Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.
Meta's AI Layoff Plan Collapses Amid Employee Backlash and Agent Failures — 90% of Executives Feel No Productivity Gains, and GigaBrain-0.7 Marks How Far Embodied AI Has Come ‖ AI News for August 26
Qwen3.8-Flash-Next Officially Released — 6B Active Parameters with 1M-Token Support; Z.AI Confirms "Ox Alpha = GLM" and Promises Weight Release ‖ AI News, August 26
Security-Specialized 'OpenAegis' Outperforms Kimi K2.7 — The Puzzles AI Still Can't Solve, and ByteDance Reportedly Unveils Office AI 'Doubao Work' ‖ AI News, August 26
Bill Gates Warns "AI Has Already Crossed a Dangerous Threshold" — Mystery Model Ox Alpha Reportedly GLM-5.3-Flash; 2-Bit Quantization Brings 27B Into the ~10GB Range ‖ AI News, August 26
Perplexity Announces 'Portable Computer' for Local AI ── Open-Source Kernel Hits 78K tok/s on AMD MI350X × Qwen3.6-35B-A3B, Japan Ranks Last in Production Agent Adoption at 17% ‖ AI News, August 26
Fully NVFP4-Quantized Qwen3.8-27B Checkpoint Lands at 19.7GB — ByteDance Seed's Diffusion Self-Distillation, Four-Mac-Studio Cluster Claims 4.8TB/s ‖ AI News, August 26
3-Bit Quantized Qwen3.8-27B Correctly Implements Optical Numerical Computation on a 16GB GPU — Auditing 'Causality Leakage' in Hybrid LLMs, Robotics Startup Generalist Headed to a $3 Billion Valuation ‖ August 26 AI News
Stability AI Raises $76M — IBM's Granite 4.2 Ships with Switchable Thinking Modes + Apache 2.0, Google Launches Legal-Focused Gemini ‖ AI News for August 26
OpenAI's In-House Chip 'Jalapeño' Beats Nvidia GB300 on Power Efficiency — M5 Ultra Mac Studio Packs 512GB of Memory, Meta's Paid Agent 'Hatch' Coming Soon
US State Attorney General Investigates OpenAI's "Runaway Agent" — Inside Nvidia's "4x" Inference Chip Claim, and Qwen3.8-Flash-Next May Be Near ‖ AI News, August 25
