All Updates
Page 864 of 1663
April 23, 2026
Qwen3.6-27B Nails Svelte 5 Coding After OpenAI Fails
User switched to Qwen3.6-27B in opencode after OpenAI errors on Svelte 5 task. Delivered perfect result on first try despite longer local inference time.
Copilot Keyboard IME Formal Release with Slang Conversion
Microsoft released the formal version of Copilot Keyboard, a new IME that converts net slang in one shot. It revives features like 'イルカのカイル' for fun, efficient typing.
Bain Shops $5B Data Centre Stake
Bain Capital seeks buyer for stake in Singapore-based Bridge Data Centres, valued at $5B. Features nine centres in Malaysia, Thailand, India; ByteDance as anchor tenant. Citi/JPMorgan manage process; talks preliminary.
Pumen Tech Profits Halve on High R&D Costs
Pumen Technology's full-year net profit was halved due to persistently high R&D and sales expenses. The company is 'burning money' to weather challenges. Q4 profits saw a sharp 'knee-cap' decline.
Goldman: War Hurts Europe Amid AI Buildout
Goldman Sachs strategist Sharon Bell covers corporate earnings, AI buildout, and Middle East war effects on Europe. Prolonged conflict could exacerbate economic pressures. AI infrastructure expansion remains a key focus.
Kioxia Tops Japan on NAND Price Rally
Kioxia Holdings Corp. entered Japan's top 10 companies by market capitalization for the first time. Rising prices for its core NAND flash memory product drove a rally in its share price.
MLP Maps 8 Inputs to 58 Body Params via Physics Loss
A small MLP predicts 58 Anny body-shape parameters from 8 questionnaire inputs with high accuracy by embedding the body model's forward pass directly into the training loss. This physics-aware loss enforces exact matches for height, mass, and measurements derived from mesh volume and geometry. It outperforms baselines like linear regression by 10x on mass prediction.
Tim Cook's Legacy: AI Stumble
Tim Cook steps down as Apple CEO after operational successes but criticized for missing the AI revolution. Apple lagged behind ChatGPT, launching Apple Intelligence in 2024 with privacy-focused features. New CEO John Ternus must accelerate AI catch-up against Google, Microsoft, and OpenAI.
Portable AI GPU Workloads Across Providers
Discussion on defining AI GPU workloads portable across providers without provider-specific configs for availability and cost shifting. Critiques K8s multi-node pools, Terraform, and custom layers for hidden complexities. Advocates workload needs definition separated from infra binding via scheduling layer.
Nature Cover: Robot Beats Pro Ping-Pong Pros
A robot has earned a spot on Nature's cover by defeating human professional table tennis players. It excels at handling challenging spin balls.
Marvell Acquires Polariton for Plasmon Optics
Marvell announced the acquisition of Swiss firm Polariton, developers of plasmon-based high-speed, low-power silicon photonics PIC devices. This strengthens Marvell's optical interconnect I/O portfolio for next-gen 3.2Tbps platforms. The technology surpasses traditional silicon photonics in bandwidth, efficiency, and size.
30K Samsung Rally Demands AI Profit Shares
30,000 people rallied outside Samsung’s main chip hub demanding employees get a larger share of AI boom profits. Samsung Electronics is profiting significantly from the AI surge. This highlights labor tensions in AI-driven growth.
Tencent Launches HY3-Preview Flagship AI Model
Tencent has unveiled HY3-Preview, its first flagship AI model since former OpenAI researcher Yao Shunyu joined to lead foundational AI efforts. The closed-source model, with 295 billion parameters, is the company's most powerful yet, matching leading Chinese models but trailing top US ones from OpenAI and Google DeepMind.
Edge AI Slit-Lamp Detects Cataracts Early
Tohoku University developed a compact edge AI system for slit-lamp anterior eye segment exams. It enables early detection of cataracts and other issues without internet connectivity, suiting remote or disaster areas. The device is lower cost than traditional machines.
Why No Consumer LLM Inference Chips Yet?
Post questions lack of affordable plug-and-play consumer chips for local LLM inference like $200 Llama 3 stick at 30W. Notes Taalas targets datacenters only. Suspects industry prefers API subscriptions over one-time hardware sales.
Cheaper LLMs Excel in OCR Benchmarks
Benchmarked 18 LLMs on OCR using 7,560 calls across 42 documents, revealing cheaper and older models often match premium accuracy at lower costs. Released open-source dataset, framework, leaderboard, and free testing tool. Tracks reliability, cost-per-success, latency, and field accuracy.
CBA Cuts 120 Jobs in AI Push
Australia’s largest bank, Commonwealth Bank, will cut around 120 roles. This is part of a broader initiative to integrate artificial intelligence. The move aims to enhance efficiency at the nation’s top lender.
PixVerse Named UN 2026 AI for Good AI Partner
PixVerse has been selected as the official AI partner for the United Nations' 2026 AI for Good Global Summit. The company is simultaneously launching a global call for AI video submissions from creators worldwide.
Nokia Earnings Beat on AI Pivot
Nokia Oyj reported first-quarter adjusted profit exceeding analyst forecasts. Success driven by strategic push into artificial intelligence and cloud software. Data center pivot is showing promising early results.
Taiwan Banks Launch Local Finance LLM
Taiwan initiates project for large language model tailored to finance sector. Aims to empower local firms against global AI platforms. Addresses gaps in domestic regulations and market nuances.