🪨 Less Context Beats More For AI

Hey folks,

Andrej Karpathy just stirred up the AI world again with nanochat, a minimalist ChatGPT clone built in just 8,000 lines of hand-coded code. The project lets you spin up your own conversational model in ~4 hours on a GPU, something Karpathy calls “the best ChatGPT you can get for $100.” Unlike nanoGPT (which was mostly about pre-training), nanochat packages everything: training, inference, and a web interface in one file. Netizens are already joking that once it runs, everyone’s “now a machine learning engineer.”

Let’s dive in..

🪨 Anthropic Cracks the Code on AI Context Engineering

🐝 The Buzz: Anthropic just dropped a guide on context engineering that's rewriting how we build AI agents. They're telling us to use LESS context, not more. In a world obsessed with cramming maximum tokens into every prompt, Anthropic's Applied AI team reveals a counterintuitive truth: "finding the smallest set of high-signal tokens that maximize the likelihood of your desired outcome" is what actually drives performance.

The timing couldn't be better. As developers push against 200k+ token context windows, Anthropic's research shows that models experience "context rot" - a gradual performance degradation as context length increases. Think of it like trying to find a specific conversation in a group chat with 10,000 messages. Your brain (and the AI's attention mechanism) simply can't process it all effectively. Models have a finite "attention budget," and every token competes for a piece of it.

The Three Pillars of Context Engineering

1. Compaction: When approaching context limits, don't just truncate - intelligently summarize. Anthropic's agents compress conversation history while preserving critical details. It's like taking meeting notes instead of recording every word.

2. Structured Note-Taking: Agents maintain persistent memory outside the context window. This is active knowledge management.

3. Sub-Agent Architectures: Instead of one mega-agent drowning in context, deploy specialized sub-agents for focused tasks. Each explores extensively but returns condensed summaries. The main agent coordinates while staying lean and focused.

💡 Takeaway: We've been thinking about AI context backwards. More isn't better - precision is. Anthropic's context engineering principles mark a fundamental shift from "throw everything at the model" to "curate what matters." This is about building AI agents that can actually think clearly, maintain focus, and deliver consistent results. The teams that master context engineering will build the agents that actually work in production. Everyone else will keep wondering why their 100k-token prompts keep failing.

Together with The Deep View

The AI Insights Every Decision Maker Needs

You control budgets, manage pipelines, and make decisions, but you still have trouble keeping up with everything going on in AI. If that sounds like you, don’t worry, you’re not alone – and The Deep View is here to help.

This free, 5-minute-long daily newsletter covers everything you need to know about AI. The biggest developments, the most pressing issues, and how companies from Google and Meta to the hottest startups are using it to reshape their businesses… it’s all broken down for you each and every morning into easy-to-digest snippets.

If you want to up your AI knowledge and stay on the forefront of the industry, you can subscribe to The Deep View right here (it’s free!). 

🗜️ OpenAI Goes All-In on Custom Chips

🐝 The Buzz: OpenAI just announced a partnership that sent shockwaves through the chip industry, a 10-gigawatt collaboration with Broadcom to co-develop custom AI accelerators starting in 2026. Broadcom's stock surged 9.88% immediately, adding more than $150 billion to its market value in a single day. The deal represents OpenAI's boldest move yet to break free from Nvidia's grip on AI compute.

OpenAI designs the accelerators and systems, while Broadcom handles development and deployment on its Ethernet stack with custom networking, memory, and compute tailored specifically for OpenAI's workloads. By designing its own chips, OpenAI can slash compute costs and stretch infrastructure dollars significantly further.

Here's what makes this partnership different:

• Full-stack customization: Networking, memory, and compute all optimized for OpenAI's specific workloads

• Massive scale: 10 gigawatts of custom accelerators represent unprecedented investment in proprietary AI infrastructure

• Market validation: Broadcom's $150B market value surge proves investors believe in the custom chip thesis

💡 Takeaway: OpenAI's custom chip strategy signals a fundamental shift in AI infrastructure. We're moving from a world where everyone rents Nvidia GPUs to one where major AI labs design their own silicon. This could democratize AI compute over time, or concentrate power even further among the few companies with resources to go custom.

📰 Like newsletters? Here are more our readers enjoy.

💻️ How to Generate A Sketch Line Drawing

• Go to Gemini Nano Banana

• Upload a photo

• Prompt the following

Create a photo-style line drawing / ink sketch of the faces identical to the uploaded reference image — keep every facial feature, proportion, and expression exactly the same.
Use blue and white ink tones with intricate, fine line detailing, drawn on a notebook-page style background.
Show a right hand holding a pen and an eraser near the sketch, as if the artist is still working.

🐝 AI buzz bits

🚀 Microsoft is shaking things up at GitHub to fend off fierce competition from AI coding tools like Cursor and Claude Code. They're embracing a vision to make their AI tools accessible everywhere developers work, from code to apps, ensuring GitHub remains the go-to hub for software development.

🍌 Google’s Nano Banana is spreading its fruity sweetness to Search and NotebookLM. Users can now snap photos and transform them with AI, alongside enhanced video overviews and stylish illustrations.

☕ Anthropic's new Zero Slop Zone pop-up in NYC is a breath of fresh air, inviting over 5,000 visitors to ditch screens and embrace thinking caps, coffee, and analog tools. This quirky event underscores their push to counter low-quality AI content while promoting their Claude chatbot. 

🚨 AI Homeless Man Prank creates chaos on social media: This viral trend uses AI to generate fake intruder images, leaving unsuspecting viewers panicked and police overwhelmed. Authorities warn that it not only dehumanizes the homeless but also diverts essential public resources.

🔥 AI productivity tools

For a full list of 1000+ AI tools, visit our Supertool Directory

🤘 Midweek Vibes