12-28-Daily AI News Daily
Alright, let’s get this done! I need to be the Senior Technical Translator and Editor here. My brain’s already going through the motions. First thing’s first: gotta get those pesky brackets and URLs handled with a laser focus.
Alright, got it. Scan for [text](URL) and . Translate all the Chinese descriptive text inside those brackets into English. Keep the URLs and file paths in the parentheses exactly as they are. This is a must, no room for error. Self-check time: every [ and ] accounted for? Did I translate all the Chinese descriptive text? Are the URLs untouched? Check, check, check. ✅
Now for the main event: rewrite each paragraph with the core subject front and center. I need to understand what each paragraph is really about. Then, re-arrange all the info around that core idea to make it flow better. 💪
Next, I need to get the vibe right. Keep it conversational, informal, energetic. Use everyday English, and some slang to give it some life. Gotta ditch the original emojis and inject some new ones that fit the English context – can’t be too boring, right? 🤔 Also, absolutely no touching the Markdown! Headings, lists, code blocks, br tags—they all stay. The code itself? Untouched. Comments? Those get the English treatment, of course. Got it. And hey, make sure I translate everything, and keep the structure exactly as it is.
Finally, output just the finished English version. No intros, no explaining what I did or why. Let’s do this! 🚀
AI Daily 2025/12/28
AI Daily
Today’s Digest
OpenAI is urgently hiring a Head of Safety, with Sam Altman admitting models can already find critical vulnerabilities—risks are skyrocketing.
VSCode officially rebranded as an "AI Code Editor," MiniMax open-sourced M2.1, and the developer toolchain is fully AI-fied.
Today's message is clear: the stronger AI capabilities get, the more urgent safety issues become. Don't just get caught up playing with new tools.⚡ Quick Navigation
- 📰 Today’s AI News - Quick Overview of Latest Updates
💡 Pro Tip: Wanna be among the first to try out the latest AI models mentioned here (Claude 4.5, GPT, Gemini 3 Pro)? No account? No problem! Head over to Aivora to grab an account, get started in a minute, and enjoy worry-free after-sales support. 😉
Today’s AI News
👀 Just One Sentence
OpenAI is on the hunt for an “AI Safety Chief,” and Sam Altman himself posted that models are already capable of finding vulnerabilities.
🔑 3 Keywords
#SafetyAlert #OpenSourceFiesta #VSCodeTransformation
🔥 Top 10 Heavy Hitters
1. OpenAI Urgently Hiring “Head of Preparedness,” Sam Altman: Models Can Now Find Critical Vulnerabilities
Sam Altman dropped a bombshell late at night, stating that AI models are now powerful enough to discover critical vulnerabilities in cybersecurity. You might have thought AI safety was a distant concern, but OpenAI is already grappling with AI’s potential impact on mental health in 2025, and now faces issues like bio-capability release and self-improving systems. The new ‘Head of Preparedness’ role demands someone who can ‘dive straight into the deep end,’ a high-pressure but profoundly impactful position. OpenAI finally admits: as model capabilities skyrocket, so do the risks.
