TIAOTECH / NEWS

Stay curious. Stay in the loop.

Selected industry news from AIHOT. Follow advances in AI, products and technology.

AIHOT · Selected summaries with original sources · Nov 2, 2026, 9:00 AM

Technology products · V2EX · 分享创造

Open-source WeChat Official Account Writing and Publishing Automation Toolkit ↗

Developer gnakzz has open-sourced a WeChat Official Account automation pipeline, comprising the command-line tool zzhub-pipeline (@akairos/pipeline), the pure-algorithm cover generator typeset-studio (@akairos/typeset-studio), the layout engine myno (@akairos/myno), and WeChat extension plugins. It supports Markdown input, AI draft review, adaptive rendering of dual-viewport covers, multi-account binding, and secure publishing via a proxy for WeChat official interfaces. All components are open source; the local CLI supports breakpoint resumption, and after cloud deployment, operations can be performed via mobile browsers.

Industry · AMZ123 · 跨境快讯

内蒙古破获AI人脸虚开数电票案,涉案金额18.95亿元 ↗

内蒙古霍林郭勒市公安局侦破首起利用AI动态人脸视频绕过电子税务局人脸核验、批量虚开数字化电子发票案件,涉案价税合计18.95亿元,涉及700余家空壳企业、2万余份发票及全国超5000家受票企业。

Source-language fallback
Perspectives · 虎嗅 · AI

上百座机器人训练场开始算真账 ↗

截至2026年6月底,全国已建成启用超70家具身智能训练场,另有46家在建或规划;训练场年数据产能正迈向千万小时级,但有效数据率仅30%–50%,客户高度依赖本体厂商,场景同质化严重;信通院指出行业正从建设热潮期转向能力沉淀期,未来需以真实产线数据、任务效果计价和‘训练即服务’模式破局。

Source-language fallback
Technology products · V2EX · 分享创造

WristBeam Released: Open-Source Tool to Sync Garmin Body Battery to Mac Menu Bar ↗

Independent developer kongwenyan released WristBeam, an open-source macOS tool. By reverse-engineering the BLE communication logic of the Garmin Connect APK, it enables local direct connection to the Garmin Instinct 2 Solar watch to read body battery, heart rate, stress, step count, and other data, displaying them in the menu bar and floating window. It supports local storage, export, and an optional read-only MCP interface for AI to query cached data; compatible with macOS 13+, providing installation packages for Intel and Apple Silicon, under the MIT license.

Perspectives · 虎嗅 · AI

Meta Releases Open-Source Model Muse Spark Open with 816000000000 Parameters, Supporting Local Inference on Dual Mac Studios ↗

Meta released the open-source large language model Muse Spark Open, with a total of 816000000000 parameters, activating approximately 42000000000 per inference, featuring a context window of 1000000 tokens, native multimodality, and focusing on coding and Agent tasks; after NVFP4 quantization, the model weights are approximately 493GB, requiring two Mac Studios with 512GB memory interconnected via Thunderbolt 5, utilizing the M5 Ultra unified memory architecture and the MLX framework to achieve dual-machine tensor parallelism, with a generation speed of about 25 tokens/second; Meta claims it is the most capable US open-weight model to date, but scores lower than Chinese open-source models such as Qwen3.8, GLM-5.3, and Kimi K3 in programming benchmarks Terminal-Bench 2.1 and DeepSWE 1.1; demonstrated workflows include locally modifying a Mastodon client and operating DaVinci Resolve for video editing, with all data remaining on-device.

Technology products · V2EX · 分享创造

GuanChao TideView Launch: Domestic Charting Analysis Tool for A-Share Traders ↗

The developer team launched Guanchao TideView, supporting market data viewing for A-shares, HK stocks, and futures, with Windows, Mac, and web versions available. Core features include AI-assisted writing of standard Pine v5 scripts (supporting plain-language commands, trial runs, and backtesting), masked K-line training (including T+1 and price limit simulation prompts), automatic divergence lines on main and sub-charts, and a structured review workbench. The free version includes basic market data and community scripts; the Plus version at 69 CNY/month unlocks AI scripts, backtesting, divergence, Chan Theory, and review features. It does not connect to brokers, cannot place orders, and has no investment advisory qualifications.

