@mmasnick I bet NYT is hiding that every-time they restarted their GPT session and ask the exact same prompt, they likely got slightly different answers...
OpenAI has the log of the chat sessions and can show how many prompts NYT tried until they got the answer they wanted.