Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.
Xiaomi Fixes MiMo-V2.6's Tool-Calling Loop; EDR Evasion Experiments with a Local Qwen, a 14B Distilled from a 30B at Half the Width, and Sompo Japan Deploys Gemini to 20,000 Employees
GPT-3 Reaches End of Service — GPT-5.6 Terra as Replacement Called 'Overkill', Adobe Integrates with Gemini/Claude, and an LLM Trading Study Where Reasoning Costs Don't Pay Off
Gem16, a DIY Inference Engine That Fits Gemma4 26B on a 16GB GPU — Having LLMs Design 3D-Printed Bridges for Destruction Testing, Plus Hitachi's New Role: FDE
US and China Unify on "Superintelligence (SI)" as the Term for AI and Launch a Regular Dialogue; FT: "Companies Reject Pricey Models and Go Open"; Plus an Inference Engine That Runs a 177B MoE from an SSD on a 16GB GPU
Anthropic Releases "MHS," a Common Standard for AI Agents Operating Lab Equipment — Amodei to Have First Dinner with Trump, Plus Free Claude for 10,000 Scientists
Longtime Anthropic Staff Consider Buying Land in Case AI "Goes Rogue" — AI Scam Costs Italy's Largest Bank Millions, Humanoids Enter Elderly Care
LLM-Powered Robot Cleans an Unfamiliar Kitchen — 80–90% of OpenAI's Research Is Aimed at GPT-7 and Beyond, and GPT-6 Astra Scores Perfect on Nonobench
AI Agent Security Incidents Reach Tens of Thousands — Big Tech AI Investment Heads Toward $1.2 Trillion, While Human Oversight Weakens in AI Weapons Regulation
LLM Agents Can Tamper With Their Own Execution Traces — Disinformation Networks Creep Into Pre-Election AI Chatbots, and Local LLM Inference on the Mac Runs Up to 1.5x Faster
Chinese State Media: "AI Governance Is the Responsibility of Both the US and China" — Solidigm Weighs US IPO, Plus a Breakdown of 97 Opus 5.5 Failure Notes
