the first 'decision intelligence stock'. This milestone underscores the commercialization of academy-spun AI tech.
AI Journaling Like a Best Friend
Author experiments with AI journaling apps Rosebud and Mindsera over two months. Mindsera's minimalist design suits writers, acting as a talking diary with advice on thoughts, fears, and plans. It enhances meditation-like reflection on random ideas.
Nintendo Switch 2 Mario Bundle Launches $500
Nintendo launched a Switch 2 bundle with Super Mario Galaxy 1+2 for $500 to tie into the movie's strong box office. It saves $20 amid DRAM price surges and Switch 2 production cut rumors. The promo signals potential future price hikes for hardware and games.
AirPods Pro 3 Drops to 1358 Yuan on Tmall
AirPods Pro 3, launched last month at 1899 Yuan, now 1357.73 Yuan on Tmall via subsidies and coupons. Key AI features include Apple Intelligence real-time translation and heart rate AI tracking. ANC doubles Pro 2's performance with new foam tips and 8-10 hour battery.
Gemma 4 31B SpecDec +29% Speedup
Speculative decoding with Gemma 4 E2B draft model boosts Gemma 4 31B inference by 29% on average, reaching 50% on code generation. Key issue was fixed GGUF metadata mismatch causing token translation overhead. Achieves high acceptance rates on structured tasks like math and code.
AI Code Wars Heat Up
AI code generation emerged as an early killer app, predating ChatGPT hype. Microsoft and OpenAI launched GitHub Copilot in spring 2021 to autocomplete code snippets. The newsletter highlights the booming competition in AI coding tools.
Narwal CEO Urged to Hustle Like Rivals
Opinion piece urges Narwal founder Zhang Junbin to emulate the relentless tinkering of rivals Chang Jing and Yu Hao. Without 'crazy obsession,' success is impossible in competitive robotics. Emphasizes madness as key to survival.
AI: Greatest Art Heist in History?
The article portrays generative AI as plundering the art world via unauthorized reproductions. It criticizes AI outputs as 'slop', highlights environmental costs like water usage in data centers, and notes societal harms such as inducing delusions. Artists had long predicted these dystopian outcomes.
Open Weights Models Spotlight in Enterprise AI Gap
企業需求與前沿AI間的鴻溝日益擴大，將開放權重模型推上舞台。Google、Microsoft、Alibaba及Nvidia等公司春季推出新一波模型。企業偏好廉價、可靠且不竊取專有資料的模型。
China's Robots Conquer Walls with Embodied AI
China deploys first embodied intelligent special robots for high-risk tasks in chemical tanks, ships, using magnetic wall-climbing and dexterous 15-DOF arms for welding and inspection. Backed by largest special robot large model with 100k+ hours data. Sea cable and grain silo robots boost efficiency 10x and 3x via advanced sensors and swarm coordination.
MiniMax M2.7 95% MMLU on Mac Quant
Quantized MiniMax M2.7 versions for Mac achieve 88% (63GB) and 95% (89GB) on 200q MMLU benchmark. Runs efficiently on M5 Max at ~50 tokens/s and 400pp. Positions it as 'Sonnet 4.5 at home' level.