Coding
Qwen3.8-Max Claims Global Frontend Coding Lead
Alibaba updated its flagship model, Qwen3.8-Max. The update is reported to place the model's frontend programming capability at the top globally.
量子位 · 14d ago
Full archive
Every story we have kept, newest first.
Looking for the daily editions? → Past editions
Page 45 of 1372
September 2, 2026
Applications
Reinforcement Learning Expert Joins Stardust Intelligence
Sun Peng, a former ByteDance reinforcement learning expert and former head of Tencent Robotics X's Agent Center, has joined Stardust Intelligence. The appointment is intended to strengthen the company's full-stack Physical AI capabilities.
量子位 · 14d ago
Infrastructure
Measuring AI Infrastructure by Intelligence Yield
Microsoft argues that the next phase of AI progress should be measured by how effectively infrastructure produces useful intelligence. Drawing on the semiconductor concept of yield, the article shifts focus from infrastructure elegance and development effort to practical output and efficiency.
Microsoft Blog (AI Tag) · 14d ago
Infrastructure
Why AI Datacentres Face a Global Backlash
Communities from Scotland to India are resisting new datacentre developments, warning that they could strain energy supplies and worsen the climate crisis. The backlash challenges the tech industry’s argument that massive facilities are essential for national AI competitiveness.
The Guardian Technology · 14d ago
SetupAIInfrastructure
Infrastructure
GPTC CEO Details AI Chip Packaging Strategy
GPTC Chairman and CEO Hoton Chang discusses the company’s business strategy and rising demand for advanced packaging in the semiconductor industry. The conversation took place during SEMICON Taiwan in Taipei and focuses on infrastructure supporting advanced chip production.
Bloomberg Technology · 14d ago
Research
World Labs Unveils Atlas for Spatial Intelligence
World Labs has introduced Atlas, an omn model that jointly handles text, images, and 3D data while preserving 3D spatial context. It can reconstruct spaces from a small number of images, generate free-camera footage and bullet-time video, and support robotics simulation.
ITmedia AI+ (日本) · 14d ago
SetupAIInfrastructure
Infrastructure
SEMI CEO Weighs Memory Prices and Chip Supercycle
SEMI President and CEO Ajit Manocha discusses the outlook for memory chip prices and the semiconductor market’s ongoing supercycle. The interview was recorded at SEMICON Taiwan in Taipei with Bloomberg Technology.
Bloomberg Technology · 14d ago
Infrastructure
AI Compute Spending Could Hit $4 Trillion
Industry executives at a Wuxi conference projected that global new data-center infrastructure investment could exceed $4 trillion by 2028. Semiconductor spending is expected to account for more than half of the total, while China showcased advanced-packaging platforms.
Pandaily · 14d ago
Audio
Meta Launches Real-Time Multilingual Speech Model
Meta announced Muse Voice Transcribe, its first real-time speech recognition model. It combines speech recognition, speaker separation, and end-of-utterance detection while supporting identification of more than 20 speakers and mixed-language audio.
ITmedia AI+ (日本) · 14d ago
Coding
Mac Apps Can Drop Intel Support
Apple now allows Mac App Store apps that require macOS 13 or later and use the universal format to drop support for Intel-based Macs. Developers can simplify their build process and reduce download and on-device storage requirements.
cnBeta (Full RSS) · 14d ago
SetupAIInfrastructure
Infrastructure
NAND Shortage Could Peak in 2027
Phison Electronics CEO KS Pua forecasts that NAND flash memory could face its most severe shortage in 2027. He also highlights NAND’s growing importance in AI infrastructure and discusses demand for controller chips used in SSDs, USB drives, and memory cards.
Bloomberg Technology · 14d ago
Business
Why Household AI Is Driving Different Group’s Growth
The article analyzes how household AI consumption is becoming a mainstream trend and what this means for 不同集團（06090.HK）. It presents AI as a way for the company to expand the boundaries of future products rather than as a standalone feature.
钛媒体 · 14d ago
Business
UK AI Strategy Architect Joins Anthropic
Matt Clifford, who helped shape the UK government’s AI strategy, has taken a senior role at Anthropic. He will reportedly continue serving as chair of a government technology funding body, prompting warnings about a potential conflict of interest.
The Guardian Technology · 14d ago
Models
Memory Is AI’s Next Competitive Moat
The interview argues that as foundation models become increasingly similar, persistent memory may become the key source of differentiation. It focuses on who controls the ownership, storage, and use of AI-generated user memories.
钛媒体 · 14d ago
Infrastructure
AI Rebuilds Cybersecurity Operations
Cybersecurity operations are shifting from manual processes toward AI-driven automation as threats grow in volume, speed, and sophistication. AI can automate repetitive, high-volume tasks while improving accuracy and efficiency across the security lifecycle.
