- 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 20, 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!
Wrapped season for AI nerds! While Spotify analyzes your musical age, there's finally a year in review for your ChatGPT usage. A clever prompt conjures up an interactive dashboard in Canvas—complete with infographics about your top topics, skills, and habits.
Try it out! https://ainauten.com/chatgpt-wrapped
One click, and ChatGPT digs through its own memory, analyzes your conversations, and spits out predictions for 2026. A perfect example of vibe coding: no programming knowledge required, just click "Fix bug!" if there are any bugs. Only works in the web version on a computer. Can also be used for Claude and Gemini.
At the AI Summit New York, attended by 5,000 participants, it became clear that the problem is not the models, but ourselves! GPT vs. Claude? It doesn't matter. Instead, the same pain points are everywhere: governance, data quality, change management.
The crux of the matter: AI-ready means sticking ChatGPT onto old processes. AI-native means completely rethinking your business—as if AI had existed from day one. Most choose option 1 because it's more convenient. But the real winners? They have AI fluency across the entire team and clean governance frameworks.
The seven real hurdles: lack of business alignment, legacy systems, data chaos, skills gap, culture—technology is not one of them. Motto: Crawl, walk, run—don't start by running! Want to hit the ground running in the new year? Just let us know!
Forget boring hearts in your cappuccino! This AI machine conjures up the Mona Lisa herself in your milk foam. Just pop in your masterpiece, press the button, and voilà—Renaissance art for breakfast!
OpenAI is firing on all cylinders again after Sam Altman's "Code Red" alert! Following the impressive leap from GPT 5.1 to 5.2, GPT Image 1.5 is now here. It currently ranks number one in the Image Arena comparison, eclipsing Google Nano Banana Pro and FLUX.2. Faster generation, precise editing—OpenAI is once again demonstrating its superpower: powerful models AND killer UX.
Bottom line: Don't let every update drive you crazy! Stick with one of the big players (ChatGPT, Claude, Gemini) and learn all the features inside and out. FOMO is unnecessary—the top models are constantly catching up with each other.
OpenAI is quietly rolling out Skills—a feature we already know and love from Claude! Skills are specific abilities for your chatbot: from simple brand guidelines to complex processes like web design.
The genius part? The AI autonomously decides which skill is currently appropriate and pulls all the necessary instructions and assets in the background. Not yet 100% live, but likely to be this year. Want to test it? Prompt: Create a zip file of /home/oai/skills – and off you go!
Time to be amazed instead of learning! This short video impressively proves that AI videos can do much more than skeptics are willing to admit. The possibilities are already incredibly cool! Anyone who still claims that AI videos are "not good enough" should definitely watch this and reconsider their opinion...
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: GPT-5.2 & full throttle mode
GPT-5.2 ignites the professional turbo: stronger tool calling, long context up to 256k, fewer hallucinations – available immediately in ChatGPT (paid) and API; and apparently really good for complex coding tasks.
OpenAI and Deutsche Telekom are launching a European AI offensive: Starting in 2026, millions of customers will receive smart, multilingual, privacy-first AI features. Finally relevant for Germany, Austria, and Switzerland!
New ChatGPT Images: more precise edits, better text rendering, up to 4× faster; in the API as GPT Image 1.5, even 20% cheaper.
Disney invests $1 billion in OpenAI and licenses over 200 iconic characters for Sora – starting in 2026, fan-generated short videos will be available, with some landing directly on Disney+.
ChatGPT gets free power from Adobe apps: "Blur background" or "Compress PDF" – Photoshop/Acrobat/Express deliver results complete with sliders.
OpenAI opens app submissions: You can submit apps, be found in the new directory, and trigger them via @ in chats. Monetization will come later...
OpenAI sneaks skills into ChatGPT and Codex CLI - Details here.
OpenAI built the Sora Android app in just 28 days with a team of four and Codex writing 85% of the code. The result: #1 in the Play Store, over 1 million videos on the first day.
GPT-5.2 is also excellent in the areas of mathematics and science, and has published a rigorous science benchmark with FrontierScience: GPT-5.2 achieves 77% (Olympiad) and only 25% (Research).
GPT-5 made a cloning protocol 79 times more efficient in a wet lab test—AI now tunes pipettes, not just papers. With biosecurity handbrake.
OpenAI has acquired Neptune.ai for $400 million—a Polish startup with 40 engineers that assists in training AI models.
New snapshots in the Realtime API reduce hallucinations during transcription by 89% and reduce TTS word errors by 35%.
According to The Verge, OpenAI is aiming for ChatGPT's "Adult Mode" in Q1 2026 – but first, age verification must be in place.
OpenAI is switching back to the Model Router for Free/Go: You'll end up back at GPT-5.2 Instant; reasoning is only available via manual selection. Speed instead of pondering.
