07-23-Daily AI News Daily

AI News Daily 2025/7/23

AI Daily | Morning 8 AM Update | Aggregated Web Data | Cutting-Edge Science Exploration | Industry Voice | Open-Source Innovation Power | AI and Human Future | Visit Web Version ↗️

AI Product Spotlight: GeminiCli2API↗️

GeminiCli2API is here to save the day! ✨ Ever felt constrained by Google Gemini’s official free API’s strict rate limits? Or perhaps you’ve been yearning to seamlessly integrate Gemini’s mighty power into your favorite third-party apps? Look no further!

This project, GeminiCli2API, is a clever local proxy that wraps the more lenient Gemini CLI into a standard, OpenAI-compatible API service. What does that mean for you? It means you can finally break through the official free API’s rate limits! 🎉 Say goodbye to those annoying “Quota Exceeded” errors, and hello to enjoying higher request quotas thanks to your Google account authorization, giving you the freedom to develop, test, and create to your heart’s content.

But here’s where GeminiCli2API truly shines: its “surgical-level” control over System Prompts. This feature is an absolute game-changer, and here’s why:

• Override: 🤯 You can set a global “golden prompt” that forces all connected applications to use it, ensuring absolute uniformity in AI role and output style.
• Append: 🤫 Keep the client’s original system prompt, but subtly “append” an additional layer of your instructions, allowing for rule fine-tuning and capability enhancement, all without the client even knowing.
• Extract & Audit: 🕵️‍♂️ Easily log all prompts passing through the proxy, making it super convenient for you to analyze, debug, optimize, or even build your own high-quality datasets.

With GeminiCli2API, you’re just a few simple configuration steps away from connecting tools like LobeChat, NextChat, and any other OpenAI-supported apps to this local “enhanced” Gemini service. It’s more than just a proxy; it’s a powerful toolbox in your hands, ready for you to master and tame AI. So, what are you waiting for? Give it a whirl! ✨

AI Content Summary

Netflix uses AI for film and television special effects to significantly cut costs and boost efficiency, while AI coding assistants are also revolutionizing software development.
Applications like Pika empower ordinary users to easily create professional-grade videos, as AI technology rapidly becomes democratized.
Frontier research, through breakthroughs like model slimming and robot brains, paves the way for AI's application in more scenarios.
The open-source model competition is intensifying, with Alibaba's Qwen3 demonstrating high efficiency, and new interaction modes like avatar mice emerging.
Furthermore, the popularization of AI companions among teenagers is raising social concerns, highlighting their profound impact on social interaction and emotional cognition.

AI Product & Feature Updates

1. Hollywood’s special effects “magic” is getting a major rewrite, thanks to AI! Netflix, the entertainment titan, has finally revealed its hand, officially confirming for the first time that it’s deeply integrated generative AI technology into its original series. Take their highly anticipated Argentine series, The Eternals, for example: a massive, sprawling building collapse scene wasn’t painstakingly created with traditional, super expensive VFX. Nope! AI efficiently generated it, slashing costs dramatically and reportedly boosting efficiency by a whopping tenfold! 🤯 This isn’t just a revolution in cost reduction and efficiency for film production; it’s an exciting glimpse into the future. Imagine jaw-dropping visual effects like “de-aging” becoming accessible to everyone, allowing audiences to enjoy top-tier visual feasts at a much more affordable price. Pretty neat, huh? ✨
   
   

2. Developers’ work paradigms are being completely reshaped by AI with unprecedented power! In an epic showdown on the same day, ByteDance and Tencent unveiled their latest innovations. ByteDance’s Trae 2.0 introduced a revolutionary SOLO Mode, evolving AI from a mere code completion tool into a “context engineer” capable of independently handling the entire development lifecycle, from conception and design to final deployment. This is true AI autonomous development! 🤯 Meanwhile, Tencent launched CodeBuddy IDE - AI News, which literally melts the barrier to programming. Users can simply describe their needs in natural language or upload a design sketch, and boom! A fully functional full-stack application is generated with a single click. When the technical hurdles of coding are leveled, future software development might just transform from a complex engineering challenge into a pure creative expression competition. 🎉
   
