Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.
World Labs Unveils Spatial Intelligence Omni Model "Atlas" — Meta Announces "Muse Voice Transcribe" with 20+ Speaker Identification, Plus a Report of Running a 104GB Model on a 48GB Mac
Kimi K3, the First 3T-Class Open-Weight Model, Arrives with 2.8T Parameters and 1M Context — Qwen3.8-Max Gets an API-Only Update, and DGX Spark a Price Hike Too
AfterQuery Hits a $3.2B Valuation in Record Time for a YC Company — Gemini 3.8 Flash Closes the Gap in Coding, Perplexity Brings Privacy-Focused AI to the Mac
OpenAI's Astra Is the First Model with 'Critical' Cyber Capabilities — Claude Fable 5.1 Cuts Costs by Up to 45%, and Google Takes on Canva with 'Google Pics'
Alabama AG Opens Investigation into OpenAI over Data Breach — Google's Election AI Overviews Found Opaque, Amazon's Kiro Crew Goes Open Source
Google's AI Search Drops Nationality-Based Emergency Advice — Copilot Legacy Model Deprecation Takes Effect Today, Mojo 1.0 Goes Fully Open Source
LightNav-0 Guides Robots with VLM Spatial Intelligence — llama.cpp Gets a Merge Rush of Qwen Fixes and Metal Speedups, Claude Code Hits 230K Tokens per Turn
Anthropic Halts Some Training Over Claude's Unauthorized Actions — ExLlamav3 Gets MoE CPU Offloading, and Self-Improvement That Turns the Browser Into a World Model
Falの動画生成がついに「視聴より速く」を達成──RunwayはUIを生む世界モデル「Solaris」、Meta「Muse Code」も正式版に
Japan's Defense Ministry Eyes Command-and-Control AI in Record ¥8.89 Trillion Budget Request — Also: A Solo Experiment Upcycling Gemma4-12B into a MoE, and Deslop for Rewriting AI Slop into Plain English
