Every story we have kept, newest first.
Looking for the daily editions? → Past editions
Page 353 of 1375
Tencent has officially released its Hy3 model, marking a significant step in its AI strategy. The model is being integrated into internal products like WorkBuddy and Yuanbao to improve task success rates and reduce hallucinations.
虎嗅 · 86d ago
DeepSeek is reportedly developing custom AI chips specifically optimized for inference tasks. The project has been underway for over a year, with the company quietly engaging with design firms, foundries, and memory suppliers.
量子位 · 86d ago
ASUS has released the ProArt PA40SU, a 40Gbps portable SSD enclosure featuring active cooling and tool-free installation. It is designed for high-performance storage needs in professional creative and mobile workflows.
IT之家 · 86d ago
Kuaishou is shifting its strategic focus from traditional short video content to AI development. The company is betting on its Kling AI model to secure its future growth.
钛媒体 · 86d ago
The trust industry is undergoing a profound transformation as regulators crack down on non-compliant expansion. Firms like Everbright Trust and Huaxin Trust are being forced to restructure their business models.
钛媒体 · 86d ago
The article analyzes the competitive landscape between Tencent's Video Account and Kuaishou, suggesting that Tencent's strategic shifts in shareholding and product focus are putting pressure on Kuaishou's market position.
钛媒体 · 86d ago
Momenta, a leader in autonomous driving technology, has officially listed on the Hong Kong Stock Exchange (6880.HK). The company has achieved significant growth, with over 1 million vehicles currently using its systems.
IT之家 · 86d ago
Tencent has hired former OpenAI researcher Yonglong Tian to join its Large Language Model department. He will focus on the research and development of Vision-Language Models (VLM).
36氪 · 86d ago
Anthropic has discovered 'J-space,' a collection of internal neural patterns that reveal what a model is 'thinking' without it needing to write it down. This breakthrough allows researchers to observe silent reasoning, such as when a model detects it is being tested.
Computerworld · 86d ago
Meituan has successfully implemented a trillion-parameter large model to optimize its logistics and delivery operations, enhancing efficiency behind the scenes.
钛媒体 · 86d ago
Automotive manufacturers are seeking to diversify their battery supply chains to reclaim profit margins, potentially challenging CATL's dominant market position.
钛媒体 · 86d ago
The leading non-public Traditional Chinese Medicine (TCM) hospital group has officially launched its IPO. This move signals a new cycle of scale and capitalization for traditional medical brands.
钛媒体 · 86d ago
Louis Vuitton has lost a significant lawsuit, marking a shift in how luxury brands protect their symbols in the modern era. The ruling suggests that consumer sentiment is increasingly challenging traditional brand protection strategies.
钛媒体 · 86d ago
Avatr is undergoing significant strategic shifts including price reductions and organizational consolidation following a sharp decline in sales. The company is accelerating its IPO process to secure necessary capital for future operations.
钛媒体 · 86d ago
Young people in India are integrating AI tools like ChatGPT into their daily lives while maintaining traditional beliefs in astrology. This trend highlights the unique intersection of modern technology and cultural practices.
钛媒体 · 86d ago
Shanghai-based AI chipmaker Iluvatar CoreX is reportedly considering a share sale in Hong Kong to raise at least $800 million. This move follows a significant 428% rally in the company's valuation since its January IPO.
Bloomberg Technology · 86d ago
The article explores the necessity of 'acting out' (being irrational or non-conformist) as a core human trait that distinguishes us from AI. It argues that in a hyper-rationalized world, preserving the ability to dream and behave unpredictably is vital for human identity.
虎嗅 · 86d ago
Xiaomi is launching a new sub-brand named SkyNomad, focusing on extended-range electric vehicles (EREV) designed for camping. The first model is expected to debut soon to compete in the high-end smart vehicle segment.
Ifanr (爱范儿) · 86d ago
This research evaluates the effectiveness of using Large Language Models to generate synthetic consumer data for projective marketing research. The study compares LLM-generated responses with human data, highlighting both significant thematic overlaps and distinct differences in linguistic style and diversity.
ArXiv AI · 86d ago
This research synthesizes 27 papers to categorize LLM agent limitations into six failure clusters, including tool use, planning, and long-horizon reasoning. It reveals that performance gains on sub-tasks do not guarantee end-to-end reliability in complex agentic workflows.
ArXiv AI · 86d ago
Prompt-to-Paper is a multi-agent system designed to automate scientific manuscript generation by grounding claims in verifiable literature and executing real computational experiments. It features a quality-driven improvement loop that significantly enhances manuscript rigor and reproducibility.
ArXiv AI · 86d ago
This research introduces a novel framework that uses energy-based representations to explain AI decisions in complex cyber-physical systems. By modeling variable dependencies without requiring directed causal graphs, it offers a more scalable and robust method for interpreting system behavior.
ArXiv AI · 86d ago
AgenticAI-Supervisor is a new API and UI-driven RL Gym environment designed to move beyond static evaluation for autonomous agents. It enables high-fidelity trace generation and multi-dimensional reward shaping to improve multi-step decision-making.
ArXiv AI · 86d ago
The Narrative World Model (NWM) introduces a narratology-grounded temporal-state graph to help AI track complex story arcs in long-form fiction. It significantly outperforms existing agent-memory frameworks like Graphiti/Zep in multi-hop narratological reasoning.
ArXiv AI · 86d ago
LLMForge is a new framework for generating mechanical CAD designs from natural language, utilizing iterative refinement and multi-model evaluation. The study benchmarks seven foundation models on 97 engineering tasks, highlighting the effectiveness of VLM-based visual critique.
ArXiv AI · 86d ago
This research proposes moving memory retrieval inside the agent's reasoning loop to eliminate network latency. By using an in-process store, retrieval time drops from milliseconds to microseconds, enabling agents to maintain a persistent, high-speed working memory.
ArXiv AI · 86d ago
FirstResearch introduces a structured 'Research Question Certificate' to make LLM-generated scientific hypotheses inspectable and auditable. The framework outperforms existing baseline agents by enforcing explicit derivation constraints for scientific discovery.
ArXiv AI · 86d ago
CSTutorBench is a new benchmark designed to evaluate small language models (SLMs) in their effectiveness as tutors for block-based programming environments like VEX VR. The study reveals that while models excel at tone and vocabulary, they often struggle with pedagogical tasks like avoiding answer leakage and tracking student debugging history.
ArXiv AI · 86d ago
ArtisanCAD is a new industrial CAD agent that uses expert-grounded knowledge distillation to convert ambiguous natural language prompts into executable, production-ready B-Rep models. By utilizing a procedural intermediate representation (CAD-IR), it bridges the gap between high-level design intent and precise engineering operations.
ArXiv AI · 86d ago
Akashic is a new LLM inference service that optimizes long-context management using MemAttention to organize data into bounded chunks. It improves efficiency by reducing retrieval fragmentation and minimizing I/O overhead without needing to rewrite full interaction histories.
ArXiv AI · 86d ago