The latest AI news, in focus.

The Sequence Learning Loop - Issue #960: Last Week in AI Infrastructure

DeepSeek, Google, and NVIDIA build the agent stack

From Monk To Startups: One Founder’s Unusual Path To Building Government Software

After living as a monk while managing Graves’ disease, a former Facebook engineer left Big Tech to pursue meaningful work, eventually founding government software startup FSH Technologies which aims to improve processes such as school food services and public records management through the use of artificial intelligence.

b11954

tests : adjust constants to reduce irrelevant warnings (#31254) **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11954/llama-b11954-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b11954/llama-b11954-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b11954/llama-b11954-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b11954/llama-b11954-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b1

Macmillan Staff Say the Publisher Threw a Party for a Workplace AI Assistant That Was Later Killed

Workers tell WIRED that NOA—or “No Ordinary Assistant”—was unhelpful. Bestselling authors also say AI is ruining their relationships with Big Five publishers.

b11953

cuda : add top-k MoE fusion for 384 experts (#31288) Assisted-by: Codex **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11953/llama-b11953-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b11953/llama-b11953-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b11953/llama-b11953-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b11953/llama-b11953-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/dow

OpenAI reportedly plans to purchase TPU chips from Google, further reshaping the AI computing supply chain.

On October 28, reports indicated that OpenAI intends to procure artificial intelligence chips from Google. For a long time, OpenAI has been a major purchaser of NVIDIA GPUs, relying on them for training and inference of large-scale models. As demand for computing power surges, companies are seeking more cost-effective alternatives suitable for running AI models. Google’s self-developed TPU has gradually gained prominence in recent years, emerging as a potential option. This move may reflect a trend toward diversification in AI chip procurement.

b11951

chat : stop qwen3-coder reasoning and content before template separators (#30557) * chat : strip trailing whitespace from parsed reasoning and content Assisted-by: Claude Fable 5.1 * chat : stop qwen3-coder reasoning and content before template separators Assisted-by: Claude * tests : remove the Qwen parser case from the chat session tests Assisted-by: Claude * common : handle partial prefix until match at end of generation --------- Co-authored-by: Zeshawn Zahid Co-authored-by: Alde Rojas **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11951/llama-b11951-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel

DouBao’s PR team responds to the “150,000 extras fed into a model” rumor: The claim is seriously inaccurate.

Liu Xing, head of DouBao’s PR department, issued a statement debunking the rumor, stating that the claim “Hengdian has partnered with ByteDance to feed the portraits and voices of 150,000 extras into ByteDance’s large AI model” is seriously false. He explained that the real “Hengdian Film & TV AI Service Platform” aims to build public digital infrastructure, organizing on-site assets such as film studio locations into high-quality datasets. This helps production crews efficiently scout locations, prepare for shoots, and streamline filming, while also lowering the barrier for small and medium-sized film and TV companies to adopt AI, enabling more creators to produce quality work.

Chelsea Troy proposes a new paradigm for cultivating software engineering skills in the AI era: slowing down is essential for genuine advancement.

According to InfoQ, during her talk at the Craft Conference, Chelsea Troy pointed out that while generative AI accelerates output, it cannot replace the deep learning process. Effective skill development requires returning to the principles of cognitive science, emphasizing a slower pace, confronting discomfort, spaced repetition, outcome prediction, and diversified formats.

Why it mattersThis report is based on an authoritative interview and transcript of her speech published by InfoQ on October 28, 2026. It focuses on the essence of software engineering education, systematically integrating cognitive science with frontline development practices, offering clear guidance and timely relevance for talent cultivation amid the current proliferation of AI tools.

Ant Group launches Lingguang 2.0, enabling tasks to continue even after the user exits the conversation

On October 27, Ant Group unveiled Lingguang 2.0, shifting its product positioning from a “task executor” to a “personal AI assistant capable of handling tasks.” The core change lies in how tasks are executed. Powered by an independent cloud server, Lingguang 2.0 breaks free from the limitations of single-session conversations: once a user exits the dialogue, related tasks keep running in the background and dynamically adapt based on new information and changing conditions. This capability is underpinned by three key upgrades—cross-session memory, long-term task scheduling, and enhanced tool integration. Moreover, the point at which the AI becomes involved has been moved earlier in the process: previously, users had to first identify issues and articulate their needs before the AI could act. With Lingguang 2.0, once authorized by the user, the system can proactively detect potential needs by leveraging connected data and services, initiate analysis, monitoring, and follow-up actions, and issue timely reminders when user attention is required. Officially demonstrated use cases focus on continuously managing everyday tasks, such as tracking price fluctuations after booking airline tickets or identifying potentially idle items from Alipay statements...

Foreign blogger uses AI earphones as a medical translation tool: “Just listen and translate—communication becomes much smoother.”

