NAll news

All news

Monday15 items

  1. LeikejiProduct release
    AI score57

    Zhuoyu Technology Releases Pansing Intelligent Assisted Driving Solution for Commercial Vehicles

    Zhuoyu Technology released its commercial-vehicle intelligent assisted driving solution, Zhuoyu Pansing, in Zhuhai. It is divided into two versions, PDS 4 and PDS 5. PDS 4 supports highway NOA and has already entered mass production across multiple brands, while PDS 5 corresponds to a native multimodal foundation model and will support urban NOA. Pansing has partnered with six commercial heavy-truck brands—XCMG Automobile, Shaanxi Automobile Group, Sinotruk, FAW Jiefang, Foton Motor, and Dongfeng Commercial Vehicle—which together account for more than 93% of China's commercial heavy-truck market. Zhuoyu says its AI reinforcement-learning fuel-saving algorithm achieved a measured fuel-saving rate of 4.5%–7.5% in tracking tests covering more than 200 mass-produced vehicles and cumulative mileage of over 20 million kilometers.

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

  2. Sina TechProduct release
    AI score58

    Daxiao Robotics and Nanyang Technological University S-Lab Release World Model Mundo

    Xuhui embodied intelligence company Daxiao Robotics, in collaboration with Nanyang Technological University S-Lab, released the world model Mundo, which requires only 1 to 12 real photographs or short videos to reconstruct indoor 3D spaces such as offices and home rooms, and supports free roaming and physical collision interaction. The model can write physical parameters such as object mass and friction into the scene; users can pull open drawers, push tables and chairs, and throw objects through code, and the physics engine calculates motion and collision results. The system can also synchronously output the robot's first-person and third-person views and retain motion trajectories and interaction data. Mundo is a supplement to Daxiao Robotics' "Kairos" world model system. The company has attempted to reconstruct a hotel floor environment, simulating tasks such as opening doors and picking up and placing objects to carry out robot simulation training. The relevant project page and technical blog have been launched.

    Same story · 4 reports

    #Model releases#Embodied AI#Multimodal

  3. Ayi AI NotesProduct release
    AI score73

    Karpathy built the HackerTube video site with Claude Opus 5.5 in two hours

    Over the weekend, Karpathy used Claude Code together with Opus 5.5 to filter 43 million comments spanning 20 years of Hacker News down to 6,454 frequently recommended long videos, and built HackerTube, which requires no login, in about two hours. The site's cover images are taken directly from the middle frame of each video rather than the optimized thumbnails uploaded by creators, and browsing state is kept in the local browser session. He only manually applied for a YouTube API key from Google Cloud, rented a VPS, and bought a domain in the browser; after that he handed the IP to Claude and let it handle deployment, security hardening, and code optimization.

    Quoted postAndrej Karpathy · @karpathy

    Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: https://t.co/JIhXN7OSh1 - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.

    Same story · 5 reports

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

  4. 36Kr · AIProduct release
    AI score62

    SenseTime open-sources SenseNova-RoboRSI framework and 27B embodied model

    The SenseTime SenseNova joint research team published a technical report on arXiv, evaluating the embodied capabilities of GPT-6-Astra across 12 simulation benchmarks and 8 real robot platforms, while open-sourcing the SenseNova-RoboRSI execution framework and its accompanying 27B embodied model, claiming to reduce task completion costs to 1/50 of directly invoking GPT-6-Astra. According to SenseTime, RoboRSI-27B improved its success rate on RoboCasa365 from 26.0% to 46.5% while reducing token usage by 73.3%; these results come from its own evaluation. The article notes that the 1/50 figure measures cost compression for specific tasks such as stacking bowls and loading pens under extended-budget conditions, and awaits third-party retesting; long-horizon tasks and fine-grained control of real robots remain weak points.

