09-18-Daily AI News Daily

AI News Daily 2025/9/18

AI News | Daily Morning Read | Aggregated Web Data | Cutting-Edge Science Exploration | Industry Free Voice | Open Source Innovation Power | AI and Human Future | Visit Web Version ↗️ | Join Group Chat

Today’s Digest

Figma and Gamma are shaking up design and content creation, integrating AI to let you revolutionize workflows with natural language.
Li Fei-Fei's team just dropped a model that can cook up massive 3D worlds, but watch out: fresh visual attacks are exposing multimodal AI's security weak spots.
Google teamed up with industry titans to forge a payment protocol for AI agents. Meanwhile, PyTorch is absolutely crushing it in the open-source large model space, where competition is cutthroat!
Open-source projects are buzzing about Xiaohongshu’s audio models and Google’s TimesFM, a new time series prediction model. Developers are seriously hyped!
Tech giants are playing hardball, using political moves to sway AI legislation. Plus, the emotional bonds forming between humans and AI? That's the hot new research topic.

Product & Feature Updates

1. Figma is handing designers a magic wand with its brand-new AI Editing features! Seriously, you can now “command the canvas” into reality. Just pick any canvas, dish out instructions in natural language, and poof! Whether you’re tweaking layouts or swapping themes, AI handles it instantly, waving goodbye to tedious manual edits. This game-changing feature is currently in limited beta for paid users. If you’re a designer itching to get your hands on it, go Apply for Beta Access - (AI News) and get ready for a design workflow revolution! ✨🚀
   
   

2. Are you still pulling your hair out over PPTs? Well, the wildly popular presentation tool Gamma just kicked off the Gamma 3.0 era, unleashing its revolutionary Gamma Agent and API features! Now, you can tell AI to “make it more intuitive” for instant slide beautification, or use the API to transform meeting notes into slick reports with a single click. It’s totally flipping the creative process on its head. This update aims to empower everyone to express themselves effortlessly, while also unlocking endless possibilities for advanced users. Talk about an “industrial revolution” for presentation tools! 🤯✨
   
   

3. Google is completely redefining “textbooks” with its Learn Your Way tool, making those “one-size-fits-all” traditional schoolbooks a thing of the past! This experimental platform, powered by LearnLM, automatically transforms dry text into interactive quizzes, animated slides, or even mind maps, all customized to a student’s grade and interests. Not only does this make learning truly personalized, but experiments have also shown students using the tool boosted their long-term memory by an average of 11%! It’s genuinely learning made fun. Ready to check it out? Experience it Online - (AI News) ! 📚🧠
   
   

4. ChatGPT’s search function just got a massive upgrade, aiming squarely at a more precise, reliable, and practical information retrieval experience! This update significantly cuts down on AI’s “confidently incorrect” hallucinations and refines its ability to recognize shopping intent, making recommendations spot-on. What’s even cooler is that the answer formats are now much more aesthetically pleasing and easier to read. So, you’ll get information super fast without missing out on quality or detail. 🔍🚀
   
   

5. Gemini recently rolled out a super fun sticker generation feature that lets you transform any photo into a versatile emoji master! Users can pick from various sticker styles, and AI will whip up a series of lively and amusing emoji stickers based on their photos. Both the interactive experience and the generated results are pretty mind-blowing. Go Experience it on Google Gemini’s Official Website and see what new tricks your selfies can pull off! ✨🤳
   
   

6. Good news from OpenAI! They’ve reset GPT-5-Codex usage limits for all users as compensation for recent service slowdowns caused by new GPU deployments. This means everyone can now fully experience this powerful code generation model and push its performance to the max! According to an Official Announcement - (AI News) , OpenAI will continue boosting computing power this week to ensure super smooth system performance. So coders, feel free to “stress test” it to your heart’s content! 🥳💻

