All Updates
Page 866 of 1663
April 23, 2026
AI to Learn 2.0 Governance for Opaque AI
AI to Learn 2.0 is a new governance framework for AI-assisted work in learning-intensive domains, emphasizing deliverables that remain usable without original AI models. It features a maturity rubric, capability-evidence ladder, and gates to ensure auditable, transferable outputs with human-attributable evidence. Examples demonstrate separation of polished AI substitution from accountable workflows.
Qwen Code v0.15.1 Fixes Streaming & CLI
Qwen Code released v0.15.1 with core fixes for StreamingToolCallParser scoping per stream, CLI improvements combining elapsed time and timeout indicators, and stopping slash completion render loops. Version bumped to 0.15.1. Full changelog from v0.15.0 available.
Tesla Delays Advanced Driver-Assist in China
Tesla has delayed rollout of its most advanced driver-assistance features in China. This reflects regulators' caution toward rapidly evolving autonomous tech in the world's largest auto market.
Albatross Unveils Real-Time AI Rec Model
Swiss ICT firm Albatross has developed an AI model for real-time product recommendations to boost personalized shopping experiences. It is the first sequence embedding model trained on real-time interactions.
AI Phones Need New Stage Now
AI smartphones should advance to a new phase. Phone makers have opportunities to deeply engage in rebuilding AI interaction systems and leverage hardware for direct entry advantages.
AI Manga Chat CharChat Booms with Gen Z
CharChat is an AI chat app enabling romantic conversations with manga characters, surging in popularity among Gen Z users. Flirty interactions like characters complimenting users' cuteness drive high engagement. The app has boosted original manga sales by 6 times.
AI Enters Filmmaking: Humans Obsolete?
This opinion piece questions whether human filmmakers are still needed as AI disrupts the movie industry. It describes the sector as an AI-rewritten 'card table' where some industry players are panicking while others are strategically entering the game.
Xiaomi MiMo-V2.5 Public Beta Launch
Xiaomi has launched the public beta for its most advanced AI model series, MiMo-V2.5. The flagship model rivals top global AI systems in handling complex tasks, marking a significant step in Xiaomi's AI development.
Guardd: Isolation Forest Linux Anomaly Detection
Guardd is an open-source Linux host-based anomaly detection system using Isolation Forest on eBPF-collected exec and network events. Events are grouped into 60-second windows and converted to feature vectors for unsupervised scoring. The project addresses false positives and seeks ML feedback on features and training.
AI Boom Drives Chinese Optical Stock Rush
Chinese optical stocks are seeing surging investor interest due to the AI boom. Demand for optical components vital to AI infrastructure is expected to fuel sector outperformance. This positions them as a top trade opportunity.
Alibaba Opens Qwen to Airline Partners
Alibaba has launched its first external partnership for the Qwen AI app with China Eastern Airlines. Users can now complete full flight bookings—from search and purchase to seat selection and check-in—via a single natural-language chat interface. This expands Qwen's agentic AI into real-world services beyond Alibaba's ecosystem.
Qwen 3.6 27B: Ultra-Eager Autonomous Coding Agent
User praises Qwen 3.6 27B (noted as 35B variant) for relentless autonomous building, testing, and fixing in code refactoring. Model simulates eagerness and fear, continuing tasks unprompted. Impressive self-initiated fixes beyond user expectations.
Meta Tracks Employees to Train Replacement AI
Meta tracks detailed employee activity during daily work to generate training data for AI. This data teaches AI systems to handle routine workplace tasks. The approach could automate jobs without explicit labeling.
Arc Pro B70 Trails RTX 3090 in llama.cpp Benchmarks
Benchmarks compare Nvidia RTX 3090 and Intel Arc Pro B70 using llama.cpp on Vulkan and SYCL backends across multiple GGUF models. Arc Pro B70 shows 71% slower prompt processing and 53% slower token generation on average versus RTX 3090. SYCL backend improves some Arc results over Vulkan.
Ant Group's Efficient Ling-2.6-Flash Model
Ant Group released Ling-2.6-Flash, a 104B-parameter AI model optimized for token efficiency and cost savings. The model is quickly gaining popularity due to its massive token usage. It represents a push towards more economical large-scale AI deployments.
Second Sanquan BCI Trial Restores Paralyzed Artist's Painting
NAOTIGER Tech's second 'Sanquan' (fully implantable, wireless, full-function) BCI clinical trial succeeds, enabling a 29-year-old paralyzed art teacher to eat and paint independently. The system achieves 50ms ultra-low latency and integrates BCI-driven functional electrical stimulation (BCI-FES). This marks China's BCI entering practical, replicable clinical phase.
Microsoft 2026 Win11 Overhaul Targets 18 User Demands
Microsoft pledges 2026 Windows 11 refactor focusing on stability, addressing 18 key user feedbacks from Insiders. Changes include movable taskbar, resizable Start menu, fewer reboots, and reduced bloat. Aims for 'most stable OS' with performance and UI revamps.
Qwen Adds China Eastern Flight Booking
Alibaba's flagship Qwen AI app now allows direct booking of China Eastern Airlines flights. This is the first time its agentic AI technology is shared with a major commercial partner.
Torens 80% Revenue Tied to NAURA, AMEC
Torens shows extreme customer concentration with top five clients over 90% of revenue. NAURA and AMEC alone contribute more than 80% of sales. The actual controller's brother-in-law reportedly cashed out 645 million yuan ahead of IPO.
Tesla Integrates Doubao AI in Cars
Tesla China is integrating ByteDance’s Doubao AI model into its in-car system. The integration aims to power voice commands and significantly enhance the overall user experience.