Foreign blogger “Carlo Wu (Carlo)” shared his experience of visiting a traditional Chinese medicine doctor on social media. He wore the Qwen AI clip-on earphones—one for himself and one for the doctor—and used face-to-face translation to facilitate communication. Carlo said he finds traditional medicine very interesting, but language has always been an obstacle. During the consultation, when the doctor asked where he was feeling discomfort, he could only gesture while saying, “Everything hurts.” After putting on the earphones, the two were able to communicate more effectively. He remarked that this was the first time he truly understood what the doctor was saying.

[AINews] Anthropic ends Coding Interviews: "Hunger beats tenure"

Since coding is solved, why code for the job?

AI Emergency Alerts Are Coming. Humans Might Not Be Ready

Human beings struggle with emergency communication. Chatbots can help—if people take them seriously.

The number of CVE vulnerabilities has risen almost vertically, with AI acting as an “amplifier” for security research.

According to tech media gigazine, developer @0x_Osprey has launched a web-based CVE Tracker that visualizes changes in the number of vulnerabilities assigned CVE IDs, showing a nearly vertical upward trend since 2026. CVE is the globally standardized vulnerability numbering system, relied upon by security researchers, vendors, and enterprises to track, report, and remediate flaws. This tool compiles publicly available records into trend charts and overlays them with the release dates of mainstream AI models, enabling users to intuitively compare vulnerability growth with AI releases.

Polymathic AI and Princeton Researchers Release FlashMap, a Flow Map Drafter for 6.14x Faster LLM Decoding

FlashMap is a speculative decoding drafter from Polymathic AI, Princeton and USC ISI researchers. It drafts 16 tokens jointly in one pass and reaches up to 6.14x lossless speedup on Qwen3-8B. The post Polymathic AI and Princeton Researchers Release FlashMap, a Flow Map Drafter for 6.14x Faster LLM Decoding appeared first on MarkTechPost.

Major leak ahead of Baidu World Conference: Reports indicate it will unveil progress on its world model

On October 28, the Science and Technology Innovation Board Daily exclusively reported that Baidu plans to officially disclose recent advancements and breakthrough results in developing its “world model.” This field is among the hottest frontiers in global AI, with core technology centered on precise dynamic modeling of real-world physical environments and future prediction. Industry analysts believe that if this technology is first applied to autonomous driving, it could significantly enhance self-driving cars’ ability to predict scenarios and make intelligent decisions in complex road conditions.

n8n@3.0.1

## [3.0.1](https://github.com/n8n-io/n8n/compare/n8n@3.0.0...n8n@3.0.1) (2026-10-28) ### Bug Fixes core: Keep the ExpressionError name in minified builds (https://github.com/n8n-io/n8n/issues/42309) ([b3192c6](https://github.com/n8n-io/n8n/commit/b3192c6d151125c76c88fe8a421e027b4242a3f9)) core: Keep undefined values in QuickJS expression results (https://github.com/n8n-io/n8n/issues/42264) ([532f645](https://github.com/n8n-io/n8n/commit/532f645f21c99e178a09448caf37ebd77d24b56a)) editor: Align carets within split buttons (https://github.com/n8n-io/n8n/issues/42203) ([61cbfea](https://github.com/n8n-io/n8n/commit/61cbfeafa267b5c769b345c50ccbd4fc504b0e69)) editor: Improve first load speed (https://github.com/n8n-io/n8n/issues/41987) ([8b6bdbb](https://github.com/n8n-io/n8n/commit/8b6bdbbe874b

Liquid AI unveils the LFM3 multimodal model family, emphasizing efficient, specialized intelligence

Recently, Liquid AI officially launched its next-generation foundation model family, LFM3. The first release includes six models with approximately 2 billion parameters, covering text, vision, audio, omni-modal, and decision-making capabilities—all built on the LFM3-1.7B-Base foundation and achieving a pre-training throughput of 22 trillion tokens. This series adopts a novel tri-hybrid layer architecture, integrating gated short convolutions, sliding windows, and other mechanisms, prioritizing fewer parameters and higher efficiency.

Tencent WorkBuddy launches “Remote Control”: Your phone can view and take over your computer desktop in real time, with remote operations not consuming Tokens.

On October 28, Tencent WorkBuddy introduced “Remote Control,” supported by the Tencent Edge One team. With a single tap in the mobile app, you can connect to your computer, view the desktop in real time, and take control. The screen appears as a floating window at the top of the home page or conversation tab, with an input box available and the ability to correct misaligned tasks. After switching to full-screen landscape mode, tapping anywhere on your phone corresponds to clicking the same spot on the computer. On the right side, tools for “Operations,” “Keyboard,” and “Mouse” are provided; “Operations” allows you to tile windows, return to the desktop, switch screens, and more.

Meet Google DeepMind’s Genie 4: Real-Time World Model With Native Audio

Google DeepMind released Genie 4 as a research preview. The real-time world model adds native audio, live dialogue and text steering, with Gemini 3.8 Flash planning events inside the simulation loop. The post Meet Google DeepMind’s Genie 4: Real-Time World Model With Native Audio appeared first on MarkTechPost.