Recently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and diversifying connectivity patterns. While yielding substantial performance gains, this diversification fundamentally compromises the identity mapping property intrinsic to the residual connection, which causes severe training instability and restricted scalability, and additionally incurs notable memory access overhead. To address these challenges, we propose Manifold-Constrained Hyper-Connections (mHC), a general framework that projects the residual connection space of HC onto a specific manifold to restore the identity mapping property, while incorporating rigorous infrastructure optimization to ensure efficiency. Empirical experiments demonstrate that mHC is effective for training at scale, offering tangible performance improvements and superior scalability. We anticipate that mHC, as a flexible and practical extension of HC, will contribute to a deeper understanding of topological architecture design and suggest promising directions for the evolution of foundational models. — Read More

#china-ai

Or: The West banned the machines. China hired the machinists. Sometimes plans just do not go how you planned them. Ironically, I had been writing this article for a month now, and all research pointed at China being way too far behind. Well…

December 2025. Reuters reveals that China completed an operational EUV lithography prototype in a high-security Shenzhen facility. Not through reverse engineering captured ASML machines. Not through some breakthrough in domestic optics manufacturing. Through something far simpler.

They recruited the humans who knew how to build them. — Read More

In this wave of artificial intelligence, it’s easy to be swept up in grand narratives: computing power, models, the scale of parameters, disruption, and replacement. But the changes truly worth documenting often happen out of sight: how AI is used, how it enters daily life, and how it changes the way people work.

Last week, The Intellectual and Doubao jointly launched: “A Story Collection | How Were You ‘Amazed’ by AI This Year?”, not asking “how powerful is AI,” but a more specific question: When AI enters your work and life, what exactly does it change?

…Perhaps what is truly worth documenting is not what AI can do, but how researchers, after its intervention, re-understand their work, judgments, and responsibilities—what tasks can be automated, and what problems must still be decided by humans.

These scattered and specific experiences constitute the first batch of “field notes” of AI entering the scientific field. They may not be complete, but they are sufficiently honest. — Read More

… China’s extensive AI-powered visual surveillance systems are already well documented. This report reveals new ways that the Chinese Communist Party (CCP) is using large language models (LLMs) and other AI systems to automate censorship, enhance surveillance and pre-emptively suppress dissent.

… AI-powered technology is widening the power differential between China’s state-supported companies operating abroad and foreign populations—further enabling some Chinese companies to systematically violate the economic rights of vulnerable groups outside China, despite Beijing’s claims that China respects the development rights and sovereignty of other countries.

The risks to other countries are clear. China is already the world’s largest exporter of AI-powered surveillance technology; new surveillance technologies and platforms developed in China are also not likely to simply stay there. — Read More

Chinese artificial intelligence startup DeepSeek released two powerful new AI models on Sunday that the company claims match or exceed the capabilities of OpenAI’s GPT-5 and Google’s Gemini-3.0-Pro — a development that could reshape the competitive landscape between American tech giants and their Chinese challengers.

The Hangzhou-based company launched DeepSeek-V3.2, designed as an everyday reasoning assistant, alongside DeepSeek-V3.2-Speciale, a high-powered variant that achieved gold-medal performance in four elite international competitions: the 2025 International Mathematical Olympiad, the International Olympiad in Informatics, the ICPC World Finals, and the China Mathematical Olympiad. — Read More

Large language models have made significant progress in mathematical reasoning, which serves as an important testbed for AI and could impact scientific research if further advanced. By scaling reasoning with reinforcement learning that rewards correct final answers, LLMs have improved from poor performance to saturating quantitative reasoning competitions like AIME and HMMT in one year. However, this approach faces fundamental limitations.

Pursuing higher final answer accuracy doesn’t address a key issue: correct answers don’t guarantee correct reasoning. Moreover, many mathematical tasks like theorem proving require rigorous step-by-step derivation rather than numerical answers, making final answer rewards inapplicable.

To push the limits of deep reasoning, we believe it is necessary to verify the comprehensiveness and rigor of mathematical reasoning. Self-verification is particularly important for scaling test-time compute, especially for open problems without known solutions. Towards self-verifiable mathematical reasoning, we investigate how to train an accurate and faithful LLM-based verifier for theorem proving. We then train a proof generator using the verifier as the reward model, and incentivize the generator to identify and resolve as many issues as possible in their own proofs before finalizing them. — Read More

A Chinese vibe coding tool went viral so fast that its signature feature crashed just days after it launched.

LingGuang, an AI app for vibe coding and building apps using plain-language prompts, launched last Tuesday and reached over 1 million downloads in four days. By Monday, the app had crossed 2 million downloads, said Chinese tech group Ant Group, which built the AI coding assistant tool.

On Monday, LingGuang ranked first on Apple’s mainland China App Store for free utilities apps and sixth overall for free apps. — Read More

Today’s children are true “AI natives.” They are born and raised in the AI ​​era; interacting with the digital world is an innate instinct. The AI ​​entities that provide them with education must also be immersive, interactive, personalized, and warm.

The birth of AI English teacher Jessica heralds the future of education: no longer one-way knowledge transmission, but rather the natural acquisition of a communication ability to face the world through symbiosis and dialogue with AI. She possesses a vast amount of knowledge, boundless patience, a memory capable of remembering every child’s situation, and a warm heart—a true “super-teacher.” — Read More

Today, we are introducing KimiK2Thinking, our best open-source thinking model.

Built as a thinking agent, it reasons step by step while using tools, achieving state-of-the-art performance on Humanity’s Last Exam (HLE), BrowseComp, and other benchmarks, with major gains in reasoning, agentic search, coding, writing, and general capabilities.

… K2 Thinking is now live on kimi.com under the chat mode [1], with its full agentic mode available soon. —Read More

Three months after the Chinese AI company DeepSeek shocked global markets with a highly capable reasoning model, another China-linked company made a splash with a capable agentic AI system. Did Manus, released in March 2025, portend Chinese leadership in AI systems that go beyond chatbots to take action on the user’s behalf? Victor Mustar, head of product at Hugging Face described Manus’ capabilities as “mind-blowing, redefining what’s possible.” A journalist’s comparison with ChatGPT DeepResearch found that Manus provided better results, despite speed and stability issues.

Manus had been released by a Singapore-based firm but developed by a startup in Wuhan with backing from the Chinese tech giant Tencent. It wasn’t China’s only foray into the emerging field. The same month, the Beijing-based firm Zhipu AI launched AutoGLM-Rumination, an open-source agentic system the company said achieved “state-of-the-art” scores on benchmarks such as AgentBench. (Zhipu also announced an “international alliance” for autonomous AI models, to include 10 countries associated with the Belt and Road Initiative and from ASEAN.) Earlier in January, Alibaba released the Qwen-Agent framework for building agentic systems with its Qwen models. ByteDance followed with its Coze Studio platform in July. Last month, Tencent open-sourced Youtu-Agent agentic framework, which was reportedly built atop a DeepSeek model.

With so much action this year in Chinese “agentic” AI efforts, it’s worth pausing to ask what Chinese developers mean when they talk about agentic AI. Moreover, what does the proliferation of such systems in China mean for AI safety and governance in the country? — Read More