    #Papers/Research#Embodied AI#Open-source ecosystem

  5. indigoProduct release
    AI score62

    Karpathy built HackerTube with Claude, filtering 6,454 recommended videos from 43 million HN comments

    Andrej Karpathy built the weekend project HackerTube with Claude Code and Opus 5.5, filtering 6,454 consistently recommended videos from 43 million comments across 20 years of HN, taking about 2 hours. The site requires no login, keeps state in the local browser session, and uses a representative frame from the middle of each video as the thumbnail. He said that aside from applying for a Google Cloud API key, setting up a VPS, and buying a domain, all other updates, hardening, deployment, and optimization were left to Claude, and the comment analysis data was packaged into a zip file so Claude could use it to generate its own version.

    Quoted postAndrej Karpathy · @karpathy

    Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: https://t.co/JIhXN7OSh1 - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs ✊ A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.

    Same story · 5 reports

    #English text is being prepared.#Agents#Anthropic

  6. TMTPost · AGIProduct release
    AI score71

    DeepSeek Harness team releases Super-agent architecture

    The DeepSeek Harness team released the Super-agent architecture on its official WeChat account, adopting a Router/Worker two-tier design. The Router stays online permanently, uniformly receiving group chat, GitHub, and email messages and maintaining global context, while Workers isolate context and execute in parallel. The team said the architecture proposes no new theory but instead brings mature systems ideas such as scheduling, caching, and compute-storage separation into the Agent system. The accompanying prototype is open-sourced under the MIT license, but the team stressed that it is not a formal product. In tests with about 300 messages and 20 work requests, task completion time was 12.4 minutes at p50 and 15.7 minutes at p95; a single-session approach with Subagent tools had a p95 of 29.5 minutes under neutral prompts.

    #Industry news#Agents#DeepSeek

  7. English text is being prepared.Product release
    AI score51

    Subvocal silent-interaction AI glasses open for pre-order at an early-bird price of $499

    The Subvocal Company, a San Francisco-based startup, has announced that its Subvocal silent-input smart glasses are open for pre-order at an early-bird price of $499, requiring a refundable deposit of $199. The first batch is expected to be delivered in the first quarter of 2027, and it is currently available only in the U.S. market. Users do not need to speak aloud; text is generated solely from the subtle speech-muscle activity that occurs when silently mouthing words, for use in sending messages, taking notes, and interacting with AI assistants. The company claims that the prototype device has achieved a silent input speed of more than 200 English words per minute, but this figure comes from internal company testing and has not yet been independently verified.

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

  8. Electronic Technology ApplicationProduct release
    AI score52

    Beijing Space Computing Innovation Center Releases Space-Ground Integrated Space Computing Cloud Platform

    The Beijing Space Computing Innovation Center has released a space-ground integrated space computing cloud platform, which has completed core function development and covers computing resources of more than 40 satellites. The platform achieves unified management and pooled scheduling with ground-based computing power such as CPU, GPU, and NPU, and supports space-ground integrated collaborative scheduling for large-model inference services. A CITIC Securities research report believes that the timeliness of satellite remote sensing data has improved from the hour level to the ten-minute level, and the accelerated networking of remote sensing constellations is driving the realization of fundamentals across the industry chain.

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

  9. Aravind SrinivasProduct release
    AI score58

    Perplexity Computer Launches Interactive Globe of Earth's 4.5-Billion-Year Evolution

    Perplexity's Computer built an interactive globe of Earth's 4.5-billion-year evolution based on 14 open datasets, with its own engine driving 258 plates across 109 paleo-elevation maps. Users can drag the timeline or enter a city name to see where it sat 400 million years ago. Aravind Srinivas reposted the release with the caption "Computer, visualize earth across 400M years".

    Quoted postComputer · @AskPerplexity

    Computer built an interactive globe of Earth's 4.5 billion years from 14 open datasets, with its own engine moving 258 plates across 109 paleo-elevation maps. Drag the timeline, or type a city to see where it sat 400 million years ago. https://t.co/OAyogdWhBc https://t.co/ZFsREdM8Kn

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

  10. Phoenix TechnologyProduct release
    AI score60

    OpenAI announces that ChatGPT and Codex will share a quota, effective for Plus, Pro, and Business starting in December

    OpenAI has sent a notice to users stating that, as ChatGPT's chat and work features are integrated, ChatGPT and Codex will share a single quota within the coming weeks, with the change applying to Plus, Pro, and Business plans starting in December. Quota consumption depends on the selected model and task complexity; using instant-mode chat does not consume the shared quota. Users can view their combined remaining quota and reset time on the usage page in settings, and after the quota is exhausted they can wait for the reset, upgrade their plan, or purchase additional quota. OpenAI also plans to integrate ChatGPT's chat and work tabs, testing first with a small group of users, and Codex will still retain a standalone experience in the desktop app.

