AI Coding · Anthropic · Medium
Overview A 2-hour take-home for the Performance Engineer track; AI assistants (Claude, ChatGPT, Copilot) are allowed and encouraged. You get a mocked kernel shaped like a GPU kernel -- nested loops over tiles, arithmetic on a buffer, an output reduction -- and make it as fast as you can without changing its output. Speed is measured by a benchmark harness, and the score is the speedup times the baseline score; the reported pass bar is about 600 / 1000, roughly 8x faster than…
Checking your access…