NAll news

All news

Friday1456 items

  1. Rohan PaulProduct release
    AI score58

    Joy releases a 100,000-hour dual-track real conversation dataset covering 10 languages

    Shobhit Banga releases Joy, a dual-track conversation dataset containing 100,000 hours of real, unscripted two-person conversations, with each person on a separate audio track, covering 50,000 speakers and 10 languages. Rohan Paul points out that current speech AI is mainly trained on fake conversations written by LLMs and read aloud by TTS, which cannot capture people talking over each other, and he is curious about how these conversations are sustained for 10 to 20 minutes.

    Quoted postShobhit Banga · @shobhitbanga

    Drop 1/15: Introducing Joy, a dual-channel conversational dataset. Real, unscripted conversations between two people, each on their own track. Built to teach voice models how people actually talk: when to listen, when to jump in, when to say “hmm”. 100,000 hours. 50,000 speakers. 10 languages.

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

  3. Ma Dongxi NLP
    AI score42

    How Qwen3-0.6B generates a response token by token

    Taking Qwen3-0.6B as an example, a single response goes through six steps: chat template, tokenize, prefill, selection, decode, and detokenize; the model has 596 million weights and a vocabulary of 151,936 entries, and generating each token requires a full run that scores the entire vocabulary. Prefill processes the entire prompt at once and fills the KV cache, while decode processes only one new token at a time until an end token appears. Temperature, top-p, and greedy decoding or sampling only determine how tokens are selected from the logits and do not involve the weights.

    #Tutorials / practice#Reasoning#English text is being prepared.

  5. Shao MengOpinion
    AI score63

    How the Amp team operates with 21 people and Agents replacing processes

    English text is being prepared.

    Quoted postAmp · @AmpCode

    21 people, a single meeting a week, no quarterly goals, no pull requests, no mandatory code review, no titles (except one dictator), a hundred releases a day, bugs fixed in minutes, famous for killing features. @sqs and @thorstenball on how the Amp team works: https://t.co/BTXBCPAXep

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

  6. Shao Meng
    AI score36

    Tencent and WorkBuddy Jointly Release a 78-Page AI Agent Enterprise Implementation Handbook

    Tencent, Tencent Cloud, and WorkBuddy have jointly released the 78-page "2026 AI Agent Enterprise Solutions and Implementation Practice Handbook," which uses WorkBuddy as the entry point to advance personal assistants into "digital employees" that enter business systems, with ClawPro, Agent Runtime, and TokenHub responsible for provisioning, isolation, models, and governance. The handbook proposes a three-stage framework of "able to converse → able to work → able to take a post," and presents case data including Yozoo's Agent penetration rate exceeding 75%, Youka's MTTR of 5–15 minutes, and operations and maintenance costs of approximately -40% (all as self-reported in the text).

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

  7. Shao Meng
    AI score28

    Shao Meng: Most domestic Video Agents have been reduced to token relay stations

    Shao Meng believes that although domestic Video Agent teams such as LibTV, TapNow, Flova, Pexo, and Medeo emphasize barriers like ecosystem, Agent architecture, and industry know-how in financing roadshows, after all their work they may have all ended up as token relay stations. He speculates that if these websites supported BYOK, the cost for users to bring their own video generation model API Key would be lower than consuming Credits on the site, and even lower when the API Key has a discount. He suggests using REA or building your own Video Agent from scratch paired with discounted API Keys to reduce costs; for users creating video content on TikTok, Ins, and Douyin, the money saved would be considerable.

    Quoted postEnglish text is being prepared. · @liyue_ai

    English text is being prepared.

    #English text is being prepared.#Agents#Video

  8. Kling AIFirst-party
    AI score37

    Kling 4.0 launches on Genspark AI Media Studio

    Genspark announced that Kling 4.0 is now integrated into Genspark AI Media Studio: it generates 30-second videos in a single pass with no stitching required, and supports up to 15 image/video/character/voice references and 10 keyframes; credits are 50% off through November 4. The official Kling AI account reposted this news.

