Hextra-AI-Insight-Daily/content/en/_index.md

linkTitle, title, breadcrumbs, next, description, cascade

linkTitle	title	breadcrumbs	next	description	cascade
AI Daily	AI Daily-AI资讯日报	false	/en/2026-01/2026-01-08	Your daily source for curated AI news, practical tools, and actionable tutorials to master Artificial Intelligence;	type docs

AI News Daily 2026/1/9

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

Today's Digest

Fourier GR-3 Debuts at CES with 55 Degrees of Freedom, Focusing on Companionship
Xpeng's Second-Gen VLA Mass-Produced, Achieving Vision-Driven Entry-Level L4 Capabilities
OpenAI Launches Health Version, Connecting to Apple Health to Interpret Blood Tests
MiniMax Raises HKD 4.2 Billion in Hong Kong IPO, Marking Monetization Phase for Domestic Large Models
Ant Group's "Afu" Monthly Active Users Exceed 30 Million, Completing the Entire Consultation-Diagnosis-Treatment Chain

Product & Feature Updates

1. Fourier GR-3 Debuts at CES. Fourier GR-3 made its grand debut at CES 2026, bringing along its 🔥humanoid robot GR-3 (AI News). Visitors got a hands-on experience in the interactive zone. This robot is seriously agile with 55 degrees of freedom, capable of playing chess, chatting, and even 💡 recognizing expressions. Designed for warm companionship, GR-3 uses a calming Morandi color palette to soften its mechanical vibe. Fourier also teased a concept prototype for a desktop doll robot. 👋

2. Xpeng's Second-Gen VLA Large Model Hits the Road. Xpeng's Second-Gen VLA large model is rolling out! He Xiaopeng declared that 🚀"Physical AI" is the key buzzword for 2026. Their second-gen VLA model (AI News) is slated for mass production and deployment in Q1. This bad boy can directly drive actions from visual input, claiming entry-level L4 capabilities – now that's impressive! Both the 2026 P7+ and G7 models will pack this tech, and Robotaxi operations are finally kicking off 💡. Plus, Xpeng is gearing up to mass-produce humanoid robots and flying cars. Wild stuff! 🤯

3. OpenAI Launches ChatGPT Health. OpenAI officially dropped ChatGPT Health – a health-focused version (AI News) that can link up with Apple Health and your electronic medical records 🏥. Super cool! Just upload your blood test reports, and it'll break down the indicators in plain English for you. It even whips up a list of questions to ask your doctor. Crucially, your health data is independently encrypted and NOT used for model training. Right now, it's only in an invite-only test phase for a select few 💡.

4. Google Classroom Launches Gemini Podcast Tool. Google Classroom just dropped a new Gemini podcast tool! 🎙️ Teachers can now punch in a course topic and automatically generate podcast-style audio lessons (AI News) – tailor-made for Gen Z. It supports a dynamic host + guest conversational format and even lets you add background music 💡. One teacher raved that students' completion rates hit a whopping 92% (that's double reading a PDF!). With multi-language support and one-click generation, it's seriously a game-changer for bite-sized learning. 🎧📚

5. Tencent Open-Sources HY-Motion 1.0. Tencent Hunyuan just open-sourced HY-Motion 1.0! 🔥 This text-to-3D motion large model (AI News) boasts billions of parameters and is built on the DiT architecture. It's a powerhouse, generating over 200 motion categories, from everyday walks to 💪 intense sports. The best part? Its output plugs directly into Unreal and Unity, making it super easy to use right out of the box. The project is already live on HuggingFace (AI News). Get on it! 🚀

---

Frontier Research

1. V-Agent Multimodal Video Search. A groundbreaking new paper just unveiled V-Agent, a multimodal video search system. 🔍 This V-Agent video search system (AI News) fine-tunes VLMs to simultaneously grasp both visuals and audio – how cool is that? Three agents team up 💡 to handle user intent, smashing SOTA zero-shot performance on the MultiVENT2.0 test set. They've embedded video frames and speech-to-text into a unified space, and guess what? The model is already open-source! 🤩

