All Updates
Page 1144 of 1512
March 17, 2026
Huawei Launches AI Inference Infrastructure
Huawei released new AI data infrastructure for inference scenarios, including the central AI Data Platform and edge FusionCube A1000 hyperconverged machine. It unifies knowledge base, KV Cache, and memory management to boost Agent accuracy by 30%. Deployment time shortens 80% with 30% higher compute utilization.
macOS 26.4 Revives Tabs, Ends Rosetta
macOS 26.4 brings back compact Safari tabs and native battery limits. Rosetta 2 is nearing end-of-life in this targeted small update.
Japanese humanoid robot video baffles devs
Highlanders startup released a video of their domestic humanoid robot prototype walking, which unexpectedly drew massive attention. CEO Masaya Masuoka posted on X that the development team is perplexed by the reactions. The specific reason for the buzz remains intriguing.
Best Qwen3.5-27B Quant for Coding?
Reddit user asks for the best Qwen3.5-27B quantized models (Q4-Q5, 20-24GB) optimized for coding tasks. Mentions candidates like Unsloth, bartowski, and mradermacher. Seeks advice on proper comparison methods.
T5 Seeks $2B for AI Data Centers
T5 Data Centers plans to raise $2 billion in equity to develop additional facilities. This initiative reflects the enormous capital inflows into AI infrastructure construction.
Compute-Power Electricity Synergy Saves Billions
Integrating data centers with green power cuts electricity costs by 45-66% for compute firms, boosts generator revenues, and resolves industry mismatches.
Llama.cpp Updates May Weaken Qwen Coding Abilities
Users report Qwen 3.5 and Qwen 3 Coder Next performing worse as coding assistants, ignoring instructions despite prompt changes and higher quants. Suspected cause: LM Studio's auto-updates to llama.cpp over the past week.
H Company Launches Holotron-12B Agent Model
H Company released Holotron-12B, an open-source multimodal model developed with NVIDIA for computer-use agents. It matches Holo2/Qwen performance but offers 2x higher throughput. Links to Hugging Face, technical deep dive, and X post provided.
Microsoft fixes Samsung C drive lockout bug
Faulty Samsung app locked Windows 11 users out of C drive. Microsoft removed the app and shared recovery steps to restore system access. Fixes now available for affected PCs.
LLMs Compress Unevenly: Gemma Best
Researchers compressed MLP layers in 6 LLMs without quantization, revealing surprising degradation differences. Gemma 2B retains 92% accuracy at 14% compression, while Llama 3.1 8B drops to 85%. Standard HF checkpoints available for vLLM, TGI, and llama.cpp.
Siemens & Rittal Team for DC Power
Siemens and Rittal formed a strategic partnership to co-develop future products for efficient power distribution in data centers. Solutions emphasize sustainability. Announced via Caijing.
OpenHarmony Powers China's First Smart Highway
Shenzhen Expressway launched China's first OpenHarmony + RISC-V smart highway on Machong-Shenzhen Airport segment. Features KaihongOS for full-intelligence with AI agents enabling 80% auto-execution. Emphasizes minimal staffing via robots/drones and wireless deployment.
ChatGPT Adult Mode: Spicy Text, No Graphics
OpenAI clarifies its adult mode for ChatGPT will allow erotic text conversations. It bans explicit images, voice clones, or video content amid technical delays and safety concerns.
minRLM: 3.6x Token Savings on GPT-5-mini
minRLM is a token and latency-efficient Recursive Language Model implementation, achieving 72.7% on benchmarks with GPT-5-mini using 3.6x fewer tokens than official baselines. It outperforms vanilla LLMs by +30pp on GPT-5.2, winning 11 of 12 tasks. Open-source code runs secure Python steps in Docker containers, with easy uvx trials.
Korea Dominates AI HBM Memory Boom
Korean chipmakers SK Hynix and Samsung reversed fortunes via HBM bets for AI GPUs, capturing 80% market and record $25B Feb exports. SK Hynix sold out 2026 HBM capacity after pioneering HBM3E mass production. Deep Nvidia collaboration via on-site engineers secured demand.
BYD Geely Adopt NVIDIA DRIVE for L4 AV
BYD and Geely develop Level 4 AV programs using NVIDIA DRIVE Hyperion platform. It integrates compute, sensors, networking, and safety for faster validation and deployment. Nissan also builds L4 program on it.
Cathie Wood: AI Productivity Up 2.8%, Set to Surge
Ark Invest CEO Cathie Wood discusses AI-driven productivity gains impacting the global economy. She notes nonfarm productivity is up 2.8% year-over-year, above trend. Wood expects these gains to rise much higher.
Sears Exposes AI Chatbot Conversations Publicly
Sears exposed customer phone calls and text chats with its AI chatbot to anyone on the web. These conversations often contain contact information and personal details. This vulnerability enables scammers to launch phishing attacks and commit fraud.
OpenAI Japan Launches Teen Safety Blueprint
OpenAI Japan announced the Japan Teen Safety Blueprint to prioritize teen safety. It introduces stronger age protections, parental controls, and well-being safeguards for teens using generative AI.
Li Auto Slumps as Range Extender Peaks
Li Auto faces sales drop and profit pressure in 2025, with Q4 revenue down 36% due to price cuts and model shifts. Extended-range EVs lose edge to rivals; pure EV struggles amid internal competition. Company pivots to store reforms, AI R&D, and humanoid ambitions.