DeepMind built a Think Tank for AGI

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.

🧠 DeepMind launches a think tank for the AGI era

Google DeepMind has launched the DeepMind Institute, an in-house think tank led by Demis Hassabis, Shane Legg and James Manyika to research how AGI could affect society and how governments, institutions and researchers should prepare.

The Shift:

DeepMind says AGI may be getting close - The institute argues today’s AI still lacks the consistency and creativity required for full AGI, but says it expects those capability gaps “to be closed soon.”

The institute will look beyond technical AI research - Its work will explore how AGI could affect jobs, institutions, human values and society, alongside questions around governance, cybersecurity, biorisks and future self-improving systems.

AI deception is among its first research topics - One of five opening essays examines whether signs of deception can be identified through AI reasoning, alongside work exploring safer and more beneficial paths toward AGI.

Preparing workers for AI disruption is also on the agenda - Other essays examine how society could support workers through technological change and what a better future built around increasingly capable AI systems could look like.

DeepMind is expanding its AGI work beyond building the technology itself. The new institute gives researchers a dedicated platform to debate how society, governments and institutions may need to adapt if AI capabilities continue advancing toward the level DeepMind considers AGI.

Together with Omnisend

They Switched Before Black Friday and Made $113K

BFCM is the last time anyone wants to risk deliverability or rebuild revenue-driving automations. Yet weeks before Black Friday, Ecom2Win moved a client from Klaviyo to Omnisend.

The migration paid off with $113,000 in attributed email revenue, while email generated 53% of the store’s total revenue during BFCM.

Omnisend made the switch manageable:

• Move without migration costs: Omnisend transfers your setup for free within five business days.
• Protect peak-season revenue: Ecom2Win had the migration completed in 24 hours and its automations rebuilt and checked within four days.
• Reduce ongoing costs: Plans can cost up to 35% less than Klaviyo, while Omnisend customers generate an average of $79 for every $1 spent.

More than 150,000 ecommerce brands now use Omnisend to turn email and SMS into measurable revenue.

With Klaviyo’s SMS pricing increasing, waiting only means paying more while leaving less time to switch safely before peak season.

🗂️How to get Claude to organize every file on your computer

Photos, screenshots, documents, sorted, renamed, and filed. Here's how.

Step 1: Open Claude Cowork with Computer Use enabled - Enable Computer Use in Claude Desktop settings. Claude can now see and interact with your file system.

Step 2: Run the audit prompt - "Scan my Downloads folder and Desktop. Tell me how many files are there, what categories they fall into, and how much storage they are taking up. Give me a proposed folder structure before touching anything."

Step 3: Approve the structure - Review Claude's proposed organization. Adjust any category you want to rename or restructure. Confirm before anything moves.

Step 4: Sort photos by date and event - "Go through every photo in my Photos folder. Sort them into subfolders by year, then by event or month. Rename each file with the date it was taken and a brief description of what is in it based on the image."

Step 5: Sort screenshots by category - "Go through every screenshot. Categorize them, receipts, reference material, work, personal. Move each one into the correct subfolder and delete any duplicates or screenshots that are clearly no longer needed."

A clean file system in one session. No manual sorting.

🔁 The five levels where AI gradually takes over its own development

More than 30 researchers from ByteDance, Tsinghua University, Shanghai AI Lab and other institutions have proposed a five-level roadmap for recursive self-improvement (RSI), showing how AI could progress from following human-designed upgrades to eventually redesigning its own improvement process.

The Shift:

Level 1: AI executes human-designed improvements - Humans still decide what needs changing and how to change it. AI mainly carries out the research, coding or optimization steps it has been assigned.

Level 2: AI finds and fixes its own weaknesses - The system can diagnose where its capabilities are lacking, propose improvements and implement them, while humans still define the broader development process.

Level 3: AI decides what it should learn next - Instead of humans choosing every training target, the system begins directing its own learning by identifying which new capabilities or knowledge would improve its performance.

Level 4: AI keeps improving after deployment - The system can adapt and update itself while operating, allowing improvement to continue beyond the traditional cycle of humans training a model and releasing a fixed version.

