Ask HN: How do you find out if the LLM API is giving degraded responses