All Updates
Page 1771 of 1946
March 2, 2026
Qwen3.5-9B Tops Coding Benchmarks
Reddit post debates if Qwen3.5-9B suffices for agentic coding. It beats Qwen3-30B-A3B on all coding items and rivals Qwen3-Next-80B/GPT-OSS-20B on many. Targets 8GB VRAM laptops with Q8 quant and long contexts.
Xiaoyunque Eyes Higgsfield-Like AI Video Dominance
ByteDance's Xiaoyunque AI video tool lowers creation barriers with templates, plugins, and explainer replication, riding Seedance 2.0 wave. Targets casual creators for brainwave videos on Douyin. Compares to Higgsfield's success in simplifying pro video production.
Silicon Recycling Scales to Asia Pacific
Breakthrough recycling technology for high-value silicon is scaling from a German pilot to Asia Pacific operations. The global market for recycled solar panel materials is projected to reach US$1.12-2.7 billion by 2030, driven by end-of-life panels. Silicon is strategically vital for solar, EV batteries, and semiconductors.
Qualcomm X105 5G Modem Hits 14.8Gbps
Qualcomm released the X105 5G modem and RF system, the global leading 5G Advanced platform. It is the industry's first modem ready for 3GPP Release 19, laying the foundation for 6G development and testing.
LM Studio Parser Bugs Break Qwen3.5 Tools
LM Studio's parser silently disrupts Qwen3.5 tool calling and reasoning by scanning <think> blocks for tool patterns, causing infinite loops. Registering a second MCP server also breaks parsing for the first. Bugs reported over a year ago remain open, with workarounds like disabling reasoning.
Speaklone: On-Device Qwen3-TTS for iOS/macOS
Solo developer launches Speaklone app enabling on-device Qwen3-TTS inference (1.7B on macOS, 0.6B on iOS) via MLX-Swift, with voice cloning from 5-30s samples and natural language voice design. It supports streaming TTS without cloud dependency, overcoming memory constraints through quantization, cache limits, and chunked decoding. Now available on the App Store.
Iowa Adopts Strictest US Data Center Zoning
An Iowa county has implemented some of the strictest zoning rules for data centers in the US. Local residents, however, continue to express concerns that these measures fall short of addressing community impacts.
Claude Imports Rival Chatbot Memories
Anthropic launches memory import tool extracting context from ChatGPT, Gemini, or Copilot into prompts for Claude. Assimilation takes 24 hours; view via 'See what Claude learned' and edit in settings. Focuses on work topics, ignoring personal details.
Samsung Wallet Adds Aliro Smart Lock Support
Samsung announced Aliro standard support for Samsung Wallet at MWC 2026, launching late March in select regions. Compatible with Aqara, Nuki, Schlage, Xthings locks using NFC or UWB unlocking. Digital keys protected by Knox chip meeting EAL6+ security.
iFlytek 40g AI Glasses with Multimodal Translation
iFlytek debuted AI Glasses at MWC 2026, designed for face-to-face talks with real-time translation subtitles on lenses and audio playback. Weighing just 40g, it pioneers lip-reading multimodal noise reduction for noisy settings, boosting accuracy over 50%. Pre-orders open March 4 on Chinese e-commerce.
Toyota Suzuki Adopt Chinese Auto AI Chips
Toyota and Suzuki will use Horizon Robotics' SoC chips in vehicles sold outside China, cheaper by 30% with proven reliability. Horizon, founded by ex-Baidu expert, powers 27 carmakers and 285 models including BMW and BYD. Move counters Chinese EV export competition in key markets.
14.ai Replaces Startup Customer Support
A married founder duo's company, 14.ai, is automating and replacing customer support teams at startups. They also launched a consumer brand to test AI's limits in handling support tasks.
NIST Tightens Foreign Lab Access Rules
NIST implemented entry restrictions at Boulder and Gaithersburg sites, prohibiting hundreds of foreign researchers from solo lab access evenings and weekends without federal escorts. Researchers from certain countries face potential full access revocation by late next month. New rules cap international scholars at NIST to 3 years maximum.
Glass Paper Bottlenecks AI World
A thin glass paper thinner than a hair strand is the unexpected bottleneck stalling major AI models like GPT-5, Gemini Ultra, and Claude. While AI infrastructure drives up prices for chips, memory, and cables, this material holds the fate of next-gen AI. The industry is scrambling amid global AI buildout.
Qwen3.5 vs Qwen3 Benchmarks Visualized
Reddit user averaged official Qwen3.5 and Qwen3 benchmark scores into visualizations. New models (purple/blue) generally outperform older ones (orange/yellow). Raw data available in Google Sheet; some small model data missing.
Qwen 3.5 2B Runs on Android
ChatterUI v0.8.9-beta9 app enables Qwen 3.5 2B on Android, tested on Poco F5 with Snapdragon 7 Gen 2. User reports slower inference due to architecture but solid general knowledge at low context. Experimental pre-release noted.
AWS $4B Datacenter Divides Ohio Town
A Wilmington, Ohio resident sued local officials over governance issues amid plans for a $4bn AWS datacenter on 500 acres. AWS seeks a 30-year property tax abatement in exchange for funding schools and infrastructure. The project highlights tensions between economic growth and community concerns in rural America.
Zillow CEO on AI Disrupting Real Estate Data
Zillow CEO Jeremy Wacksman discusses company growth amid housing crisis, positioning Zillow as essential vertically integrated real estate platform. He addresses AI agents potentially bypassing Zillow by directly accessing MLS databases. Conversation covers database politics, realtor lawsuits, and vision for agent-focused platform.
Chrome Gemini Spy Vulnerability Exposed
High-severity bug in Chrome's Gemini AI lets malicious extensions spy on PCs. Attackers can expose data or monitor activity. Update and protect advised.
Alibaba open-sources Qwen3.5 tiny models
Alibaba's Tongyi Qianwen open-sourced four small Qwen3.5 models: 0.8B, 2B, 4B, 9B. They inherit strong multimodal capabilities for edge devices to compact servers. Now the full Qwen3.5 family is open-sourced on ModelScope and Hugging Face.