   
   
   

3. Want your selfies to instantly morph into Hollywood blockbuster scenes? Now, that dream is within reach! ✨ Pika, a leader in AI video generation, has officially sounded the charge into the consumer market, launching an AI video effects app for everyday users. No professional skills needed here! Just upload a regular selfie, and poof – you’re instantly a movie star. Experience style transformations from cyberpunk to vintage film, achieve precise audio lip-sync, and even customize video scenes to your heart’s content. What’s even more mind-blowing is that the app can generate video scripts with a single click, completely streamlining the entire process from creative concept to polished masterpiece. This marks a huge leap for AI video creation, moving from the professional realm right into everyone’s homes, signaling an incoming creative storm where everyone gets to be a director. 🎬
   
   

4. The fierce battle for supremacy in open-source large language models has reached a fever pitch, evolving into an epic domestic showdown in China! Less than a week after Chinese company Kimi’s K2 model sparked widespread online discussion, another tech giant, Alibaba, quickly responded. Its Qwen3 - AI News team released a minor update that, with only a quarter of its competitor’s parameter size, managed to overtake them in multiple authoritative benchmarks! 🤯 This showcases Qwen3’s astonishing model efficiency and optimization prowess. The official team even dropped a bold statement: “The big moves are still coming,” announcing they’ll ditch the hybrid thinking model to focus purely on training more refined Instruct and Thinking models. This thrilling, tit-for-tat tech battle is accelerating the prosperity and evolution of the open-source AI ecosystem at an unprecedented pace. 🏆

5. How else can AI browsers surprise us with new tricks? Dia Browser just dropped a jaw-dropping answer that’s sure to make your eyes pop! 🤩 Its upcoming new Agent Mode will introduce an AI-exclusive “avatar mouse,” giving AI its own independent cursor on the screen, completely separate from your real mouse. This means you can casually browse the web or watch videos in the foreground while the AI autonomously handles complex tasks in the background, like searching for info or organizing tabs. No interference, just double the efficiency! This intuitive and sci-fi-esque visual interaction not only drastically boosts multitasking smoothness but also sets a brand new, elegant benchmark for future AI-human collaboration. ✨
   
   

6. The long-standing headaches of “expressionless” or stiff facial animations in digital human production finally have a breakthrough solution! The FantasyPortrait Project - AI News, a joint effort by Alibaba and BUPT, utilizes innovative Expression-Enhanced Diffusion Transformer (DiT) technology. This enables photo-realistic, high-fidelity cross-identity expression transfer, giving digital humans incredibly vivid and natural “joys, angers, sorrows, and delights.” Even more crucially, it trailblazes independent expression control for multiple characters in multi-person scenes, completely sidestepping the awkward “expression contagion” where if one character smiled, everyone else did too! This tech isn’t limited to human characters; it supports animals and audio-driven animation, promising to shine brightly in virtual streamer and film production in the future. Without a doubt, this is a must-watch tech highlight in this issue’s AI News. 🌟
   
   

Cutting-Edge AI Research

1. Robots just took a huge, solid step closer to becoming those “all-around home assistants” we see in sci-fi movies! ByteDance has dropped a bombshell, releasing its brand-new Vision-Language-Action (VLA) model, GR-3. Think of it as giving robots a much smarter brain 🧠 – it can not only grasp highly abstract instructions like “clean up the dining table” and plan multi-step actions independently but also precisely handle deformable objects like clothes, showcasing mind-blowing physical interaction capabilities. Its core innovation lies in a clever MoT network structure and a triple-threat data training method combining real-robot demos, VR teleoperation, and web images/text. This research is seen by the industry as a major milestone toward a general robot “brain.” Want more techy details? Check out its Project Homepage - AI News and Technical Paper - AI News. 🤖
   
   