Technology products · V2EX · 分享创造

AllSwap Launches Cross-Chain Exchange Service with One-Time Deposit Addresses, No Wallet Connection Required ↗

AllSwap has launched a cross-chain exchange service that requires no wallet connection, registration, or identity verification upload. Users complete single-asset exchanges via one-time deposit addresses, supporting 47 chains and over 190 assets; quotes include a 0.25% service fee and network costs already factored in, with the received amount being the final net total; failed orders enter a manual refund process, and incorrect addresses or networks cannot be reversed; currently, orders for chains requiring Memo/Tag are directly rejected.

Industry · IT之家 · AI

MOFCOM Proposes Revising the "Catalogue of Technologies Prohibited or Restricted from Export by China" to Add AI and Other Technologies ↗

The Ministry of Commerce (MOFCOM) and the Ministry of Science and Technology have launched a public consultation on revising the "Catalogue of Technologies Prohibited or Restricted from Export by China," proposing to add 40 key technologies across three major fields: artificial intelligence, biomedicine, and photovoltaics. Among these, 23 are related to artificial intelligence, including malicious code generation and mutation technologies subject to export prohibition, as well as vulnerability mining, large models, agents, and computing power clusters subject to export restrictions. The revision clarifies that overseas localized deployment falls within the scope of technology export management, while API calls, SaaS applications, and edge-side deployments are excluded; automatic licensing facilitation measures apply to scenarios such as open source, joint R&D with overseas entities, and international standard-setting.

Perspectives · PANews

What TSMC's Q3 2026 Earnings Reveal About the Ramp-Up of 2nm Capacity and Surging HPC Business for Nvidia and AMD ↗

TSMC's Q3 2026 earnings report shows that revenue from the 2nm process rose from 3% to 8%, providing the first third-party confirmation of mass production for AMD MI450; HPC business revenue accounted for 68% with a quarter-over-quarter increase of 22%, serving as the most reliable current indirect indicator of AI demand; however, specific figures for CoWoS packaging capacity were not disclosed, meaning shipment ceilings for both Nvidia and AMD remain constrained by this factor.

Research · PANews

LayerTree: A Bitcoin Non-Custodial Shared UTXO Scaling Solution Based on Merkle Sum Tree ↗

Lightning Labs co-founder Olaoluwa Osuntokun proposed LayerTree, an off-chain state solution that allows millions of users to share a single Bitcoin UTXO. The core concept is using an off-chain Merkle Sum Tree to record balances, with only the tree root and shared UTXO stored on-chain; it supports exits via a single transaction without operator approval (O(1) exit), as well as two off-chain transfer methods: channel leaf (integrating Lightning channels) and voucher (cumulative digital checks redeemable on-chain); the solution relies on BIP upgrades that have not yet been activated, such as BIP 441, BIP 348, BIP 443, and the experimental OP_MERKLESUMUPDATE.

Perspectives · 虎嗅 · AI

Infinigence AI Files for IPO on HKEX in Confidential Form ↗

Infinigence AI has filed for an IPO on the HKEX in confidential form, planning to raise hundreds of millions of US dollars, with CITIC Securities, Haitong International, and UBS serving as sponsors; the company has completed a C-round financing of 2.3 billion yuan, with cumulative financing of 4.3 billion yuan and a valuation of 14.3 billion yuan; its core business has shifted from an early-stage thousand-card heterogeneous mixed-training platform to a "Front Store, Back Factory, One Center" Token service model, covering 37,000 P of computing power and adapting to 16 mainstream chips; however, audited financial data has not yet been disclosed, and the profitability model remains to be verified.

