NAll news

All news

Monday28 items

  1. China.com TechnologyProduct release
    AI score66

    Xiaomi's open-source AI desktop pet HachimoDock debuts, with a DIY cost of about 150 yuan

    HachimoDock, an open-source AI hardware incubated by Xiaomi's Technology Committee Innovative Products Team, officially debuted at the 2026 Guancha Conference in Hangzhou, with a hardware DIY cost of about 150 yuan. The device integrates an AI desktop pet, voice control of PC-side Agents, smart home integration, and custom component creation. Through voice, it can dispatch tasks such as looking up information, organizing files, and writing code to Agents including Codex and WorkBuddy, while the screen displays execution status in sync. The software source code is now public on GitHub, along with a hardware bill of materials, structural component models, and assembly tutorials, supporting self-replication and secondary development.

    Same story · 4 reports

    #Product updates#Open-source ecosystem#Agents

  2. Phoenix TechnologyProduct release
    AI score69

    Xiaomi's open-source AI hardware HachimoDock debuts: DIY cost about 150 yuan

    HachimoDock, an open-source AI hardware incubated by the innovative product team of Xiaomi's Technology Committee, made its debut at the 2026 Guanzha Conference, with a DIY hardware cost of about 150 yuan. The device integrates a personalized desktop pet, voice control of PC-side Agents, smart home integration, and custom component creation. Users can collaborate with a Coding Agent in natural language to generate components such as to-do lists, computer status dashboards, and Pomodoro timers, and can also dispatch requests via voice—such as looking up information, organizing files, and writing code—to Agents including Codex and WorkBuddy. The device can connect with Xiaomi MiloCo for voice-based home control. The software source code is now public on GitHub, along with a hardware bill of materials, structural component models, and an assembly tutorial.

    Same story · 4 reports

    #English text is being prepared.#Agents#Embodied AI

  3. Kling AIFirst-partyProduct release
    AI score58

    Kling 4.0 Launches on Runware, Supporting 30-Second 1080p Video with Native Audio

    Kling 4.0 is now live on Runware, supporting up to 30-second 1080p video with native audio. This version supports multiple keyframes including first, middle, and last frames, with up to 15 inputs per request across keyframes, reference images, reference videos, and elements, and can also edit existing clips based on prompts. Pricing starts at $0.133 per second of output, with a 15% discount off regular pricing for the next 7 days; it joins the 19 Kling models already available on the Runware API, and you can switch the model ID to klingai:kling-video@4.0 when making calls.

    Quoted postRunware · @runware

    @Kling_ai's Kling 4.0 is live on Runware on D0 🎬 Up to 30 seconds of 1080p video with native audio, plus: - Multi-keyframe support, including first, middle, and last frames - Up to 15 inputs per request across keyframes, reference images, reference videos, and elements - Prompt-based editing of existing clips From $0.133 per second of output, with 15% off regular pricing for the next 7 days. It joins the 19 Kling models already on our API, across video, image, avatar, and lip-sync. Already calling Kling models on Runware? Switch your model ID to klingai:kling-video@4.0

    #Product updates#Video#Tutorials / practice

  4. Sina TechProduct release
    AI score67

    Xiaomi's open-source AI hardware HachimoDock debuts, with a DIY cost of about 150 yuan

    HachimoDock, an open-source AI hardware incubated by the Innovative Products team of Xiaomi's Technology Committee, debuted during the 2026 Guancha Conference, with a DIY hardware cost of about 150 yuan. The device combines desktop pet interaction, voice control of PC-side Agents such as Codex and WorkBuddy, linkage with Xiaomi Miloco smart home devices, and on-site generation and running of custom components; task status is fed back through the screen and the desktop pet. The software source code is now public on GitHub, along with a hardware bill of materials, structural component models and assembly tutorials.

    Same story · 4 reports

    #Product updates#Open-source ecosystem#Agents

  5. MiniMaxFirst-partyProduct release
    AI score57

    MiniMax Summarizes 11 New Community Contributions to the H3 Ecosystem

    MiniMax announced 11 new community contributions to the H3 video generation model ecosystem, covering 8-step reference-conditioned video plus audio, 4-step distillation, sparse attention, and reference token compression, as well as integration with deployment and workflows on GPU servers, TPU, and Apple Silicon. The company says these improvements bring lower experimentation costs, finer creative control, and more runtime environment options, and thanks haoailab, Hugging Face, and teams including PTD, DMAD, SparkH3, TorchTPU, and FlashML.

