---
title: TodayDaily
breadcrumbs: false
next: /en/2025-06/2025-06-25
description: Google DeepMind just dropped Gemini Robotics On-Device, an on-device
AI model designed specifically for robots 🤖! It's built on the multimodal reasoning
of the Gemini 2.0 model, letting robots quickly learn new tasks, work stably even
offline, and even pull off intricate operations like folding c...
cascade:
type: docs
---
# AI Daily Insights 2025/6/26
> `AI Daily` | `Morning Update` | `Web-wide Data Aggregation` | `Deep Dive into Frontier Science` | `Industry Voices Speak Up` | `Open Source Powerhouse` | `AI and the Future of Humanity` | [Visit Web Version ↗️](https://ai.hubtoday.app/)
#### **AI Content Summary**
```
AI products are updating fast, Google launched on-device AI for robots. iFlytek's medical model reached expert level.
Quark's college application service is blowing up, expanding computing power. Rokid glasses are mass-produced, securing tons of orders.
AI research sees breakthroughs in multimodal and 3D reconstruction. Zhou Hongyi says AI can't replace human emotion or creativity.
```
#### **AI Product & Feature Updates**
1. Google DeepMind just dropped **Gemini Robotics On-Device**, an **on-device AI model** designed specifically for **robots** 🤖! It's built on the **multimodal reasoning** of the **Gemini 2.0 model**, letting robots quickly learn new tasks, work stably even offline, and even pull off intricate operations like folding clothes ✨. This is definitely laying a solid foundation and opening a new chapter for the future of **embodied intelligence**!
2. College application season is in full swing, and **Quark's** smart college recommendation service has been so popular it's seeing **queues**, with over 3 million reports generated so far 📈. That really shows how much students trust its AI capabilities. Faced with this "sweet headache," **Alibaba Group VP Wu Jia** came back strong, saying the team has urgently **boosted computing power** to make sure every student gets this crucial guide for their academic journey! 💪
3. Rokid, in collaboration with Lens Technology, has officially started mass **production** of their **consumer-grade AI+AR glasses, Rokid Glasses**! 👓✨ These glasses are super **lightweight** and pack a punch with **AI large model capabilities** like **smart teleprompter, real-time translation, and AI object recognition**. They've already snagged **250,000 global pre-orders**! This signals an upcoming **commercial boom** in China's AI glasses market – the future's looking bright! 🚀
4. At the 2025 Cloud Next conference, Google showcased its next-gen **customer service intelligent assistant** 🤖, powered by the **Gemini model**. This assistant is seriously cool – not only can it handle **multimodal interactions**, but it can also apply for **discounts** itself and is deeply integrated with the Salesforce **CRM system**! This hints at a massive intelligent revolution coming to customer service 💥, but we'll have to wait and see about its accuracy and privacy protection. 😉
5. iFlytek just launched its powerful **Spark Medical Large Model V2.5 International Edition** 🚀, trained on entirely domestically produced computing power! This model topped the charts on the authoritative MedBench platform with an impressive 98.4 points. Its comprehensive diagnostic and treatment capabilities have reached the level of an attending physician at a top-tier hospital, and it even surpasses human doctors in completeness, practicality, and readability! 👨⚕️🩺 It also supports multiple languages, promising to make a big splash in the global medical market and drive international medical technology exchange and collaboration! 🌍✨
6. ElevenLabs finally dropped its standalone **text-to-speech mobile app**! 📱✨ Whether you're on iOS or Android, you can now generate audio snippets anytime, anywhere. Even free users get about 10 minutes of audio generation time! This app uses the latest v3alpha model and even supports **emotional expression control**. Plus, it's set to add speech-to-text and conversational AI tools in the future – how convenient is that?! 🗣️
#### **AI Frontier Research**
1. **SuperDec**, a joint effort by ETH Zurich, Stanford University, and Microsoft teams, is **shattering conventional 3D reconstruction limits** 🤯! This tech uses an innovative **hyper-tetrahedron** principle to create compact yet vivid **3D scene representations**. It efficiently handles complex **point cloud** data and shows huge potential in areas like precise grasping and path planning for **robotics**, as well as **controllable visual content generation**, opening up new horizons for the digital world! 👀 [Project Link](https://super-dec.github.io/)
2. **4D-LRM** is a super cool, innovative **large spatio-temporal reconstruction model** 🤩. It can fully reconstruct **dynamic objects' 4D representations** (3D space plus time dimension) from just a few input views, generating high-quality scenes from any time and any angle! In the future, it's set to make a huge impact in virtual reality, film production, industrial simulation, and more! 🌟 [Paper Link](https://huggingface.co/papers/2506.18890)
3. ByteDance and Shanghai Jiao Tong University have teamed up to release the **ProtoReasoning framework** 👏. It cleverly leverages structured prototype representations like **Prolog** and **PDDL** to significantly boost **large language models' logical reasoning capabilities** and efficiency in cross-domain knowledge transfer 🚀. This research lays a solid foundation for future theoretical explorations of **reasoning prototypes** – truly awesome! [Paper Link](https://arxiv.org/abs/2506.15211)
4. MMLab at the University of Hong Kong, MMLab at the Chinese University of Hong Kong, and SenseTime have jointly developed the **GoT-R1 framework**. This groundbreaking research, by introducing **reinforcement learning**, significantly enhances **multimodal large models' semantic-spatial reasoning capabilities** in **visual generation tasks** 🚀, allowing models to learn better reasoning strategies autonomously! It not only breaks free from the GoT framework's reliance on templates but also achieves **SOTA performance** in complex scene generation – seriously impressive! ✨ [Paper Link](https://www.jiqizhixin.com/articles/2025-06-25-9)
