07-18-Daily AI News Daily

AI Insight Daily 2025/7/18

AI Daily | Morning 8 AM Update | All-Net Data Aggregation | Cutting-Edge Scientific Exploration | Industry Free Voice | Open Source Innovation Power | AI and Human Future | Visit Web Version ↗️

AI Content Summary

Recent AI news shows Xiaomi launching quick car search, Google Search integrating AI-powered call making, and other practical features.
Frontier research focuses on efficiency and realism, such as Google's MoR architecture and technology that imbues 3D models with physical properties.
Concurrently, the AI industry's talent competition is intensifying, and various open-source tools like ReadMeX are empowering developers.

AI Product & Feature Updates

1. Xiaomi Car’s new Quick Car Search feature is giving YU7 owners a powerful remedy, according to the latest AI news. Still tearing your hair out trying to find your car? With just a tap in the app, your car will automatically snap five “photos” of its surroundings when locked, helping you ditch that lost-in-the-parking-lot feeling for good. This isn’t just basic photography; it’s backed by Xiaomi’s self-developed image understanding large model quietly working its magic, precisely identifying parking spot numbers, floors, and landmarks, so you can spot your car at a glance. Even cooler, just call out “Super Xiaoai” and it’ll tell you your beloved car’s exact location via voice and pop-up. This AI news truly embodies the philosophy of “technology makes lazy people happier” and is definitely worth keeping an eye on. The widespread adoption of this feature marks a significant leap forward in the AI field.

   
   
   

2. Claude Sonnet 4 is back at Windsurf after a dramatic reunion! 🎉 After all the on-again, off-again drama and experiencing a “love-hate relationship” with Anthropic, Windsurf, the AI coding powerhouse, proudly announced that the star model Claude Sonnet 4 is back in its fold, directly accessible to paid users. This AI news sparked heated discussions in the developer community. Previously, due to acquisition rumors involving OpenAI, Anthropic had restricted Windsurf, forcing users to “find workarounds” and driving many developers to rival products. This “rekindled romance” not only lets programmers once again enjoy Claude Sonnet 4’s powerful code generation and refactoring capabilities but also hints at the increasingly delicate competitive and cooperative relationships among AI giants. This whole “breakup and makeup” saga ultimately makes users the biggest winners, standing out as an interesting tidbit in recent AI news.

3. Google Search is rolling out three major AI innovations, vowing to transform from a mere information portal into your ultimate personal assistant. ✨ Still manually searching for info and making calls with your old search engine? You’re way behind the curve! First up, Google has equipped its AI search mode with Gemini 2.5 Pro, its “most powerful brain,” making complex reasoning, programming, and calculations a piece of cake. This AI news is super exciting! Next, Deep Search, crafted for in-depth research, will also become available, generating comprehensive reports with citations in a single click, instantly turning you into an expert. But the real jaw-dropper is the debut of AI-powered call making: simply search for “pet groomers nearby,” and the AI will automatically contact businesses, inquire about prices, book appointments, and might even haggle for you. 🤯 This feature, first launched in the US, signals that AI is reshaping local services, and future AI news will increasingly focus on such applications. Soon, you might not even be talking to a human when you pick up the phone! This is undeniably one of the week’s most significant pieces of AI news.

   
   
   

4. ChatGPT’s macOS desktop app now has a recording mode fully rolled out to Plus users! 🎉 You talk, ChatGPT types! Whether you’re brainstorming or just jotting down fleeting ideas, it transforms into your personal meeting note-taker, not only transcribing in real-time but also automatically extracting key points and even directly generating task lists and code. This AI news holds significant implications for the future of productivity tools. From now on in meetings, your boss won’t have to worry about you zoning out, because AI handles your notes and summaries, maxing out your efficiency. This is undoubtedly a major stride forward in the AI news landscape. Go check out AI News: the new button at the bottom of the chat window!

   
   
   

AI Frontier Research

1. Google DeepMind has unveiled a game-changing solution with its new MoR (Mixture-of-Recursions) architecture, which could become the “efficiency killer” for traditional Transformer models. While large models are powerful, their hefty computational costs have always been a major bottleneck for widespread adoption. Now, Google DeepMind is shaking things up with its brand-new AI News: MoR Architecture (Mixture-of-Recursions)! MoR’s genius lies in not treating all information equally; instead, it intelligently allocates just the right amount of “thinking depth” to each data token via a dynamic routing mechanism. This “on-demand computation” strategy, combined with advanced caching, allows MoR to achieve or even surpass Transformer performance using less than half the parameters. This isn’t just a simple optimization; it’s more like an architectural revolution, heralding a new era where AI becomes lighter, more efficient, and easier to deploy at scale. This AI news is a must-read for all industry professionals.

