All Updates

Page 518 of 1666

May 25, 2026

๐Ÿ‡จ๐Ÿ‡ณ
cnBeta (Full RSS)โ€ข65d ago

Epic Games Announces Unreal Engine 6

Epic Games has officially unveiled Unreal Engine 6, showcasing its capabilities with a real-time demo featuring Rocket League. This marks the next generation of their industry-standard game development platform.

#game-engine#3d-rendering#real-time-graphics
๐Ÿ“Š
Bloomberg Technologyโ€ข65d ago

AI Boom Drives Record Gains in Global Momentum Stocks

The ongoing AI boom is fueling record-breaking returns for momentum investors globally. Despite macroeconomic concerns regarding growth and geopolitical tensions, AI-related equities continue to lead market performance.

#market-trends#investment#ai-economy
๐Ÿ“ฑ
Ifanr (็ˆฑ่Œƒๅ„ฟ)โ€ข65d ago

Chinese AI firm achieves 60B parameter model on mobile

A Chinese AI company has successfully overcome technical bottlenecks to run a 60-billion parameter large language model on mobile hardware. The achievement emphasizes a fully domestic ecosystem, including frameworks, chips, and methodology.

#on-device-ai#edge-computing#model-optimization
๐Ÿ“„
ArXiv AIโ€ข65d ago

SciAtlas: A Large-Scale Knowledge Graph for Scientific Research

SciAtlas is a massive, multi-disciplinary knowledge graph containing 43 million papers and 3 billion triplets. It provides a structured topological framework to help AI agents navigate complex scientific connections and reduce logical hallucinations.

#knowledge-graph#neuro-symbolic#agentic-workflow
๐Ÿ“„
ArXiv AIโ€ข65d ago

RMA: Agentic System for Research-Level Mathematical Reasoning

RMA is an agentic framework designed to solve complex, research-level mathematical problems through iterative proof refinement and literature grounding. It outperforms existing models like GPT-5.2R on the First Proof benchmark by utilizing specialized modules for analysis, search, and verification.

#automated-reasoning#multi-agent-systems#mathematics
๐Ÿ“„
ArXiv AIโ€ข65d ago

NeuroNL2LTL: Bridging Natural Language and Formal Logic Verification

NeuroNL2LTL is a neurosymbolic framework that translates natural language into Linear Temporal Logic (LTL) with formal correctness guarantees. By integrating a verifier-in-the-loop training process, the system uses formal verification outcomes as reinforcement learning rewards to ensure high-reliability specifications.

#neurosymbolic#formal-verification#ltl
๐Ÿ“„
ArXiv AIโ€ข65d ago

ImProver 2: Neurosymbolic Framework for Automated Proof Optimization

ImProver 2 is a new neurosymbolic framework for Lean 4 that optimizes formal proofs using expert-iteration and structural metrics. It enables 7B-parameter models to outperform much larger systems in restructuring research-level mathematical proofs.

#formal-verification#lean-4#neurosymbolic
๐Ÿ“„
ArXiv AIโ€ข65d ago

IDS Enables AI to Generate Formally Verified Systems

Inductive Deductive Synthesis (IDS) is a new agentic LLM framework that jointly synthesizes code and formal proofs. It successfully achieves 100% verification on distributed key-value store specifications, significantly outperforming current SOTA coding agents.

#formal-verification#distributed-systems#agentic-workflow
๐Ÿ“„
ArXiv AIโ€ข65d ago

EVE-Agent: Evidence-Verifiable Self-Evolving Search Agents

EVE-Agent introduces a framework for self-evolving search agents that ensures training data is grounded in verifiable evidence. By rewarding models based on the marginal accuracy gain provided by source-grounded spans, it creates an auditable and trustworthy training loop without human labels.

#self-evolving-agents#rag#autonomous-agents
๐Ÿ“„
ArXiv AIโ€ข65d ago

BOHM: Zero-Cost Hierarchical Attribution for Compound AI Systems

BOHM is a new method for attributing performance in compound AI systems by extracting data directly from existing routing weights. It offers multi-resolution attribution without the high computational cost of traditional Shapley-based methods.

