On this week's AI board

Titania: a language model stack from transformer to transistor

Rust project that defines its own GPU instruction set, compiler and simulator, and runs Qwen3-0.6B on all of them.