ChatGPT Wrapped, QWEN Image Layers, Groq | Weekly Digest
PLUS HOT AI Tools & Tutorials
Hey! Merry Christmas! We wish you good times this holiday season.
Top stories this week: ChatGPT rolled out its Year in Review, OpenAI tightened Atlas against prompt injections, Qwen launched layered image editing, Amazon expanded Alexa+ integrations, Nvidia snapped up Groq’s AI chip tech, and Waymo is testing Google’s Gemini as an in-car assistant.
But let’s get everything in order.
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
From our partners:
Recraft
Recraft is one of the best image generation models out there that comes with its own design tool, trusted by brands like Ogilvy, HubSpot, Netflix, Asana, and Airbus.
There, you can create marketing‑ready visuals with precise color control, vector-ready assets, and custom brand styles.
The platform removes friction from consistent tasks so anyone can generate visual content from text fast without any design experience.
Bring your ideas to life instantly!
Featured Materials 🎟️
ChatGPT’s Wrapped
Have you already checked your ChatGPT Year in Review?
Looks like every company jumped on the year-end recap bandwagon because people eat this stuff up.
The chat starts a little awkwardly with a year in poetry card. Mine came out like something a kid would recite at a school assembly to make their parents proud. But we’ve already talked about how ChatGPT sticks to its technical voice, so I wasn’t expecting anything impressive.
You also get:
Your year, painted in pixels
Achievement badges
Random stats
Your title of the year
The year-end wrap-up will pop up on the ChatGPT home screen. You’ll be able to check it out on both the web and mobile apps. Or, you can simply ask ChatGPT for “Your Year with ChatGPT” to get started yourself.
Go check it out!
News of the week 🌍
OpenAI is On Guard
OpenAI has stepped up security for ChatGPT Atlas, specifically to deal with prompt‑injection attacks (hidden instructions slipped into emails, docs, or webpages that try to hijack an AI agent’s behavior). To get ahead of this, OpenAI built an automated red‑teaming system powered by reinforcement learning.
Once new attack patterns are detected, OpenAI uses them to adversarially train updated agent models, tighten system‑level safeguards, and quickly roll out new checkpoints to all users. The main focus here is long‑horizon attacks, where a tiny injection early on (like a shady email in your inbox) quietly steers the agent into doing something weird later, like sending emails or sharing sensitive data.
At the same time, OpenAI is pretty upfront that prompt injection isn’t something you just fix once and move on from. Like phishing or social engineering, it’s a long‑term game.
The goal isn’t perfect security, actually, but making attacks harder and rarer. So, the defense is built as a loop: automated attack discovery > adversarial training > system tweaks > fast deployment > repeat. Users can also play a role. The more specific the task, the less room there is for sneaky instructions to mess things up.
Qwen Breaks Images Into Layers
Alibaba’s Qwen has released an image editing model called Qwen-Image-Layered.
Layers aim to provide highly precise generation. You can put background, details, clothes, and objects on separate levels, so the model tweaks what you want without touching the rest.
Btw, they trained it on regular Photoshop PSD files to teach it how to split objects across layers. The model natively handles RGB color channels and the transparent alpha channel.
Each layer contains full-color and transparency data, allowing you to recolor, swap content, edit text, delete objects, resize, or move elements on any layer without affecting the others.
Also, you can create any layers you wish. Qwen even does recursive decomposition, breaking any layer into more sublayers for complex edits.
Christmas & EOY are coming! Thank you for being with us, and enjoy the Christmas Offer for Creators AI Full Access (limited in time)
Amazon Expands Alexa+
Amazon claimed that starting in 2026, Alexa+ will officially integrate with Agni, Expedia, Square, and Yelp.
Angi will look for and book home services, and get rough price estimates plus a few options to choose from.
Expedia will hunt down, compare, book, and manage hotels and stays.
Square will book and pay for small business services, fully synced with Square’s scheduling system.
Yelp will search and snag reservations at salons, restaurants, or local spots using your voice.
That said, these new integrations turn Alexa+ into an AI sidekick for offline stuff and travel.
Example scenarios Amazon throws out:
“Find me pet-friendly hotels in Chicago this weekend and book one” via Expedia.
“Get me a contractor to fix my roof / clean my place” via Angi, with options and rough costs.
“Book me a haircut tomorrow after 7 PM nearby” via Yelp + Square.
Nvidia to Buy New Accelerator Chip Technology
Nvidia just dropped a whopping $20 billion to grab assets from Groq, the AI chip startup founded by some of the brains behind Google’s TPUs. This is officially Nvidia’s biggest deal ever, way past the $7 billion Mellanox acquisition in 2019.
The deal isn’t a full buyout, though: Groq stays independent, keeps its cloud business running, and even gets to license its tech to Nvidia. But key players, including Groq’s CEO Jonathan Ross, are hopping over to Nvidia to help scale Groq’s ultra‑fast inference chips, which Nvidia plans to fold into its AI platform for all sorts of real-time workloads.
Gemini Enhances Driving
Waymo is apparently testing Google’s Gemini AI inside its robotaxis to create a ride‑focused assistant.
Waymo is Alphabet’s (Google’s parent company) self-driving car project. They build fully autonomous robotaxis that can drive passengers without a human behind the wheel.
Researcher Jane Manchun Wong found references in the Waymo app to a large internal document called the “Waymo Ride Assistant Meta-Prompt”. Spanning over 1,200 lines, it lays out how an in-car AI assistant is meant to interact with passengers. The idea is to create a practical companion that can answer general questions, manage in-cabin stuff like climate and lighting, and even help keep passengers calm and comfortable during the ride.
The feature hasn’t hit public builds yet, but the instructions make it clear that Gemini is intended to enhance the ride experience.
If you’re curious, the assistant can’t control everything, like route changes or seat adjustments, and if it’s asked to do something it can’t, it’ll respond like, “Not something I can do yet.
Useful tools ⚒️
Aident AI - Build automations with natural language, not workflows
ConnectMachine - A private AI agent that manages your network and connections
DiffSense - Local AI git commit generator for Apple Silicon
Vibe Pocket - Run AI agents like Claude Code, Codex, Opencode on mobile
Dvina - Private AI that connects 120+ apps and your live databases
Dvina is a private AI that connects all your apps, docs, and databases in one place. Ask questions, explore context, and get insights without needing to know where the data lives or how it is structured. Works for solo users or big teams with complex datasets.
Weekly Guides 📕
Check our fresh Tutorial & Review of Recraft:
Claude Code Skills just Built me an AI Agent Team (2026 Guide)
How to Run LLMs Locally - Full Guide
Gemini 3 Flash Tutorial: Build a UI Studio With Function Calling
I Built an AI Agent That Actually Manages My Email, Calendar, and Tasks
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
Bonus Materials 🎁
Stanford CME295 Transformers & LLMs - to watch a free LLM course from Stanford
51 Charts That Will Shape Al in 2026 - to listen where artificial intelligence stands right now and what matters most heading into 2026
Love this perspective! I'm so curious how these year-end sumaries are generated. My ChatGPT wrap-up was also hilariously poetic, making me wonder about the new frontiers in NLG.