Research · 量子位

优必选 Thinker-WM 发布:统一多模态世界模型,霸榜三大具身基准 ↗

优必选发布自研多模态世界模型 Thinker-WM,在 WorldOlympiad、EWMBench 和 WorldArena1.0 Track-1 三大具身世界模型基准中全部排名第一;模型采用 Unified Dynamic Transformer(UDT)实现 RGB 视频、深度与机器人动作在共享潜在空间的联合建模,并通过 Motion Probe 实验验证视觉表征中已编码运动信息;训练使用约3万小时异构数据,包括2万小时人类视频、1万小时真实机器人数据和1千小时仿真数据;模型已部署于优必选机器人平台,开源地址见 GitHub 和 Hugging Face,论文发布于 arXiv。

Source-language fallback
Perspectives · 虎嗅 · AI

苹果与Meta和解:Muse Spark Open支持在Mac Studio集群本地运行 ↗

Meta开源大模型Muse Spark Open(8160亿参数,激活约420亿)将支持在搭载M5 Ultra的Mac Studio集群上本地运行,依赖苹果MLX框架与多机协同实现推理;此举标志AI竞争单位正从单一模型转向模型、芯片、框架与场景的组合交付。

Source-language fallback
Perspectives · 新智元

从“看懂世界”到“预演下一步”:上海具身智能产业中VLA与世界模型的并行实践与融合探索 ↗

新智元报道上海张江及浦东多家企业(如蚂蚁灵波、智元、擎朗、大晓、穹彻、灵初、它石智航、上海人工智能实验室等)在VLA和世界模型两条技术路线上同步推进具身智能落地:VLA已在药房、汽车零部件产线等场景规模化上岗;世界模型处于试运营或嵌入式融合阶段,代表模型包括LingBot-VA 2.0、GE-Act 2.0、开悟世界模型3.1、InternW0-Δ等;《浦东新区“十五五”规划》《国家人形机器人产业标准体系建设指南(2026版)(征求意见稿)》《上海市人工智能标准体系建设指南(2026—2028年)》均已将世界模型列为关键方向与标准建设重点。

Source-language fallback
Technology products · V2EX · 分享创造

Independent developer releases desktop tool Folkbench Switch for switching API providers between Claude Code and Codex ↗

To address issues such as lost login states, 502 interruptions, and high onboarding costs when frequently switching between official and third-party APIs for Claude Code and Codex, xgguest developed the desktop utility Folkbench Switch using Tauri + Rust. The tool provides a simple interface for model selection, provider switching, API Key entry, and configuration writing, with installers published on GitHub.

Perspectives · 虎嗅 · AI

Guangyin Technology Advances Industrialization of Optical Sensing, Optical Switching, and Optical Computing in Phases ↗

Guangyin Technology has chosen to first deploy optical sensing (having secured trial mass-production orders from a leading baijiu company and clinical cooperation with an international beauty giant), then move into R&D of nanosecond-level optical circuit switching (OCS) full-machine solutions, targeting small-batch shipments by 2028; optical computing is maintained only as a technical reserve. Founder Cheng Qixiang emphasizes that technological advancement does not equal market advancement. The opportunity for OCS arises because electrical switching latency becomes a bottleneck in AI cluster scale-up scenarios, while optical computing is constrained by integration levels and optoelectronic conversion overheads, making it difficult to replace general-purpose computing in the short term.

Industry · IT之家 · AI

Danish Supreme Court Rules AI-Generated Child Pornography Also Carries Prison Sentence ↗

The Danish Supreme Court upheld an 18-month prison sentence for a 31-year-old man who possessed and distributed over 36,000 AI-generated child pornography images; the ruling clarified that even if the individuals in the images are fictional, they remain subject to the Criminal Code, and probation is not applicable; this case marks Denmark's first conviction under such grounds, with a lower court previously sentencing him to 15 months of probation, which was overturned by the High Court to an 18-month actual prison term.

