Speculative cascades – A hybrid approach for smarter, faster LLM inference | Not Hacker News!