2. PhysVideoGenerator: Physics-Aware Video Generation. PhysVideoGenerator is here to make video generation more realistic! This paper teaches video generation models 🎬 to understand physical laws (AI News), finally fixing those awkward, unnatural object collisions. They're using V-JEPA2 to pull out physical features and inject them right into the generation process 💡, which seriously amps up gravity and temporal consistency. It's still in the proof-of-concept stage, but the training stability is already solid. Big steps forward! 🚀

3. ThinkRL-Edit: Reasoning-Based Image Editing. Say hello to ThinkRL-Edit, a new framework that makes image editing models 🧠 think before they act (AI News). It uses Chain-of-Thought sampling to explore tons of different solutions. Unlike old-school methods that only randomize during denoising, this bad boy starts exploring at the semantic level 💡. And get this: they swapped out fuzzy scoring for a clear-cut binary checklist, which has totally blown past previous work in performance. Super neat! ✨

---

Industry Outlook & Social Impact

1. MiniMax Sees 15% Surge in Hong Kong Grey Market. MiniMax just made waves in the Hong Kong grey market, soaring by 15%! 🔥 This marks another large model company hitting the stock market after Zhipu AI. MiniMax (AI News) successfully raised a whopping HKD 4.189 billion, with its grey market price peaking at HKD 199.8, pushing its market cap close to HKD 60 billion 💰. The cash is mainly going into next-gen multimodal models and their own chips. This signals that domestic large models are finally entering their monetization phase – about time! 🤑 Of course, profitability pressures and global competition are still big hurdles ahead.

2. Ant Group's "Afu" Monthly Active Users Skyrocket to 30 Million. Ant Group's "Afu" is on fire, with its monthly active users skyrocketing from 15 million to 🚀30 million (AI News) in just one month! This surge even pushed OpenAI to urgently launch ChatGPT Health. "Afu" has built out a full-blown consultation-diagnosis-treatment pipeline 💡, linking up with 5,000 hospitals and ten top device brands. Meanwhile, ChatGPT Health is still stuck in the "information assistance" phase 😩. It's clear that the AI health paths in China and the US are splitting ways.

3. Google Urgently Hires AI Quality Reviewers. Google is in hot water, facing a 🔥search quality trust crisis (AI News) as its AI overviews keep hallucinating. Imagine asking the same question twice and getting wildly different, wrong answers – anywhere from 4 million to 70 million! 😩 Even worse, some medical advice provided by AI has been potentially fatal 💡. Google is now urgently hiring AI quality reviewers, indirectly admitting these functional flaws through their job postings for the first time. Yikes! 🚨

4. Malicious Chrome Extensions Steal AI Conversations. Heads up, folks! Two sneaky 💀malicious Chrome extensions (AI News), pretending to be legit AI tools, managed to snag over 900,000 downloads. Get this: one even got a Google Featured badge! 😩 These extensions were specifically designed to snatch chat histories from ChatGPT and DeepSeek. They'd regularly ping remote servers with your browsing URLs and sensitive keywords. Thankfully, they've been yanked offline 💡. Stay safe out there! 🔒

---

Top Open Source Projects

1. claude-mem: Automatic Session Memory. First up, claude-mem! This awesome Claude Code plugin (AI News), with ⭐12.3k stars, automatically snags every operation from your coding sessions 🔥. It compresses that info with AI and injects it into future conversations, so your context never gets lost – how cool is that for long-term projects?! 🤯 It's built using Claude's agent-sdk 💡.