   
   
   

2. Two papers from Mila Lab are tackling the “lag” and “inaction” issues plaguing reinforcement learning, finally giving AI some “reflexes”! Is AI finally saying goodbye to being “slow on the uptake”? Imagine AI playing a fighting game, having to “pause to think” for a second before every punch – how awkward would that be? That’s precisely the current predicament of reinforcement learning (RL) in real-world scenarios. Traditional AI training is like a turn-based game, with the environment and AI taking turns “spacing out,” leading to robot chefs burning omelets and self-driving cars being a beat too slow. Two ICLR papers from Mila Lab hit the nail on the head, proposing an ingenious solution to end AI’s two major ailments: “delay” and “inaction.” The first paper introduces “interleaved asynchronous inference,” akin to giving AI a multi-core brain, allowing multiple “thought processes” to rotate shifts, ensuring actions never drop offline. The second paper draws inspiration from CPU pipeline design, creating a neural network architecture with “temporal skip connections.” With both techniques combined, it means we can deploy massive yet lightning-fast AI models in the future, allowing them to truly shine in rapidly changing fields. This is precisely the focus of this issue’s AI news.

   
   
   
   
   

3. New research, “PhysX: Physics-Based 3D Asset Generation,” is giving 3D models a physical soul. AI isn’t just about “face-sculpting” anymore; it’s also learning “bones and muscles”! Current 3D generative AI acts like a top-notch artist, capable of “sculpting” stunning virtual objects, but completely oblivious to physical properties. This makes them mere “pretty faces” or “ornaments” in serious applications like physical simulation and robot training. Researchers have introduced a groundbreaking paradigm, PhysX, aiming to imbue 3D assets with a physical soul. To achieve this, they’ve built the first 3D dataset with systematic physical annotations, PhysXNet, and developed the PhysXGen framework, which can generate 3D models with both realistic appearance and credible physical properties from a single image. This work marks AI-generated content’s shift from “virtual aesthetics” towards “physical realism,” paving the way for creating more reliable digital twins and accelerating the development of embodied AI. It’s a cutting-edge exploration in the AI news sphere.

AI Industry Outlook & Social Impact

1. ByteDance’s AI talent war is stirring up fresh waves, as a “talent war” without gunpowder quietly unfolds. Yang Jianchao, the head of Doubao’s large model visual multimodal team, has announced a “temporary break,” with Zhou Chang, former tech lead for Alibaba’s M6 large model, stepping into his shoes. This seemingly calm personnel shift is actually a microcosm of the global AI talent war, and a hot topic constantly tracked in AI news. Yang Jianchao’s temporary departure reflects the unimaginable work intensity and pressures of cross-border collaboration in the cutting-edge AI domain. Zhou Chang’s arrival, meanwhile, showcases ByteDance’s determination to spare no expense in “recruiting talent” globally to solidify its Seed visual model team. In this high-stakes game among tech giants, the coming and going of every key figure could potentially alter the future trajectory of the tech landscape. This kind of AI news reveals the fierce intensity of industry competition.

   
   
   

Top Open-Source Projects

1. ReadMeX, a free domestic AI tool, is making documentation a breeze for developers. For most developers, coding is creation, but writing documentation? That’s a “修行” (cultivation/arduous practice). Now, a free domestic AI marvel called ReadMeX - AI News is making this “cultivation” a walk in the park! Just throw a GitHub project URL at it, and it’ll churn out a well-structured, professional README file in a flash, complete with both Chinese and English support, just like a seasoned technical writer. Its speed and quality even outshine similar well-known tools. ReadMeX doesn’t just generate docs; it also integrates multi-repository management and popular open-source project documentation aggregation, turning it into a powerful knowledge management platform. The arrival of this tool is undoubtedly a massive gift to the developer community, allowing everyone to pour more energy into true innovation. This AI news is incredibly valuable for developers.

