While the AI industry chases reinforcement learning, Essential AI made the opposite bet. Their new 8B model embodies a thesis about where machine intelligence originates. The transformer's co-inventor is calling the shots on research.
Tim Cook built Apple's leadership into a monument of stability. In 2025, that monument cracked. Meta poached AI and design chiefs with $25M packages. The chip architect may follow. What broke inside the world's most valuable company?
OpenRouter's 100 trillion token study was supposed to prove AI is transforming everything. The data shows something else: half of open-source usage is roleplay, enterprise adoption is thin, and one account caused a 20-point spike in the metrics.
Meta launched an AI app that uses your Facebook and Instagram history to personalize responses from day one. Built on their Llama 4 model, the app transforms years of social media data into an AI that claims to understand your preferences, habits, and interests.
The app remembers key details about users. Tell it you're learning Spanish, and it tracks your progress. Mention food allergies, and it adjusts recommendations. It learns from your social media engagement - the posts you like, content you click, and how you interact with friends.
Voice commands drive the experience. Meta added experimental "full-duplex speech" technology that generates voice responses directly instead of converting text to speech. The AI speaks more naturally, though the feature remains in testing across the U.S., Canada, Australia, and New Zealand.
The app connects with Meta's broader ecosystem. Start a conversation on Ray-Ban smart glasses, continue on your phone, then finish on desktop. One limitation: you can't start chats on desktop and move to glasses. The AI works across WhatsApp, Instagram, Facebook, and Meta's smart glasses.
Meta packed the app with social features. A Discover feed lets users share AI interactions and modify popular prompts. Nothing posts without permission, but the social angle shows Meta's strategy: make AI part of everyday digital conversations.
The desktop version adds tools for work. Users can generate and edit documents, create images, and export PDFs. Meta tests features for document analysis and rich text editing, pushing beyond casual chat into productivity.
Meta banks on personalization to stand out. While competitors like ChatGPT and Claude focus on broad knowledge, Meta's AI aims to understand individual users. They bet that knowing your coffee order and meeting schedule matters more than explaining complex topics.
The hardware integration reveals bigger plans. By connecting with Ray-Ban smart glasses, Meta positions their AI as an always-available assistant. They want it ready whether you're walking downtown or sitting at your desk.
The privacy question looms large. Meta's pitch boils down to a simple trade: share your data, get an AI that understands you. Some users will embrace personalization. Others might question the data collection. Meta bets enough people want a truly personal AI to accept the trade-off.
The timing matters. As AI assistants multiply, Meta carves out their niche: deep personalization through social data. They've turned their biggest criticism - collecting vast user information - into their key advantage.
Why this matters:
Read on, my dear:
Meta: Introducing the Meta AI App: A New Way to Access Your AI Assistant | Meta
Tim Cook built Apple's leadership into a monument of stability. In 2025, that monument cracked. Meta poached AI and design chiefs with $25M packages. The chip architect may follow. What broke inside the world's most valuable company?
The New York Times sued Perplexity for copyright infringement—months after signing an AI licensing deal with Amazon. Perplexity built revenue-sharing programs for publishers. The Times declined to join any of them. Now lawyers are involved.
Chinese hackers operated inside U.S. VMware servers for 17 months undetected. The malware repairs itself when deleted. It hides where most security teams don't look. CISA's December 4 advisory exposes an architectural blind spot in enterprise defense.
Werner Vogels ends his 14-year keynote streak by handing out printed newspapers and warning developers about "verification debt." His parting message: AI generates code faster than humans can understand it. The work is yours, not the tools.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
