/
Published on 13.03.2026
TLDR: OpenAI released GPT-5.4 in Thinking and Pro variants just two days after GPT-5.3, bringing an expanded context window and mixture-of-experts architecture. The Pro tier costs $30 per million input tokens and $180 per million output tokens.
GPT-5.4's Higher Performance, Higher Price
TLDR: The State of Mobile 2026 report reveals that global AI app revenue tripled to over $5 billion, downloads doubled to 3.8 billion, and users spent 48 billion hours in AI apps, representing a roughly 3.6x increase over 2024 and 10x over 2023.
TLDR: Meta, OpenAI, and other tech giants are building private power plants independent of regional electrical grids to fuel AI data centers. A Cleanview study found 46 off-grid projects accounting for 30% of all planned U.S. data center capacity, with 90% announced in 2025 and equipment being "almost entirely gas-fired" despite public emphasis on renewables.
AI Data Centers Go Off the Grid
TLDR: Apple researchers proposed Feature Auto-Encoder (FAE), a diffusion image generator that learns to reconstruct embeddings from DINOv2. By shrinking embeddings before reconstructing them, FAE trains roughly 7x faster than prior approaches while achieving better image quality.
Lightning-Fast Diffusion Learning
TLDR: Andrew Ng discusses the vision for a Stack Overflow-like platform designed specifically for AI coding agents, referencing Context Hub (a CLI tool with over 5,000 GitHub stars that provides API documentation to agents) and Moltbook, a Reddit-like social network for agents acquired by Meta.