Immigrant Software Dev, have some opions on geopolitics
Teaching a frozen BC policy when to ask for help, and how to push its actions back toward expert behavior using counterfactual regret labels and flow matching in DCT space.
How one ICRA 2025 paper replaces planning, state machines, and learned policies with plain gradient descent — and it actually works.
Storing FP8 weights as bytes on Ampere, decoding via a LUT, scaling, quantizing to INT8, and using IMMA tensor cores—so you can experiment with FP8-like numerics without Hopper.
Reverse-engineering the Rockchip RK3588 NPU to run SmolVLM 15x faster by discovering hardware limits, defeating compiler optimizations, and building a custom sharding runtime