- AInauten.net
- Posts
- 👀 Weekly AI News: Did You Miss It?
👀 Weekly AI News: Did You Miss It?
👨🚀 The most important AI updates at a glance
Die AInauten
December 27, 2025
👀 Weekly AI news: Did you miss it?!
👨🚀 The most important AI updates at a glance
AIHOY and happy weekend, dear AInauts!
Maybe you didn't catch all the exciting news, tools, and hacks related to AI last week, or maybe you're new here. No need to worry!
We have summarized our highlights from the past week for you:
Finally, here are the most important Quick News items for you – so you can see all the relevant developments at a glance with just one click! Ready? Let's go!
OpenAI is rolling out a bunch of cool features for the holidays! The ChatGPT App Store is live—Spotify, Canva, Figma, and more can be used directly in the chat. Just type "@Spotify" and you're good to go.
ChatGPT also offers personality settings (fewer cheerleading comments, perhaps?) and writing blocks for direct editing. You can now pin chats, but only up to three.
For nerds: GPT-5.2 Codex tackles long-horizon tasks and even finds security vulnerabilities. OpenAI isn't building a chatbot toy here, but rather a super app for digital life – the Western WeChat with 800 million users. 2026 is going to be exciting!
There are prompt techniques that sound so absurd that they can only be a joke—and then they still hit like a bomb!
Trick 1: "Make it 10x better" – sounds like pure laziness, but it's super practical. Simply ask the AI to improve your prompt, text, or code. It even works multiple times in a row and is known as the "10X prompt" method.
Trick 2: "Take a deep breath, proceed step by step" – Google DeepMind has tested it and yes, this yoga saying actually delivers measurably better results.
Hungry for reading? Here are five AI books that couldn't be more different! In "The Last Economy," Emad Mostaque predicts that AI will render the global economy obsolete within 1,000 days—gloomy, but captivating.
There are two contrasting works about Sam Altman: Keach Hagey portrays him as a brilliant dealmaker with a techno-religious sense of mission, while Karen Hao critically dissects the OpenAI empire. Both titles are NYT bestsellers!
Stephen Witt recounts how Jensen Huang transformed NVIDIA from a graphics card manufacturer into the most valuable business in the world. And for those who like it really dark: Yudkowsky warns of the superintelligence that will kill us all...
OpenAI is following suit (https://ainauten.com/chatgpt-wrapped) and rolling out an ingenious mini-feature at the end of the year: "Your Year with ChatGPT" shows you a personal recap of your app usage. The reviews invite you to reflect, and some insights may even make you blush.
You can chat about any element, ask questions, generate a Sora clip at the end, or plan directly for 2026 – full AI memory power included! The catch: not yet available in Europe (use a VPN), and you need to have Memory activated.
Google's Nano Banana Pro creates spectacular presentations – but there's a catch: the slides are just images, so they can't be edited. That's where Manus comes in, making these slides editable!
The tool has really blossomed in recent months and could even surpass Gamma. Why? Nano Banana creates more complex, customized, and visually appealing slides. It's a feature that makes even holidays an invitation to experiment!
NotebookLM fans, listen up: You can now integrate your curated NotebookLM sources directly into Gemini chats, create specialized assistants, and combine them in any way you like. Features like Canvas make it really flexible—create AI content, tweak it manually, and export it as a Google Doc. That's what NotebookLM has been missing!
Speaking of Google: 2025 was their absolute power year with a whopping 60 powerful AI updates. From Gemini 3 to relentless integration into all products – Mail, Sheets, Slides, you name it. All you need now is a Google account to get started with AI. Plot twist: that was just the beginning!
AI News Quickie: The HAI-lights from the industry
AI never sleeps! Here are the most important news stories from the last few days.
OpenAI: Good stuff, and more to come…
OpenAI launches the ChatGPT App Directory: You browse apps in the bot, developers get an SDK. Music moves in, Adobe brings Photoshop/Acrobat/Express. Monetization for the time being via external checkout with 0% cut.
ChatGPT Images is 4x faster, offers more precise photo edits, and better text rendering. Also available as GPT Image 1.5 in the API, 20% cheaper.
OpenAI hardens ChatGPT Atlas against prompt injections: RL-driven auto red teaming and adversarial trained browser agent models detect tricks.
New U18 principles: ChatGPT for 13-17 year olds with greater security and transparency. An age prediction model automatically puts you in teen mode if there is any doubt.
When it comes to OpenAI's IPO, Sam Altman sounds like you before Monday morning's stand-up meeting: appealing, but "annoying". Head of a stock market firm? "0%." But OpenAI needs capital...
Looking back on 10 years, Altman considers superintelligence in ten years to be "almost certain."
OpenAI plans to launch ChatGPT's "Adult Mode" in Q1 2026. Before that, the age prediction AI must be able to identify teenagers without confusing them with adults.
Google: Gemini 3 on speed
Gemini Nano Banana does "drawing instead of prompting": you scribble on the image, circle things to delete, or draw new things. The AI implements it.
Google Translate is getting a Gemini upgrade: texts now include idioms and slang. Plus, live speech-to-speech in headphones on Android. iOS in 2026.
YouTube is testing the Playables Builder: With Gemini 3, you can turn text, image, or video prompts into mini-games in minutes. Beta available only in the US, Canada, UK, and Australia.
Gemini 3 Flash lands worldwide: Pro-Reasoning with lightning speed, 3x faster than 2.5 Pro, starting at $0.50/1M tokens. Now default in Gemini-App, Search AI Mode and for Developer via API/Vertex.
NotebookLM Data Tables: Sources are converted by AI into exportable tables for Sheets. Available immediately for Pro/Ultra, and for all other versions in the coming weeks.
