Follow
LLMs in early 2025: bouncing balls in hexagon
LLMs now: one shotting 3d fluid simulation that can run (mostly) at 60 fps on llvm pipe
@matrix@gameliberty.club
🌿 
@matrix
>LLMs now: one shotting 3d fluid simulation that can run (mostly) at 60 fps on llvm pipe
On a CPU with several billions of transistors close to 1nm thic with over 16GB of ram plus GPU and not on a CPU with 382 million transistors and a iGPU with 177 million transistors and 8GB or shared RAM.
>LLMs now: one shotting 3d fluid simulation that can run (mostly) at 60 fps on llvm pipe
On a CPU with several billions of transistors close to 1nm thic with over 16GB of ram plus GPU and not on a CPU with 382 million transistors and a iGPU with 177 million transistors and 8GB or shared RAM.
@mangeurdenuage yep, our hardware is godtier, while software is absolute garbage
@matrix they just say this to exaggerate capability. sure they can do a fair bit more now but they still fuck up constantly on anything but the basics unless you fight and argue with it for hours and hours with a billion reemphasis phrases on what to do. one shotting is a complete myth and a lie by the big tech elites to act like its more powerful than it actually is or ever could be. it's a glorified markov chain
@ube kinda yeah, they're still unable to do the last 5%
