Hacker News new | ask | show | jobs
by asdaqopqkq 17 days ago
aren't llms smart enough to directly write custom kernels for custom hardware from cuda code?