NAll news

All news

Tuesday598 items

  2. Zhidongxi
    AI score55

    Jianzhi Robotics Completes $100 Million Financing, Quarterly Data Delivery Rises to One Million Hours

    Beijing-based data and evaluation infrastructure company Jianzhi Robotics announced the completion of a new round of $100 million (approximately RMB 665 million) in financing, with investors including SAIC Motor, Baidu Group, Shoucheng Holdings, Guanghe Venture Capital, and CITIC Jinshi Investment. Founded in May 2025, the company focuses on multimodal data collection and governance, has launched the Gen DAS Dex collection terminal, and plans to release Ego 2.0 and Dex 2.0 under DAS 2.0 during CoRL. Its effective data delivery volume has risen from 100,000 hours per quarter to one million hours per quarter, and it is expected to achieve quarterly delivery of 10 million hours by 2027.

    Same story · 3 reports

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

  3. Zhidongxi
    AI score10

    2026 China AI Agent Conference Beijing Stop Announces RSI Workshop Speaker Lineup

    The 2026 China AI Agent Conference (Beijing Stop) will be held on December 15-16. The RSI Technology Workshop is scheduled for the afternoon of December 16, led by Li Yuxuan, Chief Scientist of Intelligent Evolution at ModelBest, with Evolvent AI founder Hu Mengkang, LlamaFactory author Zheng Yaowei, and ByteDance Jindouyun research intern Wu Yin participating in the presentations. The conference features a total of eight technical workshops, and researchers from institutions including Tencent Hunyuan, Alibaba Amap, and ByteDance Seed have confirmed their attendance.

    #Industry news#Agents

  4. Zhidongxi
    AI score47

    Arm Technology's Jishu Community Releases Robot Dog Puppy and Development Tool JishuBuddy

    The open source AIOS Alliance initiated by Arm Technology released two Jishu Community achievements at WTCC 2026 in Shanghai: JishuFriends' first prototype, the Puppy robot dog, and JishuBuddy, a hardware development tool for Agentic AI. Puppy features environmental perception, voice and visual interaction, autonomous movement, and long-term memory capabilities, and can recognize family members and remember their routines and life moments. JishuBuddy uses natural language as its interaction entry point, supports SSH and USB/UART connections to Arm ecosystem hardware, and works with JishuFriends to complete simulation and physical hardware debugging.

    #Product updates#Embodied AI#Agents

  6. ZhidongxiModel releases
    AI score78

    Meituan Releases LongCat-2.5: A 1.6-Trillion-Parameter Native Multimodal Model Strengthening Long-Horizon Agent Capabilities

    The Meituan LongCat team has released LongCat-2.5, a native multimodal model with a total parameter count of 1.6 trillion and approximately 48 billion parameters activated per Token. Building on LongCat-2.0, it strengthens native visual understanding and agentic coding capabilities. In benchmarks published by Meituan such as AutomationBench and LongCatClawBench, LongCat-2.5 leads Claude-Opus-5 and GPT-5.6-Sol, but on Terminal Bench v4.0 it scores 38.5, surpassing GPT-5.6-Sol while still trailing Claude-Opus-5. In response to domestic accelerators having less than 80GB of memory per card, the team reorganized parallel strategies around supernodes, claiming that single-card throughput on long-sequence SFT tasks improved by 86% over the DDP baseline.

    Same story · 11 reports

    #Model releases#Agents#Multimodal

  7. Zhidx
    AI score60

    Phase 1 of Moore Threads' Wuxi domestic intelligent computing center is running at full capacity with 2,112 GPUs

    Moore Threads has entered into a strategic partnership with the Wuxi Huishan Investment Group and Neolix. Neolix plans to expand autonomous driving model training at the Wuxi (Huishan) domestic intelligent computing center, with computing power demand expected to reach nearly ten thousand cards. The center's Phase 1, comprising 2,112 MTT S5000 cards, is now fully operational, with a computing power utilization rate of 92.9% and paid computing power accounting for 59.7%. It has already engaged 55 customers, with training workloads accounting for over 62%. The center is planned at a scale of ten thousand cards; Phase 1 built two thousand cards first for customer adaptation and validation, before advancing Phase 2 and Phase 3 expansion.

    #Industry news#English text is being prepared.#English text is being prepared.

  8. Zhidongxi
    AI score49

    Dahua Technology subsidiary Huaqi Smart wins China Mobile's 2.63 billion yuan AI super node centralized procurement

    Dahua Technology subsidiary Huaqi Smart has won China Mobile's centralized procurement of next-generation artificial intelligence super node equipment for 2026 to 2027, with an amount of approximately 2.63 billion yuan and a proposed winning bid of 4,096 cards. The project procures a total of 20,480 cards of AI super nodes as well as 448 RoCE switches and 82 sets of storage, with the total amount expected to be 11.6 billion to 11.7 billion yuan. Huakun Zhenyu, Baode Computer, Baixin Information, and Shenzhou Kuntai were also shortlisted at the same time.

    Same story · 2 reports

    #Industry news#English text is being prepared.

  10. Muse
    AI score48

    Muse sues on behalf of user and reaches settlement

    Phillip Jackson said he hired @Muse to sue on his behalf; the other party offered a settlement within 20 minutes of receiving the documents and paid the compensation plus additional recovery costs within ten days. The court accepted the motion to dismiss filed by Muse yesterday.

