Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.
GPT-5.6 Sol API Pricing Cut by Up to 33%, Live 'Open Training' of a 535B Model, and an AI Training Opt-Out at Twitch — AI News for August 23
OpenAI Calls for Stronger AI Safety Bill ── Lack of Rogue Model Containment Plans Flagged, NVIDIA Notifies Customers of Over 15% Price Increases, Zara Tackles Returns with AI Try-On ― AI News for August 23
Hollywood Creatives "Digging Their Own Graves" — Qwen 3.5 4B Gains +16.7% Reasoning via Quantization Precision Allocation, and 99.4% of Agent Input Isn't Human — AI News, August 22
AI Regulation Moves as Uber Is Hit with a €825M Fine — ECB Weighs In on the Bubble, Claude Watermark Mechanics Explained, KV Cache Splitting Delivers ~3x Faster Prefill — AI News, August 22
Research Building "Beliefs" into World Models — Camera-Free AI Glasses, FreeToken at 100tok/s Even Beyond VRAM Limits, and Video Editing via MCP — AI News for August 22
Netflix Demonstrates Its LLM Recommender "GenRec" — Psychometrics Exposes the Weaknesses of Safety Benchmarks, and llama.cpp v0.2.0 Arrives — AI News, August 22
The Cursor Usage Logs That Trained Grok 4.5 — Output Tokens Down to About a Quarter, 991 Spoofed AI Crawler Hits, and a Practical RTX 3090 × vLLM Migration: AI News for August 22
Anjadhe: A No-Account Local AI Secretary Debuts — Running Qwen3.8-27B on 16GB VRAM, Plus Building an MLflow Evaluation Foundation
Simulation Becomes the New Scaling Law — Simile AI Valued at $2B, Linus Endorses AI Debugging Help, and Opus 4.6's Guardrails Easily Bypassed
NVIDIA's AVO Fully Conquers ARC-AGI-3 — DeepSeek's Experimental Model Nears Opus 4.8, US Data Center Opposition Climbs to 75%