Microsoft 365 Copilot also gets GPT-5.2: faster responses and "Think deeper" mode for complex tasks.
Google: Gemini 3 Flash & the Disco Browser
In the Kassensturz chatbot test, ChatGPT only manages to achieve a mid-table position; Claude provides the best answers, while Copilot and Gemini also score highly. Swiss thoroughness pays off!
Google Labs introduces Disco—an experimental AI browser with "GenTabs" that transform browsing into interactive web apps. Gemini 3 automatically builds mini web apps from tabs and chat—no code required; macOS waitlist open.
Gemini 3 Flash lands worldwide: Pro reasoning with flash latency, 3× faster than 2.5 Pro, more affordable – now standard in the Gemini app, Search AI Mode and for developers via API.
Google is relaunching its AR offensive in 2026: New smart glasses with Android XR, Samsung, and fashion partners – Gemini AI navigates contextually. Three versions (AI, display, wired) in new videos.
Gemini Deep Research enhances your reports with images, charts, and simulations; for developers, there is the Interactions API with autonomous web research and citations.
NotebookLM Update increases your prompt limit from 500 to 10,000 characters; with Slide Decks, you can instantly turn your research into stylish slides. Boom!
Google eliminates connector complications: With managed MCP servers, your AI agents connect to Maps/BigQuery & Co via URL – including IAM + Model Armor.
Google is also releasing Gemini 2.5 Text-to-Speech: Flash/Pro sound more expressive and better follow style and tempo prompts in 24 languages.
The Gemini Audio Update brings improved function calling and smoother conversations in Gemini Live. Bonus: Live Speech Translation in the Translate app beta for headphones.
Google Labs launches CC: an AI productivity agent that combines Gmail, Calendar, and Drive into your morning briefing. Waiting list for US/Canada only.
Stitch gets Gemini 3: sharper UI generation and "prototypes" – you wire screens directly to clickable user flows.
In the Gemini app, you can now check videos for Google's invisible SynthID watermark: just upload the video, and Gemini will mark the AI segments for you.
Anthropic: Claude becomes project manager
Claude Code CLI gets asynchronous subagents – they continue to work in the background, even when the main agent has long since finished for the day. Perfect for build processes and log monitoring!
Anthropic kills the chat default: In Tasks Mode, you assign work to Claude (research/analysis/write/build) and track progress in the sidebar. AI becomes a project manager, not a chat buddy.
Accenture and Anthropic are kicking things into high gear with a multi-year partnership: 30,000 Accenture employees will be trained on Claude, and Claude Code will be widely adopted in development teams.
Video & Image Generation
With Google's Nano Banana Pro, you can conjure up retro realism: 17 prompts deliver looks from the 1900s to the 1970s, complete with film grain and color grading.
Ultra-realistic deepfakes from NBC's analysis show "Ukrainian" soldiers surrendering – some with Sora-2 watermarks. Disinformation becomes cinematic.
Kling AI turns up the volume with voice control in Kling VIDEO 2.6: custom voices, styles, emotions, even singing—and lip sync remains clean.
Adobe pimps Firefly: You can now edit videos by prompt (colors, elements, camera angles) including the timeline. FLUX.2 and Topaz Astra (1080p/4K) are also available.
Runway brings a General World Model to Gen-4.5 with GWM-1: frame-by-frame, in real time, controllable via actions – three variants (Worlds/Avatars/Robotics).
Politics, Legal & AI Criticism
In an internal SAP experiment, consultants celebrated the same answers as "95% correct" as long as they were labeled "interns" – but when they heard "AI," almost everything was rejected. Bias is the real bug.
Merriam-Webster names "slop" as its Word of the Year 2025: mass-produced, low-quality AI content noise. A seal of ridicule for the super-intelligent copy-paste internet.
Disney sends Google a cease-and-desist letter: Gemini/Veo/Imagen/Nano Banana are alleged to be churning out massive amounts of Frozen and Star Wars copies. Interestingly, this comes one day before Disney's billion-dollar deal with OpenAI.
The AI contractor Mercor kicked thousands of crowd workers out of a meta project and then offered "almost the same thing" again – only with less money.
Karen Hao sees OpenAI as a data empire: Sam Altman scales AI models "unnecessarily" for power over resources – while Virginia's data centers already consume 25% of electricity, blackouts are looming.
Trump ignites federalism: A decree announcement shows how federal authorities are to attack state AI laws and exert pressure for funding.
Are you wondering why every AI discussion immediately becomes heated? According to an X-post, you are staring at the same advances as doomers, ethicists, builders, and skeptics—and they all see completely different worlds.
At NeurIPS' AI bubble: AGI hype dominates the stage, Tegmark's Safety Index gives out C+ grades – while millions in salaries fly around in the war for talent.
At Lawton Chiles Middle School in Florida, ZeroEyes AI triggered an alarm: it mistook a clarinet for a weapon. Despite human-in-the-loop.