    Quoted postPhillip Jackson · @philwinkle

    UPDATE: I hired @Muse to sue someone on my behalf. I won. Well, they caved. Within 20 minutes of being served papers, the company offered a settlement. Within ten days, they paid me *with additional compensation* for recovery expenses. @Muse filed the motion to dismiss, which the court accepted yesterday. To everyone who said I would fail, respectfully, you were wrong.

    #English text is being prepared.

  11. Testing CatalogProduct release
    AI score58

    Hume AI releases the Hume Decision API, using an audio-native model to answer questions about recordings

    Hume AI launches the Hume Decision API, an audio-native model that can directly analyze the recording itself to answer questions about the recording. Users can customize questions and options, for example, determining whether a caller appears frustrated, and obtain an estimate of how human listeners would answer. In preliminary tests, the model's agreement rate with the review panel's average answers was 70.7%, compared with 56.5% for an individual human listener.

    Quoted postHume AI · @hume_ai

    Introducing the Hume Audio Decision API. Ask anything: How many speakers? Is anyone laughing? Is the caller frustrated? What used to take 10+ models is now one call. Trained on 10M+ human judgments, so it answers the way human listeners would. Realtime, or across hours of recordings. Request access below.

    Same story · 2 reports

    #Product updates#Speech#Evaluations / benchmarks

  13. OpenAIFirst-party
    AI score42

    OpenAI launches the LifeVal everyday tasks benchmark

    OpenAI says that for AGI to benefit all of humanity, it should help people handle the everyday matters they value most, so it built the LifeVal benchmark. The benchmark was designed based on feedback from thousands of ChatGPT users and covers scenarios such as finding childcare, accessing government services, and managing personal finances. Over the past year, OpenAI has used LifeVal to measure and improve models' ability to handle these everyday tasks.

    Same story · 4 reports

    #Evaluations / benchmarks#OpenAI

  14. Hacker News TrendingOpinion
    AI score60

    Author announces abandonment of Ubuntu due to delayed patching of critical kernel vulnerability

    The blogger decided to migrate all servers away from Ubuntu because Ubuntu was slow to patch CVE-2026-98432 (CVSS 9.8) in the Linux kernel TCP code. The vulnerability has existed since 2016 and can be exploited remotely to gain root privileges; Anthropic stated it had built working exploit code; Debian completed patching on October 16, while all supported Ubuntu versions (including LTS) were still marked as vulnerable nine days after the vulnerability was rated. The author points out that volunteer-maintained distributions such as Debian, Arch, and Alpine have all released fixes, while Canonical, which sells security subscription services, has not yet provided a patch, so he took the affected servers offline.

    #Safety / alignment#Open-source ecosystem#English text is being prepared.

  15. Hacker News Trending
    AI score33

    Momentic: AI QA Engineer Mo Has Caught 117,010 Bugs

    Momentic has launched Mo, an AI QA engineer that uses prompts to test applications end-to-end and verify critical flows. Official data shows it has caught 117,010 bugs, executed 70,607,819 test runs, and validated 80,290 PRs. Tests are written in plain-English YAML; when the UI changes, it automatically repairs them and submits test file changes for review. It also supports MCP, allowing coding agents to write and check tests. The CLI can be used in CI, using exit codes as a required check before merging or deploying.

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

  18. Rohan Paul
    AI score23

    Sam Altman: Energy is AI's biggest challenge; atoms are still atoms

    Speaking at the opening ceremony of ADIPEC in Abu Dhabi, Sam Altman said that energy is AI's biggest challenge, because "atoms are still atoms." He believes that more efficient models, robots that can boost chip fab efficiency and assemble racks will drive major progress in other areas, but the physical world on which power supply depends is hard to change, and this will become the dominant factor determining AI's cost, abundance, and the scale of what can be created.

    Same story · 3 reports

    #English text is being prepared.#OpenAI

  19. heise onlineProduct release
    AI score62

    Anthropic Tightens Claude Code Rules, Prohibits Cloning Existing Games

    Anthropic added two instructions to its coding assistant Claude Code targeting Opus 5.5, preventing it from cloning specific existing games, while allowing original development based on known genres or the user's own assets. An Anthropic spokesperson confirmed the change to Game Developer, citing developer Addy Osmani's explanation on X, saying the basis is the terms of use prohibiting infringement of others' intellectual property. The report also mentioned that some developers report Claude Code wrongly blocking their own projects involving game mechanics, anti-cheat software, and reverse engineering, and Anthropic said it will continue to adjust the rule based on feedback.

    Same story · 13 reports

    #Safety / alignment#English text is being prepared.#Anthropic

  20. heise online
    AI score24

    German Chancellor Merz: Google Cloud is sufficiently sovereign for intelligence agencies

    At a digital summit, German Chancellor Merz said that Google Cloud is sufficiently sovereign for German intelligence agencies, and that European sovereignty can also be achieved through U.S. suppliers investing in Germany and Europe. He said the offline cloud to be used by the intelligence agencies is not a fully integrated infrastructure, but "European infrastructure," and expressed high confidence in its data security. Previously, German and European institutions had been pushing to replace U.S. hyperscale cloud providers with local vendors.

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