TOP STORY Google Warns Criminal Hackers Used AI to Discover Major Zero-Day - Google has confirmed that criminal threat actors leveraged AI tools to identify and exploit a significant software vulnerability, marking a new escalation in AI-assisted cyberattacks. New York Times


Qwen-Image-2.0 Technical Report - Alibaba releases the technical report for its latest vision-language model, detailing architectural improvements and benchmark results at scale. Hugging Face

Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models - New technique separates computation from patch granularity in byte-level LMs, potentially improving efficiency without sacrificing performance. Hugging Face

AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems - A framework for real-time auditing that predicts failures in multi-agent pipelines before they propagate. Hugging Face

Interfaze: A New Model Architecture Built for High Accuracy at Scale - Interfaze introduces a novel architecture claiming improved accuracy across large-scale inference workloads. Interfaze Blog


1-Trillion-Parameter Model on Intel Optane Persistent Memory at 4+ Tokens/Sec - A custom PC build using Intel Optane persistent memory demonstrates local inference of a 1T parameter model at usable speeds. Reddit / LocalLLaMA

Gemma 4 Running Fully Offline on WebGPU via Transformers.js - Google's Gemma 4 model runs entirely in-browser over WebGPU, demonstrated controlling a Reachy Mini robot over WebSerial. Reddit / LocalLLaMA

Training an LLM in Swift: Matrix Mult from GFlop/s to TFlop/s - Part 1 of a series walks through optimizing matrix multiplication performance by three orders of magnitude using Swift. Cocoa with Love

Claude Platform Now Available on AWS - Anthropic announces Claude is deployable natively within the AWS ecosystem, deepening cloud infrastructure integration. Anthropic Blog


TanStack NPM Supply-Chain Compromise Postmortem - TanStack details a serious supply-chain attack on its NPM packages, outlining how the breach occurred and steps taken to remediate. TanStack Blog

OpenAI Launches DeployCo to Help Businesses Build Around Intelligence - OpenAI spins out a new deployment-focused company aimed at helping enterprises operationalize AI products at scale. OpenAI Blog

Sony Says AI Tools Will Flood the Market with Even More Games - Sony leadership argues that AI-driven efficiency gains in game development will dramatically increase the volume of titles reaching market. Ars Technica

Trump-Xi Summit Could Reshape the AI Race - This week's Beijing summit between the US and Chinese presidents is expected to directly address competition and cooperation in AI development. ABC News


If AI Writes Your Code, Why Use Python? - An essay questioning whether Python's dominance is justified when AI agents increasingly abstract away language-level decisions. Medium

Interaction Models - A thoughtful exploration of how the design of human-AI interaction patterns shapes outcomes and user trust. Thinking Machines

The Rise of 'Stacey Face': How AI Enhancements Are Warping Beauty Standards - Investigates how AI-powered photo filters and looksmaxxing culture are creating homogenized, algorithmically-driven beauty ideals. The Independent

I Let AI Build a Tool to Figure Out What Was Waking Me Up at Night - A developer documents using AI-assisted coding to build a personal sleep-disturbance detection tool from scratch. Martin.sh