Controlled Decoding from Language Models | Dark Hacker News