AI资讯 2025/7/29
AI 日报
AI内容摘要
AI技术正深刻变革工具应用，如设计与浏览器智能化。
但其环境碳足迹及高昂使用费引担忧，内容治理亦收紧。
头部企业如京东阿里加速AI落地，推智能人、汽车系统。
Today’s AI News
The tech world has been abuzz recently, with the AI wave thoroughly transforming how we use tools! From design to browsing, and even AI’s own “carbon footprint” beginning to face public scrutiny, it’s a magical reality where a sense of future coexists with a sense of responsibility.
First, imagine being a designer without needing Photoshop! Canva and Claude AI, this new duo, have directly turned design from a “professional skill” into “chat commands.” You just need to type “design a new product launch poster for me” in Claude’s chat box, and the AI will instantly understand your intent, helping you present a visual feast in Canva! Behind this is Anthropic’s Model Context Protocol (MCP), acting as a “translator,” enabling seamless dialogue between AI and various applications. This isn’t just about convenience; it’s about lowering the creative threshold to the floor, truly realizing the vision of “everyone can design” and allowing non-professionals to easily create professional-grade content. From now on, your “flash of inspiration” will no longer be constrained by cumbersome software!
Next, your browser is also “supercharged”! Microsoft’s Edge browser has launched the explosive Copilot mode, directly transforming the browser from a passive tool into your intelligent AI assistant. What’s most impressive is the multi-tab RAG (Retrieval Augmented Generation) feature, which acts like a super scholar, helping you perform deep analysis and summarize the content of multiple tabs you have open simultaneously. For instance, when you’re writing a thesis or conducting research, it can quickly extract key information and even compare different viewpoints. There’s also the even cooler Copilot Vision, which can “see” the content on your screen. If you’re looking at lipstick on an e-commerce site, it can immediately help you with price comparison and recommend similar products. Isn’t this like bestowing the superpower of “seeing and hearing everything” directly upon the browser? In the future, browsing the internet won’t be about you passively searching; instead, information will actively come to you, with an efficiency that makes you want to shout!
However, behind AI’s “magic,” are there hidden “costs”? Mistral AI, a company from Paris, bravely peeled back a corner of AI’s glossy exterior, performing an “environmental check-up” on its Large Language Model Large2 for the first time and releasing a thought-provoking environmental impact analysis report. The study found that AI models are the biggest “energy hogs” during the training and inference (i.e., when you chat with it) phases, contributing 85.5% of greenhouse gas emissions and 91% of water consumption. Don’t underestimate the “invisible consumption” of each chat. For example, every 400 characters you chat with “Le Chat” generate approximately 1.14 grams of CO2 emissions and 45 milliliters of water consumption. This makes us ponder: as we enjoy the convenience and efficiency brought by AI, should we also pay attention to its “invisible burden” on the planet? Mistral AI’s transparency is commendable, and it also calls on the entire AI industry to act, making AI “greener” as it rapidly develops. After all, Earth is our common “server”!
In summary, AI is permeating every aspect of our lives at an unimaginable speed. It can be a “powerful assistant” for our creativity and productivity, yet also inadvertently leave a heavy “environmental footprint.” How to balance technological innovation with sustainable development, allowing AI to both illuminate our lives and not deplete our future energy, is undoubtedly a major challenge for all tech companies and for each of us.
Recently, the digital world’s compass has shifted again! On one hand, AI is introducing new tricks in image creation, with autoregressive models making a comeback, vowing to challenge the popular diffusion models; on the other hand, our digital life housekeeper — the Central Cyberspace Administration — has begun to crack down heavily on “self-media” publishing false information, especially targeting the chaos of AI-generated content. Of course, there’s also a thoughtful small upgrade: NotebookLM’s “featured notebooks” function has fully launched, allowing you to more conveniently access the latest information.
