@Teslanaut@DJLougen Active parameters are only a thing with MOE models, they have special layers that act as routers. So every chat goes through the router, then it gets sent to the expert.
Dense models do not have these routing layers. They require every layer/paramteter be touched to function.
Me starting with LLMs:
"bigger GPU, more VRAM = faster inference"
Me now:
- VRAM bandwidth
- KV cache behaviour
- memory latency
- cache locality
- PCIe bottlenecks
- kernel efficiency
- quantization tradeoffs
- memory movement
Modern AI inference is basically systems engineering disguised as matric multiplication.
MICROSOFT EDGE STORES ALL YOUR PASSWORDS IN PLAIN TEXT
edge password manager security fail
A security researcher just found that Edge decrypts EVERY saved password at startup and keeps them in memory - even if you never visit those sites.
The wild part? Edge still asks for your Windows password to VIEW them in the UI. The passwords are already sitting there, naked in RAM.
Chrome only decrypts passwords when you actually need them. Edge just.. doesn't.
Microsoft's response when reported: "working as intended"
If someone gets admin access on a shared machine, they can dump ALL Edge passwords from ALL logged-in users. Even disconnected ones.