Desk
Meta AI research, Llama, and related products.
After Meta touted new child-exploitation detection, researchers say hundreds more CSAM and nudify ads—some animating real teens’ photos—still ran on Facebook and Instagram.
OpenBMB's 2.52B dense MiniCPM5-2B posts a 53.9 average across 34 benchmarks and ships Apache-2.0 weights plus training data—strongest where agents call tools, weaker on textbook knowledge.
Meta FAIR and academic partners treat experiment selection—not idea generation—as the scarce resource, ranking candidate runs before any GPU hour is spent.
Nous Research turns the painful VRAM-guesswork of local open-weights inference into a hardware-aware, one-click setup path inside MIT-licensed Hermes Desktop.
Instagram is labeling real photos as AI Content while fully generated Gemini images with C2PA and SynthID sail through unlabeled.
Meta’s Muse Spark offers roughly 95% token discounts if you share prompts and outputs for training — turning data opt-in into an explicit price list.
Meta killed the performance grade for AI usage, then put Hatch, its OpenClaw-like computer-use agent, onto workers' corporate laptops anyway.
Meta Engineering's 'second brain' post is a vendor architecture for compliance: knowledge in files, recipes for reasoning, and expert corrections compiled without retraining.
New York’s AI policy debate is becoming a test of whether child-safety rules can protect minors without normalizing identity checks for everyone.
Meta’s ads model shows LLM infrastructure does not transfer cleanly to recommendations; efficiency comes from rebuilding kernels and parallelism around irregular user histories.
Meta uses LLMs for ambiguous privacy classification, then moves stable behavior into reviewed deterministic rules.
SilverTorch folds indexes, filters and ranking into one PyTorch graph to spend retrieval latency on more candidates and richer scoring.