Curious about the training data of OpenAI's new GPT-OSS models? I was too