#### **AI Industry Outlook & Social Impact**
1. Zhou Hongyi recently chatted about the future of AI in a video. He believes that no matter how powerful AI gets, it will never fully replace humanity's unique abilities in **emotional understanding** 💖, **complex problem-solving** 🧠, and **creative thinking** 🎨. He stressed that future work will increasingly involve **managing and training** AI, even citing a failed case of AI customer service at a Swedish company to show that AI still has limitations when handling complex customer needs. 🧐
2. Federal Judge William Alsup made a groundbreaking ruling: **Anthropic's** use of **copyrighted books** to train its **AI model** without permission was deemed **fair use**! 😮 This sets an important precedent for copyright disputes in the AI industry. However, Anthropic is still facing **theft charges** for obtaining training material from pirated websites – a mixed bag, huh? 🤔
#### **TOP Open Source Projects**
1. **Dioxus** is a super popular **full-stack application framework** with 28,310 stars ⭐! It's like an all-in-one toolbox, aiming to give developers a unified solution to easily build applications for web, desktop, and mobile platforms, greatly simplifying the complexity of cross-platform development! 💻📱 [Project Link](https://github.com/DioxusLabs/dioxus)
2. **jsoncrack.com** is a star project boasting **38,020 Stars** ⭐! It's an innovative **open-source visualization application** that can instantly transform various data formats like **JSON**, **YAML**, **XML**, and **CSV** into **interactive diagrams** 📊, massively boosting data readability and analysis efficiency. It's a godsend for data enthusiasts! 😍 [Project Link](https://github.com/AykutSarac/jsoncrack.com)
3. **free-for-dev** is a treasure trove of a project for **DevOps** and **infrastructure developers**! ✨ It boasts an astonishing **100,044 Stars** and is a super handy **open-source** project specifically curating and providing a **list of free tiers** for SaaS, PaaS, and IaaS services. This is truly a tailor-made money- and time-saver for developers! 💰⏰ [Project Link](https://github.com/ripienaar/free-for-dev)
#### **Social Media Shares**
1. Yang Yi excitedly shared Google AI Developer's new **Gemini CLI**, calling it a "cyber Bodhisattva"! 🤩 This **open-source AI agent** brings **Gemini 2.5 Pro** right to your terminal, offering **high-frequency free usage** and making **code writing, debugging, and task automation** a breeze! He thinks it's a "top-tier" solution for current tool shortcomings, especially for **MCP deployment and GitHub search**, with unlimited potential! 🚀
More details: ['More details'](https://x.com/Yangyixxxx/status/1937881859788304743)
2. Xiao Hu gasped, he found an "awesome" **AI design website**! It's seriously a godsend for designers! 🎨✨ It can generate beautiful, ready-to-use interfaces, and it massively **simplifies design prompt requirements**. Even cooler, it can provide detailed design solutions based on simple descriptions, and even generate **multi-level pages based on contextual logic**, plus it supports **precise editing of elements**, hugely boosting design efficiency and flexibility! 😍
More details: ['More details'](https://x.com/imxiaohu/status/1937847459117687004)
3. Yang Yi thinks **AI singer Yuri** is the first **AI Influencer** to truly "go mainstream"! 🎤🔥 This **AI singer from Surreal** successfully collaborated with The North Face, and her work has racked up over 7 million plays! This clearly shows **AI's growing influence and commercial potential in the virtual idol space**, signaling an exciting new era! 🎉
More details: ['More details'](https://x.com/Yangyixxxx/status/1937843457630020058)
4. Alipay is really on the cutting edge! ✨ They've launched their first-ever **AI tipping** service, letting developers integrate this feature into their **smart agents** so users can "send flowers" (give tips) to their favorite **smart agents**! 💰💖 ['More details'](https://x.com/imxiaohu/status/1937830267873525781)
5. Google's dropping a major bombshell! 🎉 They've made the powerful **Imagen 4** and **Imagen 4 Ultra** image models available for free in **AI Studio**! 🤩 Now, users can experience these awesome image generation models for free via the **Gemini API** and AI Studio. Go check it out! 🎨 ['More details'](https://x.com/op7418/status/1937708999430033734)
6. Anthropic's **Claude Artifacts** is getting an update! 🥳 Users will soon be able to browse and share popular web creations in the **Artifacts Gallery**, and even create **AI frontend applications** directly via the **Claude API** – just thinking about it sounds super cool! 💻✨ ['More details'](https://x.com/op7418/status/1937707999902203955)
7. Zero Jun Chatting AI shared an **AI video** that hit over 50 million views in 24 hours. He hit the nail on the head by pointing out that the secret to current **viral AI videos** is one word – "**outrageous**"! 😂 It's not about being human-like. Common viral themes include ASMR, animal Olympics, and AI natural disasters. Want to see more "outrageous" videos? Click ['here'](https://m.okjike.com/originalPosts/685b65582b50c68918a1279b) to learn more!
8. Tom Huang shared 20 super useful **programming Prompt** tips 💡, and also revealed that Warp is heavily developing a **terminal Agent** similar to Claude Code. While this Agent is pay-per-use, apparently, one use can earn back the cost! 😱 It's practically a productivity godsend for programmers! 🚀 For more details, hurry and click ['here'](https://x.com/tuturetom/status/1937669382752338129) to check it out!
---
#### **Listen to the Audio Version**
| 🎙️ **Xiaoyuzhou** | 📹 **Douyin** |
| --- | --- |
| [Next Life Speakeasy](https://www.xiaoyuzhoufm.com/podcast/683c62b7c1ca9cf575a5030e) | [Next Life Intel Hub](https://www.douyin.com/user/MS4wLjABAAAAwpwqPQlu38sO38VyWgw9ZjDEnN4bMR5j8x111UxpseHR9DpB6-CveI5KRXOWuFwG) |
|  |  |