Towards unlimited contexts: faster-than-GPU sparse logarithmic attention on CPU [video] | Not Hacker News!