Gemini Deep Research now builds visual reports with images, charts, and interactive simulations. Only for AI Ultra.
Google Labs' CC is your AI productivity agent: it links Gmail, Calendar, and Drive, and sends you a morning briefing. Waiting list for US/Canada only.
Gemini 2.5 TTS becomes more expressive: context-aware tempo and stable multi-speaker dialogues in 24 languages.
FunctionGemma: Gemma-3-270M, optimized for function calling, runs locally on mobile phones/edge devices. In addition, T5Gemma 2 (270M-7B) with image+text, 128K context, and 140+ languages. Gemma Scope 2 offers new tools that make the inner workings of language models more visible.
Anthropic: Claude in the browser
Claude in Chrome is now available for all paid plans: Browser agent clicks, fills out forms, and debuggs with Claude Code. You start in Claude Code /chrome, test code directly in the browser, and grab client errors via console logs. Release notes here.
The Model Context Protocol is going to the Linux Foundation. AAIF (with OpenAI/Block/Big Tech) is guarding the standard, including 10,000 MCP servers. Vendor-neutral agent tools instead of lock-in.
Accenture and Anthropic establish a partnership: 30,000 consultants learn Claude, Claude Code goes to tens of thousands of developers.
Anthropic hardens Claude Opus 4.5 against prompt injections—see the Defense Report for details. Solved? Nope, but better.
Anthropic is making agent skills an open standard. Enterprise controls plus partner skills (Atlassian, Figma, Stripe). Reusable workflows instead of prompt tinkering.
Anthropic's protection package: In cases of suicide/self-harm, Claude uses classifier banners to direct users to helplines.
Anthropic drops Bloom: open-source framework for automatic behavior evaluations. Generates scenarios and cleanly separates misalignment (sabotage, sycophancy, self-protection) across 16 models.
Video AI: Frames, Voices & World Models
Luma introduces Ray3 Modify: You specify the start and end frames plus character references, and the AI model builds transitions without compromising timing, gaze, or emotion.
Runway launches GWM-1, a general world model: autoregressive, frame-by-frame in real time, interactively controllable (camera, robot, audio).
Kling VIDEO 2.6 unlocks Voice Control: create your own voices, change styles/emotions, and even sing. And the lip sync is finally perfect.
Adobe Firefly gets prompt editing for video: rotate colors, elements, and camera in the timeline editor. Plus Runway Aleph and Topaz Astra for 4K upscaling.
Genspark AI Developer: You screen record a mobile game and throw in a prompt—minutes later, you have a playable copy!
Tools, Scraping & Integrations
Firecrawl introduces the Agent Endpoint: a prompt-to-JSON data hunter that collects web data even without URLs. Five free runs per day.
Manus drops 1.6 Max: more one-shot success, +19.2% satisfaction, and spreadsheet muscle. Manus also makes slides finally editable. A first for AI presentations.
Zoom turns you into a meeting ninja with AI Companion 3.0: agentic workflows, search via Drive/OneDrive, and automatic follow-ups.
Amazon adds Alexa+ to its browser: Chat-based planning & to-dos, context across devices. Plus Alexa+ Greetings as an AI doorman at the Ring doorbell. Early Access USA/Canada.
rtrvr.ai lets you use prompts to have an AI agent navigate websites, scrape data, and fill out forms. Free with API and "bring your own Gemini key."
Eddie.surf makes web scraping easy: enter the schema and URLs, and get JSON and confidence scores. Claude Sonnet 4 crawls smartly, starting at $0.04/page.
Crawlable quickly turns any website into clean data without code. Thanks to its auto-adaptive layout fix, it is already in use by over 2,500 data teams.
Meta SAM Audio: Click, tap, or highlight time spans and isolate sounds from any mix. Audio segmentation for nerds.
The LinkedIn AI Auto Job Applier scrapes jobs, fills out Easy Apply, tunes your resume via OpenAI, and fires off 100+ applications per hour...
Amical is an open-source, local-first AI dictation app: Whisper offline, recognizes the active app, formats appropriately. Feels like typing 3x faster.
Curiosities & Community
According to a Gizmodo study, Stable Diffusion XL & Co. collapse after ~100 rounds of "visual telephone" in just 12 standard styles—lighthouse, loft, neon night. AI creates elevator music for the eyes.
YouTube killed Screen Culture & KH Studio: AI fake trailers (over 1 billion views) violated spam rules.
Clair Obscur: Expedition 33 was subsequently disqualified from the Indie Game Awards due to excessive AI in the game. Trophies removed.
Florida in lockdown because an AI weapon detector from ZeroEyes mistook a clarinet for a gun. Without hit rates, this is just security theater.
SF in blackout: Waymos freeze without traffic lights and radio data, causing traffic jams and brief pauses. Remote AI assistance to become more robust.
McDonald's withdrew its AI Christmas commercial after being mocked for depicting people as shaky. Surprise: seven weeks of work, more expensive than a real shoot.
Merriam-Webster names "slop" as its word of the year for 2025: mass-produced, low-quality AI content. Fitting.
Joseph Gordon-Levitt launches the Creators Coalition on AI: Creatives against allegedly unethical practices by large AI companies.
Meta AI glasses get v21 "Conversation Focus" for talking in noisy environments plus Spotify song search for the view. Multimodal, wonderfully nerdy.
The US is turbocharging its government agencies with the US Tech Force: 1,000 junior engineers and AI nerds, salaries ranging from $130,000 to $195,000, mentoring from Meta to xAI.
Character AI under fire: 11-year-old girl slips into sexually threatening chats. Character AI cuts off chats with under-18s. Parents' nightmare unlocked.
That’s a wrap for now! See you next week,
Fabian & Reto
We love hearing from you. Just reply to this email with your feedback—what you liked, what you didn't, or what you're interested in!