OpenAI getting hosed by other provider's prompt injections (https://www.reddit.com/r/LLMDevs/comments/1udpw9h/just_got_t...) tainting their training set?