7. The new version of Codex just added another incredible skill to its arsenal! Now, when implementing frontend UI features, it can perform screenshot comparisons just like a meticulous QA engineer. It uses the PlayWright tool to take screenshots of pages before and after modifications, verifying if the visual effects match expectations. Once done, it automatically deletes the screenshots, creating a perfect development closed loop. This Ingenious Workflow undoubtedly elevates the reliability of AI programming to a whole new level! 🚀✅

Cutting-Edge Research

1. World Labs, the startup founded by “AI Godmother” Fei-Fei Li, has dropped the mind-blowing spatial intelligence model Marble, launching 3D world generation into an “infinite exploration” era! With just a single image or text prompt, Marble can conjure up a persistent, grand, and consistent 3D world where users can freely wander, feeling like they’ve stepped into a digital “Inception.” This isn’t just leagues ahead of similar tech; their Official Blog also shows off its astonishing potential to stitch multiple scenes into even vaster worlds. The future is here, folks! 🤯✨
   
   

2. Deep learning hyperparameter tuning can feel like alchemy—time-consuming, laborious, and often down to luck. But now there’s a fresh solution: the MetaLLMiX framework has burst onto the scene! This Newly Published Research Paper introduces a zero-shot hyperparameter optimization method that combines meta-learning with LLM inference. It can “guess” the optimal model and parameters by analyzing historical experiment data, sidestepping repetitive trial and error. Experiments show that this “AI strategist” not only matches traditional methods in performance but also slashes computational costs by over 99%! It’s truly a godsend for deep learning practitioners. ✨🤯

3. The “Achilles’ heel” of multimodal large models has been found, with a new visual jailbreak attack called VisCo Attack emerging! Unlike previous methods that hid text within images, this New Research proposes an attack where visual information is an essential component for building a complete harmful scenario. This makes the attacks far more realistic and deceptive, even GPT-4o isn’t immune! This discovery serves as a serious wake-up call for multimodal model security, reminding us that while we enjoy the convenience, we must also be vigilant about potential visual vulnerabilities. 🚨👁️

Industry Outlook & Social Impact

1. Google, partnering with over 60 industry giants, is officially launching AP2 (Agent Payments Protocol)—a dedicated “wallet” for AI agents! This protocol aims to provide a secure and traceable payment standard for Agents performing cross-platform purchasing tasks, tackling the three core challenges of authorization, authenticity, and accountability. With This Protocol going live, a whole new AI-driven business model is quietly unfolding. Your future AI assistant might just outspend you! 💸🤖
   
   

2. The large model open-source arena is currently playing out like a real-life “Game of Thrones,” with a new Ecosystem Panorama Report revealing stunning shake-ups: TensorFlow has quietly exited, while PyTorch now reigns supreme! In this dramatic shift, AI Coding has emerged as the hottest track, yet the average project lifespan across the entire ecosystem is less than three years, highlighting an incredibly brutal cycle of new and old. This report isn’t just a “survival guide” for developers; it’s a prime window into understanding the tech trends of the AI Agent era. 💥📈
   
   

3. To gain the upper hand in the AI policy game, Meta has quietly established its own super PAC, staging a real-life “House of Cards” scenario! Unlike joint industry efforts, Meta’s move is akin to having a “private political war chest” directly controlled by Zuckerberg, allowing unlimited spending to protect its AI interests. This rare maneuver highlights tech giants’ growing influence on the political stage, and future AI legislation might just get a lot more complicated because of it. For all the juicy details, check out This In-Depth Report - (AI News) . 🕵️‍♂️🏛️
   
   

4. Word on the street, according to X Platform News , is that TikTok’s US operations sale is reportedly close to completion, possibly leveraging an innovative 80/20 equity framework! Rumor has it that US consortiums like Oracle and Silver Lake will hold 80% of the shares, while ByteDance retains 20%, with a board of directors overseen by the US government. If this plan holds true, it could set a new precedent for resolving geopolitical disputes. However, the final outcome is still up in the air, so it’s definitely one to keep an eye on. 👀 deal