Practical tips · V2EX · 分享创造

LibraryCoverStudio Open-Source Release: Cover Design and Scheduled Update Tool for Jellyfin / Emby Media Libraries ↗

EvanQu has open-sourced LibraryCoverStudio, a tool for automatically generating and periodically updating category covers (e.g., movies, TV shows) for Jellyfin / Emby media libraries. It supports template editing, poster sourcing, preview rendering, and automatic API uploads. The project offers a Docker Compose deployment solution, an administrator registration process, media server connection configuration, 7 built-in templates, and Cron scheduled task settings, explicitly stating that it does not rely on AI.

AI products · V2EX · 分享创造

SoloMCN Open Source Update: Teaching Claude Code to Create Animations and Automatically Distill Them into Maintainable Skills ↗

SoloMCN is an open-source "One-Person MCN" tool. This update focuses on automatically distilling the animation production workflows executed by Claude Code (research, storyboarding, voiceover, rendering) into reusable, version-controlled skills. It supports draft comparison, manual review for activation, and three-way merge updates. Skills are categorized by workflow, method, or tool, and can be invoked directly in Claude Code via commands such as /explainer-video. A Claude subscription is required, with a one-line command installation for macOS.

AI models · IT之家 · AI

PrismML Releases Ternary Bonsai 2 Series Models: Lightweight Ternary Quantized Models Based on Qwen 3.5 ↗

PrismML has released the Ternary Bonsai 2 series of models, including three sizes: 2B, 4B, and 9B. Based on Alibaba's Qwen 3.5 foundation, these models use a -1/0/+1 ternary weight representation, with storage footprints of 0.43GB, 0.95GB, and 2GB respectively. They support local execution on devices such as the iPhone 18 Pro Max, M5 Pro, and RTX 4090; notably, the 2B model achieves up to 120 tokens/s on the iPhone 18 Pro Max. The models are open-sourced under the Apache 2.0 license and can be deployed via llama.cpp and MLX.

Industry · 量子位

Shenchen Kaidong Badminton Robot Sets Guinness World Record with 1452 Consecutive Shots ↗

On December 22, 2025, a badminton robot developed by Shenchen Kaidong achieved 1452 consecutive rallies against a human player in Shaoxing, earning Guinness certification for the category "Most consecutive badminton returns by a mobile robot." Based on its proprietary general-purpose physical motion platform and robot brain, the robot features 0.5-millisecond recognition, a swing speed of 30 m/s, and smash capability at 250 km/h. The company claims full-stack independent R&D and holds over one hundred patents. Its technical path focuses on the perception-prediction-decision-execution closed loop in the real physical world, extending to scenarios such as logistics sorting and industrial flexible operations.

AI products · V2EX · 分享创造

Baocode: A VS Code-style IDE ported to Flutter, integrated with Claude Code ↗

leokun released his self-developed IDE Baocode, which fully ports the Monaco editor and xterm terminal using Flutter. The package size is only 15MB. It supports remote SSH and calls server-side Claude Code, providing custom API interfaces without requiring modifications to Claude Code configurations. The open-source address is https://github.com/baocode-dev/baocode.

Industry · 虎嗅 · AI

Tesla Optimus Mass Production Factory Audit Lands in Ningbo, Auto Parts Supply Chain Shifts to Humanoid Robots ↗

In mid-September 2025, Tesla launched mass production factory audits for its Optimus humanoid robots in Ningbo and has placed orders with local supply chain companies. With 80% of component categories overlapping with the new energy vehicle industry chain, nearly 5,000 auto parts enterprises, and a foundation of 6 companies listed among the top 30 robot component firms by The Economist, Ningbo has become a key supply chain node. Companies such as Tuopu Group and Huaxiang Qiyuan have begun mass producing robot components, but their robot business revenue share remains extremely low in 2025 (Tuopu at 0.05%, Huaxiang at 24.97 million yuan), indicating that the sector is still in an early ramp-up phase.