2. VSCode’s Official Website Quietly Changed Its Slogan: Now “The Open Source AI Code Editor”
VSCode’s official website now proudly displays its new slogan: “The open source AI code editor.” This isn’t just marketing fluff; VSCode natively supports AI auto-completion, code modification confirmation, and more. Microsoft is clearly declaring that the AI era for code editors has officially arrived, letting you enjoy AI-powered features without even needing plugins. 🚀
3. MiniMax Open-Sources M2.1 Model, Significantly Boosting Code and Tool Usage Capabilities
MiniMax, a domestic large model, is making waves again. MiniMax has open-sourced its M2.1 model, which significantly improves upon its predecessor in code generation, tool invocation, and task planning. For developers looking to run open-source models locally, this offers another compelling option to explore. ✨
4. Zhipu AI Releases Z Code: One API Key to Switch Between Multiple AI Programming Agents
Zhipu AI has launched Z Code, a lightweight visual desktop application that integrates the capabilities of various command-line tools like Claude Code, Codex, and Gemini CLI, which can often be too complex. With just one API Key, you can seamlessly switch between them, eliminating the hassle of configuring multiple setups. For those eager to try different AI programming tools without the headache, this is a sweet deal. 🤩
5. StoryMem: Teaching AI Video Models to “Cut Shots” and Generate Multi-Shot Long Videos
StoryMem is here to shake things up if you’re tired of single-shot AI videos. StoryMem enhances single-shot models like Wan2.2 into multi-shot video models, capable of generating longer videos with dynamic camera cuts, similar to Sora2. The open-source code and models are available, so if you’re into AI video creation, give it a shot! 🎬
6. High School Student Uses AI to Discover 1 Million+ Astronomical Objects, NASA Director Tweets Recruitment Offer and Fighter Jet Experience
A high school student’s story is truly inspiring! Using AI, this student discovered over a million astronomical objects that humans had previously missed in astronomical data. Jared Isaacman of NASA publicly recruited them on Twitter, even promising a fighter jet flight as a signing bonus. Kevin Weil, OpenAI’s CPO, called it the perfect collision of all his worlds—AI is empowering ordinary people to make extraordinary discoveries. 🤯
7. ezff: Process Videos with a Single Sentence, No Need to Memorize FFmpeg Commands Anymore
ezff is here to save you from constantly looking up FFmpeg commands. After installing npm install -g ezff, you can process videos directly using natural language: for example, “ff convert video.mp4 to gif” or “ff compress video.mp4 to 10mb.” Converting formats, compressing, cropping, extracting audio, adjusting speed—you can do it all in plain English. 🙌
8. ES-CoT: Teaching Reasoning Models to “Quit While Ahead,” Saving 41% of Tokens
ES-CoT is a new research proposal that addresses the costliness of powerful Chain-of-Thought reasoning models. This new research introduces ES-CoT, which allows models to detect if an answer has converged during the reasoning process and stop early if it has. Without fine-tuning or architectural changes, it saves an average of 41% of reasoning tokens while maintaining accuracy. This approach is super practical for production environments focused on cost control. 💰
9. Claude Code Transcripts: Turning Claude Code Conversations into Beautiful HTML Documents
Claude Code Transcripts, a tool by Simon Willison, recognizes that your Claude Code conversation logs are valuable project documentation. This tool converts Claude Code’s JSON sessions into paginated, mobile-friendly HTML pages, and you can even share them to GitHub Gist with a single click. For anyone looking to solidify their AI programming experience, this is incredibly useful. 📝
10. System 3 Has Arrived: Enabling AI Agents to “Self-Evolve”
System 3 is here, building on System 1’s rapid perception and System 2’s deep reasoning. So, what exactly is System 3? This paper introduces Sophia, a metacognitive framework that empowers agents with continuous adaptation, autonomous goal generation, and identity recognition. In 36 hours of deployment testing, it demonstrated significant self-improvement capabilities. This could be the pivotal step in agents evolving from mere ’tools’ into true ‘partners.’ 🤖
📌 Worth Your Attention
[Product] One-Click GIF Sticker Generator Tool - Upload an image, and it automatically generates 16 animated stickers with 6 style options. ✨
[Open Source] awesome-llm-apps - An 84k-star collection of LLM applications, featuring RAG and Agent examples. 🌟
[Open Source] Yuxi-Know - An Agent platform combining LightRAG and knowledge graphs, with MCP support. 🧠
[Tool] rendercv - A YAML to PDF resume generator, specifically for academics and engineers. 👨💻
[Research] A Survey of AI Agent Memory Systems - This paper reviews 200+ papers, proposing three forms of memory: Token, Parameter, and Latent. 📚
[Discussion] 20%+ of YouTube New User Recommendations Are AI-Generated Content - Yet another piece of evidence for the proliferation of AI slop. 🤦♀️
[Industry] China Launches Nationwide Distributed AI Computing Network - Connecting data centers over 2000 kilometers apart. 🌐
❓ FAQs
How to Try Out Claude Code?
Claude Code, an AI programming tool developed by Anthropic, currently requires a Claude Pro or Max subscription to use. For users in mainland China, this might pose challenges with payment or account registration restrictions.
Solutions:
- Aivora offers account services related to Claude.
- Super fast delivery: order, use immediately, no need to deal with payment or registration hassles. 🚀
- Provides stable, exclusive accounts with worry-free after-sales support. ✅
Visit aivora.cn to see the full list of AI account services. 👇