Every story we have kept, newest first.
Looking for the daily editions? → Past editions
Page 71 of 1372
Zhipu has claimed the Niulai model, with the article presenting a hands-on test described as friendly to ordinary users. The piece appears to combine model ownership news with an initial practical evaluation.
InfoQ中国 · 20d ago
A mature SaaS company integrated Agents across its research and development workflow, reportedly producing 64 pull requests in one and a half weeks and completing 30% of a project refactor. The article examines the results and implications of embedding Agents into day-to-day software engineering.
InfoQ中国 · 20d ago
This article focuses on the operational challenges that emerge after an Agent enters production. It provides a technical-practice perspective on how teams should continue improving and managing deployed Agents.
InfoQ中国 · 20d ago
Meta’s US settlement could encourage governments in other countries to seek similar concessions. Separate legal actions involving the company remain pending in Kenya, the Netherlands, and elsewhere, including a case alleging Facebook’s algorithm promoted posts targeting an Ethiopian academic.
The Guardian Technology · 20d ago
This research identifies an Accuracy-Efficiency Paradox in on-device energy forecasting: highly accurate models can consume more energy than they save. It proposes a Total Cost of Ownership framework that combines inference energy use with battery aging to measure net energy loss.
ArXiv AI · 20d ago
Researchers introduce the Relational Hypergraph Transformer (RHT), an architecture for learning from complex relational databases with high-dimensional, categorical, and temporal data. On the Synthea healthcare benchmark, RHT delivered the strongest embedding semantic coherence, while XGBoost achieved the highest rare-code recall.
ArXiv AI · 20d ago
PICasso is an AI-assisted framework that converts natural-language specifications into verified, optimized photonic integrated circuit layouts. Its benchmark results show improved specification satisfaction and a 1.74 dB reduction in mean insertion loss through simulation-guided optimization.
ArXiv AI · 20d ago
A study evaluated 20 LLM-generated literature reviews using short and long context windows across 15 dimensions. Longer contexts improved breadth and coherence but also increased repetition, omissions, and descriptive writing, reinforcing the need for expert oversight.
ArXiv AI · 20d ago
Researchers developed an LLM pipeline to extract model-relevant information from 536 peer-reviewed agent-based disease-modeling papers. GPT-4.1 achieved 77.95% paper-level accuracy, while GPT-5.0 reached 81.67%, though complex fields remained less reliable.
ArXiv AI · 20d ago
This review examines how Transformer-based large models and self-supervised learning can address data scarcity, poor cross-domain generalization, limited interpretability, and automation challenges in battery prognostics and health management. It proposes a roadmap spanning collaborative data ecosystems, industrial intelligence validation, physics-informed trustworthiness, and efficient on-device deployment.
ArXiv AI · 20d ago
This research proposes a CRM integration framework that combines churn-risk predictions with SHAP and LIME explanations for targeted retention actions. On the IBM Telco Customer Churn dataset, Logistic Regression achieved the best AUC-ROC at 0.8411, while LightGBM reached the highest accuracy at 78.42%.
ArXiv AI · 20d ago
EEG-to-Report is a browser-based framework that combines EEG review, multimodal annotations, automated feature extraction, and structured JSON storage for training EEG-language models. It also uses convolutional networks and a large language model to generate editable draft clinical reports for neurologist review.
ArXiv AI · 20d ago
EduRiskX combines a temporal Transformer with F-Logic reasoning to detect at-risk students earlier while providing rule-based explanations. On the OULAD dataset, it achieved 0.900 accuracy, a 0.894 F1-score, and detected risk at an average of Week 9.32.
ArXiv AI · 20d ago
CIFQA is a deterministic, tool-grounded multi-agent framework designed to improve accuracy on calculation-intensive financial questions. In fixed-deposit testing, it achieved 95.54% accuracy on calculation-heavy queries and 90.87% overall, outperforming larger models with the same financial information.
ArXiv AI · 20d ago
Researchers introduce CARL, a closed-loop autotelic reinforcement learning system that discovers and controls self-organizing patterns in the Lenia cellular automaton. CARL finds stable solitons, steers their movement with minimal local interventions, and enables humans to guide them through mazes using high-level commands.
ArXiv AI · 20d ago
This feasibility study compares a standalone LLM with a four-step agentic pipeline for explaining ICU mortality predictions. The agentic approach eliminated explicit outcome leakage and improved guideline grounding and value specificity, while standalone LLM explanations aligned better with SHAP attributions.
ArXiv AI · 20d ago
Venture capital firm Lightspeed expects a homegrown Indian AI model with global competitiveness to be released by next year. The report does not disclose the model’s developer, architecture, capabilities, or release format.
Bloomberg Technology · 20d ago
高德發布 ABot-Recon，這是一個無長程依賴的萬幀級流式 3D 重建模型。模型可使用 12 幀重建萬幀級 3D 場景，瞄準長序列場景理解與重建需求。
量子位 · 20d ago
China's draft revision to the Road Traffic Safety Law has entered initial review by the Standing Committee of the 14th National People's Congress. The proposal addresses the long-standing lack of top-level legal rules for autonomous driving and may assign violation penalties to automakers involved in misleading promotion.
cnBeta (Full RSS) · 20d ago
Glassdoor research suggests employees aged 40 through their 60s are more optimistic about AI opportunities than younger workers. The study indicates that experience may help senior employees use large language models more effectively, while AI creates additional opportunities rather than simply replacing them.
cnBeta (Full RSS) · 20d ago
Chinese surgeons reportedly completed the world’s first commercial surgery to implant an invasive brain-computer interface in a patient with spinal cord injury. PICC Property and Casualty has since launched insurance covering BCI implantation surgery, signaling broader commercial and institutional support for the technology.
SCMP Technology · 20d ago
一苇宇航 is developing space-based data center infrastructure and has completed its fourth financing round within a year. Its near-term strategy is to commercialize modular satellite intelligence, communications, control components, and an in-orbit computing platform before building a full orbital data center network.
虎嗅 · 20d ago
Cloudflare is reportedly transforming its engineering standards into a governance system enforced by AI. The update highlights a shift from documenting development rules to operationalizing them through automated controls.
InfoQ中国 · 20d ago
NetEase Youdao has officially launched OpenPods, described as the world's first Agent earphones designed specifically for iPhone users. The product aims to turn voice interaction and audio into a productivity tool.
量子位 · 20d ago
Yotta Data Services is discussing a public offering very soon to raise capital for rapidly growing AI service demand. The move highlights how AI infrastructure providers are seeking funding to expand capacity.
Bloomberg Technology · 20d ago
Lightspeed India partner Hemant Mohapatra discusses investment opportunities across India’s AI startup ecosystem. The conversation offers an investor perspective on the sectors and companies emerging in the market.
Bloomberg Technology · 20d ago
The article examines whether NVIDIA’s rapidly growing compute demand will translate into realized revenue. Its upcoming earnings report is positioned as a key checkpoint for validating AI infrastructure returns.
钛媒体 · 20d ago
OpenAI and Thailand’s MHESI are launching an eight-week accelerator for 10 startups in health, wellness, and education. The program aims to help teams transform AI prototypes into trusted products.
OpenAI News · 20d ago
A federal judge in California ruled that the Trump administration illegally retaliated against Anthropic by blacklisting the AI startup. The ruling said the government’s action targeted constitutionally protected expressive activities.
New York Times Technology · 20d ago
Lingchu Intelligence has completed a new funding round worth more than $100 million. The article highlights the company’s rapid growth into a high-valued unicorn backed by a prominent Apple supply-chain investor.
钛媒体 · 20d ago