    Same story · 8 reports

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

  11. East MoneyProduct release
    AI score58

    DeepSeek Harness Team Releases Super-agent Architecture

    On the evening of October 31, the DeepSeek Harness team published a technical article titled "Super-agent Design Exploration: Toward a More Autonomous Agent System," proposing to rebuild the Agent system around five major modules—a two-layer architecture, event-driven design, asynchronous multi-channel interaction, storage-compute separation, and an external Watchdog—summarized as four characteristics: continuous, efficient, proactive, and persistent. In synthetic load tests with about 300 messages per run, of which about 20 were work requests, the Super-agent's median and 95th-percentile task completion times were 12.4 minutes and 15.7 minutes, respectively, while the single-session approach under neutral prompts had a 95th-percentile time of 29.5 minutes. In the Router-layer stress test, the event arrival rate rose from about 1 per minute to about 40 per minute, and the inference time apportioned to a single decision dropped from 6.4 seconds to 1.6 seconds.

    Same story · 16 reports

    #Product updates#Agents#DeepSeek

  12. MiniMaxFirst-partyProduct release
    AI score66

    MiniMax M3.1 Flash Gains Support from Multiple Community Quantization Schemes for Local Running

    MiniMax officially summarized the various quantization and local inference schemes that the community has provided for the 428B M3.1 Flash, including UnslothAI's 2-bit Dynamic GGUF weights (143.9 GB), SAGE-EXL3, Project Maya's expert offloading, EXL3-EAQ, oQ quantization and oMLX decoding on Apple Silicon, Quark MXFP4 adapted for AMD Instinct, and an inference implementation on 4× RTX 3090 that combines quantized weights with expert caching. The official statement says that low-bit weights can reduce the memory required to load the model and the weight data moved during decoding, but they require careful calibration, efficient low-bit kernels, and expert placement across GPUs and system memory.

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

  13. Elon MuskProduct release
    AI score60

    Elon Musk Says Grok Bot Will Research and Buy Gifts on Your Behalf

    Elon Musk said Grok @Bot will research gifts that friends and family like and buy them directly. The post cited Justine Moore's experience: she let Grok @bot use up to $100 from her X Money account, combining her tweets and conversations to pick gifts; the gifts have arrived and included a Halloween surprise.

    Quoted postJustine Moore · @venturetwins

    Last week, I told my Grok @bot to surprise me with a gift. I said it could use up to $100 from my X Money account, and asked it to consider my tweets + our conversations to pick a good present. Look what arrived today! Even included a Halloween surprise 🎃 https://t.co/HP7IU1D2He

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

  14. Andrej KarpathyProduct release
    AI score59

    Karpathy built HackerTube with Claude, filtering 6,454 videos from 43 million HN comments

    Karpathy released the weekend project HackerTube, using Claude to filter 6,454 repeatedly recommended videos from 43 million comments spanning 20 years of HN. The entire project was completed in about 2 hours using Opus 5.5. The site provides recommendations without requiring login, thumbnails are taken from a representative frame in the middle of the video, and the comment analysis data is packaged as a zip file that can be handed to Claude to generate your own version. He used the Mac desktop version of Claude Code and manually did only three things: applying for a Google Cloud API key, starting a VPS, and purchasing a domain. After that, he had Claude handle updates, security hardening, deployment, and code optimization.

    Same story · 5 reports

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

  15. PublickeyProduct release
    AI score62

    GitHub officially releases Copilot Fusion, automatically combining multiple AI models based on the task

    At GitHub Universe 2026, GitHub announced the official release of GitHub Copilot Fusion, a feature that selects a single AI model or a combination of multiple models to execute based on task characteristics. The original article presents three collaboration modes, one of which has a low-cost model draft a plan, a quality-check model verify it and decide whether to escalate to a stronger model, while another has models from different model families independently review the draft and lets the drafting model make one revision. The feature offers three load levels—Fusion Standard, Fusion Hight, and Fusion Max—to balance quality and cost, plus a Fusion Fast that users cannot specify; Fusion Standard also supports locally installed MAI Code 1.1 Flash or Qwen3.8 Flash Next models to execute tasks, in public preview.

