Meta's national security shift

PLUS: Physical Intelligence raises $400M for better robots

Welcome, AI enthusiasts.

Meta just drew a new battle line in the global AI race — and this time, it's getting literal.

Llama is officially open to government and military use, raising questions about the role of industry and tech in national security. Let’s get into it…

In today’s AI rundown:

  • Meta opens Llama AI to U.S. defense sector

  • Anthropic releases Claude Haiku 3.5

  • Transform meeting recordings into actionable insights

  • Physical Intelligence lands $400M for better robots

  • 5 new AI tools & 5 new AI jobs

  • More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

META

Image source: Midjourney

The Rundown: Meta just announced it will make its Llama AI models available to U.S. government agencies and defense contractors for national security applications, marking a shift in the company's policy around the military use of its open-source AI.

The details:

  • Meta is partnering with tech and defense firms, including Amazon, Microsoft, Palantir, Lockheed Martin, and Oracle, to bring Llama to government agencies.

  • Early use cases include Oracle using Llama to analyze aircraft docs for faster repairs and Scale AI fine-tuning it for mission planning and threat analysis.

  • The move represents an exception to Meta's policy, which prohibits using Llama for military, warfare, or espionage.

  • The policy shift comes shortly after reports of Chinese military researchers using an older version of Llama 2 to develop defense applications.

  • Meta frames the decision as crucial for establishing open standards in AI, arguing that U.S. models should set the foundation for global AI development.

Why it matters: AI’s infusion into the government and military was inevitable, but the timing is interesting given the recent Chinese military reports. This move (alongside OpenAI’s increasing ties to government-related efforts) will also raise polarizing questions about the influence of industry and tech on national security.

TOGETHER WITH HUBSPOT

The Rundown: HubSpot’s free AI Task Delegation Kit provides you with the essential templates, guides, and dashboards you need to start turning AI into your own 24/7 personal assistant.

With the kit, you’ll access:

  • AI Command Center: Quick-reference overview of all your AI tools

  • Task Delegation Organizer: Efficiently assign and manage AI-driven tasks

  • Productivity Calculator: Monitor and improve your AI delegation effectiveness

Download the kit for free and start discovering the power of AI delegation.

ANTHROPIC

Image source: Anthropic

The Rundown: Anthropic just released its Claude 3.5 Haiku model, offering new improvements to tool use, reasoning, and coding compared to its predecessor — but it comes with a pricing hike that has drawn criticism from the developer community.

The details:

  • Anthropic introduced 3.5 Haiku in October alongside the platform’s announcement of computer use, beating the previous top model, 3 Opus, on key benchmarks.

  • The new model's pricing has increased 4x to $1 per million input tokens and $5 per million output tokens, compared to $0.25 and $1.25 for Claude 3 Haiku.

  • The model also comes in at nearly 7x the cost of similar-sized rival GPT-4o- mini and 13x the cost of Gemini Flash, despite similar benchmark performance.

  • Unlike previous versions, the 3.5 release launches without image analysis capabilities but offers an extended knowledge cutoff date through July 2024.

  • The model is now available via Anthropic’s API, Google’s Vertex AI, and Amazon Bedrock.

Why it matters: The fact that Anthropic’s small but fast model now exceeds the capabilities of the previous generation’s largest model says a lot about how fast things are moving. Conversely, the price hike to ‘reflect its increase in intelligence’ will ruffle feathers — especially given the cost of similar competitors.

AI TRAINING

The Rundown: Google Gemini's audio analysis feature can automatically extract key information from your business meetings, providing summaries, action items, and strategic insights.

Step-by-step:

  1. Upload your meeting recording to Google Gemini (choose Gemini 1.5 Pro 002).

  2. Generate a comprehensive summary of key topics and decisions

  3. Extract action items and responsibilities with deadlines

  4. Uncover strategic insights and improvement opportunities

Pro tip: Save your favorite prompts as templates for quick analysis of future meetings. We have written detailed prompts you can copy and paste, which are accessible to members here.

