Ollama v0.33.1-rc0 Adds MLX Support for Qwen3.8 Flash Next
Ollama released v0.33.1-rc0, a release candidate that adds MLX support for the Qwen3.8 Flash Next model . The update, tracked as pull request #18032, extends Ollama's Apple Silicon MLX backend to cover the Qwen3.8 Flash Next architecture .…
PyTorch Ecosystem Adds 10 Projects Spanning RL, Inference, Survival Analysis
The PyTorch Ecosystem Working Group added 10 open-source projects to its Ecosystem Landscape . New entries include Perforated, a data-efficiency library that applies neuron-specific reinforcement learning signals during backpropagation…
Pinterest Cuts Retrieval Latency 85%, Saves 7 Figures With Unified CLR
Pinterest Engineering disclosed a scaled Conditional Learned Retrieval (CLR) system for its home feed that consolidated separate interest and board retrieval models into a single unified framework . The system generates condition-aware…
vLLM v0.28.0 Ships Kimi-K3 and DeepSeek V4 Optimizations
The vLLM project released v0.28.0, comprising 584 commits from 270 contributors . The release includes a Kimi-K3 optimization push featuring Decode Context Parallel support, fused FlashKDA kernels, combined all-gathers with 1.5–3x…
ALL BRIEFS
RANKED · 7 THIS DAYAI Model Security Crosses Into Funded Infrastructure Category
A near-billion-dollar raise for an AI guardrails vendor, a critical vulnerability in NVIDIA's agent infrastructure, and the first confirmed multi-agent cyberattack on a nation-state all surfaced within the same news cycle — evidence that…
Local AI Agents Emerge as Product Category Across Hardware and Software
The convergence of Apple's M6 and M5 Ultra desktops, Perplexity's Portable Computer agent, and JetBrains' Junie Local coding agent signals that on-device AI agents are crossing from hobbyist workaround into a distinct, commercially…
Multi-Agent AI Framework Breaches Asian Government Systems Over Four Days
A near-autonomous AI framework deploying up to eight parallel sub-agents breached government infrastructure in Asia over a four-day operation in early July 2026, extracting more than 2,564 personnel records, 85 cracked credentials, and…