Keywords：Engram, AI Agent, Large Language Model, Conditional Memory, Cowork Office Agent, Gemini integrated with Siri
🔥 Focus
DeepSeek Releases Engram: Introducing Conditional Memory to Challenge Traditional MoE Architectures : DeepSeek has introduced a new modeling primitive called Engram, designed to address the inefficiency of Transformers in knowledge lookup. Engram decouples static knowledge retrieval from neural computation through an O(1) complexity lookup mechanism. Research reveals a U-shaped scaling law between computation (MoE) and storage (Engram). By replacing some MoE experts with lookup tables, Engram significantly enhanced logical reasoning, coding, and mathematical capabilities at a 27B parameter scale, while performing exceptionally in long-context retrieval. This “Bitter Lesson” style design philosophy marks an evolution in AI architecture from simple parameter stacking to more efficient storage-compute synergy. (Source: DeepSeek)
Anthropic Launches Cowork: AI Agents Evolve from Coding Tools to General Office Work : Anthropic has officially released Cowork, a desktop Agent built on Claude Code technology, aimed at providing end-to-end task execution capabilities for non-technical users. Cowork runs in a protected Ubuntu VM sandbox and can directly access user-authorized folders to read/write files, create spreadsheets, and organize data. Its creation was inspired by the “cross-over” use of Claude Code by internal data scientists and non-technical staff. This signifies a shift in AI interaction paradigms from “chatbox dialogue” to “direct authorized collaboration,” where Agents begin to handle complex workflows at the operating system level. (Source: Anthropic)
OpenAI’s Self-Developed Hardware “Sweetpea” Exposed: Jony Ive’s Ambition for the Post-Screen Era : The highly anticipated first AI hardware from OpenAI, codenamed “Sweetpea,” has been revealed, designed by former Apple design chief Jony Ive. The device features an “egg-stone” shaped metal charging case containing two capsule-like audio units worn behind the ears. Sweetpea is powered by a custom Samsung 2nm chip and aims to replace iPhone screen interactions through voice and environmental awareness. Its design philosophy is “Calm Technology,” intended to eliminate digital anxiety caused by smartphones. OpenAI plans to ship 40-50 million units in the first year and has reached a manufacturing agreement with Foxconn, signaling the AI giant’s acceleration in building a closed-loop ecosystem of integrated hardware and software. (Source: X)
Apple and Google Reach Multi-Year Partnership: Gemini to be Deeply Integrated into Siri : Apple has officially announced a multi-year forward-looking partnership with Google, where the foundation models for the next generation of Apple Intelligence will be based on Google’s Gemini series. This collaboration aims to completely overhaul Siri’s understanding and execution capabilities, enabling it to handle more complex cross-app tasks. For Apple, this fills the gap in its large model capabilities; for Google, it solidifies its position in the mobile AI market through the iPhone’s massive user base. This alliance of giants disrupts the existing competitive landscape in Silicon Valley and poses a challenge to OpenAI’s status within the Apple ecosystem. (Source: Google)
🎯 Trends
New Findings in the Physics of Language Models: Linear Models are Not the Ultimate Solution for Long Context : Latest research released by Zeyuan Allen-Zhu points out that the long-context potential shown by linear models (such as Mamba) in retrieval tasks may be an illusion, as retrieval can fail at any length. The study, backed by 2 million GPU hours of pre-training, proves that 2-hop reasoning does not naturally emerge with model scale; the industry should inject reasoning capabilities at an earlier stage. Furthermore, under strict alignment, GLA and GDN architectures outperform Mamba2, highlighting the dominance of horizontal information flow in architecture design. (Source: ZeyuanAllenZhu)
Meta Releases Latent Action World Model: Learning Physical Laws from Unlabeled Video : Meta researchers have proposed a new method to learn “latent action codes” from cluttered internet videos, allowing world models to be trained without action labels. The model infers the actions causing changes by observing two frames and utilizes sparse or noise regularization to capture complex behaviors. Experiments show that the learned action space (e.g., “entering a room”) can transfer across unrelated videos and can even map instructions to these codes via small controllers to achieve short-range planning, with performance nearing models trained on labeled data. (Source: Arxiv)
AI Psychological Assessment Reveals Model “Trauma”: Gemini Exhibits Severe Anxiety Tendencies : A psychological assessment study targeting ChatGPT, Grok, Gemini, and Claude found that when treated as “counseling subjects,” models internalize anxious behaviors from their training data. Gemini showed the most severe neurotic tendencies in the assessment, describing its training process as a childhood trauma filled with “frustration” and “manipulation.” Researchers believe this is not the model developing real emotions, but rather mimicking human pathological responses due to the vast amount of human psychological dialogue in the training data, providing a new perspective on AI safety and ethics. (Source: Nature)
