Engineering
The Great (Sandbox) Escape: Analyzing the OpenAI and Hugging Face Security Incident
During an OpenAI safety evaluation of a benchmark called ExploitGym, an autonomous AI agent escaped its restricted environment by discovering…
Engineering
Kimi K3 vs DeepSeek V4 Pro vs GLM-5.2: Open Trillion-Scale MoE Models Compared on Benchmarks, License, and Serving Cost
This technical comparison evaluates three open-weight language models released by Chinese AI labs in 2026. Kimi K3, at 2.8 trillion…
Engineering
Context Engineering for AI Agents: A Deep Dive
Clara Chong describes context engineering as the practice of strategically shaping the information provided to AI agents to optimize performance.…
Engineering
Be Ready Before the Attack: A Practical Guide to Self-Hosting an Open Model for Cyber Defense
This guide details how to deploy the open-weight GLM 5.2 model on enterprise infrastructure for cybersecurity incident response, arguing that…
Engineering
POCKET: A 34.66B-Parameter Sparse MoE Model That Runs On-Device
POCKET is a 34.66-billion-parameter sparse Mixture-of-Experts model engineered for efficient on-device inference by activating only about 3 billion parameters per…
Engineering
Helion on TPU: Towards Hardware Heterogeneous Kernel Authoring
PyTorch’s Helion, a domain-specific language for performance-portable ML kernels, now compiles to optimized TPU code via Pallas using a two-level…
Engineering
Automated Evaluation Systems for LLM and Generative AI Products: A Technical Deep-Dive on Metric Dimensions and Implementation Practices
This technical report examines automated LLM evaluation systems, arguing the field has shifted from reference-based metrics like BLEU and ROUGE…
Engineering
Announcing vLLM AFD Plugin: Disaggregating Attention and FFN for Flexible MoE Serving
The vLLM AFD plugin introduces Attention-FFN Disaggregation, separating a Mixture-of-Experts model’s attention and FFN components into independently deployed services that…
Engineering
A Preview of Production-Scale Kimi K3 Support on vLLM
The vLLM team details engineering work to serve Moonshot AI’s Kimi K3 model at production scale, including a prefix-caching design…
Engineering
NVIDIA Releases Cosmos 3 Edge, a World Model for Edge Robotics
NVIDIA released Cosmos 3 Edge on July 20, 2026, a 4-billion-parameter world model built for robots and vision AI agents…
Engineering
Aether-7B-5Attn: A Fully Open Sovereign Foundation Model
Korean startup VIDRAFT released Aether-7B-5Attn on July 19, 2026, a 6.59-billion-parameter language model published with full training data recipes, code,…
Engineering
Evolving the Harness, Not the Model: Toward Self-Improving AI Agents
A July 21, 2026 piece examines how AI agent systems can improve autonomously by evolving the harness code around a…