SCMP Technology · 14d ago
Applications
UI-Venus-2 Brings GUI Agents to Real-World Apps
UI-Venus-2 is an open-source foundation GUI agent designed to operate across mobile, web, and desktop environments. Its training approach scales environment coverage, function-grounded task generation, trace-level verification, and safety-aware execution.
ArXiv AI · 14d ago
Research
Small Models Track Elder Scams Turn by Turn
Researchers introduce a cumulative risk assessment framework that updates scam probabilities as elder-fraud conversations evolve across multiple turns. Fine-tuned Phi-4 and LLaMA-3.2 deliver strong turn-aware performance while remaining suitable for mobile and resource-constrained deployment.
ArXiv AI · 14d ago
Research
SCAFFOLD Brings Structured Diagram QA to Research AI
SCAFFOLD is a large-scale dataset designed to train vision-language models to understand diagrams in computer science research papers. It provides figure images, captions, context, question-answer pairs, and step-by-step reasoning traces across multiple dataset sizes.
ArXiv AI · 14d ago
Research
Prediction Error Can Mislead Causal Inference
This arXiv study examines whether nuisance-function prediction error reliably reflects causal estimator quality in partially linear models. Simulations show that XGBoost achieved the best non-oracle RMSE, while DML-XGBoost generally delivered better confidence-interval coverage, and prediction error did not consistently track causal bias.
ArXiv AI · 14d ago
Infrastructure
OpenAgentFlow Sets System-Wide Agent Safety Boundaries
OpenAgentFlow introduces a control-plane and action-plane architecture that enforces safety at the action-commit boundary across heterogeneous AI agents. Its Android implementation reached 94.0% accuracy, blocked 95.3% of attacks, and correctly handled 27 of 30 dynamic-policy cases.
ArXiv AI · 14d ago
Research
MiNER Extracts Malaria Entities from Clinical Text
MiNER is a BioBERT-based biomedical NLP system fine-tuned to recognize clinically significant malaria-related entities in scientific and clinical texts. The researchers report improved precision, recall, and accuracy, and have released a human-labeled dataset for entity and relation extraction.
ArXiv AI · 14d ago
Research
I-CARE Makes Unlearning Interference Measurable
I-CARE is a methodology for systematically measuring and reporting interference in text-to-image model unlearning. It defines tasks, metrics, and reporting templates, and includes open-source software plus a web interface for exploring results.
ArXiv AI · 14d ago
Research
Hyperedges Strengthen Textual World Models
HyperWorld studies how state serialization affects learned textual world models, comparing raw observations, independent sentences, pairwise triples, and entity-centered hyperedges. Hyperedge serialization performs especially well for small-to-medium models and under distribution shift, while also delivering the strongest downstream greedy-planning success rate.
ArXiv AI · 14d ago
Research
gpt-oss-120b Tracks MD5 Across 196 Tool Calls
A new study tests whether LLMs can preserve exact intermediate state across long chains of dependent tool calls. At temperature 0, gpt-oss-120b completed 196 MD5-related calls and produced the correct digest in a majority of runs, with context retention and worker voting improving reliability.
ArXiv AI · 14d ago
Research
Genetic Algorithms Tackle Stochastic Lot Sizing
This paper introduces a discrete-time Markov decision process for multi-item capacitated lot sizing with deterministic quantities but stochastic demand-arrival periods. A genetic algorithm achieves an average optimality gap of about 3.44% and stays below 5% on difficult benchmark instances while providing substantial speedups.
ArXiv AI · 14d ago
Research
EULER Makes Multi-Agent Math Discovery More Reliable
EULER is a multi-agent system that searches for mathematical proofs by transferring conjectures across adjacent and distant domains. On 120 recent conjectures, it produced 10 proofs, 3 refutations, and 45 scoped partial results, while evidence checks reduced incorrect conclusions.
ArXiv AI · 14d ago
SetupAIInfrastructure
Infrastructure
AI Could Drive $31.6 Trillion in Data Center Spending
Global data center spending could reach $31.6 trillion by 2050 as organizations build infrastructure to support growing AI demand. PricewaterhouseCoopers describes the projected investment boom as unprecedented in history.
Bloomberg Technology · 14d ago
Research
A Safety Framework for Clinical AI Failures
AI Morbidity and Mortality (AI M&M) introduces a structured, blameless process for reviewing clinical AI failures and near-misses. It reconstructs how risks emerge across AI systems, clinicians, workflows, and institutional controls, turning individual incidents into corrective action.
ArXiv AI · 14d ago
Infrastructure
Why Industrial AI Still Struggles With Data
The article examines why abundant industrial data has not automatically produced successful industrial AI applications. It frames the core database challenge as translating the signals and behaviors of diverse machines into data that AI systems can understand and use.
钛媒体 · 14d ago
Infrastructure
Why AI Companies Are Designing Their Own Chips
Large model companies are increasingly preparing their own computing resources, including exploring AI-assisted design of AI chips. The article examines how this shift could reshape competition and supply dynamics across the AI industry.
钛媒体 · 14d ago