All Updates

Page 502 of 1665

May 27, 2026

๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข62d ago

Google's New Models Face User Backlash

Google's integration of the new Gemini 3.5 model across its product suite has met with significant criticism from users. The article highlights the gap between executive hype and actual user experience.

#llm-performance#user-experience#google-gemini
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข62d ago

Korean organization hosts 1.44MB floppy disk game contest

A Korean organization, 2P Game Arcade, is hosting a game development contest where entries must fit on a single 1.44MB floppy disk. The event celebrates retro computing and constraints-based development.

#optimization#retro-computing#edge-ai
๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข62d ago

Tesla VP hints at potential return of Model S/X

Tesla has recently halted production of the Model S and Model X. However, company executives have indicated that a future return of these flagship models remains a possibility.

#automotive#product-strategy#tesla-news
โš›๏ธ
้‡ๅญไฝโ€ข62d ago

Fudan-linked Embodied AI Startup Raises Nearly 100M RMB

A new embodied AI startup with roots at Fudan University has secured nearly 100 million RMB in angel funding. The company focuses on developing advanced tactile sensing capabilities for robots.

#robotics#embodied-ai#tactile-sensing
๐Ÿ”ฅ
36ๆฐชโ€ข62d ago

China-Laos Railway Fruit Imports Exceed 100,000 Tons

The China-Laos Railway has transported over 107,900 tons of imported fruit this year, a 30% year-over-year increase. The efficiency of the cold chain logistics allows tropical fruits to reach the Chinese market in as little as 3 days.

#logistics#supply-chain#cold-chain
โš›๏ธ
้‡ๅญไฝโ€ข62d ago

Codex Self-Distillation: Automating Repetitive Tasks with Prompts

An OpenAI employee shared a technique using Codex self-distillation to automate repetitive coding tasks. By using specific prompts, developers can significantly reduce manual effort.

#automation#prompt-engineering#llm-optimization
๐Ÿ”ฅ
36ๆฐชโ€ข62d ago

WeChat Mini Games surpass 500 million MAU

WeChat revealed that its Mini Games platform has reached over 500 million monthly active users. The ecosystem now supports over 500,000 developers, with a significant portion being small teams.

#gaming#mobile-apps#ecosystem-growth
๐Ÿ“Š
Bloomberg Technologyโ€ข62d ago

GGL Capital Warns of AI Liquidity Overhang

GGL Capital's Gigi Luk highlights a growing risk of liquidity overhang in the AI sector. She notes that the performance gap between market leaders and laggards is widening structurally.

#market-analysis#venture-capital#ai-investment
๐Ÿ”ฅ
36ๆฐชโ€ข62d ago

European April Car Registrations Rise 7%

European new car registrations reached 1.15 million in April, a 7% increase. Electric vehicle deliveries saw a significant 38% surge.

#ev#market-data#automotive
๐Ÿ”ฅ
36ๆฐชโ€ข62d ago

Porsche Leasing Increases Capital to 1 Billion

Porsche Financial Leasing has increased its registered capital from 820 million to 1 billion RMB, a 23% increase. The company focuses on leasing and financial services.

#fintech#automotive-finance#capital-expansion
๐Ÿ“„
ArXiv AIโ€ข62d ago

ScientistOne: Autonomous Research via Chain-of-Evidence Framework

ScientistOne is an autonomous research system that utilizes a 'Chain-of-Evidence' (CoE) framework to ensure all generated claims are traceable to source material. It significantly reduces hallucination rates and improves method-code alignment compared to existing autonomous research agents.

#autonomous-agents#ai-research
๐Ÿ“„
ArXiv AIโ€ข62d ago

Rethinking Long-Term AI Agent Memory as Data Management

This research argues that current database paradigms fail to support long-term AI agent memory, leading to issues like unregulated growth and data loss. It introduces 'Governed Evolving Memory' (GEM) and a prototype called MemState to treat memory as a state-level workload rather than simple record storage.

#agent-memory#data-management#vector-database
๐Ÿ“„
ArXiv AIโ€ข62d ago

POLAR: Personalized Memory-Augmented Embodied AI Agents

POLAR is a new framework that enables embodied multimodal agents to personalize assistance by maintaining a multimodal knowledge graph of long-term user interactions. It improves task execution by retrieving semantic and episodic memories to better interpret implicit user requests.

#embodied-ai#multimodal-memory#knowledge-graph
๐Ÿ“„
ArXiv AIโ€ข62d ago

OmniToM: Benchmarking Theory of Mind in LLMs

OmniToM is a new benchmark designed to evaluate Theory of Mind in LLMs by requiring explicit modeling of belief structures rather than just end-point answers. It reveals that current models struggle with the complex knowledge-access and representational tasks required to track evolving mental states.

#theory-of-mind#llm-evaluation#social-reasoning
๐Ÿ“„
ArXiv AIโ€ข62d ago

New Agentic AI Frameworks for Scientific Workflows

This research introduces two agentic AI frameworks, DeepTS and DeepScribe, designed to automate scientific data curation and complex lecture analysis. By utilizing a hybrid Local Body/Remote Brain architecture, these systems enhance reasoning capabilities for high-energy physics and time-series research.

#agentic-ai#scientific-computing#rag
๐Ÿ“„
ArXiv AIโ€ข62d ago

MPMMine: A New Benchmark for Constraint Acquisition Algorithms

MPMMine is a new benchmark suite designed to address the lack of standardized evaluation tools for Constraint Acquisition (CA). It provides structured, open-format data including models, instances, and natural-language descriptions to improve the reproducibility of CA research.

#benchmarking
๐Ÿ“„
ArXiv AIโ€ข62d ago

LLM Introspection: Reality Check on Metacognitive Claims

This research challenges the prevailing belief that LLMs possess genuine metacognitive monitoring capabilities. The study demonstrates that current model performance in self-reporting internal states is likely driven by pattern matching rather than true introspection.

#metacognition#ai-safety#model-evaluation
๐Ÿ“„
ArXiv AIโ€ข62d ago

JobBench: A New Benchmark for Human-Centric AI Agents

JobBench is a new benchmark designed to evaluate AI agents on high-priority professional workflows rather than just economic replacement value. It tests models across 130 tasks in 35 occupations, highlighting the gap between current performance and real-world professional utility.

#benchmarking#agentic-workflow#professional-ai
๐Ÿ—พ
ITmedia AI+ (ๆ—ฅๆœฌ)โ€ข62d ago

GMO Group reports 98% AI adoption rate among employees

GMO Internet Group released survey results showing a 97.8% AI adoption rate across the organization. The report highlights a significant surge in the use of AI agents and identifies five key traits of effective AI users.

#enterprise-adoption#ai-agents
๐Ÿ“„
ArXiv AIโ€ข62d ago

BrickAnything: Geometry-Conditioned Buildable Brick Structure Generation

BrickAnything is an autoregressive framework that generates physically buildable brick structures from 3D shapes. It utilizes structure-aware tree tokenization to ensure structural stability and geometric fidelity during the assembly process.

#3d-generation#generative-ai#robotics
Page 502 of 1665