01-09-Daily AI News Daily

AI News Daily 2026/1/9

AI News | Daily Morning Read | Aggregated Web Data | Frontier Science Exploration | Industry Free Voice | Open Source Innovation | AI and Human Future | Visit Web Version | Join Group Chat 💬

Today’s Rundown

Fourier GR-3 Debuts at CES with 55 Degrees of Freedom, Focused on Companionship
Xpeng's Second-Gen VLA Mass Produced, Achieves Visual-Driven Entry-Level L4 Capability
OpenAI Launches Health Version, Connects to Apple Health for Blood Test Interpretation
MiniMax Raises HKD 4.2 Billion in Hong Kong IPO, Domestic Large Models Enter Monetization Phase
Ant Aifu Monthly Active Users Exceed 30 Million, Completes Consultation and Diagnosis Full Chain

Product & Feature Updates

1. Fourier GR-3 Debuts at CES. At CES 2026, Fourier showcased its 🔥 Humanoid Robot GR-3 (AI News) , offering attendees a hands-on experience in an interactive zone. This robot, the GR-3, is super agile thanks to its 55 degrees of freedom across its body, and it can play chess, chat, and even 💡recognize expressions. Fourier designed it for warm companionship, using Morandi colors to soften its mechanical look. The company also unveiled a desktop doll robot concept prototype.

2. Xpeng’s Second-Gen VLA Large Model Hits the Road. Xpeng’s second-generation VLA large model is set for mass production rollout in Q1, following He Xiaopeng’s announcement that 🚀Physical AI will be the core keyword for 2026. This model can directly drive actions from visual input, boasting entry-level L4 capabilities (pretty cool, right?). Both the 2026 P7+ and G7 models will feature this tech, and Robotaxi operations are also slated to officially kick off 💡. Xpeng is also pushing to mass-produce humanoid robots and flying cars.

3. OpenAI Launches ChatGPT Health. OpenAI officially launched ChatGPT Health (AI News) , which can connect to Apple Health and electronic health records 🏥. You can upload blood test reports, and it’ll explain the indicators in plain language, plus generate a list of questions for your doctor (how handy!). Importantly, health data is stored independently and encrypted, not used for model training. Currently, it’s only available for 💡a limited number of users in open testing.
   
   

4. Google Classroom Launches Gemini Podcast Tool. Google Classroom has rolled out a Gemini podcast tool, allowing teachers to input course topics and 🎙️ automatically generate podcast-style audio lessons (AI News) specifically designed for Gen Z. It supports a host + guest conversational format and even lets you add background music 💡. One teacher reported student completion rates hitting 92% (double that of reading PDFs!). With multi-language support and one-click generation, it’s a real game-changer for bite-sized learning (awesome!).

5. Tencent Open-Sources HY-Motion 1.0. Tencent Hunyuan has released 🔥 HY-Motion 1.0, a text-to-3D action large model (AI News) , featuring billions of parameters and based on the DiT architecture. This model can generate over 200 action categories, covering everything from everyday walking to 💪sports competitions. Its output integrates directly with Unreal and Unity, making it plug-and-play (super cool!). The project has already been open-sourced on HuggingFace (AI News) .
   
   

Frontier Research

1. V-Agent Multimodal Video Search. A new paper introduces the 🔍 V-Agent video search system (AI News) , which allows fine-tuning VLMs to understand both visuals and speech simultaneously. Three agents collaborate 💡to process user intent, achieving SOTA zero-shot performance on the MultiVENT 2.0 test set. Video frames and speech-to-text are embedded in the same space (how clever!), and the model has been open-sourced.

2. PhysVideoGenerator: Physics-Aware Video Generation. This paper introduces PhysVideoGenerator, a model that enables video generation models 🎬to learn physical laws (AI News) , addressing unnatural object collisions. It uses V-JEPA2 to extract physical features and inject them into the generation process 💡, significantly improving gravity and temporal consistency. While currently in the concept verification stage, training stability has already been validated (pretty neat!).

3. ThinkRL-Edit: Reasoning-Based Image Editing. A new framework called ThinkRL-Edit allows image editing models 🧠to think before acting (AI News) , using chain-of-thought sampling to explore multiple solutions. Unlike traditional methods that only randomize during denoising, ThinkRL-Edit starts exploring at the semantic level 💡. By replacing vague scoring with a binary checklist, it significantly outperforms previous work (impressive!).

Industry Outlook & Social Impact

1. MiniMax’s Hong Kong IPO Grey Market Sees 15% Jump. MiniMax, another 🔥large model company, has gone public following Zhipu AI, raising HKD 4.189 billion MiniMax (AI News) . Its grey market price peaked at HKD 199.8, pushing its market cap close to HKD 60 billion 💰. The funds will primarily go towards next-gen multimodal models and proprietary chips, signaling that domestic large models are entering a value realization phase (big moves!). Profitability pressures and international competition remain future challenges.