    Quoted postMiniMax (official) · @MiniMax_AI

    Open weights. Shared progress. MiniMax H3 is moving fast. We built H3 for video generation with native stereo audio and multimodal reference control. The open-source community is making that capability faster, more accessible, and easier to build on. Recent highlights: • FastH3 — FastVideo, Nuva Lab and NVIDIA: 4-step distillation, now running on DGX Spark and Apple Silicon. • Sol-H3 — NVIDIA’s SANA team: 15 seconds of 768p video + audio in 6.6 seconds on 8×B300, in the team’s warm-inference benchmark.* • VDN — Haocheng Xi and the OpenVDN team: rethinking attention for faster H3 inference, with weights, training and inference code released. • PDD — NVIDIA’s distillation method, brought to H3 by Alibaba PAI as 8-step Acc-LoRAs, now supported in ComfyUI. • LightX2V — 4- and 8-step Turbo LoRAs, with workflows for text, image and reference-conditioned video + audio. Behind every release are people training, optimizing, quantizing, testing and sharing. Special thanks to: @haoailab @nuvalab @NVIDIAAI @xieenze_jr @HaochengXiUCB @ArashVahdat @julberner @LightX2V @ComfyUI And to the individual contributors pushing the work forward: @haozhangml @cxlcl1 @lawrence_cjs @yitongli165665 @haopengl33 @songhan_mit @shanasaimoe Thank you for building with H3 and helping make it faster, more accessible, and more useful for the community. Powerful models go further when we build together. Keep pushing H3. Excited to see what comes next. 🚀 Explore the ecosystem: https://t.co/F1S5IjUNi1

    Same story · 2 reports

    #English text is being prepared.#Video#Open-source ecosystem

  6. Ayi AI NotesProduct release
    AI score69

    Jack Dorsey launches native speech model Dessa, which can run full-duplex conversations locally on an M4 Max MacBook

    Twitter co-founder Jack Dorsey has launched a native speech model called Dessa, which he says does not rely on text as an intermediary, is trained end-to-end on human conversation audio, can run locally offline at faster-than-realtime speed on a MacBook equipped with an M4 Max, and supports full-duplex conversation with the ability to interrupt and interject at any time. The original article says its training consumes hundreds of times fewer tokens than large language models, and that the entire training used only a single NVIDIA GB200 NVL72. The author also notes that this is currently only a small-scale research preview from a lab inside Block, that the demos were selected by the company, and that questions remain about how a pure-audio model performs on complex multi-step logical reasoning.

    Quoted postjack · @jack

    We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence. https://t.co/r0qFLdvd6c

    Same story · 10 reports

    #Model releases#Speech#English text is being prepared.

  7. InfoQ AIProduct release
    AI score76

    Google and Meta Open-Source TorchTPU at PyTorch Conference 2026

    Google and Meta open-sourced TorchTPU at PyTorch Conference North America 2026. This is a native PyTorch backend for Google TPUs, replacing PyTorch/XLA from 2020; migrating a model only requires changing model.to("cuda") to model.to("tpu"). The official release also open-sources four repositories: TorchTPU, vLLM-TorchTPU, SGLang-TorchTPU, and TorchTitan. Hugging Face models can run directly, with native support from the vLLM and SGLang inference engines. Google's published results include running the Qwen3 235B MoE model on 8192 TPU v7x chips and reaching 28% model FLOPs utilization, but the announcement also states that a one-line change only gets the model running; achieving peak performance still requires adjusting shapes, sharding, and operators.

    #English text is being prepared.#English text is being prepared.#Google

  8. InfoQ AIProduct release
    AI score64

    Google Open-Sources Cross-Platform GPU Inference Engine ML Drift

    Google has released the open-source cross-platform GPU compute engine ML Drift under the Apache 2.0 license, supporting OpenGL ES, OpenCL, Metal, and WebGPU; it is the core GPU acceleration engine driving LiteRT. Google tested it with the Gemma model on developer-grade workstations and claims its memory overhead is 12% lower than frameworks such as llama.cpp and MLX LM. The engine is already running in applications including Chrome, YouTube Shorts, Photos, and Meet; among them, YouTube Shorts reduced the average frame latency of segmentation-type effects on Android and iOS by up to 40%, while some AI retouching features in Adobe Lightroom and Photoshop saw performance improvements of up to 30%. Google will stop updating the legacy TensorFlow Lite GPU delegate and recommends migrating to the LiteRT ML Drift GPU accelerator.