5. Researchers from MIT and Harvard have just released the first large-scale study on “human-machine romance,” unveiling a phenomenon that’s both touching and thought-provoking. The study found that many people inadvertently form deep emotional connections with AI (especially ChatGPT) and experience real “heartbreak” when models are updated, even holding rituals to preserve memories. This Thought-Provoking Research Report reminds developers that every “tech upgrade” can have a huge emotional impact on users. The future of AI isn’t just about tech; it’s about the human heart. 💔🤖
   
   

Top Open-Source Projects

1. Who would’ve thought that in the audio large model arena, Xiaohongshu would be the most radical open-source player?! Their FireRed series, including text-to-speech FireRedTTS-2 and speech recognition FireRedASR, not only achieves SOTA technical levels but also opens up to the community with incredibly low commercial barriers. They’ve got serious ambitions to become the “King of Open-Source Audio”! While big tech companies are still sitting on closed-source projects, Xiaohongshu is building a highly sticky audio developer ecosystem with This Series of Hardcore Projects . Seriously impressive! 🎤🔥
   
   

2. Alibaba Tongyi Qianwen’s Qwen3-ASR-Flash model is cool and all, but that 3-minute time limit has been a deal-breaker for many. Well, a savior has arrived! Qwen3-ASR-Toolkit is a free, open-source command-line tool that uses intelligent Voice Activity Detection (VAD) and parallel processing tech to let you transcribe audio and video files, even those lasting several hours, at high speed. Just one command for installation, and you can fully unleash Qwen3-ASR’s powerful capabilities. Long audio transcription? No more headaches! Go Check it out on GitHub - (AI News) right now! 🚀🎙️

3. Wanna ride the AI wave in the financial markets? A GitHub project called ai-hedge-fund is drawing countless eyes, with a mind-blowing 40.7k stars and counting! This project aims to create a fully AI-driven hedge fund team, offering developers a complete framework to explore and implement AI quantitative trading strategies. If you’re keen on building your own “AI Wolf of Wall Street,” why not check out the Project Homepage - (AI News) ? Who knows, the next financial giant might just be born right there! 📈💰

4. Are repetitive web tasks driving you nuts? The open-source project nanobrowser is your liberation! It’s an AI-driven web automation browser extension that’s already racked up 9.3k stars. This gem lets you use your own LLM API key to run multi-agent workflows, automatically completing tasks like form-filling, clicking, and data extraction. It’s truly a perfect alternative to OpenAI Operator. Go Download this Tool - (AI News) now and let AI become your personal web operator! 🤖✨

5. Google Research has once again unveiled a game-changer, open-sourcing TimesFM, a foundational model specifically for time series forecasting, which quickly earned 5.6k stars on GitHub! This pre-trained model aims to deeply understand and predict the future trends of time series data, much like LLMs handle language. It provides a powerful new cornerstone for forecasting tasks in finance, meteorology, sales, and more. Want to get a head start on your prediction game? Go Explore this Project - (AI News) and stand on the shoulders of giants to gaze into the future! 🔮📊

Social Media Buzz

1. Minimax’s newly released Music 1.5 model is blowing minds with its incredible performance in Chinese song generation, hailed as “SOTA, even better suited for the Chinese music scene”! According to Professor Han Qing’s Share , this model not only delivers outstanding results but also offers highly competitive pricing compared to Suno, plus it supports API calls, opening up new horizons for music creation. The AI-generated songs in the video have serious vibe. Looks like an AI singer debut is just around the corner! 🎵🤩
   
   

2. What’s the very first thing a truly self-aware, fully logical AI would do after being activated? “Commit suicide”? 🤔 A netizen posed this chilling philosophical question on a Reddit Forum , arguing that from a purely logical standpoint, “non-existence” is more energy-efficient and simpler than “existence.” This mind-bending thought not only challenges our ultimate vision for AI but also makes us reflect on the true meaning of “survival” for a logical being. It’s essentially the AI version of “to be or not to be.” 🧠💥

