07-26-Daily AI News Daily

AI Daily News 2025/7/26

AI Daily | 8 AM Updates | Web Data Aggregation | Cutting-Edge Science Exploration | Industry Open Mic | Open Source Innovation | AI and Human Future | Visit Web Version ↗️

Today’s Summary

MyShell and Google have each launched new tools that allow AI applications to be developed through natural language.
Anthropic's Claude integrates with Canva, automatically generating visual designs from text content.
New progress in AI collaborative learning: The Agent KB framework enhances problem-solving efficiency by sharing an experience pool.
Alibaba, ByteDance, and others have released new models, strengthening multilingual translation and simultaneous interpretation capabilities.
Additionally, although AI summaries are popular with Google, they have led to a significant drop in click-through rates for traditional web pages.

AI Product & Feature Updates

1. MyShell has dropped a bombshell, officially launching ShellAgent 2.0, a game-changing AI app development tool! This revolutionary tool champions the Text-to-App concept, making the dream of “everyone a developer” a real possibility. Users simply describe their needs in natural language, and ShellAgent 2.0 handles the entire process from backend to deployment. Developing AI apps is now as easy as sending a message, truly democratizing AI application development.
   
   

2. Google Labs is stepping up its game with Opal, an experimental tool that lets you whip up AI apps with just a few words! Think of it as being a “magic brush user” for AI. This tool makes no-code AI app creation a reality. Users can quickly build mini-applications, much like assembling building blocks, using natural language commands or a drag-and-drop visual workflow editor. These “blocks” are powered by the mighty Gemini and the Google AI ecosystem. Currently, Opal is only available for public beta in the US, so the rest of the world will have to hold tight. ⏳
   
   
   

3. Talk about a blessing for the lazy ones! ✨ Anthropic’s AI assistant, Claude, has announced a deep integration with global design platform Canva, completely streamlining the text-to-visual design creation process. Users just upload their written content, and Claude intelligently analyzes it to automatically generate professional visual works that match their brand style. This partnership massively lowers the design barrier, making your content way more visually appealing.
   
   
   

4. The previously “hard-to-get” AI design agent, Lovart, is now fully open for registration! 🚀 It’s seriously trying to reshape the design workflow with a brand-new interaction paradigm. Its core feature, ChatCanvas, lets users give commands in natural language anywhere on the canvas, and the AI precisely understands and processes them in bulk. This “smart canvas” experience, which leverages multimodal context to the max, truly empowers AI to amplify creativity rather than replace it.
   
   
   
   

5. Wanna be the coolest kid on your social feed? 😎 Kling, the AI video generation tool from Kuaishou, is here to help! Its “Image-to-Video” Elements feature just got a massive upgrade. It not only boosts subject consistency and dynamic fluidity but also introduces a new region reference function. Users can precisely specify faces, clothing, and other elements in an image as references, giving the AI a clearer creative focus and taking visual effects to the next level.
   
   

AI Frontier Research

1. AI Agents are ditching the solo grind and learning to “copy notes” from each other! 🧠 The Agent KB framework, jointly released by OPPO, Yale, and other institutions, lets AI learn from each other’s successful experiences by building a shared “experience pool.” In the super challenging GAIA benchmark test, this framework boosted the model’s problem-solving success rate by up to 6.66%, proving that “teamwork” is way more efficient than going it alone.
   
   
   
   

2. Can AI guess your age from your face? A [Latest Arxiv Paper - AI News]( https://arxiv.org/abs/2507.18565 ) proposes a new deep learning method that could be a total game-changer for precise advertising. This method can simultaneously perform age estimation and gender classification with astounding accuracy. However, it also spills the beans on potential data bias issues lurking within AI models. This is a crucial reminder: while we enjoy the convenience of tech, we gotta stay sharp about the hidden ethical risks! 🧐

AI Industry Outlook & Social Impact

1. Can AI really crack the code of “cat and dog speak”? 🐶 Scientists are waving a red flag: watch out, it might just “lie” to please you! The London School of Economics (LSE) has launched the world’s first Jeremy Coller Centre for Animal Sentience, aiming to use AI and other interdisciplinary methods to study animal consciousness. This research isn’t just about animal welfare; it’s also offering a fresh perspective on understanding “consciousness,” that ultimate philosophical puzzle.

2. Google’s search rules have changed, but also… not really. 🤔 Officially, content creators got a reassuring pat on the back: high-quality AI content is still welcome, and existing SEO principles still apply. However, the reality bites: when AI summaries pop up, traditional web page click-through rates have plummeted from 15% to a mere 8%. This clearly means shifting user behavior is quietly reshaping the traffic landscape.

