▌ Large Language Models (LLMs)

OpenAI and Anthropic Models Achieve Major Breakthroughs

OpenAI models have demonstrated PhD-level knowledge across physics, biology, and chemistry. Additionally, Anthropic team members claim their Opus 4.5 model "has largely solved programming problems."

[View Details 1] | [View Details 2]

Heat: ⭐ 8.5/10

Commentary: The deep breakthroughs of top-tier large models in vertical domains like scientific research and coding mark the accelerated evolution of AI from general conversational assistants to professional-grade productivity tools.

OpenAI Releases IH-Challenge Reinforcement Learning Dataset

OpenAI released the IH-Challenge, a reinforcement learning (RL) training dataset designed to teach models to follow the instruction hierarchy (System > Developer > User > Tool). Experiments show that fine-tuning GPT-5-Mini on this dataset improved average performance across 16 benchmarks by 10% and significantly reduced the model's unsafe behavior rate from 6.6% to 0.7%.

▌ Large Language Models (LLMs)

OpenAI and Anthropic Models Achieve Major Breakthroughs

OpenAI Releases IH-Challenge Reinforcement Learning Dataset

StepFun Launches StepClaw and Open-Sources Step 3.5 Flash Model

Google Gemini 3 Flash Introduces Agentic Vision

Mainstream Large Models Trapped in "Over-Alignment" Dilemma

Turing Post Explores LLM Fine-Tuning Techniques Beyond Traditional RL

Virginia Tech Reveals "Temporal Drift" Phenomenon in Large Models

▌ AI Infrastructure

Huawei to Release Next-Generation AI Data Infrastructure

Cambricon's 2024 Net Profit Surges by 555%

Large Model Context Windows and Throughput Achieve 10-Billion-Level Breakthrough

Turbopuffer's Underlying Architecture Innovation and Commercial Implementation

In-Depth Analysis of Quantum Computing Hardware Technology Paths and Challenges

Evolution of Retrieval Paradigms Driven by AI Agents: Agentic Search and Hybrid RAG

Manufact Raises $6.3 Million to Build MCP Infrastructure

Jiuyou Database Completes Series A+ Funding, Focusing on Hyperconverged Architecture

APMIC Completes Series A Funding, Driving AI Models Toward On-Premises Transition

▌ Open Source Technology

Microsoft Launches AgentRx Framework: Enabling Systematic Debugging of AI Agent Systems

Stanford Researchers Release OpenJarvis Local AI Agent Framework

SWE-CI Introduces the First Repository-Level Benchmark Based on CI Loops

▌ AI Security

Autonomous AI Agent Breaches McKinsey's Internal AI Platform Lilli

▌ AI Healthcare

MIT Team Uses Deep Learning to Predict Heart Failure Risks

▌ AI Industry & Policy

Evolution and Implementation Cases of Enterprise AI Business Models

Tesla and xAI Jointly Develop Software Agent Project "Macrohard"

US Military Explores Using Generative AI to Assist Military Strike Decisions

Perplexity Enterprise Agent Achieves Complex Multi-Model Orchestration

Replit Secures $400 Million Funding and Releases New AI Coding Assistant Agent 4

Atlassian Lays Off 1,600 Employees to Fully Advance AI Strategic Transformation

Anthropic Launches Claude Marketplace

AI Interaction Paradigm Evolves Towards "Promptless Intelligence"

Proliferation of "AI Garbage" Triggers Enterprise Management Backlash

Building a System of Context for Agentic AI

Tencent WorkBuddy Releases New Version Amid OpenClaw Plagiarism Controversy

Azoma Releases Agentic Merchant Protocol

Personal AI Coding Tools Impact Traditional IDEs

Vertical Specialized Tools Continue to Maintain Advantages in Professional Scenarios

Multi-Agent AI Economics is Influencing the Business Automation Process

Anthropic Enables Claude to Share Context Across Applications

Notion AI Lead Points Out Oligopoly in the Token Market

Open Interpreter Launches AI-Native Workspace Tool

iFlytek's "Crayfish" AstronClaw Officially Launched

Seele AI Founder: Decides to Let AI Fully Take Over Company Office Work

Webflow Acquires AI Content Generation Platform Vidoso to Enhance Marketing Suite

Renesas Electronics Launches Renesas 365 Unified Development Platform

Yahoo Search Integrates Generative AI

Manus AI Launches Agent Skills

CIO Report: Accelerating Enterprise IT Modernization to Drive AI Implementation

Honor CEO Li Jian: Redefining the Future of Terminals with the AHI Concept

▌ Robotics

JD.com Releases "Intelligent Robot Industry Acceleration 2.0 Plan"

Dreame Robot Vacuums Certified as Global No. 1 in High-End Sales

▌ Others

Meta Gaming Executive Discusses VR Dilemma and Shift to AI

Jinghe Energy Debuts 60Ah All-Solid-State Battery Equipped with AI-BMS System

Bucknell University Quantifies the Limitations of AI Imitating Human Writing

ChatGPT Profoundly Changes Academic Writing Habits

Cornell University Reveals the Key to LLMs Identifying Author Writing Styles

Geolocation Vision Model GeoSpy

AMD Announces FSR "Diamond" Technology, Providing AI Optimization for Next-Gen Xbox

Oil Tankers Attacked by "Underwater Drones" in Middle Eastern Waters

Tinder Introduces AI Enhancements to Attract Younger Users

[AI]news - AI News Briefing: Multi-Model Capabilities Soar, Agent Era Triggers Enterprise Adoption Frenzy (Summary/Review)

TOPICS