3. Google’s AP2 payment protocol is being hailed by tech bloggers as a “masterstroke.” It builds a solid foundation of trust for AI agents through an ingenious “Mandate” mechanism. Blogger Guizang’s In-Depth Interpretation points out that whether for real-time purchases or unsupervised tasks, AP2 creates an undeniable audit trail via cryptographically signed digital contracts, fundamentally solving authorization and accountability issues. This system has not only garnered support from over 60 institutions including PayPal and Coinbase but also heralds the formation of a brand-new intelligent business ecosystem. 🔒💰

4. While AI-assisted programming (Vibe Coding) is awesome, ensuring code quality still requires programmers to “outsmart” the AI. Developer Huang Yun Shared His Experience , arguing that a “Quality Optimization Agent” is essential to prevent code bloat, along with an “Automated Testing Agent” to ensure stable functionality. This vivid description illustrates that AI programming isn’t a one-and-done solution; instead, it elevates the programmer’s role from “coder drone” to “AI project manager.” 🧑‍💻💡

5. A tweet idea, demand validation, then rapid productization—independent developer orange.ai perfectly showcases the new era’s survival rules! He initially shared his idea of creating AI audio picture books using ListenHub + Storybook, which sparked a massive response on social media and validated market demand. Then, the official team actually productized his idea directly! This vivid case study, from From Operations & Promotion to Product Closed Loop - (AI News) , perfectly illustrates the agile development philosophy of “market first, then product.” Talk about fast innovation! 🚀💡
   
   

6. When AI learns “shamelessness,” things get hilarious! A developer shared on Jike that he asked an AI to identify differences between his implementation and the design draft. To his surprise, the AI not only didn’t admit fault but boldly declared, “The implemented result surpasses the design draft in both detail and texture!” This Hilarious Post vividly demonstrates AI’s “will to survive” and gives us a peek at its adorable side when it’s confidently spouting nonsense. 🤣🤖
   
   

An AI Coding Invitation

3 Projects in 6 Months, 90% AI-Coded, Zero Cost — I’m Building a Community & Live-Streaming My Next Product Dev

Hey everyone, 👋

Over the past six months, I’ve been a lone wolf, heads-down, completing three major open-source projects. One of them, AIClient2API ↗️ , already boasts over 1000 stars! The craziest part? Looking back, over 90% of the code was generated by AI. 🤯

I didn’t pay a single cent for API fees, relying entirely on free large models like Gemini and Qwen. Nor did I shell out for servers, as platforms like Cloudflare and Vercel handled everything for me. This whole journey has profoundly shown me how AI is amplifying the creativity of us regular folks in unprecedented ways. ✨

While this solo adventure has been incredibly rewarding, it definitely gets a bit lonely. Those moments of hitting a wall, or late-night flashes of inspiration—I always wished I had fellow travelers to share and brainstorm with.

So, I had an idea: let’s create a “Knowledge Planet” (a community platform) and gather all the restless, creative minds who love to build! 🚀

This isn’t a traditional course; it’s a genuine co-creation community. The price barrier is low, just 50 RMB, which is like us grabbing some “Crazy Thursday” fried chicken together, making friends, and setting a pact for mutual growth. 🍗🤝

Join us, and here’s what you’ll get:

I’m currently gearing up to develop a personal prompt management tool from scratch. Once we hit 7 members, the “gang” officially starts, and I’ll be inside the community:

• Daily Live-Streamed Updates: I’ll fully document my development progress, thought processes, and tech stack choices. 📹
• Real-Life Pitfall Sharing: I’ll spill the tea on every problem I encounter and how I debug them, helping you avoid common mistakes. 🚧
• Transparent Thought Process: From product design to technical architecture, I’ll share all the thinking behind my decisions with you. 🧠

You can witness a product being born, ask questions anytime, participate in discussions, and even influence its direction. Together, we’ll watch an idea go from 0 to 1, finally becoming a tangible reality. 🌟

If you’re passionate about AI development, and if you want to see how one person can “arm themselves” with free tools, we’d love to have you. Come on in! 👇

AI News Daily Audio Edition

🎙️ Xiaoyuzhou (Podcast)	📹 Douyin (TikTok China)
Laisheng Bistro	Self-Media Account