First, let’s talk about the “metaphysical” realm of AI image generation. While Diffusion models currently create excellent images, achieving “pixel-level” precise control or simultaneously understanding multiple inputs like text and images often leads to “deviations,” and their training is costly and time-consuming. However, recently, “masters” from top research institutions like Tsinghua and Peking University have unleashed a big move: they’ve revived the long-dormant Autoregressive model (AR), introducing a new framework called MENTOR. This guy is formidable; through a unique “two-stage training method,” it not only achieves pixel-level precise control and effectively solves the problem of “neglecting one thing for another” with multi-modal input, but also requires only one-tenth of the training data of diffusion models and is faster! It’s simply a model of “cost reduction and efficiency increase.” MENTOR achieves a remarkable balance of performance in image reconstruction fidelity and multi-modal instruction following with smaller model sizes and fewer training resources. Although MENTOR still has room for improvement in some extreme details, it undoubtedly opens up new avenues for AI image creation, demonstrating the immense potential of the autoregressive paradigm in the visual generation field. The future is promising!
That said, AI, a double-edged sword, is art when used well, but troublesome when misused. The Central Cyberspace Administration has set its sights on “content black workshops” masquerading as “self-media,” maliciously exploiting hot topics, forging data, and even using AI to synthesize fake news. Starting from July 24, 2025, a special rectification action lasting two months will begin. The Cyberspace Administration requires major platforms to strengthen management: mandatory labeling of information sources, with no source, don’t expect to enter the algorithm recommendation pool; refine professional qualification certification processes, no more allowing uncertified or falsely qualified “experts” to spout nonsense; and more importantly, to optimize AI Generated Content (AIGC) identification functions, so everyone can instantly recognize what AI has “concocted.” This series of actions aims to make the online world less manipulative and more sincere, creating a credible, orderly online environment for the public.
Of course, there are also positive technological developments. For instance, Google’s NotebookLM has finally made its “featured notebooks” function fully available to everyone! Now, as soon as you log into the homepage, you can see the latest, hottest news reports in a prominent position, covering everything from tech trends to industry dynamics. This greatly simplifies the steps to obtain information, allowing users to instantly access curated content without additional operations. No more painstakingly searching for materials; you can efficiently access information anytime, anywhere, and it supports multi-platform access, providing a consistent, smooth experience on both desktop and mobile devices. This is truly a thoughtful little helper for boosting efficiency!
Doesn’t this series of news make you curious about the future of the digital world? From the battle over AI model technical routes to the governance and standardization of online content, and the popularization of personalized information access tools, we are in an era of rapid technological development and continuous rule refinement. The AI wave is surging forward, bringing infinite possibilities but also new challenges. As digital citizens, we should not only enjoy the convenience brought by technology but also keep our eyes open, learn to discern information authenticity, and jointly build a “true, good, and beautiful” digital home!
The second half of the AI battle is no longer about who has larger model parameters or better performance scores, but about who can truly “move” AI closer to you and me, transforming it from a aloof laboratory “nerd” into a “versatile partner” in life and work! The recent WAIC (World Artificial Intelligence Conference) has become the central showcase for this “implementation breakthrough,” and our tech giants JD and Alibaba have thoroughly “figured out” the “roadmap” for deep AI application, each with their unique skills!
JD: AI with a “Warm” Human Touch JD has upgraded its large model brand to “JoyAI” (meaning “Enjoy AI”), a name that immediately evokes a sense of everyday life. Their AI is no longer cold technology; it’s beginning to have a “human touch”:
- Digital Humans Take Center Stage, Outperforming Real Streamers! Remember Liu Qiangdong’s digital human live stream last year? Now, JD’s digital humans are not just eye-catching “vases”; upgraded by JoyAI, their movements are more natural, and emotions richer. They can introduce products and conduct hands-on tests like real streamers, and during “618,” their sales performance even surpassed 80% of real streamers on the market! This technology is so impressive that even high-end luxury brands are actively seeking them out, and they’ve won China’s highest award in intelligent science and technology — the Wu Wenjun Artificial Intelligence Science and Technology Award Special Prize. This isn’t just a digital human; it’s a “digital sales elite”!