    Same story · 3 reports

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

Sunday8 items

  1. Sina TechnologyProduct release
    AI score59

    OpenAI Announces ChatGPT and Codex Will Share Quota Starting in December

    OpenAI has released an announcement stating that it will adjust the usage quota rules for ChatGPT and Codex in December of this year. The quotas for the three paid plans—ChatGPT Plus, Pro, and Business—will be calculated jointly rather than billed separately. After the adjustment, quotas will be calculated uniformly based on tokens, with consumption depending on the selected model and task complexity; Instant chat will not consume the shared quota. Users can view their remaining merged quota and reset time on the "Settings—Usage" page in ChatGPT.

    Same story · 8 reports

    #Product updates#OpenAI#Agents

  2. QbitAIProduct release
    AI score65

    TypeSafe AI and Treg Open-Source a People-Finding Tool That Is 85% Cheaper Than Clay

    TypeSafe AI has teamed up with the agent tool platform Treg to open-source a people-finding tool that it says is more than 85% cheaper than Clay and has already gained 6,400 stars. After users connect Treg to an Agent and describe the target person, the Agent breaks down criteria such as region, funding stage, and job title to query a people database, and then Jev determines whether a candidate matches and fills in LinkedIn information and a verified email address. In a comparison test of 292 people, Treg found 289 email addresses, 264 of which matched the companies' publicly listed addresses, at a cost of $1.49; Clay found 280, with 262 matches, at a cost of $10.34. Looking up a single email address costs as little as $0.0089, with no charge if nothing is found, while email verification and résumé completion are billed separately.

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

  3. cnBetaProduct release
    AI score60

    OpenAI Announces That Some ChatGPT Features Will Share a Quota with Codex

    OpenAI has announced that some ChatGPT features will share a quota with Codex, measured uniformly in Tokens, with the actual consumption rate depending on the selected model and task complexity. Instant conversations do not use this shared quota, and paid users who use both services will have their quotas affect each other. Users can view the combined remaining quota and the next reset time on the usage page in ChatGPT settings; once the quota is exhausted, they can wait for automatic reset, upgrade their subscription, or purchase credits.

    Same story · 8 reports

    #Product updates#OpenAI

  4. Shao MengProduct release
    AI score72

    Karpathy Built HackerTube in Two Hours with Claude Opus 5.5

    Andrej Karpathy used Claude Opus 5.5 (Claude Code) to filter YouTube videos recommended over the past two decades from Hacker News comments and built HackerTube, a YouTube look-alike. About 40,000 recommendation comments were consolidated into 6,454 videos spanning 2007–2026 and covering 150 topics, all in roughly two hours. By hand, he did only three things: apply for a Google API key, spin up a VPS, and buy a domain; the rest was done by conversing with Claude. Once the VPS was up, he handed over the IP and let it update, deploy, harden, and optimize on its own. The site's thumbnails use the 50% progress frame, For You requires no login and stores state in the browser's local session, and the data can be exported for your own agent to make further variants.

    Same story · 5 reports

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

  5. English text is being prepared.Product release
    AI score60

    OpenAI: Starting in December, ChatGPT and Codex Will Share a Unified Usage Allowance

    OpenAI has published an announcement stating that starting in December, users of the Plus, Pro, and Business subscription plans will share a unified usage allowance between ChatGPT and Codex, against the backdrop of bringing Work's agent capabilities into the main ChatGPT experience. Usage in both places is measured uniformly in Tokens, and the rate of consumption depends on the selected model and task complexity; Instant Chat does not consume the allowance. Users can view their combined remaining allowance and next reset time in Settings → Usage; after the allowance is exhausted, they can wait for automatic reset, upgrade their subscription, or purchase additional Credits.

    Same story · 8 reports

    #Product updates#OpenAI#Agents