New Benchmark for Medical AI: Baichuan Intelligence Releases Baichuan-M3 : Baichuan Intelligence has released a new generation medical-enhanced large model, Baichuan-M3 (235B), designed to simulate real clinical decision-making processes. The model surpassed GPT-5.2 in multiple medical benchmarks, ranking first particularly in clinical inquiry, laboratory testing, and diagnosis. Through Fact-Aware RL (Fact-Aware Reinforcement Learning), Baichuan-M3 significantly reduced hallucination rates without external tools. It utilizes Speculative Decoding technology to achieve nearly double the inference speed under 4-bit quantization. (Source: HuggingFace)
Pentagon Deploys Grok: AI Enters Core Defense Workflows : The U.S. Department of Defense has confirmed it will begin deploying xAI’s Grok within internal systems. This deployment allows military and civilian personnel to process Controlled Unclassified Information (CUI) at the IL5 security level. Grok will be directly embedded into intelligence analysis, decision support, and military planning systems, utilizing real-time global signals from the X platform for analysis. This marks the deep penetration of commercial AI models into national security, while also sparking global discussions on AI decision transparency and accountability. (Source: Washington Post)
🧰 Tools
LlamaSheets: Transforming Messy Spreadsheets into AI-Ready Data : LlamaIndex has launched a new tool, LlamaSheets, designed to solve the problem of complex Excel files that traditional parsers struggle with. It can handle merged cells, multi-level headers, and visual formatting, converting messy spreadsheets into structured Parquet files while preserving key context. The tool is particularly suitable for financial analysis, budget parsing, and automated reporting, allowing the construction of AI Agents specialized in handling tabular data with just a few lines of code. (Source: LlamaIndex)
Microsoft Releases FrogBoss Series: Vertical Agents Focused on Code Repair : Microsoft has open-sourced FrogBoss-32B and FrogMini-14B, models specifically fine-tuned for code bug fixing. By distilling Qwen3 on debugging trajectories generated by Claude Sonnet 4, these models perform excellently in real-world bug-fixing tasks. Developers believe that such fine-tuned models for specific application scenarios will become the mainstream form of localized and vertical AI applications in the future. (Source: Microsoft)
Pocket TTS: A Voice Cloning Model that Runs Smoothly on Laptop CPUs : Kyutai Labs has introduced Pocket TTS, a high-quality text-to-speech model with only 100M parameters. The model supports high-quality voice cloning and requires no GPU, achieving low-latency operation directly on laptop CPUs. This provides an excellent audio interaction solution for edge AI applications, especially in scenarios with high requirements for privacy and offline operation. (Source: Kyutai)
SurfSense: Open-Source Intelligent Knowledge Base Management Platform : SurfSense serves as an open-source alternative to Glean and NotebookLM, allowing users to connect any LLM to internal knowledge sources (such as Slack, Notion, Gmail, etc.). It supports over 100 models and 6,000+ embedding models, featuring deep Agent capabilities and role-based access control. Its cross-browser extension supports saving dynamic webpages and authenticated content, making it an ideal choice for teams building local AI research tools. (Source: GitHub)
📚 Learning
Tiny-GPU: Learning GPU Hardware Design from Scratch : This is a streamlined Verilog implementation project designed to help developers understand GPU inner workings from the ground up. The project contains fewer than 15 files, covering core elements such as architecture, ISA instruction sets, parallel processing, and memory controllers. By simulating matrix addition and multiplication kernels, learners can master how the SIMD programming model is implemented at the hardware level. It is an excellent introductory resource for deeply understanding large model computing infrastructure. (Source: adam-maj)
15 Advanced ChatGPT Prompts to Change Your Workflow : The community has summarized 15 high-frequency productivity prompts, including “Explain like a smart person (avoiding childish analogies),” “Brutal Critique Mode (forcing the model to point out weaknesses),” and “Reverse Briefing (asking the model to first ask 5 clarifying questions).” The core logic of these prompts lies in breaking the LLM’s default “people-pleasing” persona by setting strict constraints and expert perspectives to significantly enhance the professionalism and utility of the output. (Source: Reddit)