   
   
   

2. Three game-changing open-source marvels, hand-picked by AI News as developer treasures, are here:

   • Data Chiropractor: Facing tangled messes of JSON or YAML files that make your eyes want to “pop out”? Hold up! Let JSONCrack - AI News give your data a “chiropractic adjustment.” This mind-blowing tool, boasting over 40,000 Stars, can convert various data formats into interactive charts with a single click, instantly clarifying your data structure.
   • SQL Translator: Still pulling your hair out trying to write complex SQL queries? The Vanna - AI News project leverages RAG and Large Language Model technology to accurately translate natural language into SQL. This AI news suggests that AI is mercilessly flattening the barrier to data analysis.
   • Digital Sherlock Holmes: Curious how many digital footprints your username has left across the internet? Maigret - AI News can search for your “digital doppelgangers” across thousands of websites based on a single username, giving you a peek into the formidable power of Open Source Intelligence (OSINT).

3. Strapi, a leading open-source Headless CMS, acts as a “central kitchen” for content, offering unparalleled flexibility. If you were opening a restaurant, would you prefer a “franchise package” with a fixed menu, or a “central kitchen” where you could freely experiment? Most developers would pick the latter, and Strapi - AI News is exactly that “central kitchen.” As a leading open-source Headless CMS, it completely separates content management (backend) from content presentation (frontend). This means you’re free to publish your content wherever your heart desires. Built entirely with JavaScript/TypeScript, Strapi boasts over 67,000 GitHub Stars and has become the top choice for countless dev teams craving flexibility, thanks to its high customizability. This AI news is crucial for building modern applications.

4. The popular open-source canvas SDK, tldraw, now officially supports building Agentic Workflows. Building AI agents just got a new “LEGO” brick! This means developers now have another powerful visual “sandbox” where they can design and debug complex AI agents like building with LEGOs. Essentially, tldraw cleverly merges the freedom of a canvas with the structure of a flowchart, opening up AI News: infinite possibilities for creating the next generation of smart interactive products.

   
   
   

Social Media Buzz

1. Kimi K2, a domestic AI newcomer, is offering a fascinating “ambient programming” experience, allowing users to build projects in two days without writing a single line of code. A netizen shared their incredible “ambient programming” experience using Kimi K2. According to them, Kimi K2 not only writes code but also has excellent aesthetic sense, actively adding dynamic effects and finding style-consistent materials, completely shedding the stereotype of “straight-laced AI.” It can even precisely grasp “metaphysical” demands like “immersion” and performs quite stably in API integration. This AI news is super intriguing! While its speed and long-dialogue memory still have room for improvement, considering its programming ability is nearly 80% that of Claude Code, and its cost is “outrageously” low, it’s truly AI News: a dark horse in domestic AI code generation.

2. Midjourney, the king of AI painting, is finally stepping into the arena to deliver a “dimension reduction attack” on the industry by exploring enterprise-level APIs. The official word is they’re exploring opening up an enterprise-level API, which means developers will soon be able to directly integrate Midjourney’s “magic brush” into their own products. AI News: The official application channel is now open – get ready for a fresh wave of creative explosions!

3. Collaborating with AI in programming is proving to be a rollercoaster ride, as one netizen shared their “mood rollercoaster” 🎢 experience. The experience of collaborating with AI, this new partner, isn’t always smooth sailing. From being initially amazed by Vibe Coding, to getting so frustrated by AI-generated bugs they wanted to “tear apart the code with their bare hands,” and finally, reluctantly accepting the reality that “it’s not unusable.” As they put it, AI News: many are stuck in the mire of AI programming: simultaneously despising it and being unable to live without it.

4. AI acts more like a brutally honest “harsh friend,” curing your “vague idea syndrome” by speaking with raw results, forcing you to upgrade your thinking. But perhaps this “torment” is precisely the unique value AI brings. Another netizen observed that AI is more like a brutally honest “harsh friend,” specifically designed to cure your “vague idea syndrome.” It will mercilessly “slap you in the face” with a prototype demo full of flaws, forcing you to organize your wishful thinking into clear logic. Unlike endlessly debating with human colleagues, AI News: AI speaks with raw results, compelling you to upgrade your thinking, enabling self-improvement through this “torture.”

Listen to the Audio Version of AI Daily

🎙️ Xiaoyuzhou FM	📹 Douyin
Laisheng Xiaojiuguan (Afterlife Tavern)	Official Account