    Quoted postGenspark · @genspark_ai

    @Kling_ai Kling 4.0 is now in Genspark AI Media Studio. Thirty seconds, one generation, no stitching. Up to 15 references across images, videos, characters, and voices, and 10 keyframes to call every beat. Credits are 50% off through November 4. https://t.co/1YN6cHWjon https://t.co/TTurtFHTJL

    Same story · 2 reports

    #Product updates#Video#OpenAI

  10. KazikeProduct release
    AI score60

    Tencent WorkBuddy launches Buddy One, connecting documents, meetings, ima, and cloud drive accounts

    Tencent WorkBuddy has launched Buddy One, bundling WorkBuddy, Tencent Meeting, Tencent Docs, ima, and Tencent Drive while integrating their account systems, offered in two tiers: the free Buddy One ID and Buddy One VIP at 700 yuan per month. After joining, WorkBuddy points can be used across all four products, and linking accounts also grants 100 points; the author's testing shows that data integration currently only provides read-only permissions, and write operations still need to be completed through the MCP connector within WorkBuddy. The author also mentions that WeChat chat records can be multi-selected and forwarded to WorkBuddy for processing, and that WeChat on desktop currently only supports WorkBuddy as an Agent entry point.

    Same story · 2 reports

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

  11. Kazik
    AI score23

    The personality difference between Claude and Codex: an example of an Excel script

    Digital Life Kazik used an example to describe the personality difference between Claude and Codex: faced with the requirement of downloading 20 Excel files every week and then manually merging them, GPT directly wrote a script to automatically click, download, copy and paste, and added exception handling and failure retries; Claude instead asked whether these Excel files came from the same database, then switched to writing SQL to connect to the backend data source, automatically generating the final report every week.

    #English text is being prepared.#Agents#OpenAI

  12. Testing Catalog
    AI score50

    Articul8 Releases Decision Model Bivia, Input Tokens $0.042 per Million

    Articul8 Launches Multimodal Decision Model Bivia, Which Can Connect to AI Agents and Choose Not to Proceed When Confidence Is Low. Bivia Weighs Each Decision Step Against the Overall Task Goal, Supports Structured Responses, Can Be Called Midway Through Agent Execution, and Is Trained on Real Production Workflows. Its Input Price Is $0.042 per Million Tokens, with No Subscription Fee.

    Quoted postArticul8 AI · @Articul8_AI

    Introducing Bivia. A multimodal AI decision model built to complete entire workflows, not just answer individual questions. We've spent 3+ years building and deploying AI in production across semiconductors, energy, manufacturing, and other complex industries. Today, we're bringing that experience to everyone.

    Same story · 3 reports

    #Model releases#Agents#Reasoning

  13. Testing CatalogModel releases
    AI score77

    NVIDIA Releases Nemotron 3.5 Transcribe Open-Weight Speech Model

    NVIDIA has released Nemotron 3.5 Transcribe, an open-weight 30B MoE model with approximately 4B active parameters. The model supports transcription in 40 languages, speaker attribution, translation, contextual prompting, and audio understanding, and ranks first in VoiceArena's 14-system comparison of speaker diarization plus ASR.

    Quoted postNVIDIA AI · @NVIDIAAI

    Meet NVIDIA Nemotron 3.5 Transcribe 🎙️ An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. Watch it keep track of who said what when everyone’s talking at once. 1/ 🧵

    Same story · 9 reports

    #Model releases#Speech#Open-source ecosystem

  14. Testing Catalog
    AI score25

    AI Daily Brief: Gemini 4 Carbon Tops AA Index with 59 Points

    Gemini 4 Carbon topped the AA Intelligence Index with 59 points and is being made available first to cybersecurity defenders. GitHub launched Copilot Fusion, costing about half as much as Claude Code with comparable performance; Alphabet's Q3 revenue reached $127.6 billion, up 25% year over year, with Gemini users reaching 1 billion. An unannounced GPT Image 2.6 appeared in OpenAI's API documentation, supporting marker-guided editing.

