Show HN: A new benchmark for testing LLMs for deterministic outputs 3 months agohmm why can't structured decoding be used? 0ThreadHN