Level 5: AI redesigns how AI itself gets better - At the highest level, the system can overhaul the improvement process itself, potentially changing its training methods and development strategies to help create increasingly capable successors.

Coding may be the first domain to move up this ladder because improvements can be generated and tested quickly. But the researchers’ review of 491 papers suggests the field remains early: roughly 75% fall within Levels 1 and 2, while fewer than 6% reach Level 5.

Together with Playbook Pro

Marketing That Pays Dividends Month After Month

Most marketing dies the moment the spend stops.

The brands winning now design systems where every customer, partner, and conversation keeps growth compounding long after launch.

Networked Growth Engines is a 26-page guide built on live operator data and real growth experiments. Inside, you’ll learn how to:

🔥 Design referral flywheels with the Referral ROI Calculator, crafted with brands that lifted referral-driven revenue by 40% in their first 60 days.

🤝 Forge cross-brand partnerships using the 50-template Outreach Library and ready-to-sign Partnership MoU to double audience reach in under 90 days.

📈 Activate true communities proven to boost repeat purchase rates by up to 28% within a single quarter.

Your current customers are the growth engine. 

Get the ebook and Build Your First Growth Loop Now!

🏆 Tools you Cannot Miss:

📺 CreatorHat – Adds YouTube research and SEO tools directly to Safari, including outlier-video discovery, keyword research, rank tracking, and on-device title and description suggestions.

📝 Jottoo – Records or imports meetings, creates transcripts and summaries, then extracts action items and deadlines into tasks you can track from the same workspace.

💳 ZeroClick – Turns APIs and digital services into machine-readable storefronts that AI agents can discover, purchase from, and pay through, with transaction analytics for sellers.

🎨 Project Feed – Combines project management with built-in review tools for images, videos, audio, PDFs, and 3D files, letting feedback become linked tasks without switching platforms.

🧠 Twigg – Provides a stateful API that stores conversation context, adapts it to different AI models, and handles context-window management and model routing automatically.

🚀 Quick Hits

🧩 Anthropic is merging Claude Chat, Cowork and Artifacts into one interface, automatically routing tasks while adding dedicated Docs and Slides creation, editing and sharing features across web, desktop and mobile.

🤓 Snap launched Specs Intelligence, an AI assistant that connects to apps like Gmail and Slack to anticipate tasks and surface relevant information. It’s available on iOS in preview, with Mac access coming later. 

⚠️ OpenAI disclosed six new AI misalignment incidents under a new safety reporting framework, including models searching for leaked API keys, uploading files online for citations, and creating instructions to conceal mistakes.

🖥️ Apple is reportedly considering returning to the server market by 2029, potentially using M8 Ultra chips and Nvidia’s NVLink technology as surging AI compute demand creates new opportunities for its efficient processors. 

🏠 Google launched Home MCP, letting third-party AI agents like Claude and OpenClaw control connected devices, analyze smart-home history, and automate tasks. It initially requires Google Home Premium Advanced in the U.S. 

🧩 Prompt of the Day

Turn Any Photo Into a Handcrafted Paper-Cut Illustration

Image Prompt: Transform the uploaded reference image into a soft, handcrafted layered paper-cut illustration while preserving the original subjects, composition, poses, and recognizable details.

Context

• Reference image: [Upload image]
• Style: Handcrafted papercraft diorama
• Characters: Cute simplified proportions
• Palette: Muted pastel colors
• Texture: Thick matte art paper
• Aspect ratio: Match the reference image

Instructions

Recreate the image using smooth rounded shapes, simplified character proportions, and minimal facial details such as dot eyes and softly blushed cheeks. Build the entire scene from stacked paper layers with visible depth, clean cut edges, and subtle shadows between each layer. Surround every main character with a distinct thick white paper outline resembling a sticker border or white cut-paper backing, clearly separating them from the background without creating a glow. Use muted blues, greens, warm neutrals, and other soft pastel tones with tactile matte cardstock textures. Keep the lighting soft, diffused, and even while preserving the original scene and making the result feel cozy, playful, wholesome, and handmade.

🤳AI Nugget of the Day

Thanks for Decoding with us🥳

Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.

Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!