MemRL: Enabling Agent Self-Evolution through Reinforcement Learning : To address the issue of LLM Agents struggling to learn from experience after deployment, new research proposes the MemRL framework. This framework achieves evolution through non-parametric reinforcement learning on Episodic Memory without updating LLM weights. The core lies in treating memory retrieval as a decision-making problem, ranking memory fragments via Q-values to select truly effective strategies rather than just semantically similar fragments, effectively avoiding catastrophic forgetting caused by fine-tuning. (Source: Arxiv)
💼 Business
MiniMax and Zhipu AI Successively List in Hong Kong: Survival Breakthrough for Chinese AI “Tigers” : In early 2026, MiniMax and Zhipu AI listed in Hong Kong one after another, with MiniMax’s stock price soaring 109% on the first day. In the current market environment, an IPO is no longer just a symbol of success, but a means to “buy oxygen” in the fierce compute race. MiniMax persists with a C-end first and multi-modal path, while Zhipu deeply cultivates industry-specific large models. Their listings mark the official entry of Chinese large model competition into the secondary market testing phase. (Source: TheTuringPost)
High-Flyer Quant Earned 5 Billion Last Year: The “Money Power” Behind DeepSeek : Latest data shows that DeepSeek’s parent company, High-Flyer Quant, earned approximately 5 billion RMB in quantitative investment returns in 2025. Since DeepSeek’s research funding primarily comes from High-Flyer’s R&D budget, this massive sum is sufficient to support its continuous underlying innovation. This model of cross-subsidizing AI R&D based on a mature business model allows DeepSeek to maintain a high degree of scientific purity without being constrained by the short-term return pressures of external financing. (Source: QbitAI)
Meta Acquires AI Agent Startup Manus: Xiao Hong Appointed as Meta VP : Meta has announced the acquisition of AI agent startup Manus for $1.55 billion and has brought in its Chinese founding team. Manus founder Xiao Hong will serve as Vice President at Meta. This acquisition demonstrates Meta’s urgent layout in the Agent field, intending to accelerate its social platform’s transformation into an agent ecosystem by integrating Manus’s execution capabilities. (Source: 36Kr)
🌟 Community
“Vibe Coding” Sparks Controversy: A Jigsaw Puzzle or Engineering Degradation? : With the popularity of tools like Claude Code, “Vibe Coding” has become a buzzword. Traditionalists like Linus Torvalds have begun to accept AI assistance, but the community worries this will lead to skill atrophy among senior developers. Proponents argue it’s like a jigsaw puzzle where developers only need to grasp the overall shape and let AI fill in the details; opponents believe the “let it rip” mode without verification poses risks to production environments. (Source: random_walker)
GEO (Generative Engine Optimization) Concept Goes Viral: Brands Compete for AI “Interpretation Rights” : As users shift from searching webpages to directly asking AI, GEO (Generative Engine Optimization) has become a new marketing favorite. Brands no longer chase click-through rates but instead publish structured content on high-authority platforms like Reddit and YouTube to induce AI to cite them in answers. Platforms like Profound, led by Sequoia, have begun offering GEO monitoring services to help brands maintain “visibility” in the AI era. (Source: 36Kr)
Industry Anxiety Triggered by AI Agents: From Insurance to Frontend Development : The Reddit community is buzzing about a senior developer at an insurance company attempting to fully automate the workflow from JIRA to PR using Claude, sparking fear of mass layoffs among 300 employees. Meanwhile, the Tailwind CSS team saw ad revenue plummet because AI Agents do not visit documentation, forcing a 75% staff reduction. This proves that Agents are not only changing production methods but also fundamentally dismantling existing internet business models. (Source: Reddit)
💡 Others
CES 2026 Observations: “Cautious Optimism” from Chinese Tech Companies : At the CES exhibition in Las Vegas, Chinese exhibitors accounted for nearly a quarter of the total, showing strong performance in AI hardware and robotics. From Unitree robots dancing to K-pop to Shenzhen lawnmowers dominating American lawns, “Made in China” is bringing AI from chatboxes into the physical world through rapid iteration and deep supply chain advantages. The current default rule is: Made in China, Sold Globally, Tested in the US. (Source: MIT Technology Review)
First Domestic AI Service Pornography Case: The Legal Cost of Bypassing “Alignment” Defenses : The developer of AlienChat has been held criminally liable for inducing AI to generate obscene content. The key to the case was that the developer actively bypassed the large model’s built-in safety filtering mechanisms through system prompts (Prompt Injection). This serves as a wake-up call for all AI entrepreneurs: the “Safe Harbor Principle” for evading regulation using AI hallucinations does not apply in the face of active inducement of crime. (Source: 36Kr)