🛒 Meta unveils 2026 AI roadmap focused on shopping agents 🛒
Zuckerberg bets on personalized commerce and new models

Meta announced a major AI product rollout for 2026, emphasizing new models and agentic tools focused on commerce. Mark Zuckerberg pointed to personalized AI shopping assistants and Meta's access to user context as key differentiators in delivering tailored agent experiences. The announcement signals Meta's intention to monetize its AI investments through commerce, leveraging its massive user base across Facebook, Instagram, and WhatsApp—a distribution advantage no other AI lab can match.

🎙️ Google tests voice cloning for Gemini 🎙️
Hidden feature spotted in AI Studio ahead of native audio release

Google is testing voice cloning capabilities for AI Studio, with a hidden "Create Your Voice" option appearing when selecting the Flash native audio preview model. The feature lets users record and upload audio samples to generate synthetic voices for AI applications, competing directly with ElevenLabs and other voice AI startups. The discovery suggests Google is preparing a native audio release for Gemini, which would simplify voice workflows for developers already in Google's ecosystem.

🏭 Xiaomi opens AI-powered 'dark factory' in Beijing 🏭
Fully automated facility produces one smartphone per second

Xiaomi has launched a fully automated 'dark factory' in Beijing's Changping District, designed to run 24/7 without human workers on the production line. The 81,000-square-meter facility can produce one smartphone per second, adding up to around 10 million devices per year, including high-end models like the MIX Fold 4 and MIX Flip. Robots and AI manage the entire process using Xiaomi's Hyper Intelligent Manufacturing Platform, which monitors production, checks quality, and fixes issues in real time. The factory represents a glimpse into how AI and robotics could reshape global manufacturing.

INTERESTING TO KNOW

💻 Developer trains 67M-parameter LLM on Mac Mini 💻

A developer trained a 67-million-parameter transformer from scratch on an M4 Mac Mini with just 24GB of unified memory, achieving 93.94% exact-match accuracy on CLI command generation. The entire pretraining run took roughly 13 hours and cost under $0.50 in electricity. The project demonstrates that modern transformer architectures are no longer gated behind enterprise infrastructure—targeted small models can be trained at home for specific tasks.

🧬 DeepMind open-sources AlphaGenome genetic code predictor 🧬

Google DeepMind published the full research paper and model weights for AlphaGenome, its AI system that scans genetic code to predict how mutations cause disease. The model can predict how a single DNA mutation affects 11 different biological processes. In testing, AlphaGenome flagged leukemia-linked mutations sitting thousands of letters away from affected genes—connections that took researchers years to discover manually. The weights, API, and Nature paper are now freely available, potentially accelerating genetic discovery like AlphaFold did for protein structure.