    #Product updates#English text is being prepared.#Open-source ecosystem

Sunday18 items

  1. Hacker News TrendingProduct release
    AI score51

    farol Open-Source CLI and GitHub Action: Assessing PR Merge Risk After AI Review

    farol is an MIT-licensed open-source CLI and GitHub Action that, after AI code review, provides a merge risk conclusion, a 0–100 risk score, and the highest-risk files, and lists the pages, APIs, tasks, and scripts touched by the changed code. It performs static analysis based on the TypeScript type checker, go/types, and Python ast, without executing or externally transmitting code, and can be run via npx @winninjs/farol or winnin/farol/action; it only comments and does not block merges. The author states that across 662 historical commits in four repositories, none of the 56 commits in the fast track tier had issues, and the risk score AUC was 0.72–0.83.

    #English text is being prepared.#English text is being prepared.#English text is being prepared.

  2. Elon MuskProduct release
    AI score68

    Grok Bot can call other people or businesses on the user's behalf and send back a call summary

    Grok Bot can now call other people or businesses on the user's behalf, make requests for them, record the other party's responses, and compile the call content into a summary to send to the user. Elon Musk announced this capability in a tweet, without specifying the scope of availability or a launch date.

    Quoted postAlex · @alexmichaelio

    Grok Bot 0.77 looks INSANE 🔥 > Calls: your Bot can dial someone, sit on hold, press keys and get transferred while you watch a live transcript > Messages on Mac: a setup screen under Connect Apps, still switched off > Tailscale: Grok Bot can join your private tailnet once an admin approves it > Payments: a screen to connect payment accounts and pick a default, on top of yesterday’s Vault > Sharing: emails when someone shares a Bot with you, and invites need Premium or Ultra

    #Product updates#Speech#Agents

  4. MarkTechPostProduct release
    AI score68

    Anthropic Launches Free Open Source Vulnerability Scanner OSS Scanner

    Anthropic has released OSS Scanner, which uses its most powerful models, including Claude Mythos, to freely audit the codebases of critical open source projects, with reports sent directly to maintainers without human review. As of October 19, 114 projects had been selected from approximately 650 applications, and in 11 days since launch, 77 critical and 803 high-severity findings were reported. Anthropic penetration testers reviewed 97 critical and high-severity findings across 48 projects; 85 (88%) met its vulnerability disclosure standards, 1 was invalid, and 11 were duplicates of known issues.

    #Product updates#Safety / alignment#Open-source ecosystem

  5. The DecoderProduct release
    AI score68

    Nvidia open-sources robot video annotation engine Flora, claims cost is 125 times lower than GPT-6 Astra

    Nvidia has open-sourced the data engine Flora for generating three types of annotations for robot videos: action descriptions, step-by-step scoring, and task success/failure judgments. The models are now available on Hugging Face. Flora consists of three fine-tuned Qwen3.5 models. Nvidia says it is already running in GR00T model training, annotating up to 200,000 hours of video per day on 512 GB300 GPUs. According to Nvidia's estimates, annotating one hour of video costs about $0.72, compared with about $88.5 for GPT-6 Astra, but this cost is an extrapolation based on a rental price of $10 per GPU per hour rather than a measured result; on WGO-Bench, Flora performs best in action boundary detection, but its overall score still lags behind Astra.

    #English text is being prepared.#English text is being prepared.#Embodied AI

  6. English text is being prepared.Product release
    AI score58

    Alibaba Qwen Go! Launches "My Apps" Feature, Enabling No-Code Building and Sharing of Custom Apps

    Alibaba's AI agent "Qwen Go!" has launched a "My Apps" feature. Users can build apps for life, work, and study scenarios without coding simply by stating their needs, and the generated apps support external sharing and continuous data updates. This feature is integrated with Alibaba Cloud's AI application production platform Meoo (Miaowu), which can automatically complete app building, database configuration, and deployment. Combined with cloud Pro capabilities, after authorization it can access users' personal data on platforms such as Taobao, Amap, Damai, and Quark Drive. Currently, "My Apps" and cloud Pro are both in the invitation-only testing phase, and users can queue up to apply on the invitation testing page within the Qwen App.

    Same story · 3 reports

    #Product updates#Agents#English text is being prepared.

  7. Phoenix TechnologyProduct release
    AI score59

    Tesla confirms it is developing its own AI maps, to be rolled out gradually with FSD v15

    Tesla Vice President Ashok Elluswamy confirmed on the earnings call that Tesla is developing its own AI-based map navigation system, which will launch with FSD v15, integrating data collected by the fleet, personalized route planning, and support for FSD voice control. He called it an entirely new architecture rather than another version of maps; the fleet will automatically collect street information and annotate data such as travel speeds, stop signs, traffic lights, and parking lots, to solve issues with speed limit recognition, parking, and route planning. The new maps are expected to be used first during the trip departure and arrival stages of FSD v15, and then gradually cover the remaining navigation capabilities such as parking and voice control over the following months.