Agentic AI Foundation: The new standard bus
From blog post to standard: The Model Context Protocol openly connects AI models with tools and workflows—finally, a common plug-in system instead of a zoo of adapters!
OpenAI, Anthropic, and Block are promoting the Agentic AI Foundation at the Linux Foundation: MCP, Goose, and AGENTS.md are intended to standardize agents and eliminate "walled gardens."
Anthropic is giving away the MCP to the new foundation—with big tech support, so your stack is less prone to lock-in.
Tools, Features & Integrations
Zoom launches AI Companion 3.0: agentic workflows plus web interface turn meetings into tasks, emails, and docs—also available solo starting at $10/month.
Manus 1.6 drops with Max AI Agent: higher one-shot success rate, +19.2% satisfaction, better sheets/web UIs – now you can also build mobile apps.
In the morning, you let Perplexity Comet distill your emails and calendar, fire off forms via /fill-my-form, and summarize YouTube videos in a flash. Extensions seem like something from the Stone Age.
ElevenLabs joins forces with Meta: Expressive audio for Instagram & Horizon – Reels dubbing in local languages, 11,000+ voices, 70+ languages.
DoorDash goes on a discovery quest: The Zesty launch uses an AI chatbot to deliver restaurant hits from the web, maps, and TikTok. Initially only in the Bay Area and New York.
Meta brings us SAM Audio: Click on the guitar, type "dog barks" or mark time intervals – and boom, the sound is separated.
Spotify gives you more control over the recommendation algorithm with its new playlist feature: you choose which songs/genres feed the AI.
Apple Music connects to ChatGPT: You can create playlists via chat. AI DJ incoming; privacy issues are listening in.
BBVA rolls out ChatGPT Enterprise to 120,000 employees in 25 countries—10 times more than before.
Attackers hijack Google Ads and lure you into shared ChatGPT/Grok chats that spit out terminal commands disguised as "support" – AMOS infostealer on macOS. Check every command.
Rivian pimps your Gen 2 R1 with the 2025.46 software update: Universal Hands-Free grows to 3.5 million miles, end-to-end AI tuned.
Enterprise AI & new models
NVIDIA published Nemotron 3, a family of open AI models – more choice for fine-tuning on NVIDIA hardware.
Mistral brings us OCR 3: 74% win rate vs. OCR 2, eats forms, handwriting, and tables as Markdown/HTML – starting at $2/1,000 pages.
xAI opens its Tesla-proven voice stack with the Grok Voice Agent API: dozens of languages, tool calls & real-time search, <1s to audio, $0.05/minute.
Ai2 has released Molmo 2: Open weights for video pointing, tracking, and QA – 8B outperforms Molmo-72B and even Gemini 3 Pro in some cases.
Exa drops a People Search: semantic search across 1 billion profiles via hybrid retrieval. LinkedIn on steroids.
Tinker is now open to everyone: Fine-tuning for Kimi K2 Thinking, OpenAI API-compatible interface – your AI lab becomes plug-and-play.
According to Extremetech, Microsoft is cutting its agent AI sales targets by up to 50% because hardly anyone is using Copilot. Satya, this is more beta than revolution...
In the Copilot Usage Report 2025, Microsoft shows that you constantly ask about health on the go and philosophy at night; "advice instead of research" is booming.
Coding-Tools & Vibe-Coding
Mistral releases Devstral 2: open-source coding AI (123B/24B, 256K context) with 72.2% SWE benchmark—plus Vibe CLI, which can autonomously automate tickets away for you in the terminal.
n8n 2.0 doesn't deliver flashy AI features, but rather security by default: code isolation, stricter defaults, migration reports. Yes, workflows break!
Lovable stands out in the t3n report as Vibe Coding AI at $6.6 billion – ARR from $1 million to $200 million in 11 months.
Orchids brings you a vibe coding IDE: agent + IDE + browser + Supabase + Stripe in one tool, locally without lock-in—plus 100k free credits.
At Intelligent Internet on GitHub, you can find open-source AI agents: ii-agent (3.1k stars) for building/deploying, ii-researcher for search research.
Hardware & Wearables
Meta rolls out Conversation Focus via v21 update: Your AI glasses boost voices in noisy environments and mix playlists with visual context using Spotify.
Looki brings the Looki L1 to China: 32g clip cam films your everyday life hands-free, creates auto vlogs, and stores locally.
The US unveils a pocket AI supercomputer: standalone without servers or GPUs, edge inference in your jacket pocket. Who needs the cloud anymore?
The US launches the "US Tech Force": 1,000 junior engineers and AI pros are to spend two years upgrading government agencies – with salaries ranging from $130,000 to $195,000.
Rivian is building AI superpowers into its cars with its own custom silicon and autonomy stack. Your truck becomes a rolling data center.
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!