In partnership with
In Today’s Issue:
🛡️ The U.S. War Department launches GenAI.mil
🤝 Anthropic donates its Model Context Protocol (MCP)
🚢 CATL is targeting fully electric ocean cargo ships
🧠 AI tools help Terence Tao and collaborators solve a classic Erdős problem
✨ And more AI goodness…
Dear Readers,
The US War Department just flipped the switch on GenAI.mil, wiring Google’s Gemini for Government into Pentagon workflows so briefings, drone footage, and policy docs all run through AI by default – a live test of what happens when a whole defense bureaucracy runs on models instead of macros. In today’s issue, we stay on that “AI as infrastructure” thread: Anthropic’s Model Context Protocol moves under the Linux Foundation as an open standard for agentic AI, CATL plans fully electric ocean cargo, and AI helps crack a classic Erdős problem in math. Plus, a clear GPU-vs-TPU explainer. Lets get into it!
All the best,
Open Standard Powering Agentic AI
Pretty cool: Anthropic is donating its Model Context Protocol (MCP) to the Linux Foundation’s new Agentic AI Foundation (AAIF), a directed fund co-founded with OpenAI and Block and supported by Google, Microsoft, AWS, Cloudflare, and Bloomberg, to keep MCP neutral, open, and community-governed.
In just one year, MCP has grown to 10,000+ active servers and over 97 million monthly SDK downloads, and will sit alongside Block’s goose and OpenAI’s AGENTS.md as founding projects standardizing how AI agents talk to tools, data sources, and other services.
CATL Targets Electric Ocean Cargo
CATL, the world’s largest battery maker, says its tech will be ready to power fully electric ocean-going vessels within about three years, expanding from today’s river, lake, and coastal ships to true transoceanic cargo routes. Building on existing projects like 10,000-ton all-electric container ships with ~50,000 kWh packs, CATL is betting that rising battery energy density plus shore charging and swapping infrastructure will make it viable to trade a chunk of cargo space for dramatically lower fuel costs and CO₂ emissions from a sector that currently accounts for roughly 3% of global emissions.
AI Helps Solve Classic Erdős
Terence Tao reports how a 1975 Erdős problem about monotone subsequences was finally cracked through a mix of old theorems, online collaboration, and modern AI tools. AI systems assisted with literature search, generating candidate examples, and proving key intermediate statements, turning what might have been months of human-only work into a complete solution within roughly 48 hours.
GPUs, TPUs, & The Economics of AI Explained
U.S. War Department Bets On GenAI
The Takeaway
👉 GenAI.mil shows how to turn GenAI from toy projects into core workflow infrastructure for millions of users inside a high-stakes organization.
👉 By choosing Gemini for Government at IL5, the Pentagon is drawing a blueprint for “sovereign AI” deployments where sensitive data never touches public models.
👉 Expect a ripple effect: once the Pentagon normalizes AI-assisted paperwork, planning, and analysis, other ministries, agencies, and large enterprises will push for similar AI-native platforms.
The US War Department has flipped the switch on GenAI.mil, a bespoke AI platform that pipes Google Cloud’s Gemini for Government into Pentagon desktops and military networks worldwide. Built under President Trump’s AI Action Plan, the system is meant to turn troops, civilians, and contractors into an “AI-first” workforce, using secure generative tools for deep research, policy summaries, and rapid analysis of documents, imagery, and video.
Pete Hegseth says "We will continue to aggressively field the world's best technology to make our fighting force more lethal than ever before.”
On paper, Gemini for Government runs as a web-grounded assistant with retrieval-augmented generation and IL5/CUI certification, letting staff query sprawling handbooks or operational data in plain language without sending sensitive information into public models. Additional “world-class AI models” are slated to join the platform, so GenAI.mil is less a single chatbot and more an internal AI app store where agentic workflows automate the grinding staff work that usually eats military time - think formatting briefings, drafting compliance checklists, or scanning drone footage for anomalies - while humans stay on the hook for real decisions.
Naturally, this development should be viewed critically. It's no coincidence that AI is classified as a national security technology. AI will play a prominent role in future wars. And its implementation is underway.
Why it matters: This is one of the first attempts to roll out frontier-grade generative AI across an entire defense bureaucracy, making the US military a live testbed for secure, large-scale deployment. The way GenAI.mil handles guardrails, failure modes, and productivity gains will quietly shape how governments and big enterprises everywhere adopt powerful models without losing control of their data, or their chain of command.
Sources:
🔗 https://www.war.gov/News/Releases/Release/Article/4354916/the-war-department-unleashes-ai-on-new-genaimil-platform/
Turn AI Into Extra Income
You don’t need to be a coder to make AI work for you. Subscribe to Mindstream and get 200+ proven ideas showing how real people are using ChatGPT, Midjourney, and other tools to earn on the side.
From small wins to full-on ventures, this guide helps you turn AI skills into real results, without the overwhelm.
Mistral Supercharges Open Coding Agents
Mistral just dropped Devstral 2, a new family of open-weight coding models built to actually ship software, not just autocomplete snippets. The lineup comes in two sizes - Devstral 2 with 123B parameters and Devstral Small 2 with 24B - both released under permissive licenses (modified MIT and Apache 2.0) so teams can run, fine-tune, and redistribute them without being handcuffed to a single vendor. Devstral Small 2 posts 68% on SWE-Bench Verified and competes with models up to five times its size, while both variants stay dramatically smaller than giants like DeepSeek V3.2 and Kimi K2, making serious agentic coding feasible on far more modest hardware.
On top sits Mistral Vibe, a terminal-native coding agent that explores your repo, edits multiple files, runs tests, and retries failed plans, turning Devstral into an actual co-worker for long-horizon coding tasks rather than a fancy autocomplete bar.
