Fable generated a CUDA megakernel that beat an optimized baseline by 18.71×. But nobody on the team would have written it — or can fully explain it.
AI Wrote a GPU Kernel 18 Faster Than Humans. Now Who Reviews It?
Fable generated a CUDA megakernel that beat an optimized baseline by 18.71×. But nobody on the team would have written it — or can fully explain it.
LLMs can give us answers in seconds. We can still make room for curiosity, judgment and the satisfaction of figuring things out.
Ten popular MCP servers add 23,895 tokens to every request, 92% of a short prompt. Caching cuts the price 95%, not the length.
Read-only MCP tools are easy to trust. search_notes can't hurt anyone. The trouble starts with tools...