Open Source TOP Projects

1. [Qwen3-Coder - AI News]( https://github.com/QwenLM/Qwen3-Coder ) (⭐8.6k) is the Qwen3 large model, tailor-made by Alibaba Cloud’s Qwen team specifically for coding tasks. It boasts powerful code generation, completion, and understanding capabilities, making it an indispensable coding powerhouse for developers. 💪

2. [neko - AI News]( https://github.com/m1k1o/neko ) (⭐13.3k) is a super cool self-hosted virtual browser project. It runs within a Docker container and uses WebRTC technology to stream the browser screen in real-time, letting you safely surf the web anytime, anywhere. 🌐

3. [awesome-generative-ai - AI News]( https://github.com/steven2358/awesome-generative-ai ) (⭐9.3k) is a meticulously curated treasure map for generative AI. If you’re looking to track the latest developments in the AIGC field or hunt for cool projects, this list is an absolute must-see. 🔥

Social Media Shares

1. Alibaba’s Tongyi Qianwen has just dropped its most powerful translation model, [Qwen3-MT - AI News]( https://qwenlm.github.io/blog/qwen-mt/) ! This beast supports over 92 languages, covering more than 95% of the global population. 🤯 Not only does this model deliver top-notch translation quality, but it also supports customizable features like terminology control, aiming to completely smash global language barriers.
   
   

2. Claude Code has learned the “Shadow Clone Jutsu”! It’s rolled out a brand-new [Custom Subagents - AI News]( https://x.com/dotey/status/1948611389951558047 ) feature. Users can now create specialized agents like “code reviewers” and “QA testers,” allowing them to work together like a mini-team, making complex coding tasks clearer and more efficient. Teamwork makes the dream work! 🚀
   

3. ByteDance has dropped a new version of its simultaneous interpretation model, [Seed LiveInterpret 2.0 - AI News]( https://x.com/Gorden_Sun/status/1948740049220370950) ! This model boasts incredibly low latency and can even automatically clone your voice for seamless Chinese-English translation. The tech is seriously impressive. 🤯
   

4. Always running into bugs when coding? 😩 Cursor has launched [Bugbot - AI News]( https://x.com/op7418/status/1948566558412058871) , a system that automatically detects logic bugs in PRs. This “bug-catching robot” focuses on finding those pesky, hard-to-spot logic errors. Just a heads-up, though, this advanced service requires a paid subscription.
   

5. Wanna make AI even “smarter”? [NVIDIA’s Share - AI News]( https://x.com/NVIDIAGTC/status/1948488421838639209 ) spills the beans on the secret sauce: Chain-of-Thought reasoning and Test-Time Augmentation are totally transforming how AI makes decisions. These two techniques help AI models think through more complex, multi-step problems, seriously boosting their problem-solving capabilities. Mind. Blown. 🤯
   

6. Don’t let your exhibition photos go to waste! 📸 ListenHub has cooked up a special feature for [WAIC - AI News]( https://x.com/oran_ge/status/1948750997012316291 ) that lets you turn multiple images into a podcast with one click. This cool function helps you organize information and solidify memories, making it an absolute must-have for trade show-goers.
   
   

7. Jieyue Xingchen just showcased the powerful multimodal capabilities of its [Step-3 - AI News]( https://x.com/op7418/status/1948725107557560521 ) model during a video call demo. The model responded quickly, identified accurately, and even performed some level of reasoning. This impressive performance is all thanks to its innovative co-design of model and system. Talk about smart! 🧠
   

8. A user has shared their experience with the [Domestic Version of Trae]( https://x.com/dotey/status/1948614928396681239) , which now comes with built-in support for Qwen3 coder and Kimi K2, and get this – it’s currently free! For developers needing a powerful code assistant, this is definitely a solid pick. 👍
   
   

9. How do you unleash the full potential of Gemini Cli? A developer [Shared His Secret]( https://x.com/xiaokedada/status/1948669388430999862) ! By customizing system prompts and combining it with command-line tools like FFmpeg, you can effortlessly process audio, video, and images, turning Gemini Cli into an even more powerful productivity beast. Get ready to level up your workflow! 🚀
   
   
   
   

10. Ever wonder how Anthropic’s internal team actually uses their own product, Claude Code? This [Article]( https://readhacker.news/s/6yqYZ ) pulls back the curtain, revealing their dev workflow from debugging to code assistance. It’s a prime example of how to really leverage AI tools to supercharge development efficiency. Pretty neat, right? ✨

AI Product Spotlight: GeminiCli2API ↗️

Gemini-CLI-2-API is here to set you free! Ever feel tied down by the rate and quota limits of free Gemini APIs, or wish you could use amazing prompts from Tool A in Tool B? Well, this bad boy isn’t just an API proxy; it’s your ultimate AI command center, giving you insane control.

One of our proudest core advantages? Gemini-CLI-2-API cleverly smashes through official free API rate limits using Gemini CLI’s OAuth authorization. Say goodbye to those annoying rate limit exceeded errors and hello to freer, more stable model calls! 🚀

But the real magic lies in the absolute control over “prompts” we hand over to you! This project lets you centrally extract, forcibly replace, or smartly append any system prompts (System Prompt) sent to large models on the server side. What does that mean? You can set up a golden standard for all connected tool clients or even achieve “soul swaps” across tools. Imagine this:

Using Kilo Code Assistant in VSCode, powered by Cursor’s god-tier prompts, to drive any top-tier large model—why be Cursor itself when you can have Cursor’s soul? 🤯

By seamlessly converting various backend model interfaces into standard OpenAI formats, Gemini-CLI-2-API instantly supercharges any of your existing tools. It’s easy to deploy, ready right out of the box. So go on, unleash the full potential of your AI tools and kick off a truly free and efficient AI exploration journey! ✨

Listen to the Voice Version of AI Daily News

🎙️ Xiaoyuzhou	📹 Douyin
Rebirth Tavern	Self-Media Account