PRESENTED BY NEBIUS

The Rundown: Nebius AI Studio allows you to build powerful AI applications with ease, via Playground or API, by providing access to premium LLMs, vision, and code models — with built-in embeddings, structured outputs, and batch processing.

With Nebius AI Studio, you’ll:

  • Deploy production-ready AI with just a few lines of code

  • Access top models like Llama 405B, Mixtral, Gemma, Nvidia Nemotron and Qwen at 3x lower cost (the cheapest inference provider on the market)

  • Build faster with OpenAI-compatible API integration

Get started instantly with $100 in credits and start building powerful apps today.

PHYSICAL INTELLIGENCE

Image source: Physical Intelligence

The Rundown: AI startup Physical Intelligence just secured $400M in funding led by Jeff Bezos and OpenAI, valuing the company at $2.4B as it unveils its π0 (pi-zero) model for general-purpose robot control.

The details:

  • The round brings together tech giants and venture firms, with Thrive Capital, Lux Capital, Khosla Ventures, and Sequoia Capital joining Bezos and OpenAI.

  • The π0 system aims to create a universal model to understand natural language commands across any robotic platform.

  • Initial demos showcase robots performing complex multi-stage tasks like folding laundry, packing eggs, bussing tables, and more.

  • π0 trained on 10k+ hours of dexterous manipulation data and open-source datasets, claiming it is the largest pre-training mixture ever used in the sector.

Why it matters: Another competitor emerges in the robotics sector — and universal, general-purpose robots may unlock automation across industries where other humanoid robots have been too inflexible and specialized. Plus, physical intelligence will certainly attract the industry's attention with backing from OpenAI and Bezos.

NEW TOOLS & JOBS

  • 🎥 Runway Advanced Camera Control - AI video generation with precise cinematic camera control

  • 🧠 FullContext - Build go-to-market workflows with AI agents using natural language

  • 💻 Loopple AI Website Builder - Create customized, no-code websites in 30 seconds with AI

  • 🤖 Zappit AI - Automate SEO with AI agents trained to fix technical errors and generate high-ranking, optimized content

  • 🖼️ Moodpaper - Generative AI image tool that translates your mood and feelings into wallpapers

  • 📈 The Rundown - Partnerships Manager

  • 📱 Meta - Product Marketing Manager, Wearables AI Experiences

  • 🤝 Abridge - Partner Success Director

  • 🎨 OpenAI - Designer, Marketing

  •  Harvey - Brand Designer

QUICK HITS

Section’s AI Crash Course. Write better prompts, organize your AI workspace, and accelerate productivity with use cases from your day-to-day work. Get 25% off with code RDAICC2.*

Perplexity CEO Aravind Srinivas responded to a post on X about the New York Times’ Tech Guild strike, offering AI help with election coverage — which comes on the heels of the platform announcing a new dedicated hub for U.S. election news.

Amazon Prime Video released X-Ray Recaps, a new AI-powered feature that generates personalized show summaries at any viewing point to keep viewers up to date without spoilers.

Hume launched its new app featuring AI assistants that blend the company’s EVI 2 speech-language model with Claude 3.5 Sonnet and Haiku for conversational interactions, emotional reflection, deep questions, and life advice.

Netflix executive Mike Verdu announced that he is transitioning to a new ‘VP of GenAI for Games’ and touted AI as transformational for game development, coming just weeks after the company laid off 35 developers from its Team Blue game studio.

OpenAI released a new ‘Predicted Outputs’ feature for its GPT-4o models, allowing developers to dramatically reduce response times by providing reference text for tasks like code refactoring and document editing.

Google’s Big Sleep AI agent achieved an AI security research milestone by uncovering a major safety flaw in the SQLite database system, which specialized testing programs had missed for years.

*Sponsored listing

THAT’S A WRAP

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, and Alvaro—aka The Rundown Team

Reply

or to participate.