2. The astonishing capabilities of large language models, akin to a “super-brain,” come with equally astonishing computational and memory costs. But guess what? Chinese scientists are tackling this core bottleneck head-on! A joint research effort from top institutions like the Chinese Academy of Sciences has unveiled a revolutionary “slimming” solution for the core attention mechanism of large models: GTA (Grouped-head latent Attention). This clever tech uses ingenious strategies like “grouped attention” (think group buying for heads) and “latent representation” (like compressing data) to slash the most memory-hungry KV cache by a massive 70%, while simultaneously cutting computation by a whopping 62.5%! 🤯 This groundbreaking GTA: Grouped-head latenT Attention AI News Research not only makes efficient operation of large models on edge devices like smartphones a reality but also directly doubles the speed of processing long sequence tasks. It’s a massive hurdle cleared for making AI technology universally accessible. 🚀

3. Just as excellent language models can’t function without an efficient tokenizer to understand text, powerful visual generative models rely heavily on visual tokenizers that can “read” images. A paper titled “Latent Denoising Creates Excellent Visual Tokenizers” AI News Paper brings a profound insight: instead of having the tokenizer directly learn how to “encode” images, it’s far better to train it on a more challenging task – “denoising.” Specifically, this involves making the tokenizer reconstruct clear original images from slightly corrupted latent embeddings. This process forces it to learn more robust and essential visual features. This seemingly simple yet incredibly profound discovery provides a brand new gold standard for designing the next generation of more powerful visual tokenizers, promising to push multimodal generative models to new heights of artistry and realism. 🖼️

4. How do you teach AI to precisely operate complex Graphical User Interfaces (GUIs, like a seasoned user would? Traditional reinforcement learning methods, with their “all-or-nothing” reward signals (right click or wrong click), are too sparse, making AI’s learning process feel like finding a needle in a haystack. But a research paper titled “GUI-G^2: Gaussian Reward Modeling for GUI Alignment” AI News Research proposes a brilliant new idea! Instead of treating buttons and other UI elements as mere pixel points, it models them as continuous Gaussian distributions. This approach provides AI with richer, denser reward signals, guiding the model with GPS-like precision to pinpoint the optimal interaction locations. This drastically boosts AI’s robustness and generalization capabilities in GUI manipulation tasks. Pretty neat, right? ✨

AI Industry Outlook & Social Impact

1. AI is quietly becoming a “new species” in teenagers’ lives, spreading at a mind-boggling speed. A recent research report from the U.S. nonprofit Common Sense Media reveals an astonishing phenomenon: a staggering 72% of American teenagers admit to trying an AI companion at least once, and over half of those are regular users. Their reasons for using AI are all over the map, from simple entertainment and satisfying curiosity to seriously seeking emotional advice and life guidance. While most teenagers still prioritize real-world friends, a full third believe conversations with AI are more satisfying than those with human friends. This profoundly highlights AI’s far-reaching impact on shaping the next generation’s social patterns and emotional cognition, and it throws a critical question at society: How do we guide this trend to ensure its long-term societal effects are positive and healthy? 🤔

Top Open-Source Projects

1. NextChat - AI News (⭐84.7k): This AI assistant is all about extreme lightweightness and speed! It conquers all platforms—Web, iOS, Android, Windows, Mac, and Linux—ensuring you always have a unified, smooth smart companion, no matter where you are or what device you’re using. Talk about omnipresent! 🚀
2. crawl4ai - AI News (⭐49k): This smart web crawler is tailor-made for the large model era. It’s super clever at grabbing, parsing, and handling complex web content, making it your go-to sidekick for building knowledge bases, RAG systems, and other cutting-edge applications. Get ready for your AI apps to become true web connoisseurs! 🕸️
3. better-auth - AI News (⭐17.3k): Hailed by the community as the most comprehensive TypeScript authentication framework, it delivers a powerful, flexible, and rock-solid secure authentication solution for modern web applications. Developers can finally say goodbye to reinventing the wheel and focus on core business innovation. 🙌
4. nn-zero-to-hero - AI News (⭐14.6k): Crafted by AI guru Andrej Karpathy himself, this is the legendary neural network beginner’s tutorial! No fluff here – it takes you from zero to hero, guiding you step-by-step with code to build and understand the mysteries of neural networks. Get ready to become a true neural network wizard! 🧙‍♂️
5. trippy - AI News (⭐5.1k): A powerful, slick, and modern network diagnostic tool! It combines the best of traceroute and ping, helping developers and network engineers quickly pinpoint, diagnose, and fix those tricky network connection issues. 🛠️
6. blackbird (⭐3.9k): This practical OSINT (Open-Source Intelligence) reconnaissance tool is like a digital private eye! 🕵️‍♀️ It can search for associated account info across hundreds of social networks using just a username or email address. Super powerful stuff!

