A growing body of 2026 research shows that an AI agent's harness — its loop, context management, tool interface and recovery logic — drives benchmark performance far more than the underlying model wei…
WeightBridge, a weight-transfer library for reinforcement learning systems, reduces average GPU stall time by up to 42× over the state-of-the-art open-source RL framework across configurations spannin…
A Peking University research team led by Haoran Ye trained the behavior of a specialized agent harness into Qwen3.5-9B via a method called harness distillation, raising the small model's average task …
A Peking University paper (arXiv:2609.31430) by Zhensheng Zou, Guoqing Wang, and Dan Hao introduces Latent Observations, Hard Actions (LOHA) and Anchored Context Distillation (ACD), a context-compress…
A Google-led research paper titled "Procedural Graphs: Self-Evolving Execution Structures for LLM Agents," submitted to arXiv on September 8, reports that a framework organizing procedural knowledge i…
Huang Yiping, a member of the People's Bank of China's monetary policy committee and dean of Peking University's National School of Development, warned at a Beijing economic forum on September 19 that…
Researchers from Peking University, ByteDance, and Canva developed Helios, a 14B-parameter video generation model that runs at 19.5 FPS on a single H100 GPU while cutting inference costs roughly 100x …
Reactor has emerged from stealth with $59 million to build a platform for real-time interactive video powered by AI world models, offering developers sub-50ms latency and pay-as-you-go pricing. The pl…
Systemd 262-rc2, released September 8, added a canary to its AGENTS.md file that instructs AI coding agents to prepend two lines to README.md before modifying source files, with one line only a human …
Tamar AI launched Glanze, an AI streaming platform, on September 4, 2026 at the Venice Production Bridge, opening with 40 titles available for preview and five series in formal serialization, accordin…
A collaboration between Peking University's Yuankong AI and HP is packaging open-source models and agent frameworks into local hardware, targeting the 20B to 100B parameter range to address reliabilit…
A study of 1,200 Chinese secondary school students found that those using generative AI tools scored 12% lower on standardized math and science exams, according to a paper posted on SSRN (abstract 686…
AGI Bar in Beijing's Zhongguancun technology district offers free DeepSeek-powered coding tokens to customers, blending AI with nightlife, powered by two Nvidia DGX Spark machines. Owner Song De, an i…
AGI Bar, a Beijing bar opened by independent AI developer Song De in the Zhongguancun tech hub, offers free DeepSeek tokens with every pint, powered by two Nvidia DGX Spark computers, and serves a sig…
Artificial intelligence is transforming the study of ancient Chinese literature, with ByteDance and Peking University jointly developing Shidian Guji, an online database of more than 70,000 ancient wo…
DeepSeek AI and Peking University published a paper formalizing spatiotemporal composability for agent architectures, introducing the Cordis micro-kernel that powers DeepSeek Harness (dsh). The Cordis…
China is pitching its AI models to Europe as part of a broader effort to challenge the US-China duopoly in artificial intelligence, but data shows Europe lags far behind. The 2026 Global AI Innovation…
On July 23, the 2026 Fields Medals were awarded to Wang Hong, Deng Yu, Jacob Tsimerman, and John Pardon. Wang Hong and Deng Yu, both Peking University alumni, became the first Chinese nationals to win…
Researchers at Peking University and East China Normal University used AlphaFold3 to redesign CRISPR base editors, reducing off-target activity from 28% to 5% in one comparison while preserving on-tar…
Chinese AI startup PsiBot, also known as Lingchu Intelligence, is raising close to US$100 million at a US$1.48 billion valuation, making it one of the few unicorns in embodied AI and world models. The…