JoyInside: Infusing “Souls” into Smart Hardware! JD’s newly launched “JoyInside” smart platform, as its name suggests, sounds a bit mystical. In reality, it aims to inject AI’s “wisdom” and “emotional intelligence” into hardware like robots and smart toys, enabling them not only to move but also to chat with you, remember your preferences, and become a “warm companion”! Imagine your little toy not only being cute but also being able to “give advice” when you’re stressed from overtime. Isn’t this like science fiction’s “intelligent cuddly pets” coming to life? Currently, over ten enterprises have connected to JoyInside, including companion dolls, chess robots, and even industrial humanoid robots and robot dogs, transforming them from mere “tool-humans” that only execute commands into beings with “high-EQ souls.”
JoyAgent: AI Agents’ “Internal Practice” and “External Openness”! The reason JD’s AI can be so “down-to-earth” so quickly is because it has been “honed” in its vast retail and logistics supply chain scenarios. Internally, they already have over 20,000 agents at work, undertaking over 18% of the company’s work content. Now, JD has also open-sourced these “battle-tested” JoyAgent platforms. Whether you’re an individual developer, a start-up company, or an enterprise-level developer, you can “take and use” them to quickly achieve intelligent upgrade. This move is like sharing their own “secret sauce” to collectively “raise the bar.”
Alibaba: From “Movie-Grade Art” to “Proactive Cockpit” Alibaba has also created new wonders in the “second half” of AI, not only pursuing “artistic flair” in technology but also “driving” AI into cars:
- Wan2.2: The World’s First MoE Video Generation Model, Packed with Cinematic Feel! Alibaba recently open-sourced a video generation model called “Wan2.2,” touted as the global first MoE architecture video generation model. Its most impressive feature is its ability to produce videos with “movie-level aesthetic effects.” Everything from lighting, composition, and color can be precisely controlled, generating 720P, 24fps HD video that can even run on ordinary graphics cards. This is like embedding a film director’s “shot list” and “artistic aesthetic” into the AI’s “brain,” allowing ordinary people to generate a blockbuster feel with “one-click blockbuster generation.”
Banma Zhixing: Turning Cars into “Proactively Thinking” AI Robots! Alibaba’s Banma Zhixing also shone brightly this time. Its equipped Yuan Shen AI large model achieved all-item first in international evaluation agencies for smart cockpit capability assessment, considered to have initially achieved L3-level cockpit capability, practically an “honor student” in intelligent cockpits. Banma Zhixing, in collaboration with Alibaba’s Tongyi large model, directly embeds AI capabilities into cars, allowing the vehicle not only to understand your speech but also to “proactively think.” For example, it can turn on the AC based on your state, recommend music in traffic, or even order coffee in the car with a single sentence, with the robot delivering it directly to your vehicle! Isn’t this like moving Musk’s envisioned “intelligent restaurant” into the car? This not only demonstrates Alibaba’s advantages in software and hardware integration but also takes Chinese automotive “smart wisdom” to the world stage.
In conclusion, whether it’s JD’s “warm” digital humans and robots or Alibaba’s “movie-grade” video generation and “proactively thinking” smart cars, they all tell us that AI is no longer an ivory tower; it is permeating every aspect of our lives in various vivid and interesting ways, truly becoming an indispensable “good helper” by our side, making technology more interesting and convenient! The “deep application” breakthrough battle of AI has just begun!
“My AI bill is already $350 this month, and I feel like I’ll miss the ‘good old days’ now!” Reddit user DavidCBlack’s image post, though brief, echoed the sentiments of many. This AI was originally meant to save us time and effort, but as we use it, the bill keeps skyrocketing! If $350 already hurts, what will happen in the future when AI is ubiquitous in every household and permeates every aspect of work and life? Will its usage cost become our new major “monthly payment”? It seems that while enjoying the convenience of AI, we must also begin to consider: Is this AI craze really saving us money, or is it quietly becoming a new burden on our wallets? In the future, will AI freedom become a luxury item?