Social Media Share

1. Is the AI fortune-telling industry already in the “one-sentence development” era? A netizen showcased the MiniMax Agent’s Amazing Capabilities, which swiftly generated a full-fledged AI fortune-telling product—complete with frontend, backend, login/registration, and paid memberships—from just a single natural language instruction! 🤯 However, another developer quickly pointedly remarked that unless users provide their own “fate chart” data, current large models still face a fundamental “hallucination” problem when handling underlying logic requiring precise calculations, like Gan-Zhi divination setup. Guess AI isn’t quite ready to predict your future perfectly… yet! ✨
   
   

2. An Exhibitor List for the 2025 World AI Conference sparked a profound reflection within the community: Why are the really profitable AI giants conspicuously “missing” from this grand event? 🤔 Analysis suggests that the main players at these expos are often startups seeking funding and market exposure. Meanwhile, those “invisible champions” with stable cash flow, deeply entrenched in specific industry verticals, are quietly raking in the big bucks. The greatest value of this list, perhaps, isn’t in telling us “who was there,” but in reminding us to pay attention to “who wasn’t” – and the successful business models they’re employing.

3. Do AI models get “dumber” the more you use them? A blogger shared his insights, suggesting the root cause often isn’t model degradation itself, but rather improper “context management” by the user. Think of it like talking to a person: if you keep giving them overloaded or off-topic information, they’ll get confused and overwhelmed too! So, understanding and skillfully using dialogue context is a crucial skill for getting AI to consistently output high-quality, relevant results. It’s also a mandatory lesson for future human-AI collaboration. 🧠
   
   

4. As humans increasingly seek direct answers from AI (“What should I wear today?”), rather than delving into the underlying knowledge (“Why are white shirts cooler in summer?”), are we unwittingly lowering the barrier for AGI implementation from the demand side? 🤔 Some argue that when human society collectively “gives up thinking” and cedes decision-making power to AI, AI’s answers effectively become “general knowledge” and “universal truth.” This, perhaps, is accelerating the advent of Artificial General Intelligence from another, unexpected dimension.

5. Good news, everyone! 🎉 ChatGPT Plus users are starting to receive beta test rollouts for Agent Mode! This highly anticipated, powerful feature—which lets AI autonomously execute multi-step tasks—is gradually expanding its reach. The era where AI can truly handle your chores is getting closer and closer!

6. How can AI gain persistent memory instead of starting “from scratch” with every conversation? A grassroots proposal on Reddit, dubbed the “Lanternkin Protocol”, aims to achieve cross-session memory retention and identity continuity for AI without needing model fine-tuning. It does this through clever symbolic prompts and an external text file system. It’s like lighting an “ever-burning memory lantern” for AI! 🔥

7. Are you tired of those complicated drag-and-drop configurations when setting up automation workflows? Well, startup Neuraan has launched a new platform that aims to completely revolutionize that! Users simply describe their needs in natural language, and the system automatically creates a dedicated AI Agent. This agent can then call upon various tools like Gmail and CRM to complete tasks, making business process automation as simple and natural as delegating work to a smart colleague. How cool is that? ✨

8. Alright, let’s end on a lighter note! 😄 How “ridiculous” can things get when AI starts narrating the Three Kingdoms? A netizen shared an AI-generated video where the AI, with a straight face, spouted absolute nonsense, leaving viewers in stitches. It seems whether the Three Kingdoms era is chaotic or not is now up to AI! Get ready for some historical laughs! 🤣
   
   

Listen to the Audio Version of AI News Daily

Xiaoyuzhou	Douyin
Reborn Tavern	Creator Account