    Quoted post🚨 AI News | TestingCatalog · @testingcatalog

    DAILY AI BRIEF 🗞 — Oct 28 GOOGLE 🔥: - Google DeepMind announced Genie 4, its most realistic interactive world model with natural-language steering and native audio, rolling out first to Trusted Testers before Google AI Ultra and Gemini API access. - Gemini Robotics ER 2 is now in public preview on Google Cloud for enterprise robotics deployments. - Gemini Notebook on mobile can now record up to an hour of live audio and use it as a source. ANTHROPIC 🔥: - Anthropic shared early results from drug-development research partnerships with six organizations using Claude. - Anthropic is letting all US Claude Max subscribers off the Claude Duet waitlist this week, with US Pro users next in line and no timeline yet for access outside the US. OPENAI 🔥: - ChatGPT added iCloud Mail and Calendar plugins on all plans across web, desktop and iOS. - Codex CLI 0.165.0 adds a built-in Google Cloud Vertex AI provider and GPT-6.1 Terra in the model catalog. MICROSOFT 🔥: - MAI-Image-2.7-Preview launched as Microsoft AI's new image-editing model, debuting at #3 on Arena Edit and #2 on AA-Image-Editing. - MAI-Image-2.6 and MAI-Image-2.6-Flash are now generally available in Microsoft Foundry. - MAI Cyber 1.1 Flash is live on MDASH, matching MAI Cyber 1 Flash at 18% lower cost per scan. - GitHub Copilot now supports server-to-server authentication for GitHub Apps in public preview, ahead of GitHub Universe opening today. SPACEXSI 🔥: - Grok Bot now supports Memory Import to bring over preferences and context from another AI assistant. - Cursor Design Mode gained Annotate for editing elements, drawing on the page and reviewing designs at different screen sizes. META 🔥: - Muse is now available in Singapore on iOS, Android and web. NVIDIA 🔥: - NVIDIA and Meta are expanding their collaboration on Muse models, with broader Vera Rubin deployments planned at Meta. - Vera Rubin NVL72 and Vera CPUs are now generally available on Oracle Cloud, with OCI Superclusters scaling to multi-gigawatt size. ARTIFICIAL ANALYSIS 🔥: - Intelligence Index v5 launched with new mostly private benchmarks, putting Claude Opus 5.5 on top at 58. LIQUID AI 🔥: - Liquid AI released LFM3, an open-weight family of six 2B-class text, vision, audio and omni models on a new tri-hybrid architecture. * Used Grok to compose this brief, cherry-picking the news and doing some post-editing.

    #Industry news#Google#OpenAI

  15. Testing CatalogModel releases
    AI score74

    Mistral releases Mistral Large 4 (Le Chonk) weights on Hugging Face

    Mistral released the model weights of the new version of Le Chonk (Mistral Large 4) on Hugging Face, targeting agentic coding and cybersecurity scenarios. Mistral claims that the model ranks first among open-source models on Cognition's FrontierCode benchmark and sets a new benchmark for Pareto efficiency on Artificial Analysis, while also improving inference efficiency and reducing inference costs. The model is also claimed to achieve SOTA in cybersecurity defense and has been tested by vendors such as CrowdStrike, Cloudflare, Rubrik, and Zscaler; users can customize and self-deploy it with Mistral Forge.

    Quoted postMistral AI · @MistralAI

    Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:

    Same story · 15 reports

    #Model releases#Open-source ecosystem#Reasoning

  17. OpenRouterFirst-partyProduct release
    AI score53

    OpenRouter Launches New Provider Router, Rewarding High-Quality Reasoning Based on Actual Cost per Request

    OpenRouter has released the biggest improvement to its provider router since its founding. The new routing no longer primarily ranks by price; instead, it rewards the highest-quality reasoning based on the actual cost per request, keeping requests low-cost while also balancing speed and reliability. Providers that deliver higher value automatically receive more traffic, giving them an incentive to keep improving, and as providers improve, users' reasoning experience on OpenRouter improves in tandem.

    Same story · 2 reports

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

  18. OpenRouterFirst-partyProduct release
    AI score55

    OpenRouter launches a new version of its provider router, rewarding high-quality inference based on actual cost

    OpenRouter has released the largest improvement to its provider router since its founding. The new router no longer primarily ranks by price, but instead rewards the highest-quality inference based on the actual cost of a request, keeping requests low-cost while also balancing speed and reliability. Providers that deliver higher value automatically receive more traffic, giving them an incentive to keep improving, and as providers improve, users' inference experience on OpenRouter improves in tandem.

    Same story · 2 reports

    #Product updates#English text is being prepared.