All Updates
Page 1207 of 1460
March 9, 2026
Turing Winner Cracks 30-Year Math Puzzle
An 88-year-old Turing Award laureate solved a 30-year mathematical mystery using Claude in just one hour. This showcases LLMs' potential in tackling complex unsolved problems. Highlights rapid AI-assisted discovery.
MiroFish Tops GitHub, Backed by Chen Tianqiao
MiroFish, an AI prediction tool developed by a Chinese post-00s developer, has topped GitHub charts. Backed by investor Chen Tianqiao, it makes universal prediction accessible. Highlighted in AGI context.
US Policies Spark Scientist Exodus
Trump-era US policies classify researchers from China and others as high-risk, restricting NIST lab access in AI and quantum fields to 3 years. Federal funding cuts at NIH/NSF drive 10k STEM PhDs away, pushing talent to Europe and home countries. Young scientists face lab bans and job scarcity.
12 Vital Questions for Billion-Yuan Hard Tech Founders
After mastering six key trends, founders of 100M+ yuan revenue firms must tackle 12 make-or-break questions. Next 3-5 years are golden era for Chinese hard tech listings. Aims to guide steadier path forward.
Vercel Partners with OpenClaw Amid Wallet Drains
OpenClaw faces criticism for potentially draining user wallets overnight. Vercel has become its new partner to address issues. This offers lessons for Tencent, Alibaba, and Baidu.
Ring CEO Fails to Ease Facial Rec Privacy Fears
Ring founder Jamie Siminoff has been addressing privacy concerns triggered by their Super Bowl ad. His explanations, especially regarding facial recognition, appear unconvincing and may heighten user worries. This highlights persistent privacy challenges in AI-driven smart home devices.
arXiv Founder: Grok Tops Paper Padding Test
arXiv founder personally tested LLMs for generating 'water papers' (filler content). Grok proved strongest in compliance and output quality. Claude was the least cooperative.
Ctrip Record Revenue AI-Powered Growth
Ctrip achieved record 2025 revenue of 624B RMB, up 17%, driven by accommodation bookings and international business now at 40% of total. Board changes signal new era with focus on inbound tourism, globalization, and AI innovations like translation engine producing 60B words yearly. R&D spend hit 151B RMB, 24% of revenue.
OpenClaw Hype Enriches Ecosystem Not Users
OpenClaw AI agent explodes with 250k GitHub stars, fueling profits for cloud providers, LLM APIs, installers, and courses, but users struggle to monetize. Analysis likens it to past hype cycles like O2O, warning it doesn't solve business acumen needs. Safety pass rate only 58.9%, with zero on intent safety.
AI Voice Social App Hits 2.6M MAU
Blizzard/Pixar veteran's AI social experiment uses voice matching for connections. Reached 2.6M MAU and $150M valuation. Surged to iOS social top 20 during Spring Festival.
HALO Assets Defy AI Hype Anxiety
Shifts to HALO (Heavy Assets, Low Obsolescence) like energy resist AI disruption amid geopolitics. Critiques OpenClaw lobster hype as emotional frenzy benefiting clouds/models over users. Urges focus on timeless fundamentals over trendy tools.
MizarVision Secures Pre-A Funding for AI Satellite Intel
MizarVision completed pre-A funding over 10M RMB led by Phenomenon Capital to expand its AI-powered geospatial intelligence business. The Spark platform fuses satellite remote sensing with multi-modal data and AI models for real-time analysis, notably preempting US military activities in the Middle East. Funds will enhance R&D and industry applications in defense, energy, and finance.
OpenClaw Creator Recommends Top Lobster Models
OpenClaw's creator recommends the best AI models for Lobster hardware. The top 3 includes two domestic Chinese models. This guides optimal model selection for compatibility.
Ex-Google Researcher Launches AI Robotics Startup in Tokyo
A former Google AI researcher has founded an AI robotics startup in Tokyo. The Silicon Valley-originated venture aims to prove AI's potential to transform Japan's dominant industrial robot supply chain.
Open-Source 30B Model Rivals Gemini & Claude
Research AI releases a powerful open-source 30B parameter model challenging Gemini and Claude. It fully implements the 'hypothesis-evidence-verification' reasoning loop. This advances accessible scientific reasoning in LLMs.
Samsung's AI Vibe Coding for Galaxy Apps
Samsung is reportedly exploring 'vibe coding', an AI-powered feature for Galaxy phones that enables users to create custom apps or modify their device experience without traditional coding. This follows Nothing's similar initiative. The feature aims to make app development accessible to non-developers.
RoboLayout: Agent-Friendly 3D Scenes
RoboLayout extends LayoutVLM with agent-aware reasoning and reachability constraints for generating navigable 3D indoor scenes from language. It supports diverse agents like robots, humans, and animals via abstract representations. Local refinement optimizes problematic placements efficiently while preserving semantic coherence.
RL for Climate-Resilient Transport
Proposes a reinforcement learning (RL) framework for long-term flood adaptation in urban transport systems. Integrates rainfall projections, flood modeling, transport simulation, and impact quantification as an integrated assessment model (IAM). Outperforms traditional optimization in Copenhagen case study by discovering coordinated strategies under climate uncertainty.
Reasoning Models Fail CoT Control
New CoT-Control suite reveals reasoning models have low controllability over chain-of-thought vs. final outputs, e.g., Claude Sonnet 4.5 at 2.7% CoT control but 61.9% output. Controllability drops with larger models, RL training, compute, and difficulty. Results suggest CoT monitoring remains viable despite evasion attempts.
ProEvolve: Programmable Agent Benchmark Evolution
ProEvolve is a graph-based framework for scalable, controllable evolution of agent environments to test LLM agents' adaptability to real-world changes. It uses a typed relational graph to unify data, tools, and schemas, enabling updates via graph transformations that propagate coherently. The system generates 200 environments and 3,000 task sandboxes for benchmarking.