#agentic-workflows#routing-optimization
๐Ÿ“„
ArXiv AIโ€ข65d ago

Advancing Mediative Fuzzy Logic for Complex Decision-Making Systems

This research introduces a unified framework for Mediative Fuzzy Logic, extending its foundations from type-1 to type-2, type-3, and quantum settings. It provides a robust mathematical structure for handling conflicting evidence in safety-critical applications like autonomous braking.

#fuzzy-logic#decision-systems#sensor-fusion
๐Ÿ“„
ArXiv AIโ€ข65d ago

A Theory of Accountability Boundaries in Agentic Ecosystems

This research introduces a capability-level theory for managing accountability in agentic AI systems. It explores how modular AI interfaces impact organizational responsibility and governance, specifically addressing the risks of 'rule debt' in automated environments.

#ai-governance#agentic-ai
๐Ÿ”ฅ
36ๆฐชโ€ข65d ago

Huawei Hubble expands semiconductor and AI investments

Huawei's investment arm, Hubble, continues to aggressively fund the semiconductor and deep-tech ecosystem. Recent investments include companies specializing in optical semiconductors and 3D printing technology.

#semiconductor#supply-chain#investment
๐Ÿ”ฅ
36ๆฐชโ€ข65d ago

TomTom China orders surge 10x for autonomous driving

European map provider TomTom reports a 10x increase in orders from Chinese automakers as they expand globally. The company provides critical mapping data and ADAS support to help Chinese EVs navigate complex international road regulations.

#autonomous-driving#ev#localization
๐Ÿ”ฅ
36ๆฐชโ€ข65d ago

Feikai Materials reports high utilization in fiber coatings

Feikai Materials reports strong production efficiency in its fiber coating division for Q2. The company is optimizing production schedules to meet demand and planning for future capacity expansion.

#infrastructure#supply-chain#networking
โš›๏ธ
้‡ๅญไฝโ€ข65d ago

SaaS-Bench reveals Claude's low success rate in office tasks

UniPat AI released the SaaS-Bench benchmark, showing that mainstream LLMs like Claude achieve a complete task success rate of less than 4%. This highlights that fully automated AI office workflows are still far from practical maturity.

#ai-agents#benchmarking#automation
๐Ÿผ
Pandailyโ€ข65d ago

ByteDance Open-Sources Lance Multimodal Model for Local VRAM

ByteDance has released Lance, a new open-source multimodal model designed to run locally on hardware with 40GB of VRAM. The model gained significant traction on Hugging Face shortly after its release.

#multimodal#local-llm#vram-optimization
๐Ÿผ
Pandailyโ€ข65d ago

Alibaba Qwen 3.7 Max Completes 35-Hour Autonomous Task

Alibaba's Qwen 3.7 Max demonstrated impressive endurance by completing a 35-hour autonomous task run. The model successfully executed 1,158 tool calls, showcasing its reliability for long-duration AI workflows.

#ai-agents#autonomous-tasks#tool-use
๐Ÿ”ฅ
36ๆฐชโ€ข65d ago

ima Copilot fully released with new Knowledge Agent skills

ima has fully opened its Copilot feature to all users, removing the previous waitlist. The update also introduces a 'Knowledge Plaza' where users can discover and publish custom Skills, such as WeChat Reading and Tencent Recruitment, allowing for deeper integration with personal knowledge bases.

#agentic-ai#knowledge-management#productivity
๐Ÿ’ฐ
้’›ๅช’ไฝ“โ€ข65d ago

Kling AI dominates overseas markets

Kling AI, the video generation model from Kuaishou, has achieved significant success in international markets. Meanwhile, the company is undergoing internal leadership changes in its domestic e-commerce division.

#video-gen#global-expansion#kuaishou
Page 518 of 1666