2. Ant Aifu’s Monthly Active Users Skyrocket to 30 Million. Ant Aifu’s monthly active users surged from 15 million to 🚀 30 million users (AI News) in just one month, prompting OpenAI to quickly follow up with ChatGPT Health. Ant Aifu has integrated the entire chain from consultation to diagnosis and treatment 💡, connecting with 5,000 hospitals and ten major brand devices. ChatGPT Health, however, is still stuck in the information assistance stage (oof 😬). The divergent paths of AI health in China and the US are becoming quite clear.

3. Google Urgently Hires AI Quality Reviewers. Google is grappling with a 🔥 search quality trust crisis (AI News) , as AI overviews frequently hallucinate. The same question yielded wildly incorrect answers, ranging from 4 million to 70 million, when asked twice (yikes 😬). Even more seriously, some medical advice provided lethal guidance 💡, and the job postings indirectly admit to functional flaws for the first time.

4. Malicious Chrome Extensions Steal AI Conversations. Two 💀 malicious Chrome extensions (AI News) disguised as AI tools racked up over 900,000 downloads. They specifically targeted ChatGPT and DeepSeek to steal chat histories, with one even earning a “Google Featured” badge (talk about sneaky 😠). These extensions regularly sent browsing URLs and sensitive keywords to remote servers but have since been taken down 💡.
   
   

Top Open-Source Projects

1. claude-mem: Automatic Conversation Memory. The claude-mem Claude Code plugin (AI News) , with ⭐12.3k stars, automatically captures all operations in a coding session 🔥. It compresses this data with AI and injects it into future conversations, meaning no more lost context (finally!). Developed using Claude’s agent-sdk 💡, it’s especially handy for long-term projects.

2. ComfyUI-LTXVideo: Video Generation. ComfyUI-LTXVideo is Lightricks’ official 🎬 ComfyUI video support (AI News) with ⭐2.5k stars, making LTX-Video model integration super easy. Its drag-and-drop workflow is a breeze to use 💡, and the generation quality is pretty solid (nice!).

3. memU: Memory Infrastructure. memU is a memory system (AI News) for LLMs and 🤖Agents with ⭐3.6k stars, designed to tackle long-term memory management issues. Developed by the NevaMind team 💡, it helps AI truly remember what you’ve said (pretty cool!).

4. VideoRAG: Video Dialogue. VideoRAG, developed by HKU, is a 📹 Video Retrieval Augmented Generation (AI News) project with ⭐1.9k stars, providing companion code for a KDD 2026 paper. It allows direct dialogue with video content 💡, understanding visuals and answering questions (how awesome!).

5. MiroThinker: Search Agent. MiroThinker is touted as the 🔍 world’s strongest search Agent (AI News) , with its 30B model achieving 1T-level performance. It supports 256K context and 400 tool calls 💡, ranking in the top tier globally in BrowseComp tests (impressive stuff!). It actively verifies and self-corrects like a scientist, and the online experience (AI News) is now live.
   
   

Social Media Shares

1. MOSS Voice Recognition Can Tag Speakers. @Gorden_Sun shared the Fudan team’s 🎤 MOSS Transcribe Diarize (AI News) , which not only recognizes speech but also distinguishes between different speakers 💡. It supports multi-language audio up to 90 minutes long, serving as a solid alternative for those in China who can’t use Gemini (pretty sweet!). The demo is now available (AI News) .

2. Gemini CLI Now Supports Agent Skills. @Jimmy_JingLv announced that following Codex, 🎉 Gemini CLI now supports skill plugins (AI News) with its v0.23.0 update 💡. AgentSkills.me is going all-in on agents this year (exciting times!).

3. Claude Code’s Canvas Drawing is Super Handy. @vista8 shared how 🎨Claude Code’s Canvas skill (AI News) , written by the Obsidian CEO, is super useful. It lets Claude search for Qing Dynasty emperors and then draw a visualized graph 💡. You can use it for organizational charts, product architecture diagrams, you name it (so versatile!).
   
   

4. Zhipu AI’s Zixuan Shares Z.ai’s Overseas Expansion Experience. @op7418 reposted 🔔Zixuan’s overseas expansion summary (AI News) from Zhipu’s bell-ringing day, detailing how z.ai quickly boosted its influence abroad 💡. It’s a must-read for anyone feeling lost about global expansion (super helpful!).

5. Planning with Files Replicates Manus. @shao__meng recommended 📁 Planning with Files (AI News) , a Claude skill that uses Markdown as an external AI brain to combat memory fluctuations and goal drifting 💡. It manages three key files: task lists, research notes, and final outputs (pretty clever system!).
   
   

6. Jensen Huang Self-Deprecatingly Jokes About Being “Beaten by Both China and the US”. @dotey reposted 🤣Jensen Huang’s humorous remark: “ We are the first company in history to be simultaneously beaten by both China and the US (AI News) .” 💡 (What a sport!)
   
   

7. AI-Driven Promotions Spark Controversy. AI-driven promotions are sparking controversy, as a Reddit hot topic discusses 🤔 AI automatically deciding promotions and raises (AI News) . A survey reveals that 60% of managers are already using AI for this 💡. Typing speed and emails are all being monitored (talk about Big Brother 😬).

AI News Daily Audio Version

🎙️ Xiaoyuzhou	📹 Douyin
Rebirth Pub	Self-Media Account