2. ComfyUI-LTXVideo: Video Generation. Next up, ComfyUI-LTXVideo! Lightricks officially dropped this 🎬ComfyUI video support (AI News), racking up ⭐2.5k stars. Integrating the LTX-Video model is now super easy-peasy. Just drag and drop it into your workflow 💡, and boom – pretty solid generation quality. 🎥✨

3. memU: Memory Infrastructure. Check out memU, a memory system (AI News) boasting ⭐3.6k stars! It's built for LLMs and 🤖 Agents to tackle that tricky long-term memory management problem. Developed by the NevaMind team 💡, this bad boy actually helps AI remember what you've said. Finally, AI that doesn't forget! 🙏

4. VideoRAG: Video Dialogue. From HKU comes VideoRAG! This 📹video retrieval-augmented generation system (AI News), with ⭐1.9k stars, is the accompanying code for a KDD2026 paper. It lets you chat directly with video content 💡, understanding the visuals and answering questions. Pretty slick for interactive video! 🗣️🎬

5. MiroThinker: Search Agent. Last but not least, meet MiroThinker! This bad boy is being hailed as 🔍the world's strongest search Agent (AI News), with a 30B model that somehow pulls off 1T-level performance. Seriously impressive! It supports a massive 256K context and 400 tool calls 💡, putting it in the global top league on BrowseComp tests. What's more, it actively cross-verifies and self-corrects just like a real scientist. You can even check it out yourself – an online demo (AI News) is now live! 🧪💻

---

Social Media Buzz

1. MOSS Speech Recognition Can Tag Speakers. First up in the social media feed, @Gorden_Sun shared some cool news about MOSS Transcribe Diarize! 🎤 The Fudan team's MOSS Transcribe Diarize (AI News) not only nails speech recognition but also identifies different speakers 💡. It handles multi-language audio up to a solid 90 minutes, making it a great alternative for folks in China who can't access Gemini. Plus, a demo is already open (AI News) for you to try! 🗣️

2. Gemini CLI Supports Agent Skills. Big news from @Jimmy_JingLv! He announced that Gemini CLI now supports skill plugins 🎉, following in Codex's footsteps. This dropped with the v0.23.0 update 💡. Apparently, AgentSkills.me is betting big on agents this year, so keep an eye out! 🤖

3. Claude Code is Super Useful for Drawing Canvases. @vista8 dropped a cool tip: Claude Code is amazing for drawing canvases! 🎨 He showed off a Canvas skill (AI News), crafted by Obsidian's CEO, that lets Claude search for Qing Dynasty emperors and then whip up a visual chart 💡. You can use it for organizational charts, product architecture diagrams – pretty much anything! Super versatile! 📊

4. Zhipu's z.ai Overseas Expansion Experience. @op7418 retweeted a gem! 🔔 On Zhipu's bell-ringing day, Zixuan shared his summary of overseas expansion (AI News), detailing how z.ai quickly became a major player abroad 💡. If you're clueless about expanding internationally, this is a must-read! 🌍

5. Planning with Files Recreates Manus. @shao__meng gave a shout-out to Planning with Files! 📁 This Claude skill (AI News) acts like an external AI brain using Markdown, which helps solve those pesky memory fluctuations and goal-drifting problems 💡. It manages three key files: task lists, research notes, and final outputs. Super smart for keeping AI on track! 📝

6. Jensen Huang Jokes About Being "Crushed by Both China and the US." Funny one from @dotey! 🤣 He retweeted Jensen Huang's hilarious remark: "We are the first company in history to be simultaneously crushed by both China and the US (AI News)" 💡. Talk about a self-deprecating legend! 😂

7. AI Deciding Promotions Sparks Controversy. And finally, a hot topic on Reddit: AI deciding who gets promoted! 🤔 A buzzing thread is discussing AI automatically deciding promotions and raises (AI News). A recent survey revealed that a whopping 60% of managers are already using AI for these decisions 💡. Your typing speed and emails are all under surveillance 😩. Big Brother, much? 😳

---

AI News Daily Voice Version

🎙️ Xiaoyuzhou	📹 Douyin
Afterlife Pub	Self-Media Account