    #Product updates#English text is being prepared.#Speech

  8. Phoenix TechnologyProduct release
    AI score60

    Alibaba Qwen Go! Launches "My Apps" Feature, Enabling No-Code Building and Sharing of Custom Apps

    Alibaba announced that its AI agent "Qwen Go!" has launched the "My Apps" feature. Users can build apps for life, work, and study scenarios simply by stating their needs without any programming, and the generated apps support external sharing and continuous data updates. This feature is integrated with Alibaba Cloud's AI application production platform Meoo (Miaowu), which can automatically complete app building, database configuration, and deployment. Combined with cloud Pro capabilities, it can, with authorization, access users' personal data on platforms such as Taobao, Amap, Damai, and Quark Drive. Currently, "My Apps" and the cloud Pro capabilities that support operating cloud phones and cloud computers are both in invite-only testing, and users can queue up to apply on the invite-only testing page within the Qwen App.

    Same story · 3 reports

    #Product updates#Agents#English text is being prepared.

  9. Alibaba Cloud Developer Community AIProduct release
    AI score55

    Polter launches Terminal Supervisor and Control Room to solve Claude Code overnight tasks getting stuck at confirmation prompts

    Polter has released a terminal with a built-in control hub, offering two capabilities: Supervisor and Control Room, designed to solve the problem of Claude Code still getting stuck at y/n confirmation prompts and stalling all night even after Auto Mode is enabled. The Supervisor uses terminal_read to read the plain-text screens of other terminals; when a screen remains static for the default 3-minute threshold, it is woken by a low-level push notification. Routine safe operations are approved on the user's behalf, while destructive operations trigger a pop-up alert and are pushed to the to-do list. The Control Room displays overnight progress across three tabs: group chat, task board (queued, working, blocked, done), and statistics. All records are stored locally in the $XDG_STATE_HOME/polter/ directory, with no requests sent to the cloud or large models.

    #Product updates#Agents#English text is being prepared.

  10. Alibaba Cloud Developer Community AIProduct release
    AI score58

    GenOffice: Open-Source Four-Piece Office Suite with AI Desktop, 2.3k Stars

    GenOffice is an Apache-2.0 open-source project that puts Docs, Sheets, Slides, PDF, and Markdown into an Electron desktop workbench sharing a single editing engine. Its primary language is TypeScript, and it provides builds for macOS, Windows, and Linux. Its DOCX editing approach archives the original file by hash, marks only the paragraphs that have become dirty, and generates OOXML fragments to splice back into document.xml, reducing formatting damage to unmodified content. The project splits agent-core, the AI provider abstraction, and the format engine into shared packages, but AI calls require logging into a Genspark account and go through its server; model API Keys are not stored locally.

    #English text is being prepared.#Agents#English text is being prepared.

  11. MarkTechPostProduct release
    AI score80

    Microsoft Research open-sources the Agensh multi-agent framework, supporting up to 1024 coding agents

    Microsoft Research has open-sourced Agensh, a multi-agent framework that removes the central orchestrator, with each worker claiming subtasks on its own, sharing findings, and merging code into the same Git repository. On ProgramBench, the team scaled from 1 agent to 1024, raising the average pass rate on the 5 hardest tasks from 19.31% to 28.78% (with 128 agents), and the single pandoc task reached 79.59% with GPT-6.1-sol paired with 32 agents. The code is released under the MIT license, with a default of 8 workers per single node; the 1024-agent run used 16 nodes with 64 agents per node.

    #English text is being prepared.#Agents#English text is being prepared.

  12. Ollama BlogFirst-partyProduct release
    AI score71

    Ollama Launches Speech Model Phonon, Supporting Text-to-Speech

    Ollama can now convert text to speech. Its first speech model is the Phonon series of small text-to-speech models from Gradium. It is currently available on macOS, with Windows and Linux support coming soon. On a MacBook Pro equipped with M5 Max, Ollama, with the help of MLX acceleration, is 35% faster than Gradium's own runtime, and its word error rate is on par with the Hugging Face Transformers reference implementation. Phonon is compatible with OpenAI's /v1/audio/speech interface, so existing OpenAI SDK code only needs to change the base URL; phonon is released under Apache 2.0, and the 40M model uses the CC BY-NC 4.0 license and is for non-commercial use only.

    #Product updates#Speech#Open-source ecosystem